JOZEF HRITZ
Brno 2021
FACULTY OF SCIENCE
Department of Chemistry
Dynamical Features of
Biomolecular Complexes
HABILITATION THESIS
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
2
Bibliographic record
Author: JOZEF HRITZ
Faculty of Science
Masaryk University
Title of Thesis: Dynamical Features of Biomolecular Complexes
Habilitation Field: Physical Chemistry
Year: 2021
Number of Pages: 72+176
Keywords: biophysical chemistry, computational simulations, molecular
dynamics, biomolecules, statistical physics, free energy, binding
affinity, enhanced sampling, nuclear magnetic resonance
spectroscopy, intrinsically disordered proteins
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
3
Bibliografický záznam
Autor: JOZEF HRITZ
Faculty of Science
Masaryk University
Název práce: Dynamické Vlastnosti Biomolekulárnich Komplexů
Obor habilitačního řízení: Fyzikální Chemie
Rok: 2021
Počet stran: 72+176
Klíčová slova: biofyzikální chemie, výpočetní simulace, molekulová dynamika,
biomolekuly, statistická fyzika, volní energie, vazební afinita,
zesílené vzorkování, nukleární magnetická rezonanční
spektroskopie, přirozene neuspořádané proteíny
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
5
Commentary to habilitation thesis
This habilitation thesis, submitted to the Dean of Faculty of Science at the Masaryk University
as a part of the application for the academic title “Docent” (Associated Professor),
documents the most important research activities of the applicant after his Ph.D.
study. The main research activities described here cover a variety of novel methods
developed in the field of biomolecular simulations and the corresponding experimental
studies that could be used to experimentally validate parts of the computational predictions
of macromolecules in solution. Biomolecular simulations are very useful tools
to rationalize experimental findings and to direct future experiments. Computer simulations
offer insight at an atomic resolution and a femtosecond timescale, which are
usually beyond experimental means. On the other hand, computational simulations use
a range of approximations and simplifications and are restricted to simulating only
very short time periods. Therefore, it is crucial to validate the results obtained from
computational simulations by comparison to experimentally obtained data as far as
possible. To achieve this the applicant deployed mostly solution NMR and biophysical
techniques such as measurements of thermostability and binding affinities to obtain
experimental data.
From the 29 research articles published after the applicant's Ph.D. thesis defense, 15
were selected to be included as the attached reprintsP1-P15 and are briefly commented
in four scientific chapters (3-6). Out of these 15 scientific peer-review publications,
the applicant was the first author in 6 of them and the corresponding author in 5
of them. The first scientific chapter here, “3. Efficient incorporation of plasticity and
water molecules into the molecular docking approach”, describes our developed
method combining ensemble ligand docking and molecular dynamics (MD) simulations,
incorporating the plasticity and flexibility of cytochrome P450 2D6 structures in
a highly efficient way and leading to a significant increase in the reliability of predicted
binding poses.P1 The approach was extended to include the unbiased addition of the
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
6
most important explicit water molecules in the docking procedure.P2 The developed
techniques were successfully applied also for the ligand docking into another cytochrome
P450sP3 as well as for RNA-protein docking.P4 The described docking approaches
are restricted by the limited sampling potential of standard MD. This can be
significantly increased by applying some enhanced sampling approach such as temperature
or Hamiltonian replica-exchange MD (T/H-REMD). This is the content of the second
scientific chapter “4. Methodological developments of enhanced sampling
methodology H-REMD” where we describe H-REMD using soft-core interactions.P5 In
contrast to T-REMD, H-REMD needs a priori knowledge of the origin of energy barriers
in the system, which implies that different systems require different H-REMD schemes.
For this particular problem, we introduced the “fast mimicking” procedure for providing
the sought parameters.P6 The applicability of described methods to produce the canonical
ensemble efficiently was shown using GTP analogs. This chapter also describes
an approach combining H-REMD with the distance-field distance restraints leading to
a powerful tool for the unbiased simulation of binding pathways and conformational
changes induced by phosphopeptides binding or unbinding from the 14-3-3zeta protein.P7
The reliable sampling of binding/unbinding events also allowed the calculation
of the corresponding binding affinity of ligands to the protein. Unfortunately, this approach
is computationally very expensive. In many applications rather than calculating
the absolute binding affinity, it is sufficient to calculate the relative free energy differences
by alchemical free energy calculations between quite similar compounds. However,
the efficiency may drop dramatically for compounds that have multiple stable
conformations separated by high energy barriers. This problem is addressed in the
third scientific chapter “5. Alchemical (relative) free energy calculations”.P8-P10 Another
class of biomolecules that represent a challenge for conformational sampling are
intrinsically disordered proteins (IDPs), not because of high energy barriers, but because
they possess a huge number (>thousands) of conformational states. They are the
topic of the fourth scientific chapter. “6. Structural and interaction properties of
IDPs determined by experimental NMR and computational studies”.P11-P15 Here
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
7
we describe the combination of experimental NMR and MD techniques in a closely coordinated
fashion allowing the characterization of this important class of proteins.
My contribution to the selected 15 articles to this habilitation thesis is summarized in
the following tables with special attention to the experimental work, supervision of
students, manuscript preparation, and research direction.
[P1]1
Hritz, J.; de Ruiter, A.; Oostenbrink, C. Impact of plasticity and flexibility on
docking results for Cytochrome P450 2D6: a combined approach of molecular dynamics
and ligand docking. J. Med. Chem. 2008, 51, 7469-7477 (IF=4.9)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
80% 70% 60% 50%
[P2] Santos, R.; Hritz, J.; Oostenbrink, C. The role of water in molecular docking simulations
of Cytochrome P450 2D6. J. Chem. Inf. Model. 2010, 50, 146-154 (IF=3.8)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
20% 70% 40% 50%
[P3] Oostenbrink C.; de Ruiter A.; Hritz J.; Vermeulen N.P.E Malleability and versatility
of Cytochrome P450 active sites studied by molecular simulations.
Curr. Drug Metab. 2012, 13, 190-196 (IF=4.2)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
- 20% 20% 20%
[P4] Byeon I-J. , Ahn J., Mitra M., Byeon C-H., Hercík K., Hritz J., Charlton L., Levin J.,
Gronenborn A.M. NMR structure of human restriction factor APOBEC3A reveals substrate
binding and enzyme specificity. Nat. Commun. 2013, 4, 1890 (IF=10.7)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
10% - 10% [P5]
Hritz J., Oostenbrink C. Hamiltonian replica exchange molecular dynamics using
soft-core interactions. J. Chem. Phys. 2008, 128, 144121 (IF=3.1)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
80% - 70% 50%
1 Bibliographic record of a published scientific result, which is part of the habilitation thesis.
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
8
[P6] Hritz J, Oostenbrink C. Optimization of Replica Exchange Molecular Dynamics
by Fast Mimicking. J. Chem. Phys. 2007, 127, 204104 (IF=3.1)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
80% - 70% 50%
[P7] Nagy,G.; Oostenbrink, C.; Hritz, J.*: Exploring the Binding Pathways of the 14-3-
3 Protein: Structural and Free-Energy Profiles Revealed by Hamiltonian Replica Exchange
Molecular Dynamics with Distance Field Distance Restraints. PLoS ONE
2017,12(7), e0180633 (IF=2.8)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
10% 80% 40% 50%
[P8] Hritz, J.; Lappchen T.; Oostenbrink, C. Calculations of binding affinity between
C8-substituted GTP analogs and the bacterial cell-division protein FtsZ.
Eur.Biophys. J. 2010, 39, 1573-1580 (IF=2.4)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
70% - 60% 50%
[P9] Hritz, J.; Oostenbrink, C. Efficient free energy calculations for compounds with
multiple stable conformations separated by high energy barriers.
J. Phys. Chem. B 2009, 113, 12711-12720 (IF=3.5)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
80 - 70 50
[P10] Jandova, Z.; Trosanova, Z.; Weisova, V.; Oostenbrink, C.*, Hritz, J.*:Free energy
calculations on the stability of the 14-3-3 protein. BBA - Proteins and Proteomics,
2018, 1866, 442-450 (IF=2.5)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
20% 50% 30% 40%
[P11] Hritz J.; Byeon I-J.; Krzysiak T.; Martinez A.; Sklenář V.; Gronenborn A.M. Dissection
of binding between a phosphorylated tyrosine hydroxylase peptide and 14-3-3ζ:
a complex story elucidated by NMR. Biophys. J. 2014, 107, 2185-2194 (IF=4.0)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
70% - 70% 50%
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
9
[P12] Louša, P.; Nedozrálová, H.; Župa, E.; Nováček, J.; Hritz, J.*: Phosphorylation of
the regulatory domain of human tyrosine hydroxylase 1 monitored using non-uniformly
sampled NMR. Biophysical Chemistry 2017, 223, 25-29 (IF=1.9)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
10 100 40 60
[P13] Zapletal, V.; Mládek, A.; Melková, K.; Louša, P.; Nomilner, E.; Jaseňáková, Z.; Kubáň,
V.; Makovická, M.; Laníková, A.; Žídek L.; Hritz, J.* Choice of force field for proteins
containing structured and intrinsically disordered regions. Biophys. J. 2020, 118,
1621–1633 (IF=3.9)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
10 50 30 40
[P14] Pavlíková Přecechtělová, J.; Mládek, A.; Zapletal, V.; Hritz, J. Quantum Chemical
Calculations of NMR Chemical Shifts in Phosphorylated Intrinsically Disordered Proteins,
JCTC 2019, 15, 5642-5658 (IF=5.0)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
- 30 20 50
[P15] Jansen, S.;Melková, K.; Trošanová, Z.; Hanáková, K.; Zachrdla, M.; Nováček, J.;
Župa, E.; Zdráhal, Z.; Hritz, J.*; Žídek, L.*: Quantitative Mapping of MAP2c Phosphorylation
and 14-3-3ζ Binding Sites Reveals Key Differences Between MAP2c and Tau. J.
Biol. Chem. 2017, 292, 6715-6727 (IF=4.0)
Experimental work (%) Supervision (%) Manuscript (%) Research direction (%)
10 30 30 20
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
10
Komentář
Tato habilitační práce, předložená děkanovi Přírodovědecké fakulty Masarykovy univerzity
v rámci žádosti o akademický titul "Docent" (Přidružený profesor), dokumentuje
nejdůležitější výzkumnou činnost uchazeče po jeho doktorském studiu. Zde popsané
hlavní výzkumné činnosti zahrnují řadu nových metod vyvinutých v oblasti biomolekulárních
simulací a odpovídajících experimentálních studií, které by mohly být
použity k experimentální validaci částí výpočetních předpovědí makromolekul v
roztoku. Biomolekulární simulace jsou velmi užitečnými nástroji pro racionalizaci experimentálních
nálezů a pro detailní návrh budoucích experimentů. Počítačové simulace
nabízejí přehled v atomovém rozlišení a současně v femtosekundovém časovém
měřítku, které jsou obvykle mimo experimentální prostředky.
Na druhou stranu výpočetní simulace používají řadu aproximací a zjednodušení a jsou
omezeny na simulaci pouze velmi krátkých časových období. V tomto případě je nezbytné
co nejvíce ověřit výsledky získané z výpočetních simulací ve srovnání s experimentálně
získanými daty. K dosažení tohoto cíle žadatel použil převážně roztokovou
NMR spektroskopii a biofyzikální techniky, jako jsou měření termostability a vazebné
afinity. Z 29 výzkumných originálních článků byly vybrány 15( přiložené reprintyP1-P15)
a jsou stručně komentovány ve čtyřech vědeckých kapitolách(3-6). Z těchto 15 vědeckých
recenzovaných publikací byl žadatel prvním autorem v 6 z nich a korespondujícím
autorem v 5 z nich.
První vědecká kapitola zde, "3. Efektivní začlenění plasticity a molekul vody do molekulárních
dokovacích metod", popisuje nami vyvinutou metodu kombinující dokování
ligandů do souboru proteinových struktur a simulace molekulární dynamiky
(MD), zahrnující plasticitu a flexibilitu cytochromových struktur P450 2D6 vysoce
účinným způsobem a vedoucí k významnému zvýšení spolehlivosti předpokládaných
vazebných pozic.P1 Tato metodológie byla dál rozšířena tak, aby zahrnovala dynamické
přidání nejdůležitějších explicitních molekul vody do dokovací procedury.P2 Vyvinuté
výpočetní techniky byly úspěšně použity rovněž pro dokování ligandů do jiných
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
11
cytochromů P450P3 a rovněž pro dokování proteinů flexibilní RNA molekulou.P4 Popsané
dokovací přístupy jsou omezeny omezeným vzorkovacím potenciálem standardní
MD. To lze významně zvýšit napr. použitím MD se zámenou replik(REMD). To
je obsahem druhé vědecké kapitoly "4. Metodický vývoj Hamiltonovskej REMD (HREMD)",
kde popisujeme H-REMD využívajíci tzv. změkčené interakce.P5 Na rozdíl od
teplotní REMD (T-REMD), nastavení H-REMD vyžaduje předchozí znalost původu energetických
bariér v systému, což znamená, že různé systémy vyžadují různá schémata
H-REMD. Pro tento konkrétní problém jsme popsali efektivní postup "rychlého mimikování"
pro poskytování hledaných interakčních parametrů.P6 Použitelnost popsaných
metod pro efektivní generování kanonického strukturního souboru byla prokázána na
sade GTP analogů. Nasledovná podkapitola popisuje přístup kombinující H-REMD s varirováním
vzdálenosti distančního pole vedoucím k výkonnému nástroji pro nezaujatou
simulaci vazebných drah a konformačních změn vyvolaných vázáním fosfopeptidů
do 14-3-3ζ proteinu.P7 Dostateční vzorkování vazebných/nespojitých příhod rovněž
umožnil výpočet odpovídající vazebné afinity ligandů k proteinu. Bohužel tento přístup
je výpočetně velmi nákladný. V mnoha aplikacích spíše než výpočet absolutní vazebné
afinity stačí vypočítat relativní volné energetické rozdíly alchymickými výpočty volné
energie mezi poměrně podobnými sloučeninami. Účinnost však může dramaticky klesnout
u sloučenin, které mají více stabilních konformací oddělených vysokými energetickými
bariérami. Tento problém je řešen ve třetí vědecké kapitole "5. Alchemické
výpočty (relativní) volné energie".P8-P10 Další třídou biomolekul, které představují
výzvu pro konformační odběr vzorků, jsou přirozene neuspožádané proteiny (IDP), nikoli
kvůli vysokým energetickým bariérám, ale proto, že mají obrovské množství (>tisíce)
konformačních stavů. Jsou tématem čtvrté vědecké kapitoly: "6. Určení strukturních
a interakčných IDP proteinů určovaných kombinaci experimentální NMR
a výpočetních studii".P11-P15 V této kapitole je detailne popsána úzka kombinace experimentální
roztokové NMR a MD simulací, která umožňuje charakterizaci této důležité
třídy proteinů.
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
12
Declaration
I hereby confirm that I have written the habilitation thesis independently, that I have not
used other sources than the ones mentioned and that I have not submitted the habilitation
thesis elsewhere.
In Brno, June 10, 2021 .......................................
JOZEF HRITZ
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
13
ACKNOWLEDGMENTS
First of all, I would like to thank my family. My parents and brother have been practical
people and did not have much appreciation for scientific research. Therefore they were
not pleased when I decided to study physics and later biophysics at the University of
P.J.Safarik in Kosice. However, when they understood how much science means to me
they fully supported me in my decision. My scientific career has not been easy either
for my wife Renata to whom belongs my warmest thanks because it meant to live in
several different countries where she followed me and had to raise our kids there: our
son Alex who was born in the Netherlands, our daughter Daniela in the USA and now
they attend their schools in the Czech Republic. Thank you for all your support and
understanding of my scientific mission.
I feel sorry there is not enough space for thanking explicitly all my colleagues, students,
and supervisors that supported me at various stages of my career. Still, I wish to rise
the warmest thanks to at least five supervisors that are great researchers but at the
same time great people who helped me greatly many times: to prof. Chris Oostenbrink
was my supervisor during my post-doctoral stay at Vrije University in Amsterdam for
4.5 years. He always had time for a little chat if needed. Thank you for all your excellent
ideas, your trust, and great collaboration! To prof. Angela Gronenborn was my supervisor
during my post-doctoral stay at the University of Pittsburgh for 2.5 years. Thank
you for that great opportunity to work in your team! I thank prof. Vladimir Sklenar
allowed me to join CEITEC-MU in Brno. And to prof. Lukas Zidek who is my current
boss and colleague at CEITEC-MU. I feel very thankful and privileged that these great
scientists still do collaborate with me and my research team and that I always feel longterm
support from their side. Regarding my pedagogical activities, I wish to thank prof.
Libuse Trnkova who supported my interest to lecture the courses from Biophysical
Chemistry from the beginning and that we could shape this interdisciplinary study program
together.
DYNAMICAL FEATURES OF BIOMOLECULAR COMPLEXES
14
Last but not least, I thank all students under my supervision. Not only because we can
perform very exciting research together but they contributed to a very nice and
friendly atmosphere. They also let me realize how much supervision and lecturing activities
fulfill me.
As mentioned before there are more than a hundred scientists and friends that I feel
very thankful to and despite I cannot list them all here explicitly let me do it at least in
this implicit way THANK YOU ALL VERY MUCH!.
TABLE OF CONTENTS
15
Table of Contents
List of Figures 17
Abbreviations 18
1 Introduction 19
2 General Theoretical Background – Molecular Dynamics 23
2.1 Ensemble averages – link to the experimental data........................................... 23
2.2 Replica-exchange molecular dynamics (REMD) .................................................. 26
3 Efficient incorporation of plasticity and water molecules into the molecular
docking approach 31
4 Methodological developments of enhanced sampling methodology H-REMD
38
4.1 H-REMD using soft-core interactions....................................................................... 39
4.2 Umbrella sampling........................................................................................................... 41
4.3 H-REMD using distance (field)restraints................................................................ 42
5 Alchemical (relative) Free energy calculations 47
5.1 Efficient calculations of relative free energies...................................................... 48
6 Structural and interaction properties of intrinsically disordered proteins
(IDPs) determined by experimental NMR and computational studies 54
6.1 Intrinsically disordered proteins (IDPs)................................................................. 54
6.2 Structural ensembles of IDPs and IDRs verified by NMR data ....................... 55
6.3 NMR investigation of IDP/IDRs binding properties........................................... 59
7 Conclusions 65
Bibliography 69
LIST OF FIGURES
17
List of Figures
Figure 1: Schematic example of Hamiltonian replica exchange molecular dynamics (HREMD)
with ten replicas.......................................................................................................................27
Figure 2: Structure of GTP and 8-Br-GTP in syn conformation.............................................38
Figure 3: Time dependence of dihedral angle around the glycosidic bond during two
MD simulations for GTP and 8-Br-GTP...........................................................................................39
Figure 4: Schematic representation of the energy profile as a function of the softness
of nonbonded interactions. .................................................................................................................40
Figure 5: Schematic representation of distance-field restraints. .........................................43
Figure 6: Thermodynamic cycle used for the computation of the difference in binding
free energies between compound X and Y into the same protein target (P)...................48
Figure 7: NMR relaxation data for δ subunit of RNA polymerase........................................57
Figure 8: Binding scheme of dp_hTH1_50 phosphorylated on positions 19 and 40 with
the 14-3-3ζ dimer, represented by a letter P. ..............................................................................58
ABBREVIATIONS
18
Abbreviations
AAAH – aromatic amino acid hydroxylases
ACT – aspartate kinase, chorismate mutase and TyrA
A-EDS – accelerated enveloping distribution sampling
BioEN – Bayesian ensemble reweighing
CD – circular dichroism / catalytic domain
cryoEM – cryo electron microscopy
CS – chemical shift
EDS – enveloping distribution sampling
FRET – Förster Resonance Energy Transfer
H-REMD – Hamiltonian replica exchange molecular dynamics
IDP – intrinsically disordered protein
IDR – intrinsically disordered region
Map2c – microtubule associated protein 2c
MD – molecular dynamics
MM – molecular mechanics
MTSL – methanethiosulfonate
NMR – nuclear magnetic resonance
PKA – protein kinase A
PRE – paramagnetic relaxation enhancement
RD – regulatory domain
RDC – residual dipolar coupling
REMD – replica exchange molecular dynamics
SAXS – small-angle X-ray scattering
TD – tetramerization domain
TyrH – tyrosine hydroxylase
INTRODUCTION
19
1 Introduction
I decided on the title of my habilitation thesis “Dynamical features of biomolecular complexes”
due to three factors: i) it is the common denominator of the majority of the research
papers and projects I was involved in after my Ph.D. ii) it is included in multiple lectures
that I give during the Biophysical Chemistry course when describing theoretical or experimental
methods of this interdisciplinary field iii) interestingly, the topic of the dynamics of
biomolecular complexes seems to be viewed quite differently in computational biophysical
chemistry from the view in structural biology.
The final statement above reminds me to place a disclaimer here, namely – opinions presented
here are subjective and based on my experience when working in both research fields.
Researchers may disagree with some of the statements presented. While I was involved in
computational physical chemistry during my first post-doctoral stay at the research group of
prof. Chris Oostenbrink, Ph.D. at Vrije University in Amsterdam; I worked in the field of
structural biology during my second post-doc in the research group of prof. Angela Gronenborn,
Ph.D. at University of Pittsburgh (USA). At Masaryk University, I try to combine and
explore experiences from both of those fields in my running research at CEITEC-MU in the
group of prof. Lukas Zidek, Ph.D. as well as in my lectures from the division of Biophysical
Chemistry led by prof. Libuse Trnkova, Ph.D. I am involved in the following courses: Fundamentals
of Biophysical Chemistry; Experimental methods of Biophysical Chemistry I;
Advanced theoretical methods of Biophysical Chemistry; Experimental methods of Biophysical
Chemistry II; and contribute to the Protein Expression and Purification courses.
For my undergraduate and graduate studies, I decided on Biophysics because I always loved
and admired physics but at the same time wished to apply its powerful arsenal of methods
and rigorous view on complex entities such as biomolecular complexes, that I was taught
can explain the biological function. I must confess that I never liked chemistry during my
high school and undergraduate studies because of the need to memorize tons of names, only
later did I acknowledge the beauty and usefulness of chemistry and its subfields such as
INTRODUCTION
20
biochemistry. Another confession is the fact that I am not fond of strict definitions between
the different fields in natural sciences. In the current interdisciplinary world of science, I
tend to say to students – focus on your scientific problem and use the benefits of the most
suitable methods and technologies you know about. This spirit lies in the name of the study
program Biophysical Chemistry which covers three traditional study fields (biology, physics
and chemistry) but in my personal view also mathematics, computational sciences, bioinformatics
and partially medicine.
There should be a combination of not only the different fields but also of theoretical computational
and experimental approaches. I think that almost every theoretician or computational
scientist experiences frustration when obtaining interesting results from computational approaches
but facing disbelief from experimentalist colleagues.
Scientific joke: There is a running joke among scientists that when a theoretician
comes with a new theory there is only one person that believes in it – him/herself.
On contrary – experimental results are typically trusted by the whole community
except the experimentalist who performed the measurements and gave their interpretation.
He/she is aware of all the technical problems faced and the simplifications
used in the interpretation of measured data.
I still remember confusion at my undergraduate lectures when first hearing that proteins are
flexible and dynamic but at the very same time all functional features we were learning were
explained by one static figure of the corresponding 3D structure. We learned that NMR in
contrast to the “artificial” X-ray crystallography allows the determination of the real structural
ensemble of the studied protein in solution. When I learned it, I immediately downloaded
some of such ensembles from the pdb database but have to admit my disappointment.
It seemed to me like some small fluctuations around a central structure. Even, today I have
the feeling that many structural biologists have the feeling that one static structure is sufficient
for correlation within the structure-function paradigm.
A much more dynamic view of biomolecules tends to prevail within computational biophysics
and biophysical chemistry. “Bio” is here quite important because computational chemists
INTRODUCTION
21
often study small quite rigid molecules by quantum mechanical (QM) methods. And often
consider a molecule by single structure sometimes extended with small fluctuations in the
harmonic approximation. More complex systems such as biomacromolecules or in general
biomolecular complexes in an explicit water environment are instead treated by classical
molecular mechanics (MM) methods following classical Newtonian laws. There are some
exceptions in which one needs to apply QM methods e.g. when dealing with enzymatic activity
or in general when electronic properties should be considered explicitly. However,
when this is not the case – classical MM is typically used despite the fact that QM would be
more precise than classical MM. The underlying reason is that conformational sampling is
more important than “precision” in most cases.
Pedagogical note: In my lectures for undergraduate students I compare this choice to the
following problem – you should determine the average height of buildings in Brno (or the
lowest one) within 24 hours where you can choose between three of the following measuring
tools. The first measuring tool has micrometer precision but one measurement takes 12
hours. The second measuring tool has decimeter precision but one measurement takes only
1 minute and allows systematic scanning of building from north and row by row from west
to east. The third measuring tool is similar to the second one, i.e. has a precision of one
decimeter but it does not allow systematic scanning through the city – it rather randomly
selects buildings in Brno but thanks to this factor one measurement takes just one second.
(note: In my example, an American city would be used rather than Brno due to their planned
design consisting of perpendicular streets and avenues …). The fourth approach is even
faster – one measurement taking the only ms and in addition, it can quite quickly recognize
different areas with a larger variety of buildings. However, its (un)precision is only 10m.
Which of these four tools is most suitable for the given task? Probably the first tool may be
needed when installing satellite or GSM antenna for the mobile network on the particular
building. The second tool may be rather useful when comparing how heights of buildings
are changing along with particular districts or along the streets or avenues. The third approach
may be best when one cares about finding reliable values of the building heights in
the whole city thanks to its high speed. Sure, the fourth method is even much faster and can
INTRODUCTION
22
provide a useful overview picture or a larger variety of buildings in a variety of districts and
hope to find the lowest ones. But in the case of typical buildings, its inability to distinguish
family homes from four-story apartment buildings is in many applications a discriminating
factor. In this pedagogical example, I tried to present the following analogy: first tool – QM;
second tool – MD; third tool – enhanced sampling MD; fourth tool – ligand-protein and
protein-protein docking methods using scoring functions.
Because the dynamics of biomacromolecules and their complexes are understood differently
in different research fields here is my view of the meaning of this term. There are multiple
layers to the dynamics of biomolecular complexes:
i) with few exceptions, proteins should fold and unfold reversibly
ii) association and dissociation of biomolecular complexes (protein/ligand)
iii) the flexibility of protein side-chains around a stable folded conformation
iv) transitions between multiple conformational states
v) plasticity of the binding site allowing it to adapt to a variety of ligands (e.g. druglike
molecules)
vi) induced fit, conformational selection
vii) regardless of the higher or lower rigidity of biomolecules, there are always highly
mobile water molecules surrounding studied molecules at room or physiological
temperatures
viii) Conformational ensembles. Most of the above factors have a direct consequence
on conformational ensembles and thus affect the thermodynamic/statistical physics
properties such as entropy, enthalpy, and free energy.
The habilitation thesis is divided into four scientific chapters summing up 15 selected peerreview
papers of the applicant after his Ph.D. defense till this date. Given that each of these
chapters’ uses the molecular dynamics method in some way – it is described with the related
issues in the general theoretical background section preceding the scientific ones.
GENERAL THEORETICAL BACKGROUND – MOLECULAR DYNAMICS
23
2 General Theoretical Background – Molecular Dy-
namics
P0: Jozef Hritz* & Arnost Mladek: Plant Structural Biology: Hormonal Regulations
2018, 295-322, (ch: Computational Molecular Modeling Techniques of
Biomacromolecular Systems), Eds. Hejátko, J., Hakoshima, T.
MD simulations allow the calculation of the trajectory of all atoms in the simulated system
during a period of time, where the time evolution of a set of interacting atoms is followed by
integrating their equations of motion. By following the dynamics of a molecular system in
space and time, we can obtain information about molecular geometries and energies, mean
atomic fluctuations, local fluctuations (like formation/breakage of hydrogen bonds), protein/ligand
binding, free energies, and the nature of various types of concerted motions. Classical
mechanics is sufficiently precise for most practical cases, except for those cases where
is not possible to neglect QM effects. At the heart of molecular dynamics simulations lay the
empirical force-fields that allow calculating the potential energy of the simulated system as
a function of the nuclear coordinates. An overview of the commonly used biomolecular
force fields, and how they were developed over the last five decades, was reviewed in
e.g.1 The applicant also recently reviewed in the book chapterP0 - several aspects of MD
in the context of convergence of produced conformational ensembles and enhanced
sampling methods. In the following part of this general theoretical chapter, only the
most relevant sections fromP0 are listed in the view of scientific chapters. For more general
aspects such as empirical force-fields, periodic boundary conditions, geometry optimizations,
MD algorithms and protocols, the reader is referred either to the book chapter itself P0
or to the textbooks describing a variety of molecular modeling techniques from QM level
through the classical MM to the fluid dynamics, e.g.2–5.
2.1 Ensemble averages – link to the experimental data
The MD simulation is in many aspects very similar to the real experiment. It is the reason,
why it is often called a computational experiment or experiment in the white and
why the calculation of averages of properties of our interest is very similar. In the case
GENERAL THEORETICAL BACKGROUND – MOLECULAR DYNAMICS
24
of an experimentally measured value of the property 𝑋(𝑟 𝑁
, 𝑝 𝑁
), it is the average value
for the duration of the experiment. In principle, the real average value would be gained
in the measurement in the limit of infinite time duration:
We know from experience, that the determination of the average value with sufficient
precision measurement within a finite time duration is sufficient (this time has to be
much longer than the relaxation time of the measured quantities). Similarly, in the MD
simulation with the high enough number of steps M we can “measure” many properties
of a system as the average value of discrete values in the individual steps:
The dilemma appears to be that one can calculate time averages using MD simulation,
but the experimental observables are assumed to be ensemble averages as outlined
above. Resolving this leads us to one of the most fundamental axioms of statistical mechanics,
the ergodic hypothesis, which states that the time average equals the ensemble
average in case of infinite simulations:
The basic idea is that if one allows the system to evolve in time indefinitely, that system
will eventually pass through all possible states. One goal, therefore, of a molecular
dynamics simulation is to generate enough representative conformations such that this
equality is satisfied. If this is the case, experimentally relevant information concerning
structural, dynamic and thermodynamic properties may then be calculated using a feasible
amount of computer resources. Because the simulations are of fixed duration, one
must be certain to sample a sufficient amount of phase space.
𝑋 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 = lim
𝜏→∞
1
𝜏
∫ 𝑋
𝜏
𝑡=0
(𝑟 𝑁⃗⃗⃗⃗ (𝑡), 𝑝 𝑁⃗⃗⃗⃗⃗ (𝑡)) 𝑑𝑡
(2.1)
𝑋 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 ≅
1
𝑀
∑ 𝑋
𝑀
𝑖=1
(𝑟 𝑁⃗⃗⃗⃗ 𝑖, 𝑝 𝑁⃗⃗⃗⃗⃗
𝑖
)
(2.2)
〈𝑋〉 𝑒𝑛𝑠 = 〈𝑋〉 𝑡𝑖𝑚𝑒
(2.3)
GENERAL THEORETICAL BACKGROUND – MOLECULAR DYNAMICS
25
In general, the ensemble average may be written as:
where 𝑋(𝑟 𝑁
, 𝑝 𝑁
) stands for the observable of interest expressed as a function of positions
𝑟 𝑁
and momenta 𝑝 𝑁
of 𝑁 particles, in the case of classical MD simulations of all
the atoms, a system is comprised of. Note that the summation runs over all possible
values of the variables 𝑟 𝑁
and 𝑝 𝑁
, or, in other words, overall possible system states.
The second term in equation (2.4) is the ensemble probability density reflecting the
intuitive fact that each microstate of the system is adopted with a different probability.
According to the Boltzmann law, the canonical ensemble probability density is given
by:
where 𝐸(𝑟 𝑁
, 𝑝 𝑁) is the total energy of the system including potential energy (𝐸 𝑝(𝑟 𝑁
))
and kinetic (𝐸 𝑘(𝑝 𝑁
)) as a function of positions and momenta, 𝑇 is the absolute temperature
of the system, 𝑘 𝐵 is the Boltzmann constant. Thanks to the fact that Boltzmann
factor is factorizable, i.e.:
we can write:
Considering that probability density has to be normalized we get:
〈𝑋〉 𝑒𝑛𝑠 = ∑ 𝑋(𝑟 𝑁
, 𝑝 𝑁
)𝜌(𝑟 𝑁
, 𝑝 𝑁
)
𝑟 𝑁,𝑝 𝑁
(2.4)
𝜌(𝑟 𝑁
, 𝑝 𝑁)~𝑒
−𝐸(𝑟 𝑁,𝑝 𝑁)
𝑘 𝐵 𝑇
(2.5)
𝑒
−𝐸(𝑟 𝑁,𝑝 𝑁)
𝑘 𝐵 𝑇 = 𝑒
−𝐸 𝑝(𝑟 𝑁)
𝑘 𝐵 𝑇 𝑒
−𝐸 𝑘(𝑝 𝑁)
𝑘 𝐵 𝑇
(2.6)
𝜌(𝑟 𝑁)~𝑒
−𝐸 𝑝(𝑟 𝑁)
𝑘 𝐵 𝑇
(2.7)
𝜌(𝑟 𝑁) =
1
𝑍
𝑒
−𝐸 𝑝(𝑟 𝑁)
𝑘 𝐵 𝑇
(2.8)
GENERAL THEORETICAL BACKGROUND – MOLECULAR DYNAMICS
26
where 𝑍 is the partition function and serve here as a normalization factor:
2.2 Replica-exchange molecular dynamics (REMD)
In addition to MD, there are also other methods that allow the generation of the canonical
ensemble. Another commonly used method is the Monte-Carlo (MC) approach. In
the first step of MC random step/move (e.g. of an atom) is generated and subsequently
𝑍 = ∑ 𝑒
−𝐸 𝑝(𝑟 𝑁)
𝑘 𝐵 𝑇
𝑟 𝑁
(2.9)
Pedagogical note: An exercise from the advanced theoretical methods of
Biophysical Chemistry, students are expected to calculate <Uharm> for
the quadratic form of Uharm
𝑈ℎ𝑎𝑟𝑚(𝑥) ≡ 𝑈0 +
𝜅
2
(𝑥 − 𝑥0)2 (2.10)
by using equation:6
〈𝑈ℎ𝑎𝑟𝑚〉 =
∫ 𝑈ℎ𝑎𝑟𝑚(𝑥)𝑒
−
𝑈ℎ𝑎𝑟𝑚(𝑥)
𝑘 𝐵 𝑇 𝑑𝑥
∞
−∞
∫ 𝑒
−
𝑈ℎ𝑎𝑟𝑚(𝑥)
𝑘 𝐵 𝑇 𝑑𝑥
∞
−∞
(2.11)
after several steps of integration, they finally derive 〈𝑈ℎ𝑎𝑟𝑚〉 = 𝑈0 +
𝑘 𝐵 𝑇
2
Students are impressed that this final result does not depend on spring
constant 𝜅, i.e. the width of potential energy. This is content of
equipartition theorem – energy tends to distribute equally among
available modes (roughly degrees of freedom), with.
𝑘 𝐵 𝑇
2
allocated to
each. Also, kinetic energy that depends harmonically, averages as
𝑘 𝐵 𝑇
2
per degree of freedom.
GENERAL THEORETICAL BACKGROUND – MOLECULAR DYNAMICS
27
Figure 1: Schematic example of Hamiltonian replica exchange molecular dynamics (H-REMD)
with ten replicas.
Different Hamiltonians of individual replicas are controlled by the parameter λ.
this state 2 (having potential energy Ep(2)) is accepted or rejected based on Metropolis
criterion:
The MC method can be very efficient at simulating liquids or systems in a vacuum. However,
simulations of complex biomolecular systems in explicit water environments suffer
from the fact that most of the trial moves are rejected because they lead to a high
increase of potential energy. This is the very problem that the replica-exchange molecular
dynamics method (REMD) tries to solve.7
The principle of REMD combines the ideas and advantages of MD and MC methods.8,9
In REMD simulations, multiple non-interacting replicas of the system are simulated in
𝑝(1 → 2) = {
1 𝑓𝑜𝑟 𝐸 𝑝(2) ≤ 𝐸 𝑝(1)
𝑒
−
𝐸 𝑝(2)−𝐸 𝑝(1)
𝑘 𝑏 𝑇 𝑓𝑜𝑟 𝐸 𝑝(2) > 𝐸 𝑝(1)
(2.12)
GENERAL THEORETICAL BACKGROUND – MOLECULAR DYNAMICS
28
parallel by MD, each at a different condition. Conservation of the Boltzmann canonical
structural ensemble for individual conditions is guaranteed by the application of the
Metropolis criterion for exchange trials of complete configurations between neighboring
replicas. Schematically it is presented in Fig. 1. The REMD methods differ in the
chosen parameter that changes among the simulated replicas. There are two types of
REMD distinguished by varying condition: temperature REMD (T-REMD)10 and Hamiltonian
REMD (H-REMD).11–13
T-REMD can be applied straightforwardly to any system, although the maximal temperature
difference between neighboring replicas depends inversely on the square
root of the number of degrees of freedom, essentially prohibiting the application of TREMD
for systems in explicit solvent. H-REMD has a much better potential for systems
in explicit solvent as the Hamiltonians of individual replicas may differ only in selected
interactions, resulting in a smaller number of replicas to be used. Once well-chosen
modifications of the interactions contributing most to high energy barriers are made,
H-REMD can provide a sampling efficiency that is several orders of magnitude higher
as compared to T-REMD.7 However, efficient H-REMD needs a priori knowledge of the
origin of energy barriers in the system, which implies that different systems require
different H-REMD schemes.14
Temperature replica-exchange molecular dynamics (T-REMD)9,10,15
A T-REMD scheme starts with choosing the temperatures. Values of temperature must
cover a wide range including the biologically relevant ones (e.g. 37°C) as well as temperatures
at which sampling is significantly enhanced, i.e. elevated temperatures. Intuitively,
systems at higher temperatures can overcome energy barriers easily, compared
to a situation at low temperatures. However, the temperature does not only change
free energy barriers, but it also changes the free energy values of individual minima.
This is secured by the replica-exchange part. At certain time intervals, for example,
every picosecond, energies of neighboring replicas are compared. A replica simulated
GENERAL THEORETICAL BACKGROUND – MOLECULAR DYNAMICS
29
at a lower temperature tends to have lower potential energy than the one simulated at
a higher temperature. However, as temperature differences between replicas are small,
it may happen that the potential energy of the high-temperature replica is lower. In this
case, coordinates of replicas are exchanged and the simulation of low-temperature continues
at higher temperature and vice versa (probability of exchange is 1). If not, replica-exchange
probability is calculated from the potential energy difference as:
where 𝐸𝑖 and 𝐸𝑗 are energies of the i-th and j-th replica, respectively. Similarly, 𝑇𝑖 and
𝑇𝑗 are respective absolute temperatures of the i-th and j-th replica.16 The computer
then generates a random number in the range from 0 to 1. If this number is lower than
the calculated probability, replicas are exchanged and the simulation continues. At the
end of the simulation, it is possible to collect all coordinate snapshots at a certain temperature.
The exchange criterion is designed in the way that such collected snapshots
sample the studied system canonically at a given temperature. This means that it is
possible to calculate probabilities of different conformational families and the free energy
surface, not only for the biological temperature.
Hamiltonian replica-exchange molecular dynamics (H-REMD)11,17
As the temperature-based REMD method suffers from the need of high numbers of replicas
to be efficient, the Hamiltonian replica-exchange (H-REMD) is often the method of
choice.12,18 The H-REMD method is grounded on the consideration that, since the different
parallel simulations do not interact (i.e. independent simulations), there is no
need to use the same Hamiltonian for all of the replicas. In H-REMD the different replicas
are usually simulated at a constant temperature, while the Hamiltonian of the system
is used as the replica coordinate. While the standard part of the Hamiltonian (the
kinetic and the potential energy) is usually common for all simulated replicas, the
𝑃 = exp [(𝐸𝑖 − 𝐸𝑗) (
1
𝑘 𝐵 𝑇𝑖
−
1
𝑘 𝐵 𝑇𝑗
)]
(2.13)
GENERAL THEORETICAL BACKGROUND – MOLECULAR DYNAMICS
30
additional bias term is replica-dependent. The switching probability 𝛼𝑖𝑗 is determined
based on the energy difference following the Metropolis criterion:
where Hamiltonian 𝐻 is defined as the sum of the force field and the bias potential.
There are various ways to modify the Hamiltonian. For protein-protein interactions, it
is often advantageous to complement the standard Hamiltonian with a distance restrain
potential.19 To overcome energy barriers, it is possible to use variable soft-core
potentials for the interactions that contribute most to the energy barriers.14 It should
be stressed, however, that the setup of the replica-coordinate parameters may be nontrivial
and it is often necessary to go through a trial and error tuning process. Another
possibility is to utilize some kind of optimization algorithm to get the most appropriate
parameter values.20
Summing 15 selected peer-review papers in 4 scientific chapters:
𝛼𝑖𝑗 = min {1, 𝑒
−
𝐻𝑖(𝑞 𝑗)+𝐻𝑖(𝑞 𝑖)
𝑘 𝐵 𝑇 𝑖
+
−𝐻 𝑗(𝑞 𝑖)+𝐻 𝑗(𝑞 𝑗)
𝑘 𝐵 𝑇 𝑗 }
(2.14)
EFFICIENT INCORPORATION OF PLASTICITY AND WATER MOLECULES INTO THE MOLECULAR
DOCKING APPROACH
31
3 Efficient incorporation of plasticity and water molecules
into the molecular docking approach
Paper 1: Hritz, J.; de Ruiter, A.; Oostenbrink, C. Impact of plasticity and flexibility
on docking results for Cytochrome P450 2D6: a combined approach of molecular dynamics
and ligand docking. J. Med. Chem. 2008, 51, 7469-7477
Paper 2: Santos, R.; Hritz, J.; Oostenbrink, C. The role of water in molecular docking
simulations of Cytochrome P450 2D6. J. Chem. Inf. Model. 2010, 50, 146-154
Paper 3: Oostenbrink C.; de Ruiter A.; Hritz J.; Vermeulen N.P.E Malleability and
versatility of Cytochrome P450 active sites studied by molecular simulations.
Curr. Drug Metab. 2012, 13, 190-196
Paper 4: Byeon I-J. , Ahn J., Mitra M., Byeon C-H., Hercík K., Hritz J., Charlton L.,
Levin J., Gronenborn A.M. NMR structure of human restriction factor APOBEC3A reveals
substrate binding and enzyme specificity. Nat. Commun. 2013, 4, 1890
Ligand-protein and protein-protein docking are very popular techniques because of
their high speed and because they provide a reasonable variety of bound conformations.
The majority of ligand-protein docking approaches treat the proteins as
mostly rigid and ligands as flexible. Typically, their limitations lie in lacking the possibility
to incorporate the induced fit effect or explicit water molecules in the expected
binding cavity. This limitation is quite obvious in the case of the crystal structure of the
apo form of Cytochrome P450 2D6 (pdb code: 2F9Q),21 an important target of pharmaceutical
companies trying to avoid drugs for which human liver metabolism is solely
dependent on only this enzyme. The reason is that about 10% of the western population
lacks this enzyme and thus may easily be overdosed if the drug is degraded only
by Cytochrome P450 2D6. For this reason, companies like to exclude those cases in the
early stages of drug development preferably in initial computational screening utilizing
ligand docking amongst other approaches. Here the problem is that the traditional ligand-protein
docking approaches lead to only a 20% reliability of prediction of the site
of metabolism (SOM). In P1 we introduced an approach combining ensemble ligand
EFFICIENT INCORPORATION OF PLASTICITY AND WATER MOLECULES INTO THE MOLECULAR
DOCKING APPROACH
32
docking and molecular dynamics (MD) simulations, incorporating the plasticity and
flexibility of cytochrome P450 2D6 structures in a highly efficient way and leading to a
significant increase in the reliability of the predicted binding poses.
The presented approach selects the three most suitable structures out of several thousand
generated by MD. We have also developed a binary decision tree to decide
which protein structure to dock the substrate into, such that each ligand needs
to be docked only once, leading to a successful site of metabolism prediction in
80% of the substrates. This approach has wide applicability and became quite popular
in the community as is documented by over a hundred citations of this paper. Another
encouraging fact was that based on an existing crystal structure of CYP2D6 in the
apo form, our computational results revealed different conformational states of
Phe483 for different sets of substrates. Four years later our data were confirmed by
crystallography studies of CYP2D6 in complex with various compounds.22
While being quite successful in the incorporation of induced fit and flexibility, our effort
to incorporate the presence of critical water molecules into the docking approach for
CYP2D6.23,P2 was less successful. The paper of my master student describes a possible
way to consider possibly critical water molecules but then let a docking algorithm “decide”
whether to keep it or not based on the scoring function.P2 Although this approach
works, the reliability of outcomes was increased by only 1-2 % which considering the
complexity of the proposed approach is probably not worthwhile for the application
field. Rather the very direct and much faster approach that was already described
aboveP1 would be preferred. The comparison of a variety of computational approaches
applied in the field of Cytochrome P450 was reviewed by our group inP3.
Here it is important to emphasize that the described incorporation of induced fit and
flexibility in molecular docking has very general applicability outside of the Cytochrome
P450 field. We for example used a version of this approach for the prediction
EFFICIENT INCORPORATION OF PLASTICITY AND WATER MOLECULES INTO THE MOLECULAR
DOCKING APPROACH
33
of the complex structure between highly flexible oligonucleotides in complex with human
restriction factor APOBEC3A protein. In this approach, we first generated a relatively
large structural ensemble of the most flexible parts of the studied system, performed
molecular docking, and then refined them through the use of NMR data. This
approach has been developed during my post-doctoral stay in the research group of
Prof. Angela Gronenborn.P4
While the modifications listed in this chapter provide improved reliability in structural
information, they were still unsatisfactory in terms of the prediction of binding affinities
from docking simulations due to the limiting potential of scoring functions. Methods
allowing for more reliable predictions of absolute and relative binding affinities
and related to this, sufficient conformational sampling, are addressed in the following
two chapters.
EFFICIENT INCORPORATION OF PLASTICITY AND WATER MOLECULES INTO THE MOLECULAR
DOCKING APPROACH
34
Paper 1
Hritz, J.; de Ruiter, A.; Oostenbrink, C. Impact of plasticity and flexibility on docking results for Cytochrome
P450 2D6: a combined approach of molecular dynamics and ligand docking.
J. Med. Chem. 2008, 51, 7469-7477
Subscriber access provided by Library | Vrije Universiteit Amsterdam
Journal of Medicinal Chemistry is published by the American Chemical Society.
1155 Sixteenth Street N.W., Washington, DC 20036
Article
Impact of Plasticity and Flexibility on Docking
Results for Cytochrome P450 2D6: A Combined
Approach of Molecular Dynamics and Ligand Docking
Jozef Hritz, Anita de Ruiter, and Chris Oostenbrink
J. Med. Chem., 2008, 51 (23), 7469-7477 • Publication Date (Web): 11 November 2008
Downloaded from http://pubs.acs.org on December 15, 2008
More About This Article
Additional resources and features associated with this article are available within the HTML version:
• Supporting Information
• Access to high resolution figures
• Links to articles and content related to this article
Impact of Plasticity and Flexibility on Docking Results for Cytochrome P450 2D6: A Combined
Approach of Molecular Dynamics and Ligand Docking
Jozef Hritz, Anita de Ruiter, and Chris Oostenbrink*
Leiden-Amsterdam Center for Drug Research, Section of Molecular Toxicology, Department of Chemistry and Pharmacochemistry,
Vrije UniVersiteit, De Boelelaan 1083, 1081 HV Amsterdam, The Netherlands
ReceiVed August 8, 2008
Cytochrome P450s (CYPs) exhibit a large plasticity and flexibility in the active site allowing for the binding
of a large variety of substrates. The impact of plasticity and flexibility on ligand binding is investigated by
docking 65 known CYP2D6 substrates to an ensemble of 2500 protein structures. The ensemble was generated
by molecular dynamics simulations of CYP2D6 in complex with five representative substrates. The effect
of induced fit, the conformation of Phe483, and thermal motion on the accuracy of site of metabolism
(SOM) predictions is analyzed. For future predictions, the three most essential CYP2D6 structures were
selected which are suitable for different kinds of ligands. We have developed a binary decision tree to
decide which protein structure to dock the ligand into, such that each ligand needs to be docked only once,
leading to successful SOM prediction in 80% of the substrates.
Introduction
Cytochrome P450s (CYPs)a
form a superfamily of hemethiolate-containing
proteins which play a crucial role in the
metabolism of drugs and other xenobiotics.1
CYPs generally
detoxify potentially hazardous compounds, but in a number of
cases nontoxic parent compounds are bioactivated into toxic
metabolites or procarcinogens into the ultimate carcinogens. The
human isoform cytochrome P450 2D6 (CYP2D6) constitutes
only ∼2% of the hepatic CYPs but is responsible for the
metabolism of 15-20% of currently marketed drugs.2,3
CYP2D6
substrates and inhibitors are commonly characterized by a
protonated basic nitrogen and an aromatic system, which are
also common features in drugs addressing the central nervous
or cardiovascular system.4
Inhibition of CYP2D6 may easily
lead to adverse drug-drug interactions. Large interindividual
differences exist in CYP2D6 activity, due to gene multiplicity
and genetic polymorphisms, emphasizing the need to include
metabolism prediction early in the drug discovery process.5,6
Early considerations of AMDET properties (absorption, metabolism,
distribution, excretion, and toxicology) include attempts
to model the CYP activity in silico.7-9
For many years,
structure-based approaches had to rely on homology models10
until the first (apo) structure of the enzyme was published in
2006.11
Many CYPs seem to show a larger extent of active site
plasticity and flexibility when compared to many other proteins.
This may be explained from the ability of CYPs to bind and
metabolize a large variety of substrates.12,13
Even if standard
docking approaches are quite reliable for less flexible proteins,
special care has to be taken for proteins with very malleable
active sites. This paper describes the impact of different kinds
of flexibility of CYP2D6 on the prediction of the site of
metabolism (SOM). We will describe strategies to incorporate
the most essential structural variety into efficient ligand docking
methods. We use a variant of ligand docking into an ensemble
of protein structures generated by molecular dynamics (MD)
simulations. This approach was introduced by Pang and Kozikovski
in their docking study of huperzine A to 69 snapshots of
a MD trajectory of acetylcholinesterase.14
The rationale behind
this approach is supported by recent advances in the description
of ligand-protein binding. The traditional lock-and-key picture
seems to be gradually replaced by a model in which an ensemble
of protein conformations is described and a ligand selectively
binds to the most appropriate shape of the binding site.15,16
In silico structure-based predictions of drug metabolism are
usually considered to be made up of (1) the binding orientation
of the substrate in the active site, placing the reactive group in
close vicinity of the heme iron atom, and (2) the intrinsic
reactivity of this group.17,18
The relative importance of these
two issues will differ between different CYPs. The substrate
binding pose will play a crucial role for the tight binding site
of CYP1A2 but seems to be less relevant for the large binding
site of CYP3A4. Here we address the prediction of the site of
metabolism (SOM) based on the most likely poses of substrates
within the active site of CYP2D6. In particular, we will address
the effects of protein plasticity and flexibility on such predictions.
For this, many thousands of docking experiments are
performed, based on which a small number of protein structures
will be selected which are most optimal for SOM predictions.
The only available crystal structure of CYP2D6 is an apo
structure, for which the crystallographers already observed that
the active site is too small to fit known CYP2D6 substrates.11
Also, other groups reported better results when docking into
CYP2D6 structures refined from MD simulations with ligand
bound.19,20
The active site needs to be opened up, to accommodate
a variety of substrates. This raises the questions how
much the active site needs to be enlarged and, more importantly,
if different substrates should be fitted into the same shape of
the active site. Because of their versatile function in the
metabolism of diverse compounds, CYPs are known to show
large plasticity in their active sites, being able to accommodate
many different substrates.12
In this work, we distinguish three
different kinds of protein flexibility.
* To whom correspondence should be addressed. Phone: +31 20
5987606. Fax: +31 20 5987610. E-mail: c.oostenbrink@few.vu.nl.
a
Abbreviations: CYP, cytochrome P450; CYP2D6, cytochrome P450
isoform 2D6; EDR, (R)-3,4-methylenedioxy-N-ethylamphetamine; LPBS,
ligand probing binding site method; MD, molecular dynamics; MMC,
7-methoxy-4-(aminomethyl)coumarin; PPD, (R)-propranolol; CHZ, chlorpromazine;
TMF, tamoxifen; SOM, site of metabolism.
J. Med. Chem. 2008, 51, 7469–7477 7469
10.1021/jm801005m CCC: $40.75  2008 American Chemical Society
Published on Web 11/11/2008
(1) Structurally different substrates will induce different
conformations of the protein active site or, alternatively,
specifically bind to different conformations from the overall
structural ensemble of the protein.16
(2) Within the conformations suitable to accommodate a
specific substrate, one may still distinguish different subclasses
of conformations, which are separated by relatively high
energetic barriers. In the case of CYP2D6 two such conformations
are observed for the side chain of Phe483, as will be
explained below.
(3) All atoms constituting the active site are continuously in
thermal motion, leading to smaller fluctuations in the shape of
the active site. We will show that also thermal fluctuations may
have a strong effect on SOM predictions.
To incorporate these three types of plasticity and flexibility,
we generate a library of 2500 protein structures, which covers
the various conformations of the CYP2D6 active site. Subsequent
docking experiments on all members of this library allows
for an analysis of the effects of the different phenomena on the
accuracy of SOM predictions. Finally, we will select the
essential ensembles of protein structures and propose an optimal
set of protein structures to be used in docking experiments. A
simple decision tree model is also presented to select for any
given putative substrate, which protein structure is most likely
to yield an accurate SOM prediction.
Results
The library of protein conformations was built up in the
following way. To account for induced fit effects, five representative
substrates (Figure 1) were taken from a database of
65 known CYP2D6 substrates. Forty-five substrates could be
grouped into five clusters, represented by the substrates in Figure
1. The remaining 20 substrates were placed in the sixth cluster.
See Figure S1 in the Supporting Information for the complete
set of substrates, how they are clustered, and an indication of
the major SOM. We remark that for some substrates multiple
SOMs were considered. One of the hypotheses to be investigated
here is that the active site adopts a conformation around the
representative substrates which is similar to the conformation
it adopts around the other members of the cluster.
Opening of CYP2D6 Catalytic Site. Docking the smallest
representative substrate (EDR) into the apo structure of CYP2D6
resulted in the SOM being placed at 4.5 Å from the heme iron.
Subsequently, the system was solvated and electroneutralized
by adding counterions and relaxed by molecular dynamics (MD)
simulation. All waters, counterions, and EDR atoms were
removed from the structure obtained after 100 ps of unrestrained
MD, which was subsequently used for the docking of the larger
representative substrate MMC. This procedure was repeated for
the other representative substrates of increasing size (PPD, CHZ,
TMF), thereby inducing CYP2D6 conformational changes to
accommodate a wider range of substrates.
Phe483 Conformations. The largest conformational changes
of the protein (mainly side chains) around the representative
substrates occurred during the 100 ps of MD described in the
previous paragraph. Subsequent MD simulations were performed
to include thermal motions in equilibrium. If different conformations
present at room temperature are separated by a lowenergy
barrier, then they should be adequately sampled within
0.5 ns of MD, but in the case of larger barriers the sampling
within this time may be inadequate.
MD simulations of CYP2D6 in complex with each of the
representative substrates revealed that two conformations of
the Phe483 side chain are possible but that transitions take place
on a time scale (∼1 ns) that is too long to accurately determine
the relative populations of these conformations. Therefore, we
have performed 1 ns MD simulations for each of the representative
substrates in which the χ1 angle of Phe483 (χ1
483
) was
restrained to either 70° or 170°. This leads to 10 structural
ensembles of active site conformations, from each of which we
stored 250 structures, sampled from the last 500 ps of the
simulations. We refer to the ensembles by the name of the
representative substrate used to generate them, followed by
the approximate value of χ1
483
: EDR_170 represents the
ensemble of 250 CYP2D6 structures which were obtained from
MD of CYP2D6 in complex with EDR in which the dihedral
angle χ1 of the Phe483 side chain was restrained to 170°. The
complete structural library now contains 2500 protein structures.
Docking into CYP2D6 Structural Ensembles. Sixty-five
substrates of CYP2D6 with known sites of metabolism were
docked into 250 protein structures for each of the 10 ensembles.
An example of SOM prediction for the representative substrates
for one particular docking simulation to one protein structure
is shown under “single docking” in Table 1. This table presents
the distance of the SOM from the heme iron atom (third column)
for the first ranked pose (score in the fourth column).
As described in the Experimental Section, five docking
simulations were performed for every protein structure, leading
to the final prediction for one particular protein structure shown
under “five dockings” in Table 1. It does not only show the
SOM prediction (fifth column) but also the corresponding
consensus between the five docking simulations (sixth column).
These data were averaged over 250 CYP2D6 structures for every
ensemble. The columns labeled as “250 frame ensemble” in
Table 1 present for each representative substrate the fraction of
frames for which the binding mode was predicted with the SOM
within 6 Å from the heme iron atom. The same percentages for
all 65 substrates docked into the PPD_70 ensemble giving the
most reliable SOM prediction (as shown below) are presented
in Figure 2. This figure is divided into six parts; the first five
parts of the table contain the substrates belonging to clusters
characterized by the listed representative ligands. Ligands which
do not belong to any of these five clusters are listed in the sixth
part of the figure. Tables with the same statistical analysis
extended with consensus results for the highest ranked pose,
over the five highest ranked poses, and over all 10 docking poses
for all 10 ensembles are available in the Supporting Information
(Tables S1-S10). For all 10 ensembles we note that the average
reliability of binding mode prediction selecting the binding pose
with the highest score from five independent dockings (third
column, labeled “first rank”) is slightly higher than the average
Figure 1. Chemical structures of the five representative substrates:
(R)-3,4-methylenedioxy-N-ethylamphetamine (EDR), 7-methoxy-4(aminomethyl)coumarin
(MMC), (R)-propranolol (PPD), chlorpromazine
(CHZ), and tamoxifen (TMF).
7470 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 23 Hritz et al.
over all first ranked poses of the individual dockings (fourth
column, labeled “consens”).
Discussion
The structural ensemble of CYP2D6 covers a wide range of
active site plasticity and flexibility. The effect of thermal motion,
induced fit, and the conformation of Phe483 on the reliability
of the SOM prediction is discussed in the following sections.
Impact of Thermal Fluctuations on CYP2D6 Docking
Predictions. Figure 3 shows the variation in the number of
substrates for which the first ranked pose places the SOM within
6 Å distance from the heme iron atom for the PPD_70 ensemble.
A sampling time of 2 ps between subsequent structures allows
for only small conformational changes. Still, it can be seen that
even such small changes in the protein structure have an impact
on the SOM prediction. For most ensembles the deviations from
the average values (summarized in Table 2) are within 10%,
but more extreme fluctuation can be observed. For instance, in
the EDR_70 ensemble, the percentage of substrates that is
correctly placed in the active site changes from 0% in structure
239 (indicated as EDR_70_fr_239) to 41.5% in the next
structure (EDR_70_fr_240). We do not see any obvious
structural difference between these protein structures (Figure
4), which indicates that very subtle differences in protein
structures can have tremendous effects on the SOM prediction.
In Figure 4, we show the binding mode prediction for substrate
MMC into both frames, which demonstrates the observation
that many substrates are bound to a subpocket between Gln244
and Ala300 when docked to the EDR_70_fr_239 structure. It
seems that such erratic behavior in the pose predictions can be
traced to the fact that the scoring functions are not continuous,
so that small structural changes can have big effects on the
prediction of the scores. We also emphasize that a correct
inclusion of entropic effects would prevent rare events from
becoming the first ranked predictions. This example demonstrates
the risk of blindly selecting a single protein structure
and strengthens us in the belief that one should consider the
values that are averaged over the structural ensemble (last two
Table 1. Best Scored Results from One Particular Docking Simulation, Final Prediction Based on Five Independent Docking Simulations into the Last
Frame, and Averages over the Complete PPD_70 Ensemble of CYP2D6 Structures, Showing Only the Five Representative Substrates
single dockingc
five dockingsf
250 frame ensemblei
codea
nameb
distd
scoree
distg
perch
(%) first rankj
(%) consensk
(%)
EDR R-MDEA 3.5 24.4196 3.5 80 52.0 53.5
MMC MAMC 3.5 26.7666 3.9 100 98.0 98.1
PPD (R)-propranolol 4.6 36.1397 3.2 100 99.2 98.2
CHZ chlorpromazine 3.5 41.9164 3.5 100 31.2 39.8
TMF tamoxifen 3.3 42.7726 3.7 100 86.4 86.6
totall
46/65 44/65 70.2 62.0 61.8
a
Three letter code for the substrate. b
Name of the substrate. c
First ranked docking pose over 10 poses from a single docking simulation. d
Distance from
the SOM to the heme iron in the first ranked docking pose. e
Value of the Chemscore docking score for the first ranked pose. f
Results based on five
independent dockings into the same frame. g
Distance from the SOM to the heme iron in the best ranked docking pose over all 50 poses. h
Fraction of the
first ranked poses in individual dockings (10 poses) giving the same SOM prediction as the overall best ranked pose. i
Results based on five independent
dockings into the ensemble of 250 frames. j
Percentage of frames (over 250 frames) where the highest scored binding pose (over five docking simulations;
50 poses) places the SOM within 6 Å distance from the heme iron atom. k
Same percentage as in footnote j, averaged over the first ranked poses of all
docking simulations. l
Statistics for all 65 substrates; number of substrates for which the SOM was predicted within 6 Å from the heme iron for the best
scoring pose in one docking simulation, over five docking simulations; the last three columns are averages over corresponding columns.
Figure 2. Percentage of frames within the PPD_70 ensemble in which the highest scored binding pose (over five docking simulations) places the
SOM within 6 Å distance from the heme iron atom. Chemical structures and names of all 65 substrates corresponding to the three letter codes are
listed in Figure S1 (a-f) and Tables S1-S10, respectively, in the Supporting Information. Substrates are divided into six groups according to five
clusters named by representative substrates, indicated with different colors in this figure.
Plasticity and Flexibility in Cytochrome P450 2D6 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 23 7471
columns of Table 1 and Tables S1-S10) or make a careful
selection of single frames as shown below.
Impact of Induced Fit Effects on CYP2D6 Docking Predictions.
Table 3 presents the ligand-probe binding site (LPBS)
distance (see Experimental Section) between the 10 structural
ensembles, based on the percentages of correct predictions in
Tables S1-S10. The largest difference of 36% is observed
between the MMC_70 and PPD_70 ensembles. Different
structural ensembles obtained from simulations of different
representative substrates are not equally well able to accommodate
the various substrates. This can be seen from the
percentage of frames in which individual substrates are positioned
with their SOM within 6 Å from the heme iron atom
(Tables S1-S10). We can determine if a substrate is indeed
most often correctly placed in the ensemble generated for the
representative substrate of the cluster to which it belongs. Table
4 shows how the percentages of protein structures in which a
correct pose was predicted for the representative substrates
docked into all ensembles. It can be seen that for MMC, PPD,
and CHZ we indeed obtain the highest reliability when docking
into protein structures refined from MD of CYP2D6 in complex
with themselves. For TMF a higher reliability (∼90%) is
obtained when docking into the CHZ_170 ensemble. Still, the
TMF_70 ensemble itself also leads to a quite high reliability
(∼70%). On the other hand, the reliability of docking EDR into
the EDR_70 and EDR_170 ensembles is significantly lower
(<40%) than when docking to the PPD_170 ensemble (∼63%).
Table 4 also shows that one of the most promiscuous
substrates is MMC because a high reliability is obtained when
docking to most of the structural ensembles. On the other hand,
E-doxepine seems to be very sensitive to the shape of the protein
binding site. Successful binding poses were only obtained to a
significant level (70%) when it was docked into the MMC_70
ensemble.
Impact of the Conformation of Phe483 on CYP2D6 Docking
Predictions. From the ensemble tables (Tables S1-S10)
and from Table 4, it can be seen that substrates show different
sensitivity toward the side-chain conformation of Phe483. The
most significant difference is observed for substrates belonging
to the MMC cluster, which are significantly more often placed
correctly in the MMC_170 ensemble than in the MMC_70
ensemble (Table 5). A closer inspection revealed that Phe483
with χ1
483
≈ 170° does not stabilize the correct binding poses
but rather destabilizes the incorrect binding pose (Figure 5). It
is interesting to note that an experimental study on the effect
of the F483A mutation on the metabolism of four substrates
shows the largest effect for MMC. The F483A mutant does not
show any metabolism of this substrate anymore.21
The molecular
explanation of this observation may very well be that the mutant
binds MMC in binding poses similar to the not catalytically
active pose (green) in Figure 5, as this is no longer destabilized
by Phe483.
Essential Conformational Ensembles. The 65 CYP2D6
substrates can be reclustered according to the ensemble of
structures that accommodates them most often in catalytically
active poses (Table S11 in the Supporting Information). This
reclustering shows that, for almost all substrates, at least one
ensemble exists that correctly predicts the SOM with high
reliability, leading to an overall reliability of 75.4%. We also
investigated which are the essential conformational essembles;
i.e., how many ensembles should be included to approach the
value of 75.4%. Table 2 presents the average reliability over
all substrates when considering single ensembles. The ensemble
leading to the largest number of substrates with their SOM
within 6 Å from the heme iron atom are the PPD_70 and the
PPD_170 ensemble (∼62%), followed by the CHZ_170 and
the CHZ_70 ensemble (∼58%). Because these are averages over
250 protein structures and over 65 substrates, the conformation
of Phe483 has only a small effect on these percentages. Next,
Figure 3. Docking into individual frames of the PPD_70 ensemble of
CYP2D6 structures. Solid line: The fraction of substrates for which
the binding pose with the overall highest score (from five docking
simulations) positions the SOM within 6 Å from the heme iron atom.
Dashed line: The average value when considering the first ranked poses
for individual docking simulations separately. Frame 216 (indicated
by the vertical line) was finally taken as the single CYP2D6 structure
giving the highest docking prediction reliability.
Table 2. Average Percentage over All Structures and 65 Substrates for
Each of the 10 Generated Ensembles
CYP2D6/
EDR (%)
CYP2D6/
MMC (%)
CYP2D6/
PPD (%)
CYP2D6/
CHZ (%)
CYP2D6/
TMF (%)
χ1
483
≈ 70° 45.4 42.2 62.0 57.4 47.4
χ1
483
≈ 170° 48.6 48.2 62.0 59.6 51.4
Figure 4. Binding mode prediction for MMC (ball and stick) when
docked into the EDR_70_fr_239 structure (shown in green) and in the
very similar EDR_70_fr_240 structure (shown in cyan). No successful
correct poses were observed for any of 65 substrates in the green
structure, but many occupy a subpocket between Gln244 and Ala300.
In the cyan structure 27 substrates docked with the SOM within 6 Å
from heme iron (indicated by the yellow dashed line).
7472 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 23 Hritz et al.
we have analyzed all combinations of two and three ensembles
and select for every substrate the ensemble for which the highest
value was obtained. This gives an upper limit to the averaged
reliability over all substrates. The combinations with the best
overall prediction are given in Table 6, with the optimal
reliability given in the third row. The assignment of the
substrates to the ensembles for all combinations is available in
Table S12 of the Supporting Information. The upper bound to
the SOM predictions converge to the overall upper bound based
on ensembles of 75.4%, with a value of 72.3% when including
only three ensembles. Note that even though all combinations
of ensembles were considered, the optimal combinations always
contain the same ensembles as the combination formed with
one ensemble less (Table 6).
For almost all substrates there is at least one structural
ensemble for which the SOM prediction is quite good. The only
exceptions are (R)- and (S)-fluoxetine with a maximum reliability
of only ∼8%. These substrates show a trifluoro group
and have their SOM at the positively charged nitrogen atom.
For most CYP2D6 substrates the positively charged nitrogen
atom is expected to interact with the negatively charged side
chain Glu216 (see, e.g., Figures 4 and 5), and the major SOM
is elsewhere in the molecule. This is also the case for the five
representative substrates used in this study. However, there are
also CYP2D6 substrates following different binding modes (e.g.,
substrates metabolized close to the positively charged nitrogen
atom (N+
)) which would not be found if one would constrain
the Glu216-N+
distance during docking simulation. For all
substrates that are metabolized close to the positively charged
nitrogen atom, other than (R)- or (S)-fluoxetine, there is at least
one ensemble that still shows a reasonable SOM prediction. In
most cases, this is the TMF_70 ensemble, in which the active
site was opened up most, allowing the substrate to dock in a
larger variety of poses. The fact that successful ensembles are
found for most substrates indicates that the CYP2D6 structural
library is extensive enough to dock substrates that are quite
distinct from the representative substrates.
Correlation of Molecular Properties with Essential Conformational
Ensembles. The observation that the reliability of
SOM predictions can be as high as 72.3%, using only three
structural ensembles, is promising, but it is far from trivial to
Table 3. Ligand-Probed Binding Site (LPBS) Distance between the 10 Protein Ensemblesa
EDR_70
(%)
EDR_170
(%)
MMC_70
(%)
MMC_170
(%)
PPD_70
(%)
PPD_170
(%)
CHZ_70
(%)
CHZ_170
(%)
TMF_70
(%)
TMF_170
(%)
EDR_70 0 9.3 17.1 17.3 30.3 28.4 26.7 26.8 21.7 20.6
EDR_170 0 18.4 17.1 28.1 24.7 24.6 23.5 20.7 18.0
MMC_70 0 23.0 36.0 33.5 29.6 30.8 21.6 23.3
MMC_170 0 27.2 29.4 30.3 30.4 25.2 23.1
PPD_70 0 16.0 26.5 26.8 30.8 27.6
PPD_170 0 18.4 18.0 27.9 21.7
CHZ_70 0 10.9 26.8 18.0
CHZ_170 0 26.5 16.3
TMF_70 0 21.0
TMF_170 0
a
For a description of the LPBS distance, see the Experimental Section.
Table 4. Fraction of Structures for Which the Substrates Dock
Correctly into the Different Structural Ensemblesa
CYP2D6
ensemble
EDR
(%)
MMC
(%)
PPD
(%)
CHZ
(%)
TMF
(%)
E-doxepineb
(%)
EDR_70 36.8 58.8 40.4 56.0 41.2 10.0
EDR_170 39.6 78.0 39.2 36.8 37.6 18.4
MMC_70 16.0 59.6 62.8 62.0 9.2 70.0
MMC_170 29.2 98.4 50.0 39.6 63.2 5.2
PPD_70 52.0 98.0 99.2 31.2 86.4 1.6
PPD_170 62.8 97.2 97.6 78.8 60.4 0.4
CHZ_70 62.4 74.0 91.2 98.0 52.4 0.4
CHZ_170 60.8 95.6 86.0 100.0 90.4 0.0
TMF_70 6.8 71.6 96.0 4.0 65.2 4.8
TMF_170 41.2 85.2 84.0 86.0 70.0 1.6
a
A docking pose is considered to be correct if the SOM of the substrate
is placed within 6 Å of the heme iron atom. b
E-doxepine is the substrate
which is the most sensitive to the structural ensemble to which it is docked.
Table 5. Effect of the Phe483 Side-Chain Conformation on the Fraction
of Structures for Which the Substrates Dock Correctly into Ensembles
of Structuresa
codeb
namec
MMC_70 (%) MMC_170 (%)
MMC MAMC 59.6 98.4
DMM diMAMC 30.4 93.2
EMP EMAMC 45.2 90.4
PMP PMAMC 39.2 67.6
BMM BMAMC 27.2 55.6
VER (R)-venlafaxine 8.8 24.0
VES (S)-venlafaxine 7.2 18.0
COD codeine 3.2 13.6
DXM dextromethorphan 18.4 76.4
average% 26.6 59.7
a
A docking pose is considered to be correct if the SOM of the substrate
is placed within 6 Å of the heme iron atom; data shown for the substrates
belonging to the MMC cluster where the strongest effect was observed only.
b
Three letter code of the substrate. c
Name of the substrate.
Figure 5. The number of frames into which MMC is incorrectly
docked is significantly higher for the MMC_70 ensemble (40.4%) (in
green) than for the MMC_170 ensemble (1.6%) (in cyan). The side
chain of Phe483 with χ1
483
≈ 170° seems to destabilize the incorrect
binding mode.
Plasticity and Flexibility in Cytochrome P450 2D6 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 23 7473
determine a priori to which ensemble a ligand should be docked.
PPD-like compounds are best docked to the PPD_70 ensemble,
and CHZ-like compounds are best docked to the CHZ_170
ensemble, but there are also some stereoisomers that give better
results when docked to the PPD_70 ensemble. As noted before,
the TMF_70 ensemble seems most suited for substrates which
are (atypically) metabolized close to the protonated nitrogen.
For new, possibly more diverse, sets of compounds it would
very advantageous if general molecular descriptors could be used
to predict the optimal ensemble of protein structures. Binary
decision trees were developed to divide the substrates over two
(PPD_70 and CHZ_170) or three (PPD_70, CHZ_170, TMF_70)
ensembles. The molecular weight was found to be the most
important descriptor to determine if substrates should be docked
into the PPD_70 or the CHZ_170 ensemble. A further improvement
of the overall reliability can be obtained, including the
TMF_70 ensemble and using the number of hydrophobic atoms
to divide the substrates between the CHZ_170 and TMF_70
ensembles. This decision tree is schematically drawn in Figure
6, and the average number of successful SOM predictions
following this decision tree is given in the fourth row of Table
6. Note that the increase in these values is considerably lower
than the upper bounds to the predictions discussed in the
previous section, which shows that the separation according to
the decision tree is still quite far from the optimal one.
Single Frames with the Highest Reliability of Prediction. The
previous sections describe the selection of essential CYP2D6
ensembles to dock substrates with a high reliability. For many
applications, such as virtual screening of large compound
libraries, it is not possible to perform docking into complete
structural ensembles. Therefore, it is of enormous practical
interest to select only very few protein structures, which lead
to accurate SOM predictions. To find the single CYP2D6
structure that offers the best SOM predictions, we first selected
from the library of 2500 structures the 15 protein structures with
the highest fraction of correct SOM predictions. Two additional
docking simulations (each of five docking runs) were performed
for these structures, to obtain a higher consensus between the
different runs. Two single frames were found in which >71%
of the substrates docked correctly. These originated from the
PPD_70 ensemble (structure 216: PPD_70_fr_216) and from
the PPD_170 ensemble (structure 173: PPD_170_fr_173).
The fact that a single structure can be selected is extremely
important if one wishes to perform docking experiments for
many putative inhibitors or substrates and no specific knowledge
is available for the ligand. These two structures offer the most
reliable docking poses out of all 2500 available protein structures
even though we note that these structures are only slightly better
than the 15 initially selected protein structures. We stress that
it is important to select protein structures carefully. As was
mentioned above, docking into the EDR_70_fr_239 structure
did not position the SOM of any substrate within 6 Å of the
heme iron atom within the first ranked pose. The choice of the
optimal structures will depend on the docking protocol and
scoring function that is used. It is unlikely that when using,
e.g., a different docking program this protein structure will fail
completely, but different structures may perform slightly better.
We compare the single protein structures that were selected
here to a homology model that was developed several years
ago in our group.10,22
The 65 substrates were docked to this
model using the same settings as before, and for 60% of the
substrates the SOM was placed within 6 Å of the heme iron
atom for the first ranked pose. This is significantly lower than
the 71% obtained here for the newly refined structures. When
docking directly to the original apo crystal structure of CYP2D6,
this percentage is only 20%. Using CYP2D6 structures from a
MD simulation starting from this apo structure without adding
a substrate decreases the reliability of SOM predictions to
0-9%, depending on the chosen frame (data not shown).
Essential Single Protein Structures. Similar to what was
described for the ensembles, we can search for a small set of
essential single protein structures. The single structures selected
in the previous section represent an essential set of size 1. Sets
of two and three single protein structures were selected from
the 2500 structures as well and are given in the fifth row of
Table 6. For the combination of three structures, an exhaustive
search involves searching through 25003
combinations. Rather,
we have limited our search to sets of three structures that contain
the set of two optimal protein structures. From an exhaustive
search, this set already contains the PPD_70_fr_216 structure,
which is the best single structure. The sixth row of Table 6
gives an upper bound to the docking reliability when using these
Table 6. Percentage of Correctly Docked Substrates for Essential Groups of Ensembles and Single Structuresa
essential groups 1 2 3 all
ensemblesb
PPD_70 PPD_70 PPD_70 all 10 ensembles
CHZ_170 CHZ_170
TMF_70
optimal reliabilityc
(%) 62.0 70.3 72.3 75.4
reliability based on decision treed
(%) 62.0 63.6 64.2
single structurese
PPD_70_fr_216 PPD_70_fr_216 PPD_70_fr_216 65 structures
CHZ_170_fr_79 CHZ_170_fr_79
TMF_70_fr_3
optimal reliabilityf
(%) 71.3 86.2 89.8 100
reliability based on decision treeg
(%) 71.3 79.4 80.3
a
A docking pose is considered to be correct if the SOM of the substrate is placed within 6 Å of the heme iron atom. b
Sets of the essential ensembles.
c
Maximum reliability over 65 substrates where for each substrate the optimal value toward the ensembles in the given set is taken. d
Averaged reliability
over 65 substrates where for each substrate the ensembles are selected based on the decision tree in Figure 6. e
Sets of the essential structures. f
Maximum
reliability over 65 substrates where for each substrate the optimal value toward the structures in the given set is taken. g
Averaged reliability over 65
substrates where for each substrate the structures are selected based on the decision tree in Figure 6.
Figure 6. Binary decision tree using the molecular weight and the
number of hydrophobic atoms to separate substrates into two or three
groups corresponding to different individual structures or ensembles.
This way, the optimal ensemble or single protein structure can be
selected for a substrate a priori.
7474 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 23 Hritz et al.
sets of structures, by selecting for every substrate an optimal
structure. Table S13 in the Supporting Information specifies for
each of these sets which substrate is best docked in which
protein structure. Note that the essential single structures stem
from the essential ensembles determined earlier, although this
was not imposed in the search algorithm. Therefore, it is not so
surprising that the binary classification tree that was developed
to divide the substrates over the single protein structures is
the same as the one described for the essential ensembles (Figure
6). The seventh row of Table 6 presents the reliability in
the docking predictions, which increases from 71.3% for a single
structure to 79.4% and 80.3% when taking two or three
structures into account, respectively. This increase is significantly
higher than observed for the essential ensembles (fourth
row of Table 6). We stress that these values can be obtained
without any additional increase of computational demands as
compared to docking into a single protein structure, as every
substrate is docked only once to a protein structure selected
based on the decision tree (Figure 6).
On the other hand, if substrates are docked to all three
essential protein structures, the consensus between predictions
can be used. The reliability of the binding mode prediction is
90% if the best frame is selected for every substrate. This means
that if substrates are docked to all three structures, and we obtain
the same binding mode, one can be almost sure that it will be
correct. From the 65 substrates, 29 showed complete consensus
when docking to the three essential structures. For 23 of these,
the SOM is within 6 Å from heme iron atom. Two of the
remaining six substrates are (R)- and (S)-fluoxetine, which were
already described above as failing to dock in the majority of
the 2500 protein structures. Among the 29 substrates, there are
11 substrates for which the consensus between the five
individual docking runs (second part of Table 1) was 100%.
All of these have the SOM within 6 Å from the heme iron atom.
In cases where no consensus is reached between the different
protein structures, we recommend to consider the two or three
docking poses for further analysis (one of them is most likely
correct). As mentioned earlier, if the substrate is docked to the
TMF_170_fr_3 structure and places a charged nitrogen close
to the heme iron atom, this docking pose deserves attention,
even if such poses are not observed for the other two protein
structures.
Conclusions
Structural refinements were performed on the crystal structure
of cytochrome P450 2D6 (CYP2D6). The active site in this apo
structure is too small to be used directly for predictions of the
site of metabolism (SOM) in substrates (only 20% reliability),
so we have adopted the active site to five representative
substrates. Ten structural ensembles of 250 protein structures
each were generated using molecular dynamics (MD) simulations
of CYP2D6 in complex with each of the five representative
substrates, in which two different conformations of the side
chain of Phe483 were considered (χ1
483
∼70° and χ1
483
∼170°).
Docking experiments were performed for 65 substrates into all
2500 protein structures. Statistical analysis of the obtained
docking poses revealed that even though thermal motion
generally involves only small conformational changes, these may
have a dramatic effect on the resulting docking poses. This
strongly warns against performing docking experiments on any
single (MD) protein structure and rather suggests to consider
averages over structural ensembles or carefully selected protein
structures only.
The effect of the side-chain conformation of Phe483 was seen
to be strongest for the cluster of substrates represented by MMC.
We suggest that Phe483 destabilizes substrate orientations far
away from the heme iron, in accordance with site-directed
mutagenesis data. The ensembles obtained from MD of CYP2D6
in complex with PPD and CHZ display the highest number of
substrates with their SOM placed within 6 Å from the heme
iron atom in the first ranked pose. The ensembles PPD_70,
CHZ_170, and TMF_70 were determined to be the most
essential ones.
These ensembles may be used for SOM prediction, but it is
probably computationally too demanding to screen large databases
for putative inhibitors or ligands by docking into complete
ensembles. Therefore, we have selected a single CYP2D6
structure (PPD_70_fr_216) which offers a SOM prediction
reliability of 71%. This reliability can be theoretically further
increased up to 90% if part of the substrates are docked into
the CHZ_170_fr_79 and TMF_70_fr_3 structures. A simple and
robust decision tree was developed increasing the SOM prediction
reliability to 80% without any additional computational
costs. This offers a very powerful tool to perform SOM predictions
efficiently, where every compound needs to be docked
to one protein structure only.
The effects that we described here are not only valid for CYPs
but are highly relevant for a much wider range of proteins. The
large sensitivity of docking reliability not only due to larger
conformational changes but also due to thermal motion should
be kept in mind in any docking experiment, regardless of the
protein target. Also, we have suggested techniques to open the
binding site of apo structures and to select the most appropriate
structures for high-throughput docking experiments. These
findings are relevant for docking studies on any protein target.
Experimental Section
Clustering and Selection of Representative Substrates. A set
of 65 known substrates of cytochrome P450 2D6 that was
previously used in docking experiments was taken from the
literature.22
These substrates were clustered based on their expected
binding mode. We assume that a conformation of the CYP2D6
active site that accommodates one representative substrate will also
be able to accommodate the other substrates from the same cluster.
Clustering was based on the observation that most substrates contain
a positively charged nitrogen in a flexible tail, connected to a rigid
body, usually consisting of several aromatic rings. The width of
the substrates perpendicular to the flexible tail was used to cluster
the compounds into five groups, according to Table S1 in the
Supporting Information. The representative substrates for these five
clusters are given in Figure 1. Twenty substrates could not be
assigned to any of these clusters.
CYP2D6 Structure Preparation. The crystal structure of
CYP2D611
was downloaded from the Protein Data Bank (www.
pdb.org; code 2F9Q). We selected chain A of this apo structure, in
which some atoms of a few side chains were missing. These were
modeled using the program MOE.23
Three mutations required for
the crystallization were reverted to the wild-type amino acids
(Asp230Leu, Arg231Leu, and Met374Val). Coordinates for the
missing loop between amino acids 42 and 51 were taken from an
earlier homology model.10
Atoms for which new coordinates were
assigned were minimized and equilibrated for 10 ps of molecular
dynamics (MD) simulation at 300 K, while position restraints were
assigned to the rest of the protein structure.
Molecular Dynamics Simulations. Molecular dynamics simulations
of CYP2D6 in complex with the five representative substrates
(EDR, MMC, PPD, CHZ, TMF) were performed using the
GROMOS05 simulation package24
in combination with the GROMOS
45A4 force field.25,26
For every substrate two simulations of
1 ns were performed in which the χ1 angle of the Phe483 was
restrained to either 70° or 170° using a force constant of 30.0 kJ
mol-1
deg-2
. All bonds were constrained, using the SHAKE
Plasticity and Flexibility in Cytochrome P450 2D6 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 23 7475
algorithm27
with a relative geometric accuracy of 10-4
, allowing
for a time step of 2 fs used in the leapfrog integration scheme.28
The system was solvated in 20292 explicit SPC water molecules29
and 7 Na+
ions in a periodic rectangular simulation box. After a
steepest descent minimization to remove clashes between molecules,
initial velocities were randomly assigned from a Maxwell-Boltzmann
distribution at 300 K, according to the atomic masses. The
temperature was kept constant using weak coupling to a bath of
300 K with a relaxation time of 0.1 ps.30
The CYP2D6-substrate
complex and the solvent (including the ions) were independently
coupled to the heat bath. The pressure was controlled using isotropic
weak coupling to atmospheric pressure with a relaxation time of
0.5 ps.30
Van der Waals and electrostatic interactions were
calculated using a triple range cutoff scheme. Interactions within a
short-range cutoff of 0.8 nm were calculated every time step from
a pair list that was generated every five steps. At these time points,
interactions between 0.8 and 1.4 nm were also calculated and kept
constant between updates. A reaction field contribution was added
to the electrostatic interactions and forces to account for a homogeneous
medium outside the long-range cutoff, using the relative
permittivity 61 of SPC water.31
Docking. From the last 500 ps of each of the 10 MD simulations
described above, 250 structures were stored to disk. The GROMOS
45A4 force field is a united atom force field, meaning that aliphatic
hydrogen atoms are treated implicitly. Coordinates for the implicit
hydrogen atoms were added, all waters, counterions, and the
representative substrates were removed, and the resulting files were
converted into mol2 file format using Sybyl (version 6.8) and
the standard Tripos atom and bond types for the amino acids and
the heme group.
Docking was performed by GOLD (Genetic Optimization for
Ligand Docking)32
version 3.3.1 in combination with the Chemscore
scoring function33
parametrized for heme-containing proteins.34
This
scoring function outperformed the Goldscore scoring function in
our preliminary work (data not shown). Ten docking runs with
maximally 1000 genetic algorithm operations were performed using
a population of 100 genes. The center point for the docking was
placed in the middle of the cavity in between residues Phe120 and
Phe483. The radius from this point was set to 18 Å to include the
solvent channel in the accessible volume for the docking. We stress
that a shorter radius will automatically result in predicted binding
modes that are closer to the heme and thus positively bias the SOM
predictions. No water molecules, or a sixth ligand on the heme
iron atom, were taken into account.
Docking Reproducibility. Because the genetic algorithm used
for docking utilizes a considerable amount of randomness, results
from individual docking simulations (consisting of 10 runs) are not
reproducible. For about 20% of substrates, the evaluation of the
first ranked docking pose (see below) changed when two independent
docking experiments were performed. Therefore, we have
performed five independent docking experiments (10 docking runs
each) for all 65 substrates in all 2500 protein structures. For every
experiment 10 docking poses were obtained, and the SOM
prediction was based on the first ranked pose. The final SOM
prediction is subsequently based on the binding mode with the
highest score over all 50 binding poses. Furthermore, we assigned
a weight (sixth column in Table 1) for the prediction, based on the
consensus with the prediction according to the other four first-ranked
poses. Note that this approach is not the same as performing a single
experiment with 50 docking runs, because the docking program
explicitly searches for new solutions that differ from previous runs.
Analysis. Any docking pose of a substrate in a protein structure
is considered to be correct if the known SOM is within 6 Å of the
heme iron atom.22
This relatively wide criterion accounts for thermal
fluctuations and dioxygen binding. In order to use this criterion in
predictions for new substrates, it is important that one only needs
to consider the first ranked pose, which should correspond to the
binding pose with the highest affinity. Unfortunately, scoring
functions are often not accurate enough to correctly predict the first
ranked pose. Also, the highest affinity binding pose does not
necessarily have to be the catalytically active pose. For these
reasons, we have performed the distance analysis not only for the
first ranked pose, but we also present the shortest distance between
the heme iron atom and the substrate’s SOM when considering the
five highest ranked poses or all 10 generated poses in a docking
experiment (see Table 1).
Ligand Probing of the Binding Sites (LPBS). We want to
emphasize that the effects of conformational changes of the protein
will depend on both the exact region in the protein that is modified
and the substrate under consideration. Therefore, a characterization
of the conformational changes in terms of an atom-positional rootmean-square
deviation (rmsd) after superposition will not reflect
the chances in the docking results. Rather, we calculate the rootmean-square
distance between the structural features of a set of
substrates when docked to two protein structures. To describe the
structural differences between two protein structures, we assign to
each of the 65 substrates a value of 1 if the SOM lies within 6 Å
of the heme iron atom and a value of 0 otherwise. The ligands
probed binding site (LPBS) distance between the two protein
structures is now expressed as the rmsd over the 65 assignments
in both proteins. Similarly, the LPBS distance between two
ensembles is calculated using the percentage of structures in which
the substrates are successfully docked. The LPBS distance truly
reflects the conformational changes of the protein that make a
difference for this set of substrates. Of course, any structural feature
may be used to calculate such a LPBS distance.
Decision-Tree Classification. The quality of the predictions of
the SOM of substrates varies strongly between the different
ensembles of protein structures and between the individual protein
structures involved. In order to be able to predict the SOM for new
putative substrates, it is highly advantageous if the ensemble or
protein structure that most likely leads to the correct prediction
could be determined a priori. For the 65 substrates under
consideration here, the major SOM is known, and we can easily
determine which ensemble or individual protein structure is most
successful in predicting its SOM. We have used this information
to train a decision tree using QSAR classification with 2-fold crossvalidation
as implemented in MOE. Our database of 65 substrates
was randomly divided into two mutually exclusive subsets of size
33 (learning set) and 32 substrates (test set). As initial set of
descriptors, we have considered the number of atoms, hydrophobic
atoms, hydrogen atoms, halogens, rings, hydrogen bond acceptors,
and hydrogen bond donors, as well as SlogP, molecular volume,
total hydrophobic surface area, water-accessible surface area,
weight, water-accessible surface area divided by the weight, and
the first, second, and third kappa shape indexes (Kier 1, Kier 2,
Kier 3).35
Several decision trees were generated at different
reliability levels and complexity. To avoid overfitting, only classification
trees of depth 1 or 2 were considered. In the end the
most simple, robust decision tree was selected (Figure 6), which
was valid for both the selection of the best performing ensemble
of protein structures (from a selection of the three essential
ensembles) and for the selection of the best performing protein
structures (from a selection of the three essential structures).
Acknowledgment. We gratefully acknowledge financial
support from The Netherlands Organization for Scientific
Research, Horizon Breakthrough grant 935.18.018 (JH) and
VENI Grant 700.55.401 (CO).
Supporting Information Available: Figures of 65 substrates,
with indications of their SOM, clustered into six groups; tables
containing the docking results averaged over the 10 protein structure
ensembles; substrates reclustered according to SOM predictions into
ensembles and into single protein structures; and force field
parameters for the representative substrates. This material is
available free of charge via the Internet at http://pubs.acs.org.
References
(1) Ortiz de Montellano, P. R. Cytochrome P450: Structure, Mechanism,
and Biochemistry, 3rd ed.; Kluwer Academic/Plenum Publishers: New
York, 2005.
7476 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 23 Hritz et al.
(2) Williams, J. A.; Hyland, R.; Jones, B. C.; Smith, D. A.; Hurst, S.;
Goosen, T. C.; Peterkin, V.; Koup, J. R.; Ball, S. E. Drug-drug
interactions for UDP-glucuronosyltransferase substrates: A pharmacokinetic
explanation for typically observed low exposure (AUC(i)/
AUC) ratios. Drug Metab. Dispos. 2004, 32, 1201–1208.
(3) Bazeley, P. S.; Prithivi, S.; Struble, C. A.; Povinelli, R. J.; Sem, D. S.
Synergistic use of compound properties and docking scores in neural
network modeling of CYP2D6 binding: Predicting affinity and
conformational sampling. J. Chem. Inf. Model. 2006, 46, 2698–2708.
(4) Rendic, S. Summary of information on human CYP enzymes: Human
P450 metabolism data. Drug Metab. ReV. 2002, 34, 83–448.
(5) Pirmohamed, M.; Park, B. K. Cytochrome P450 enzyme polymorphisms
and adverse drug reactions. Toxicology 2003, 192, 23–32.
(6) Ingelman-Sundberg, M. Pharmacogenetics of cytochrome P450 and
its applications in drug therapy: The past, present and future. Trends
Pharmacol. Sci. 2004, 25, 193–200.
(7) de Groot, M. J.; Kirton, S. B.; Sutcliffe, M. J. In silico methods for
predicting ligand binding determinants of cytochromes P450. Curr.
Top. Med. Chem. 2004, 4, 1803–1824.
(8) De Graaf, C.; Vermeulen, N. P. E.; Feenstra, K. A. Cytochrome P450
in silico: An integrative modeling approach. J. Med. Chem. 2005, 48,
2725–2755.
(9) Stjernschantz, E.; Vermeulen, N. P. E.; Oostenbrink, C. Computational
prediction of drug binding and rationalisation of selectivity towards
cytochromes P450. Expert Opin. Drug Metabol. Toxicol. 2008, 4, 513–
527.
(10) de Graaf, C.; Oostenbrink, C.; Keizers, P. H. J.; van Vugt-Lussenburg,
B. M. A.; van Waterschoot, R. A. B.; Tschirret-Guth, R. A.;
Commandeur, J. N. M.; Vermeulen, N. P. E. Molecular modelingguided
site-directed mutagenesis of cytochrome P450 2D6. Curr. Drug
Metab. 2007, 8, 59–77.
(11) Rowland, P.; Blaney, F. E.; Smyth, M. G.; Jones, J. J.; Leydon, V. R.;
Oxbrow, A. K.; Lewis, C. J.; Tennant, M. G.; Modi, S.; Eggleston,
D. S.; Chenery, R. J.; Bridges, A. M. Crystal structure of human
cytochrome P450 2D6. J. Biol. Chem. 2006, 281, 7614–7622.
(12) Guengerich, F. P. A malleable catalyst dominates the metabolism of
drugs. Proc. Natl. Acad. Sci. U.S.A. 2006, 103, 13565–13566.
(13) Ekroos, M.; Sjogren, T. Structural basis for ligand promiscuity in
cytochrome P450 3A4. Proc. Natl. Acad. Sci. U.S.A. 2006, 103, 13682–
13687.
(14) Pang, Y. P.; Kozikowski, A. P. Prediction of the binding-sites of
huperzine-A in acetylcholinesterase by docking studies. J. Comput.Aided
Mol. Des. 1994, 8, 669–681.
(15) Ma, B. Y.; Shatsky, M.; Wolfson, H. J.; Nussinov, R. Multiple diverse
ligands binding at a single protein site: A matter of pre-existing
populations. Protein Sci. 2002, 11, 184–197.
(16) Carlson, H. A. Protein flexibility and drug discovery: How to hit a
moving target. Curr. Opin. Chem. Biol. 2002, 6, 447–452.
(17) Cruciani, G.; Carosati, E.; De Boeck, B.; Ethirajulu, K.; Mackie, C.;
Howe, T.; Vianello, R. MetaSite: Understanding metabolism in human
cytochromes from the perspective of the chemist. J. Med. Chem. 2005,
48, 6970–6979.
(18) Afzelius, L.; Arnby, C. H.; Broo, A.; Carlsson, L.; Isaksson, C.; Jurva,
U.; Kjellander, B.; Kolmodin, K.; Nilsson, K.; Raubacher, F.; Weidolf,
L. State-of-the-art tools for computational site of metabolism predictions:
Comparative analysis, mechanistical insights, and future applications.
Drug Metab. ReV. 2007, 39, 61–86.
(19) Ito, Y.; Kondo, H.; Goldfarb, P. S.; Lewis, D. F. V. Analysis of
CYP2D6 substrate interactions by computational methods. J. Mol.
Graphics Model. 2008, 26, 947–956.
(20) Saraceno, M.; Coi, A.; Bianucci, A. M. Molecular modelling of
human CYP2D6 and molecular docking of a series of ajmalicineand
quinidine-like inhibitors. Int. J. Biol. Macromol. 2008, 42, 362–
371.
(21) Lussenburg, B. M. A.; Keizers, P. H. J.; De Graaf, C.; Hidestrand,
M.; Ingelman-Sundberg, M.; Vermeulen, N. P. E.; Commandeur,
J. N. M. The role of phenylalanine 483 in cytochrome P450 2D6 is
strongly substrate dependent. Biochem. Pharmacol. 2005, 70, 1253–
1261.
(22) De Graaf, C.; Oostenbrink, C.; Keizers, P. H. J.; Van der Wijst, T.;
Jongejan, A.; Vermeulen, N. P. E. Catalytic site prediction and virtual
screening accuracy of cytochrome P450 2D6 substrates by consideration
of water and rescoring in automated docking. J. Med. Chem.
2006, 49, 2417–2430.
(23) Molecular Operating Environment, v., Chemical Computing Group,
Montreal, Canada.
(24) Christen, M.; Hunenberger, P. H.; Bakowies, D.; Baron, R.; Burgi,
R.; Geerke, D. P.; Heinz, T. N.; Kastenholz, M. A.; Krautler, V.;
Oostenbrink, C.; Peter, C.; Trzesniak, D.; Van Gunsteren, W. F. The
GROMOS software for biomolecular simulation: GROMOS05. J. Comput.
Chem. 2005, 26, 1719–1751.
(25) van Gunsteren, W. F.; Billeter, S. R.; Eising, A. A.; Hu¨nenberger,
P. H.; Kru¨ger, P.; Mark, A. E.; Scott, W. R. P.; Tironi, I. G.
Biomolecular simulation: The GROMOS96 manual and user guide,
Vdf Hochschulverlag AG an der ETH Zu¨rich: Zu¨rich, 1996.
(26) Schuler, L. D.; Daura, X.; van Gunsteren, W. F. An improved
GROMOS96 force field for aliphatic hydrocarbons in the condensed
phase. J. Comput. Chem. 2001, 22, 1205–1218.
(27) Ryckaert, J.-P.; Ciccotti, G.; Berendsen, H. J. C. Numerical integration
of cartesian equations of motion of a system with constraints:
Molecular dynamics of n-alkanes. J. Comput. Phys. 1977, 23, 327–
341.
(28) Hockney, R. W. The potential calculations and some applications.
Methods Comput. Phys. 1970, 9, 136–211.
(29) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; Hermans,
J. Interaction models for water in relation to protein hydration. In
Intermolecular Forces, Pullman, B., Ed.; Reidel: Dordrecht, The
Netherlands, 1981; pp 331-342.
(30) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; DiNola,
A.; Haak, J. R. Molecular-dynamics with coupling to an external bath.
J. Chem. Phys. 1984, 81, 3684–3690.
(31) Tironi, I. G.; Sperb, R.; Smith, P. E.; van Gunsteren, W. F. A
generalized reaction field method for molecular-dynamics simulations.
J. Chem. Phys. 1995, 102, 5451–5459.
(32) Jones, G.; Willett, P.; Glen, R. C.; Leach, A. R.; Taylor, R.
Development and validation of a genetic algorithm for flexible docking.
J. Mol. Biol. 1997, 267, 727–748.
(33) Eldridge, M. D.; Murray, C. W.; Auton, T. R.; Paolini, G. V.; Mee,
R. P. Empirical scoring functions 0.1. The development of a fast
empirical scoring function to estimate the binding affinity of ligands
in receptor complexes. J. Comput.-Aided Mol. Des. 1997, 11, 425–
445.
(34) Kirton, S. B.; Murray, C. W.; Verdonk, M. L.; Taylor, R. D. Prediction
of binding modes for ligands in the cytochromes P450 and other hemecontaining
proteins. Proteins: Struct., Funct., Bioinf. 2005, 58, 836–
844.
(35) Hall, L. H.; Kier, L. B. The molecular connectivity chi indices and
kappa shape indices in structure-property modeling. In ReViews in
Computational Chemistry, 2nd ed.; Lipkowitz, K. B.; Boyd, D. B.,
Eds.; VCH Publishers: New York, 1991; p 367.
JM801005M
Plasticity and Flexibility in Cytochrome P450 2D6 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 23 7477
EFFICIENT INCORPORATION OF PLASTICITY AND WATER MOLECULES INTO THE MOLECULAR
DOCKING APPROACH
35
Paper 2
Santos, R.; Hritz, J.; Oostenbrink, C. The role of water in molecular docking simulations of Cytochrome
P450 2D6. J. Chem. Inf. Model. 2010, 50, 146-154
Role of Water in Molecular Docking Simulations of Cytochrome P450 2D6
Rita Santos, Jozef Hritz, and Chris Oostenbrink*
Leiden-Amsterdam Center for Drug Research, Section of Molecular Toxicology, Department of Chemistry
and Pharmaceutical Sciences, Vrije Universiteit, De Boelelaan 1083, 1081 HV Amsterdam, The Netherlands
Received August 7, 2009
Active-site water molecules form an important component in biological systems, facilitating promiscuous
binding or an increase in specificity and affinity. Taking water molecules into account in computational
approaches to drug design or site-of-metabolism predictions is currently far from straightforward. In this
study, the effects of including water molecules in molecular docking simulations of the important metabolic
enzyme cytochrome P450 2D6 are investigated. The structure and dynamics of water molecules that are
present in the active site simultaneously with a selected substrate are described, and based on this description,
water molecules are selected to be included in docking experiments into multiple protein conformations.
Apart from the parent substrate, 11 similar and 53 dissimilar substrates are included to investigate the
transferability of active-site hydration sites between substrates. The role of water molecules appears to be
highly dependent on the protein conformation and the substrate.
INTRODUCTION
Water is a highly versatile component at the interface of
biomolecular complexes because of its unique physical
chemical properties: it can act both as a hydrogen-bond donor
and as a hydrogen-bond acceptor; it imposes few steric
constraints on bond formation; and it is able to form
hydrogen-bond networks, occupying less space than the polar
side chains of a protein. Therefore, water can offer a high
level of adaptability to a surface, allowing promiscuous
binding, or at the same time, it can provide increased
specificity and affinity to an interaction.1
The role of water
molecules can be as a solvent, involved in the stabilization
of a biomolecular complex; as a bridge between two
molecules; as a reagent, being a substrate or product of many
enzymatically catalyzed reactions; as a lubricant, through the
formation of a network linking distant residues or by
promoting the flexibility required for the events of the
catalytic cycle;2
and as building block of macromolecules,
being part of the receptor binding site and, thereby, altering
the topological surface and recognition determinants.3
Because of the importance of water molecules in biological
systems, discussion has arisen in the literature as to whether
these molecules should or should not be included in
computational approaches for structure-based drug design,
mainly in molecular docking simulations.4-10
However, no
consensus has been reached about the role of water molecules
in such simulations. For instance, Huang et al. reported that
the docking accuracy of 12 targets from the DUD database
was improved by including water molecules,11
whereas Birch
et al. described the opposite in a study of neuraminidase
structures.12
Cytochromes P450 are a family of heme-containing
oxygenases and are considered to be highly important in
phase I biotransformation.13,14
There are several human
subfamilies of cytochromes P450, mainly involved in the
synthesis of critical signaling molecules for homeostasis of
the organism and in the metabolism of endogenous (fatty
acids, steroids, prostaglandins) and exogenous (natural
products contained in plants, drugs, environmental pollutants)
compounds. By rendering the compounds more watersoluble,
their excretion is facilitated.13
Human cytochrome P450 2D6 (CYP2D6) is considered
to be an important drug-metabolizing enzyme, despite the
fact that it corresponds to only approximately 2% of the
cytochrome P450 liver content. CYP2D6 is responsible for
the metabolism of approximately 15-20% of the current
drugs on the market, such as beta blockers, neuroleptics,
antidepressants, and others.15
Genetic polymorphisms of
CYP2D6 are known. For instance, 7% of the Caucasian
population does not possess this functional enzyme, resulting
in the defective metabolism of many important drug mol-
ecules.16
Therefore, the assessment of the metabolism of
drugs under development is desirable as early as possible to
prevent adverse drug effects. Metabolism prediction in silico
can speed up the identification of compounds that might be
relevant for further investigation. We distinguish methods
that are based on the molecular structure and reactivity of
substrates solely from methods that take the protein structure
explicitly into account.17
The commercial program MetaSite
combines both approaches.18
As an alternative, molecular
docking simulations are most commonly used to predict the
orientation of substrates in the enzyme active site.
The apo crystal structure of human cytochrome P450 2D6
was recently resolved by Rowland et al.,19
and so far, this
is the only available crystal structure of CYP2D6. Eleven
water molecules can be found in the crystal structure of
CYP2D6; however, they are positioned more than 1.0 nm
away from the heme iron. A previous study using a
homology model of CYP2D6 indicated that strategically
placed static water molecules in the active site can significantly
influence the binding pose of 65 substrates.20
On the
* Corresponding author phone: +31 20 5987606; fax: +31 20 5987610;
e-mail c.oostenbrink@few.vu.nl.
J. Chem. Inf. Model. 2010, 50, 146–154146
10.1021/ci900293e CCC: $40.75  2010 American Chemical Society
Published on Web 11/09/2009
DownloadedviaMASARYKUNIVonJanuary30,2020at09:48:33(UTC).
Seehttps://pubs.acs.org/sharingguidelinesforoptionsonhowtolegitimatelysharepublishedarticles.
other hand, we showed recently that active-site flexibility
and plasticity could account for similar effects.21
It is unclear
what the role of active-site water molecules is in the
orientation of the substrates in the active site. Should water
molecules be included explicitly in molecular docking
simulations, or is their effect negative? Is this role substratedependent,
and can the selection of water molecules in the
active site be automated?
In this study, we tried to address some of these questions
by analyzing the behavior of water molecules in the active
site of cytochrome P450 2D6 in a complex with R-3,4methylenedioxy-N-ethylamphetamine
(MDEA) using molecular
dynamics (MD) simulations. The dynamic properties
of these water molecules were determined for a few selected
protein conformations, and the water molecules that seemed
to influence the position of the ligand most were identified.
It was our aim to identify the role of these waters and their
influence on the reliability of predictions of the site of
metabolism (SOM) by molecular docking simulations for
MDEA, along with sets of similar and dissimilar compounds.
MDEA was chosen as a representative because, in our
previous work, we found that MD-generated structures with
this substrate can accommodate many substrates while still
leaving room for improvement of the SOM predictions.21
MATERIALS AND METHODS
All docking experiments were performed using protein
structures that were extracted from a molecular dynamics
(MD) simulations of cytochrome P450 2D6 in complex with
the substrate (R)-3,4-methylenedioxy-N-ethylamphetamine
(MDEA). Initial coordinates of the protein were obtained
from the protein databank (www.pdb.org) entry code 2F9Q.19
For details concerning the preparation of the complex
structure and simulation settings, we refer to our earlier
publication.21
In short, the protein was solvated in 20292
SPC water molecules22
and 7 Na+
ions. Using the GROMOS
simulation package,23
a 10-ns simulation was performed at
a constant temperature of 300 K and a pressure of 1 atm.
The temperature and pressure were kept constant by weak
coupling, using relaxation times of 0.1 and 0.5 ps, respec-
tively.24
The isothermal compressibility was set to 4.575 ×
10-4
(kJ mol-1
nm-3
)-1
. All bond lengths were constraint
to their optimal values using the SHAKE algorithm25
with
a relative geometric accuracy of 10-4
. All interactions were
calculated according to the GROMOS force field, parameter
set 45A4.26,27
Nonbonded interactions were calculated using
a triple-range cutoff scheme. At every time step (2 fs),
nonbonded interactions at distances shorter than 0.8 nm were
calculated using a pair list that was constructed every five
steps. Upon pair-list construction, interactions at distances
up to 1.4 nm were also calculated and kept constant between
pair-list updates. A reaction-field contribution28
was added
to the electrostatic interactions and forces to account for a
homogeneous medium outside the cutoff sphere, using a
relative dielectric permittivity of 61.29
No additional constraints
were added to the simulation.
Coordinates were stored every 10 ps. Starting at 2 ns, eight
structures (A-H) were selected every 1 ns to comprise the
test set for docking experiments. For each of these structures,
the water behavior was analyzed by considering the 500 ps
before and the 500 ps directly following the snapshot. All
protein structures were superposed based on the backbone
atoms prior to analysis. The distribution of water molecules
in the active site was obtained by placing the protein on a
regular grid with a spacing of 0.05 nm and monitoring the
number of water molecules closest to the grid points. For
every static protein conformation, the water molecules
occupying clusters of grid points with a probability at least
30 times larger than the probability of finding a water
molecule at a similar grid point in bulk water (0.4% for SPC
at a density of 970 g/L) were selected initially. Even though
the affinity of the water molecules was not calculated
explicitly,30,31
the high probability of occurrence at specific
sites can be considered a thermodynamic measure. Hydrogen
bonds were analyzed for these water molecules using a
geometric criterion. A hydrogen bond was defined as having
a minimum donor-hydrogen-acceptor angle of 135° and a
maximum hydrogen-acceptor distance of 0.25 nm. The
diffusion of water molecules relative to the protein structure,
D, was calculated from the mean-square displacement using
the Einstein equation
where r0 and r(t) are the water positions in a reference
configuration and at time t, respectively. Nd is the number
of dimensions that are being considered, which is 3 here.
From the protein structures A-H, the substrate, ions, and
water molecules that were not selected were removed.
Coordinates for hydrogen atoms that were implicitly treated
in the simulations were added, and the resulting files were
converted into mol2 file format using the standard Tripos
atom and bond types for the amino acids and the heme group.
Docking was performed using GOLD (Genetic Optimization
for Ligand Docking),32
version 3.3.1, in combination with
the Chemscore scoring function33
parametrized for hemecontaining
proteins.34
The center point for docking was
placed in the middle of the cavity between residues Phe120
and Phe483. The radius from this point was set to 1.8 nm to
include the solvent channel in the accessible volume for the
docking. At most, 100000 operations in the genetic algorithm
were performed using a population of 100 genes. At least
five independent docking simulations were performed to
reach statistical significance. At least 50 poses were stored
from every docking simulation, except if the best three
docking poses had root-mean-square deviations smaller than
0.15 nm. To prevent an early convergence, a relative pressure
of 1.1 was specified, and to account for diversity, the number
of niches was set to 2. Ligand flexibility was specified as
follows: The flipping of the free corners of ligand rings, the
flipping of amide bonds, and the flipping of planar nitrogens
were allowed. Intramolecular hydrogen bonds were also
allowed. GOLD offers the possibility of automatically
determining whether a specific water molecule should be
bound or displaced by turning its interactions ON or OFF
during the docking simulation.35
In addition, the water
molecules can be allowed to spin around their three principle
axes. Because this increases the number of degrees of
freedom, GOLD developers advise that it be done for at most
three water molecules at a time (private communication).
For structures in which more than three waters were selected
based on local densities, preliminary docking experiments
D ) lim
tf∞
|r0 - r(t)|2
2Ndt
(1)
ROLE OF WATER IN CYP2D6 J. Chem. Inf. Model., Vol. 50, No. 1, 2010 147
were performed using MDEA, to determine the three water
molecules for which the possibility of toggling between ON
and OFF was most crucial for finding the correct binding
pose. A water molecule was, therefore, considered to be
critical if it increased the reliability of the docking prediction
when determined to be ON and decreased the reliability
prediction when determined to be OFF. In subsequent
docking simulations, the effects of these water molecules
on the binding poses for 11 substrates that are similar to
MDEA and for 53 substrates that are dissimilar to MDEA
were explored.21
For each of these substrates, the site of
metabolism (SOM) was determined experimentally; for
complete references, see ref 20. As before, we considered a
docking pose to be incorrect if the substrate site of
metabolism was farther from the heme iron atom than 0.6
nm.20,21
The final SOM prediction was subsequently based
on the binding mode with the highest score among the five
independent docking simulations.
RESULTS
Identification of Hydration Sites. Analysis of the trajectory
of CYP2D6 with MDEA (Figure 1a) revealed hydration
sites in the active site of CYP2D6 close to the substrate
(Figure 1b). To discriminate between genuine hydration sites
and false-positive hydration sites, the regions identified for
protein conformation A were analyzed by visual inspection
with VMD.36
A genuine hydration site was defined as a
position where a water molecule would remain for at least
80% of the time, which corresponds to an occurrence at least
30 times larger than that for a similar site in bulk water.
Water molecules that were within 0.15 nm of a grid point
belonging to a hydration site were initially selected for each
protein conformation. In this way, dynamic changes in the
shape and size of the hydration site were taken into
consideration. This increase in the hydration site region also
accounts for the dynamics of the water molecules, as even
static water molecules still move considerably around a
certain position (0.05-0.1 nm).
Twelve hydration sites were identified in protein conformation
A (Figure 1c and Table 1). Between 10 and 17
hydration sites with dynamics similar to those described for
protein conformation A could be identified in the other
protein conformations.
Selection of Water Molecules for Docking. Despite our
attempt to select only the most static water molecules from
the molecular dynamics simulations of CYP2D6, the number
of water molecules selected was still too high to reliably
include in the docking experiments. To reduce the number
Figure 1. (A) Snapshot (denoted conformation A) of the molecular dynamics simulations of the complex of cytochrome P450 2D6 with
MDEA. (B) Identification of hydration sites in the cavity of cytochrome P450 2D6 by determining the probability of occurrence over a
grid. (C) Selection of the most static hydration sites in protein conformation A. Selected residues in the binding site are highlighted, and
the hydration sites are numbered in accordance with Table 1.
148 J. Chem. Inf. Model., Vol. 50, No. 1, 2010 SANTOS ET AL.
of waters to be use for docking, preliminary docking studies
were performed to identify the 3 water molecules that most
influenced in a positive way the docking poses of MDEA.
Therefore, water molecules that have a tendency to be ON
when the SOM is further than 0.6 nm from the iron and at
the same time have a tendency to be OFF when the SOM is
within 0.6 nm from the iron, were removed from the
complex, and in the end, only the water molecules responsible
for improved docking poses remained. For some protein
conformations, it was not clear which water molecules to
remove from the initial selection, in these cases a visual
inspection of the preliminary docking poses was carried out
to identify the water molecule(s) responsible for a pose where
the SOM was far from the iron. The improvement observed
in the preliminary docking results for protein conformation
A, could not be achieved by using the water molecules from
the 3 hydration sites with the highest occupancy (data not
shown).
The selected waters for protein conformation A were
HOH470, HOH472, HOH476 belonging to cluster 5, 7, and
11 respectively (Figure 1c and Table 1).
Dynamic Properties of Selected Water Molecules. On
the 1-ns time scale, the hydration sites identified are quite
static positions, as can be seen for protein conformation A
in Table 1. Between different protein conformations, the
identified hydration sites could differ though. The spot
density ratios are at least 10 times larger than for bulk water.
The same trend can be observed for the diffusion constants
of the water molecules within the hydration sites. A more
thorough analysis of bulk water simulations revealed that
no hydration site could be identified with a density greater
than 640 molecules per cubic nanometer, and the hydration
sites in the active site of CYP2D6 were found to be even
more pronounced than that (minimum of 960 molecules/nm3
).
However, only a few hydration sites revealed a high
probability to form hydrogen bonds between the water
molecules contributing to the hydration sites and the protein,
and even fewer for the ligand. This indicates the possibility
of a hydrogen-bond network between water molecules, or
the existence of trapped water molecules in the cavity of
CYP2D6 with a lubricant role. Because of the low resolution
of the sampling times in MD, the average residence time
can not be calculated with precision, but it was observed
that the average residence time was above 10 ps only in some
cases, being below 10 ps in the majority of the cases. For
70% of the hydration sites, it was found that more than one
water molecule contributed to it over 1 ns. On average, three
water molecules contributed to the same hydration site, with
a maximum of five water molecules per site. Even though
the density of water molecules at the hydration sites was
quite high, the water molecules themselves appeared to be
quite mobile.
ON vs Toggle. Preliminary docking studies with MDEA
and protein conformation A were done in order to determine
the best treatment for water molecules during the simulations.
The results are displayed in Table 2 and indicate that
allowing the constant presence of water molecules during
the docking runs (HOH ON) does not alter the docking
results for MDEA, but it decreases the performance for
MDEA-like and non-MDEA-like compounds when compared
to the possibility of displacing these water molecules
during the run (HOH toggle). Therefore, in this project, we
allowed GOLD to determine during the docking run whether
the water molecules would be displaced by the ligand.
Table 1. Parameters That Reflect the Dynamics of the Water Molecules Identified in Protein Conformation A
hydration site NHS
a
(%) VHS
b
(10-4
nm3
) rHS
c
(nm) HBpro
d
(%) HBlig
e
(%) HBHOH
f
(%) fp
g
HOHh
Di
(10-3
nm2
ps-1
)
1 14 3.75 0.054 100 0 0 11.7 466 0.094
2 18 5.00 0.060 5.6 5.6 100 11.3 467 0.148
3 26 7.50 0.068 0 96.2 3.8 10.8 468 0.238
4 27 7.50 0.068 81.5 0 50 11.3 469 0.098
5 22 6.25 0.064 0 0 100 11.0 470j
0.224
6 39 10.00 0.075 100 0 0 12.2 471 0.064
7 32 8.75 0.072 0 0 90.6 11.4 472j
0.154
8 18 3.75 0.054 0 16.7 100 15.0 473 0.048
9 16 5.00 0.060 6.3 0 100 10.0 474 0.082
10 16 3.75 0.054 81.3 0 56.3 13.3 475 1.37
11 35 8.75 0.072 91.4 0 88.6 12.5 476j
0.156
12 15 3.75 0.054 0 0 100 12.5 477 0.338
bulk water 0.4 1.25 - - - - 1.00 - 2.3k
a
Hydration site occupancy (NHS), defined as the sum of all occupancies of the grid points belonging to the hydration site. b
Hydration site
volume (VHS), defined as the volume of one grid point (1.25 × 10-4
nm3
) multiplied by the number of grid points contributing to the hydration
site. c
Hydration site radius (rHS), defined as the radius of a sphere with the same volume as the hydration site. d
Occurrence of hydrogen bonds
between the water molecules in the cluster and the protein. e
Occurrence of hydrogen bonds between the water molecules in the cluster and the
ligand. f
Occurrence of water-water hydrogen bonds. g
Spot density ratio, fp ) FHS/FB, where FHS is the spot density for the hydration site, FHS
) NHS/VHS, and FB is the bulk water density in molecules per nm3
(32 nm-3
, calculated from the density of SPC water, 970 g/L). h
Sequence
numbers of the water molecule occupying the site in protein conformation A. i
Relative diffusion constant (D) determined using eq 1. j
Selected
water molecules to be used for molecular docking simulations. k
Value taken from ref 41.
Table 2. Comparison of Molecular Docking Results for MDEA, 11
MDEA-like Compounds, and 53 Non-MDEA-like Compounds in
Protein Conformation A, When the Water Molecules Were
Specified as OFF, Toggle, or ON
compound water scenario correctly dockeda
(%)
MDEA HOH OFF 0
HOH toggle 100
HOH ON 100
MDEA-like HOH OFF 90.9
HOH toggle 90.9
HOH ON 81.8
non-MDEA-like HOH OFF 52.8
HOH toggle 56.6
HOH ON 54.7
a
Percentage of correctly docked substrates based on the
highest-ranked pose over the five independent simulations.
ROLE OF WATER IN CYP2D6 J. Chem. Inf. Model., Vol. 50, No. 1, 2010 149
Docking Results for MDEA. Docking results for MDEA
back into the active site of CYP2D6, in the (possible)
presence and absence of water molecules (Table 3), showed
that, overall, the presence of water molecules improved the
reliability of the docking prediction for some protein
conformations and did not decrease the reliability for others,
except in protein conformation G. In this conformation, the
results apparently worsen. However, this is due to a pitfall
of the 0.6 nm rule, which only monitors the distance between
the SOM and the iron and not the geometrical similarity
between docking poses. In this protein conformation, no
water molecule was switched ON by GOLD, and when
analyzing the docking poses, we observed that this pose was
just above the limit (0.6 vs 0.61 nm), but that the poses were
quite similar (Figure 2).
A distribution of the effect of water molecules on the
docking results of MDEA over all protein conformations with
different degrees of confidence can be found in Figure 3A.
A positive value on the x axis indicates an improvement of
the reliability of SOM predictions when water is included.
Larger values indicate a better reproducibility between
individual docking simulations and, thus, a larger statistical
significance. A negative x value indicates a worsening of
the results when water is included, again with more statiscal
confidence for larger values. A slightly higher frequency
toward the positive side of the graph can be seen, meaning
that the presence of water molecules might increase the
accuracy of the docking results for MDEA or at least not
alter them, given that the highest frequency is at 0.
Docking Results for MDEA-like and Non-MDEA-like
Compounds. Analysis of the docking results of similar and
dissimilar compounds in the active site of CYP2D6 (Table
3) shows that there is a greater effect for MDEA-like
compounds and almost no effect for non-MDEA-like compounds.
Further analysis reveals that there is a slight tendency
for an improvement of the docking results for MDEA-like
compounds when water molecules are included, because the
distribution is slightly shifted to the positive side of the
histogram (Figure 3B, C).
Effect of Water Molecules on the Reliability of SOM
Predictions. As can be seen from the results above, the effect
of water molecules in the docking results is ambiguous, being
highly dependent on the protein conformation and substrate.
In Figure 4, we identify four effects of water molecules:
improvement, worsening, no effect, and no improvement.
In Figure 4A, there is an improvement of the docking results
for MDEA, as water molecules occupy a position that
Table 3. Molecular Docking Simulation Results for MDEA,
MDEA-like and Non-MDEA-like Compounds Per Protein
Conformationa
MDEA MDEA-like non-MDEA-like
protein
conformation
HOH
OFF
HOH
toggle
HOH
OFF
HOH
toggle
HOH
OFF
HOH
toggle
A 0 100 90.9 90.9 52.8 56.6
B 100 100 81.8 90.9 62.3 54.7
C 0 0 36.4 72.7 45.3 50.9
D 0 0 45.5 63.6 41.5 49.1
E 0 100 90.9 63.6 62.3 58.5
F 0 0 45.5 90.9 45.3 41.5
G 100 0b
45.5 54.6 58.5 54.7
H 100 100 100 100 56.6 54.7
a
Percentage of correctly docked substrates based on the
highest-ranked pose over the five independent simulations in the
(possible) presence and absence of water molecules. b
No water was
decided to be ON. The inconsistency arises because of a threshold
problem; see text and Figure 2 for details.
Figure 2. MDEA in frame G is a good example that the 0.6-nm
rule might not be ideal for monitoring the correctness of the docking
poses. By monitoring only the distance and not the poses, it seems
that the presence of water worsens the SOM prediction (going from
2/5 correct to 0/5), whereas, in reality, the poses are very similar
but the Fe-SOM distance increases just slightly over the threshold
(0.61 nm with water vs 0.60 nm without water). Red and blue poses
are achieved without water and with water, respectively. The sphere
represents the SOM.
Figure 3. Distribution of reliability difference over all protein
conformations between docking simulations both including possible
water molecules and excluding water for (A) MDEA and (B)
MDEA-like and (C) non-MDEA-like compounds. The scale of the
x axis is as follows: A value of +5 indicates that, in five independent
simulations, the substrate always docked in the first ranked pose
with the SOM within 0.6 nm of the iron in the presence of water
but always farther from the iron in the absence of water; -5
indicates that, in five independent simulations, the substrate always
docked in the first ranked pose with the SOM far from the iron in
the presence of water but within 0.6 nm of the iron in the absence
of water; and 0 indicates no difference in the reliability of the
docking prediction, regardless of whether it was docking with the
SOM far or near the iron in the absence of water. On the y axis,
the frequency is represented.
150 J. Chem. Inf. Model., Vol. 50, No. 1, 2010 SANTOS ET AL.
prevents the ligand to dock in a subpockect between Ser217
and Phe483. HOH476 is the main water molecule responsible
for this improvement and is involved in a hydrogen bond
with the protein 91% of the time. In Figure 4B, the substrate
TRP no longer finds a catalytically active pose because water
molecules are occupying a position that prevents the ligand
from forming a hydrogen bond essential for GOLD to find
a correctly docked pose. In Figure 4C, the docking of MRP
is not affected by the (possible) presence of water molecules,
as the poses are perfectly superimposed. Finally, in Figure
4D, the best-ranked pose of AMI is altered in the presence
of water molecules, but the new poses still do not bring the
SOM closer to the iron. Similar cases were observed for other
protein conformations as well.
DISCUSSION
Water molecules play an important role in biological
systems. Studies of the water molecules in the ligand-binding
cavities of cytochromes P450 indicate that their high mobility
facilitates the movement of the substrates and products into
and out of the active site.37
In this study, we identified
hydration sites in the cavity of human CYP2D6 with an
occupation probability at least 30 times larger than that in
bulk water. However, water molecules contributing to it have
an average residence time below 10 ps, being quite mobile,
and having a low probability of forming hydrogen bonds
with the protein or ligand. Rather, they are involved in
hydrogen-bonded networks with other water molecules in
the vicinity, occupying the empty space in the cavity. The
dynamic nature of water molecules might add to the
promiscuity of CYP metabolism and to the reported malleability
of the active site.38
The position of the hydration
sites changes quite a lot from one protein conformation to
the other, supporting the idea that the water molecules are
quite mobile and considerably change the hydration shell
over the span of 1 ns. However, regions with high probability
of finding water molecules can be identified, and inclusion
of these water molecules in a docking protocol leads to an
improvement of the reliability of the SOM prediction for
the compound that was used to optimize the selection of
water molecules (MDEA) in several protein conformations
and does not alter the results for the others. For compounds
that are similar to MDEA, an effect on the pose prediction
is also observed when water molecules are included, leading
to a slight improvement in the results. There is virtually no
effect when water molecules are included for compounds
that are unlike MDEA. The water molecules seem to offer
yet another way to allow CYPs to catalyze the metabolism
of a wide range of structurally diverse compounds.21,38
Figure 4. Effect of water molecules on the docking poses. Red poses and blue were achieved without water and with water, respectively.
The sphere represents the experimentally determined SOM. (A) Improvement, MDEA in frame A: water molecules are preventing the
substrate from binding in a region far above the heme group, leading to an improvement of 100%. (B) Worsening, TRP in frame A: water
molecules are forming hydrogen bonds with the substrate, which leads to incorrect poses. (C) No effect, MRP in frame A: the presence of
water molecules does not influence the poses. (D) No improvement, AMI in frame A: the presence of water molecules leads to different
poses, but not to any with the SOM within 0.6 nm of the iron.
ROLE OF WATER IN CYP2D6 J. Chem. Inf. Model., Vol. 50, No. 1, 2010 151
It is also clear from the results that the effect of including
water molecules in the docking protocol is rather dependent
on the protein conformation and substrate. For instance, in
protein conformations A and E, there is a clear improvement
of the docking results for MDEA, but for the other protein
conformations, there are no alterations in the results, except
for protein conformation G. In protein conformation G, no
water molecule was decided to be ON by GOLD, and
therefore, the inconsistency of the results arises as a result
of boundary effects of the 0.6-nm rule used to discriminate
between correct and incorrect docking poses. This approach
monitors only the SOM-Fe distance and does not account
for the geometrical dissimilarity between poses. Because the
metabolic site of the substrates is the only available
experimental indication of the binding orientation, a rootmean-square
geometrical measure cannot be used to distinguish
“correct” and “incorrect’ poses”.
There are, however, a few pitfalls in this approach, that
might influence the accuracy of the results and, consequently,
our conclusions: (i) the GOLD parametrization
to determine whether a water molecule should be displaced
or bound, (ii) the fairness of comparing simulations in
the presence and absence of water molecules using the
same maximum number of genetic algorithm operations
when the number of degrees of freedom differs, and (iii)
comparison of docking results with different numbers of
water molecules.
According to the literature, a water molecule is determined
to be ON by GOLD if the intrinsic binding affinity
of the water molecule outweighs the free energy associated
with the loss of rigid-body entropy upon binding to the
target.35
The free energy associated with the loss of
entropy upon binding to the target varies for different
water binding sites, as tightly binding water molecules
will loose more rigid-body entropy than loosely binding
water molecules. However, this term in GOLD is treated
as a constant for simplification. This constant is optimized
for a set of water molecules in 58 crystal structures such
as HIV-1, FXa, TK, and OppA, which include both highly
structured (e.g., HIV-1) and rather promiscuous hydration
sites (e.g., OppA). It might be too high to be used
accurately with relatively mobile water molecules in the
promiscuous CYP2D6, selected from molecular dynamics
simulations. A more accurate determination of the water
binding thermodynamics31
is too computationally expensive
to perform within a docking approach.
When docking in the possible presence of water
molecules, the number of degrees of freedom increases
(water ON/OFF, rotation around the principal axes), so
to ensure sufficient sampling, the maximum number of
genetic algorithm operations (maxops) needs to be increased.
This does not hold when docking without water,
because the number of degrees of freedom is smaller.
However, no reliable estimate could be made of the value
of maxops that would lead to a fair comparison because
the degrees of freedom are very ligand-dependent and the
exact relation between the number of configurations and
the number of degrees of freedom is unknown.
Finally, it has been mentioned in the literature that an
accurate comparison of the energy scores seems to rely
on an equally homogeneous environment.8
Therefore, the
unequal numbers of water molecules among the docking
poses might introduce some noise in the comparison,
because it does not mimic the full hydration shell that is
expected in aqueous solution. However, currently, no
solution has been proposed for this problem.8,39
Taking these pitfalls into account, it still seems clear
that water molecules do influence the docking orientations
of the substrates in the CYP2D6 active site. The relevant
water molecules and their presence, however, seem to be
strongly dependent on the protein conformation and the
substrate under consideration.
In contrast, de Graaf et al.20
came to a different conclusion.
The reliability of the SOM predictions for CYP2D6 based
on docking simulations were improved by including explicit
water molecules (specified as ON) for all 65 compounds
(MDEA, MDEA-like, and non-MDEA-like). The positions
of the water molecules were based on grid-based energy
calculations, in which a region surrounding the heme iron
atom was disallowed for water molecules. Because of the
constant presence of water molecules during the docking
simulations, the search space was restricted, and the substrate
was pushed toward the heme, thereby increasing the chance
of a successful SOM prediction. The improvement that was
observed might not be due only to the presence of water
molecules and their favorable interactions with the substrate
or the protein, but might also be caused by space restrictions.
This hypothesis is supported by our preliminary docking
results for MDEA in protein conformation A of CYP2D6,
where we see a decrease in the performance if water
molecules cannot be displaced by the MDEA-like and nonMDEA-like
sets of substrates. Overall, the role of water
molecules in a promiscuous protein such as CYP2D6 does
not appear to be straightforward. From the results for the
MDEA and MDEA-like compounds, it is clear that water
molecules should not be excluded from the calculations
completely. On the other hand, for new substrates, different
water molecules might be relevant. Therefore, we do not
recommend to use a single set of water molecules for all
substrates.
Analysis of the docking poses allowed us to classify the
role of water molecules in four groups: improvement,
worsening, no effect, and no improvement. For instance,
water molecules can lead to improvement by preventing the
substrate from being docked in subpockets or by forming
hydrogen bonds that are involved in the stabilization of the
correct pose. However, water molecules can also lead to
worsening by forming hydrogen bonds that destabilize the
correct pose or by forming hydrogen bonds with the protein
that subsequently prevent a favorable hydrogen bond with
the substrate that is essential for the correct pose to occur.
However, water molecules can also not have any impact on
the results, or they can simply not lead to an improvement
of the docking results, meaning new poses can occur, but
they do not bring the SOM closer to the heme. This effect
of water molecules has been observed earlier. Some docking
studies in proteases and kinases have also pointed out that,
indeed, the inclusion of water molecules can lead to
improvement, worsening, or no effect depending on the
protein under study and its conformation.8
We observed
similar effects previously for a crystallographic water
molecule in cytochrome P450 1A2.40
152 J. Chem. Inf. Model., Vol. 50, No. 1, 2010 SANTOS ET AL.
CONCLUSIONS
In this study, we demonstrated that hydration sites can be
found in the cavity of CYP2D6, by monitoring the positions
of water molecules in molecular dynamics simulations of
CYP2D6 with MDEA. Even though the probability of finding
a water molecule at these sites was at least 30 times larger
than expected for bulk water, the water molecules themselves
were rather mobile and dynamic, possibly adding to the wellknown
promiscuity of CYPs. Inclusion of selected water
molecules in molecular docking simulations of CYP2D6 had
an effect on the reliability of the site of metabolism
prediction. However, their role is not unique, sometimes
leading to a slight improvement or to no overall alterations
depending on the protein conformation and the substrate. The
larger effect was observed for the substrate that was used to
select the water molecules. The same waters seemed to be
transferable to similar substrates, still leading to a small
overall improvement. However, for dissimilar compounds,
no net effect could be observed. Our study sheds light on
the relevant yet highly versatile role of water molecules in
the CYP2D6 active site.
Abbreviations. AMI, amiodarine [(2-{4-[(2-butyl-1-benzo-
furan-3-yl)carbonyl]-2,6-diiodophenoxy}-ethyl)diethylamine];
CYP, cytochrome P450; CYP2D6, cytochrome P450 isoform
2D6; MDEA, R-3,4-methylenedioxy-N-ethylamphetamine; MD,
moleculardynamics;MRP,R-mianserine(2-methyl-1,2,3,4,10,14bhexahydrodibenzo[c,f]pyrazino[1,2-a]azepine);
SOM, site of
metabolism; TRP, p-tyramine (4-hydroxyphenethylamine).
REFERENCES AND NOTES
(1) Ladbury, J. E. Just add water! The effect of water on the specificity
of protein-ligand binding sites And its potential application to drug
design. Chem. Biol. 1996, 3, 973–980.
(2) Helms, V. Protein Dynamics Tightly Connected to the Dynamics of
Surrounding and Internal Water Molecules. ChemPhysChem 2007, 8,
23–33.
(3) Cozzini, P.; Fornabaio, M.; Mozzarelli, A.; Spyrakis, F.; Kellogg, G. E.;
Abraham, D. J. Water: How to Evaluate Its Contribution in ProteinLigand
Interactions. Internat. Quant. Chem. 2006, 106 (3), 647–651.
(4) Poornima, C. S.; Dean, P. M. Hydration in drug design. 1. Multiple
hydrogen-bonding features of water molecules in mediating proteinligand
interactions. J. Comput.-Aided Mol. Des. 1995, 9, 500–512.
(5) Poornima, C. S.; Dean, P. M. Hydration in drug design. 2. Influence
of local site surface shape on water binding. J. Comput.-Aided Mol.
Des. 1995, 9, 513–520.
(6) McConkey, B. J.; Sobolev, V.; Edelman, M. The performance of
current methods in ligand-protein docking. Curr. Sci. 2002, 83 (7),
845–856.
(7) Sousa, S. F.; Fernandes, P. A.; Ramos, M. J. Protein-ligand docking:
Current status and future challenges. Proteins 2006, 65 (1), 15–26.
(8) Roberts, B. C.; Mancera, R. L. Ligand-Protein Docking with Water
Molecules. J. Chem. Inf. Model. 2008, 48 (2), 397–408.
(9) Moitessier, N.; Englebienne, P.; Lee, D.; Lawandi, J.; Corbeil, C. R.
Towards the development of universal, fast and highly accurate
docking/scoring methods: A long way to go. Br. J. Pharmacol. 2008,
153 (S1), S7–S26.
(10) Corbeil, R. C.; Moitessier, N. Docking ligands into flexible and
solvated macromolecules. 3. Impact of input ligand conformation,
protein flexibility, and water molecules on the accuracy of docking
programs. J. Chem. Inf. Model. 2009, 49, 997–1009.
(11) Huang, N.; Shoichet, B. K. Exploiting Ordered Waters in Molecular
Docking. J. Med. Chem. 2008, 51 (16), 4862–4865.
(12) Birch, L.; Murray, C. W.; Hartshorn, M. J.; Tickle, I. J.; Verdonk,
M. L. Sensitivity of molecular docking to induced fit effects in
influenza virus neuraminidase. J. Comput.-Aided Mol. Des. 2002, 16
(12), 855–869.
(13) Denisov, I. G.; Makris, T. M.; Sligar, S. G.; Schlichting, I. Structure
and Chemistry of Cytochrome P450. Chem. ReV. 2005, 105, 2253–
2277.
(14) Sono, M.; Roach, M. P.; Coulter, E. D.; Dawson, J. H. HemeContaining
Oxygenases. Chem. ReV. 1996, 96 (7), 2841–2888.
(15) Williams, J. A.; Hyland, R.; Jones, B. C.; Smith, D. A.; Hurst, S.;
Goosen, T. C.; Peterkin, V.; Koup, J. R.; Ball, S. E. Drug-drug
interactions for UDP-glucuronosyltransferase substrates: A pharmacokinetic
explanation for typically observed low exposure (AUCi/
AUC) ratios. Drug Metab. Dispos. 2004, 32, 1201–1208.
(16) Ghoneim, M. M.; Korttila, K.; Chiang, C. K.; Jacobs, L.; Schoenwald,
R. D.; Mewaldt, S. P.; Kayaba, K. O. Diazepam effects and kinetics
in Caucasians and Orientals. Clin. Pharmacol. Ther. 1981, 29, 749–
756.
(17) Stjernschantz, E.; Vermeulen, N. P. E.; Oostenbrink, C. Computational
prediction of drug binding and rationalisation of selectivity towards
cytochromes P450. Exp. Opin. Drug Metab. Toxicol. 2008, 4 (5), 513–
527.
(18) Cruciani, G.; Carosati, E.; DeBoeck, B.; Ethirajulu, K.; Mackie, C.;
Howe, T.; Vianello, R. MetaSite: Understanding Metabolism in Human
Cytochromes from the Perspective of the Chemist. J. Med. Chem.
2005, 48 (22), 6970–6979.
(19) Rowland, P.; Blaney, F. E.; Smyth, M. G.; Jones, J. J.; Leydon,
V. R.; Oxbrow, A. K.; Lewis, C. J.; Tennant, M. G.; Modi, S.;
Eggleston, D. S.; Chenery, R. J.; Bridges, A. M. Crystal Structure
of Human Cytochrome P450 2D6. J. Biol. Chem. 2006, 281 (11),
7614–7622.
(20) de Graaf, C.; Oostenbrink, C.; Keizers, P. H. J.; van der Wijst, T.;
Jongejan, A.; Vermeulen, N. P. E. Catalytic Site Prediction and Virtual
Screening of Cytochrome P450 2D6 Substrates by Consideration of
Water and Rescoring in Automated Docking. J. Med. Chem. 2006,
49 (8), 2417–2430.
(21) Hritz, J.; de Ruiter, A.; Oostenbrink, C. Impact of Plasticity and
Flexibility on Docking Results for Cytochrome P450 2D6: A
Combined Approach of Molecular Dynamics and Ligand Docking.
J. Med. Chem. 2008, 51 (23), 7469–7477.
(22) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; Hermans,
J. Interaction models for water in relation to protein hydration. In
Intermolecular Forces; Pullman, B., Ed.; Reidel: Dordrecht, The
Netherlands, 1981; pp 331-342.
(23) Christen, M.; Hu¨nenberger, P. H.; Bakowies, D.; Baron, R.; Bu¨rgi,
R.; Geerke, D.; Heinz, T. N.; Kastenholz, M. A.; Kra¨utler, V.;
Oostenbrink, C.; Peter, C.; Trzesniak, D.; van Gunsteren, W. F. The
GROMOS software for biomolecular simulation: GROMOS05. J. Comput.
Chem. 2005, 26, 1719–1751.
(24) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; DiNola,
A.; Haak, J. R. Molecular dynamics with coupling to an external bath.
J. Chem. Phys. 1984, 81 (8), 3684–3690.
(25) Ryckaert, J.-P.; Ciccotti, G.; Berendsen, H. J. C. Numerical Integration
of Cartesian Equations of Motion of a System with Constraints:
Molecular Dynamics of n-Alkanes. J. Comput. Phys. 1977, 23 (3),
327–341.
(26) van Gunsteren, W. F.; Billeter, S. R.; Eising, A. A.; Hu¨nenberger,
P. H.; Kru¨ger, P.; Mark, A. E.; Scott, W. R. P.; Tironi, I. G.
Biomolecular Simulation: The GROMOS96 Manual and User Guide;
vdf Hochschulverlag AG an der ETH Zu¨rich: Zu¨rich, 1996.
(27) Schuler, L. D.; Daura, X.; van Gunsteren, W. F. An improved
GROMOS96 force field for aliphatic hydrocarbons in the condensed
phase. J. Comput. Chem. 2001, 22 (11), 1205–1218.
(28) Tironi, I. G.; Sperb, R.; Smith, P. E.; van Gunsteren, W. F. A
generalized reaction field method for molecular dynamics simulations.
J. Chem. Phys. 1995, 102 (13), 5451–5459.
(29) Heinz, T. N.; van Gunsteren, W. F.; Hu¨nenberger, P. H. Comparison
of four methods to compute the dielectric permittivity of liquids from
molecular dynamics simulations. J. Chem. Phys. 2001, 115 (3), 1125–
1136.
(30) Abel, R.; Young, T.; Farid, R.; Berne, B. J.; Friesner, R. A. Role of
the active-site solvent in the thermodynamics of factor Xa ligand
binding. J. Am. Chem. Soc. 2008, 130, 2817–2831.
(31) Michel, J.; Tirado-Rives, J.; Jorgenson, W. J. Prediction of the water
content in protein binding sites. J. Phys. Chem. B 2009, 113, 13337–
13346.
(32) Jones, G.; Willett, P.; Glen, R. C.; Leach, A. R.; Taylor, R.
Development and Validation of a Genetic Algorithm for Flexible
Docking. J. Mol. Biol. 1997, 267, 727–748.
(33) Eldridge, M. D.; Murray, C. W.; Auton, T. R.; Paolini, G. V.; Mee,
R. P. Empirical scoring functions: I. The development of a fast
empirical scoring function to estimate the binding affinity of ligands
in receptor complexes. J. Comput.-Aided Mol. Des. 1997, 11 (5), 425–
445.
(34) Kirton, S. B.; Murray, C. W.; Verdonk, M. L.; Taylor, R. Prediction
of binding modes for ligands in the cytochromes P450 and other hemecontaining
proteins. Proteins 2005, 58, 836–844.
(35) Verdonk, M. L.; Chessari, G.; Cole, J. C.; Hartshorn, M. J.; Murray,
C. W.; Nissink, J. W. M.; Taylor, R. D.; Taylor, R. Modeling Water
Molecules in Protein-Ligand Docking Using GOLD. J. Med. Chem.
2005, 48 (20), 6504–6515.
ROLE OF WATER IN CYP2D6 J. Chem. Inf. Model., Vol. 50, No. 1, 2010 153
(36) Humphrey, W.; Dalke, A.; Schulten, K. VMD: Visual molecular
dynamics. J. Mol. Graphics 1996, 14 (1), 33–38.
(37) Rydberg, P.; Rod, T. H.; Olsen, L.; Ryde, U. Dynamics of Water
Molecules in the Active-Site Cavity of Human Cytochromes P450. J.
Phys. Chem. B 2007, 111 (19), 5445–5457.
(38) Guengerich, F. P. A malleable catalyst dominates the metabolism
of drugs. Proc. Natl. Acad. Sci. U.S.A. 2006, 103 (37), 13565–
13566.
(39) de Graaf, C.; Pospisil, P.; Pos, W.; Folkers, G.; Vermeulen, N. P. E.
Binding Mode Prediction of Cytochrome P450 and Thymidine
Kinase Protein-Ligand Complexes by Consideration of Water and
Rescoring in Automated Docking. J. Med. Chem. 2005, 48 (7), 2308–
2318.
(40) Vasanthanathan, P.; Hritz, J.; Taboureau, O.; Olsen, L.; Steen
Jorgensen, F.; Vermeulen, N. P. E.; Oostenbrink, C. Virtual Screening
and Prediction of Site of Metabolism for Cytochrome P450 1A2
Ligands. J. Chem. Inf. Model. 2009, 49 (1), 43–52.
(41) Harris, K.; Woolf, L. Pressure and temperature dependence of the self
diffusion coefficient of water and oxygen-18 water. J. Chem. Soc.,
Faraday Trans. 1 1980, 76, 377–385.
CI900293E
154 J. Chem. Inf. Model., Vol. 50, No. 1, 2010 SANTOS ET AL.
EFFICIENT INCORPORATION OF PLASTICITY AND WATER MOLECULES INTO THE MOLECULAR
DOCKING APPROACH
36
Paper 3
Oostenbrink C.; de Ruiter A.; Hritz J.; Vermeulen N.P.E Malleability and versatility of
Cytochrome P450 active sites studied by molecular simulations.
Curr. Drug Metab. 2012, 13, 190-196
190 Current Drug Metabolism, 2012, 13, 190-196
Malleability and Versatility of Cytochrome P450 Active Sites Studied by Molecular
Simulations
Chris Oostenbrink1,2,*
, Anita de Ruiter2
, Jozef Hritz3
and Nico Vermeulen1
1
Leiden-Amsterdam Centre for Drug Research, Division of Molecular and Computational Toxicology, De Boelelaan 1083, 1081 HV
Amsterdam, The Netherlands; 2
Institute of Molecular Modeling and Simulation, University of Natural Resources and Life Sciences,
Muthgasse 18, 1190 Vienna, Austria; 3
Department of Structural Biology, School of Medicine, University of Pittsburgh, 3501 Fifth
Avenue, Pittsburgh PA 15260, USA
Abstract: As the most important phase I drug metabolizing enzymes, the human Cytochromes P450 display an enormous versatility in
the molecular structures of possible substrates. Individual isoforms may preferentially bind specific classes of molecules, but also within
these classes, some isoforms show remarkable levels of promiscuity. In this work, we try to link this promiscuity to the versatility and
malleability of the active site at the hand of examples from our own work. Mainly focusing on the flexibility of protein structures and the
presence or absence of water molecules, we establish molecular reasons for observed promiscuity, determine the relevant factors to take
into account when predicting binding poses and rationalize the role of individual interactions in the process of ligand binding. A high
level of active site flexibility does not only allow for the binding of a large variety of substrates and inhibitors, but also appears to be important
to facilitate ligand binding and unbinding.
Keywords: Site of metabolism prediction, protein flexibility, molecular docking, molecular dynamics simulations, replica exchange.
INTRODUCTION
The superfamily of Cytochrome P450 (CYP) enzymes is known
to be highly diverse and versatile in its substrate specificity. This
allows the enzymes to metabolize a wide range of xenobiotics. This
holds true for the large diversity of the various isoforms for which
general descriptions of typical substrates may be described, e.g. in
the form of pharmacophore descriptions [1]. Individual isoforms,
however, still show varying degrees of promiscuity towards potential
substrates. The level of promiscuity has been attributed to properties
of the active site, such as size, flexibility and malleability [2].
With the increased number of crystal structures of mammalian
CYPs, a structural rationalization becomes feasible based on computer
visualizations or molecular dynamics simulations to monitor
the structure and dynamics of their catalytic sites [3-5].
CYP3A4 is probably the most promiscuous of the drug metabolizing
enzymes and is known to have the largest active site, which
in addition appears to be highly flexible [6]. CYP2C9 and CYP2D6
have smaller active site pockets and have a more tightly defined
substrate specificity, including mostly negatively and positively
charged substrates, respectively. However, the enzymes are not
restricted to substrates bearing a full charge, and even within this
restriction, a large variety of substrates have been reported. The 2A
subfamily of CYPs is among the more specific drug metabolizing
enzymes, mostly acting on flat aromatic compounds. Even within
this class considerable variability exists [1].
Here, we report and summarize some of our own rationalizations
for the versatility observed in various CYP isoforms, using
computer simulations of the molecular interactions between the
proteins and small molecule inhibitors or substrates. Molecular
simulations offer insight at a resolution that is often not reachable
by experimental means [7]. The molecular detail available in molecular
dynamics simulations of proteins and protein-ligand complexes
allows us to understand the subtle differences that enable the
protein to discriminate between ligands. At a first level, the binding
of ligands to the enzymes is explained in terms of the binding
*Address correspondence to this author at the Institute of Molecular Modeling
and Simulation, University of Natural Resources and Life Sciences,
Muthgasse 18, 1190 Vienna, Austria; E-mail: chris.oostenbrink@boku.ac.at
complementarity, according to the classical lock-and-key model.
However, in enzymes as flexible as Cytochrome P450, this model
needs to be extended to include induced fit effects. Here, we use the
term flexibility to describe the overall degrees of freedom that are
accessible to the protein. We use the term malleability to specifically
refer to the possibility of accommodating different substrates
or inhibitors as a result of the protein flexibility.
It is useful to describe the binding in the context of conformational
selection. This latter paradigm considers a protein to consist
not of a single static structure, but rather to be present in an ensemble
of different conformations [8]. By considering multiple protein
conformations to understand the binding of small molecule inhibitors
or substrates, the complexity of the models increases. However,
if predictions are to be performed at a quantitative rather than qualitative
level, the protein-ligand interactions are to be described in
terms of binding free energies, including both enthalpic and entropic
contributions [9]. All contributions from direct protein-ligand
interactions, loss of conformational freedom and desolvation of the
ligand and the protein should be taken into account in calculations
of the binding free energies.
Low-energy conformations of the protein are more often observed
in the ensemble of protein structures, while higher-energy
conformations will occur only rarely. A ligand that binds to the
active site may only be able to do so if the protein is in a specific
conformation. If this happens to be a low-energy conformation,
then the protein-ligand interactions and desolvation effects may be
sufficient to explain most of the binding free energy. However, for
tightly binding ligands, the favorable ligand-protein enthalpy may
be sufficient to stabilize the protein when it is in a high-energy
conformational state. From the perspective of the protein, this is an
unfavorable enthalpic contribution. As the ligand now locks the
protein in this conformational state, its flexibility is also reduced,
leading to an additional unfavorable entropic contribution.
In the following, we shortly describe the methods used to place
substrates into the active site and to monitor the flexibility of the
enzymes. In particular, we highlight Hamiltonian replica exchange
molecular dynamics (H-REMD) simulations using distance restraining
potentials. This approach allows for sufficient conformational
sampling along the ligand exit pathway to calculate a potential
of mean force (PMF). Following is a description of the effect of
1875-5453/12 $58.00+.00 © 2012 Bentham Science Publishers
Malleability and Versatility From Molecular Simulations Current Drug Metabolism, 2012, Vol. 13, No. 2 191
active site flexibility and the presence of water molecules on the
accurate prediction of binding poses and an analysis of the versatility
of protein-ligand interactions. Finally, the role of Phe483 in
substrate orientation and ligand unbinding in CYP2D6 is analysed
in more detail as an example of active site flexibility.
METHODS
Unless stated differently, all simulations were based on the
following X-ray structures available in the protein databank[10]:
CYP1A2 (PDB-entry 2HI4 [11]); CYP2C9 (PDB-entry 1R9O
[12]); CYP2D6 (PDB-entry 2F9Q [13]) for details on the modeling
of missing loops or side-chains and specific back-mutations, we
refer to our original publications [14-16].
Automated docking experiments described in this work were
performed with the program GOLD (Genetic Optimization for
Ligand Docking) [17] version 3.2 or 3.3.1 in combination with the
Chemscore scoring function [18]. For the docking experiments in
the X-ray structures of CYP2D6 and CYP2C9, the optimized parameters
for heme-containing proteins was used [19]. The radius for
docking was set to 1.8 – 2.0 nm around a point in the center of the
active site. Using a population of 100 genes, docking runs were
performed with maximally 1000 genetic algorithm operations. Water
molecules were either absent, included in the active site, or
switched on and off by the program [20]. In the docking experiments
the Heme group was modeled without a sixth ligand on the
iron atom.
All molecular dynamics simulations described in this work
were performed using a modified version of the GROMOS simulation
package [21, 22]. Typically, the protein structure, with or without
docked substrate was solvated in rectangular periodic boxes
containing SPC water molecules [23]. The minimum solute-to-wall
distance was at least 0.8 nm. After careful thermalization and
equilibration the simulations were performed under a constant temperature
of 298 K using the weak-coupling method with a relaxation
time of 0.1 ps [24]. The pressure was kept constant at 1 atm
using weak-coupling with a relaxation time of 0.5 ps and an estimated
isothermal compressibility of 4.575.
10-4
(kJ mol-1
nm-3
)-1
[24]. Bond lengths were restrained using the SHAKE algorithm,
allowing for a time step of 2 fs [25]. Nonbonded interactions up to a
distance of 0.8 nm were calculated every time step using a groupbased
cutoff that was constructed every fifth time step. At these
times, the interactions up to 1.4 nm were also calculated and kept
constant between pairlist updates. A reaction-field contribution [26]
was added to the energies and forces to account for a homogeneous
medium with relative dielectric constant of 61 outside the cutoff
sphere [27].
In order to monitor the egress of substrate 7-methoxy-4(aminomethyl)-coumarin
(MAMC) out of the active site of
CYP2D6 [28, 29], we used the following approach. Hamiltonian
replica exchange (H-REMD) simulations [30] were performed using
26 parallel MD simulations in which distance restraints were
applied between the Heme Fe-ion in Cytochrome P450 2D6 and the
centre of geometry of 4 atoms in the ligand MAMC using reference
distances of 0.720, 0.820, 0.910, 1.020, 1.101, 1.170, 1.220, 1.269,
1.340, 1.390, 1.430, 1.470, 1.539, 1.580, 1.620, 1.701, 1.800, 1.881,
1.950, 2.020, 2.110, 2.180, 2.270, 2.350, 2.440, and 2.530 nm, respectively.
The force constant for the distance restraints was 1000
kJ mol-1
nm-2
. Initial structures of these replicas were obtained from
a simulation in which the distance restraint was gradually increased
from 0.4 to 2.4 nm. In this simulation, the MAMC molecule was
observed to leave the active site through the solvent channel (according
to the nomenclature of Wade [31]). After an initial equilibration
of 25 ps, switching attempts were performed every 5 ps
between neighbouring replica’s using the metropolis criterion to
accept or reject the switch. All replicas were simulated for 0.4 ns.
This approach allows the individual MAMC molecules to move in
and out of the active site, potentially also exiting through alternative
egress channels. Simultaneously, the replica exchange approach
ensures that at all conformations sampled at the individual restraining
lengths correspond to a correct ensemble in a statistical mechanical
sense, i.e. have the proper likelihood of occurrence given
the Hamiltonian of the (restrained) system. This allows us to calculate
a potential of mean force using the weighted histogram analysis
method (WHAM) [32, 33].
PREDICTION OF BINDING POSES
In the context of Cytochromes P450, the binding orientation of
substrates may directly influence their regiospecific metabolism.
The reaction that takes place should not only be energetically feasible,
the proper site of the substrate should also be in the vicinity of
the reactive centre [9, 34-36]. A very simple geometrical test to see
if binding orientations are in agreement or disagreement with catalytic
activity is to measure the distance between the site of metabolism
(SOM) in the substrate to the Heme Fe ion. As an empirical
criterion, a pose is considered to agree with the observed metabolism
if this distance is shorter than 6 Å, while larger distances indicate
that the substrate orientation is incompatible with its experimentally
determined metabolism [37, 38].
This 6 Å rule, however crude it is, allows us to check for known
substrates how well automated docking simulations perform and to
check the effect of slight modifications in the protein structure or
the presence of water molecules in the active site.
CYP1A2
The active site of CYP1A2 is relatively narrow and contains an
aromatic cluster of three Phenylalanine residues, allowing mostly
flat aromatic compounds to bind to it. In the X-ray structure, the
protein forms a complex with the inhibitor -naphthoflavone, with
a structured water molecule bridging a hydrogen bond to the active
site [11]. 20 known substrates were docked to the active site using
the GOLD docking program [17]. For three substrates (15%) no
docking solutions were found in which the SOM was placed within
6 Å of the Heme iron using the chemscore scoring function and
including the water molecule. For 13 substrates the top-ranked pose
did correspond to a catalytically feasible pose, while for the remaining
4 substrates the second ranked pose needed to be included to
understand the metabolism [39]. Interesting results were obtained
when the presence of the water molecule was determined based on
the scoring function, in the course of the docking experiment [20].
Overall, the number of ‘correct’ predictions slightly decreased to 10
first ranked poses, 6 substrates for which the second or third ranked
pose needed to be considered and 4 substrates for which no catalytically
active pose could be identified. As an example, the binding
poses of imipramine were studied in more detail. When the water
molecule was not taken into account, the first ranked pose corresponded
to an orientation in agreement with the major metabolite of
this molecule. In the presence of the water molecule, alternative
orientations were found, in agreement with the formation of the
minor metabolite. When the presence or absence of the water molecule
was determined by the docking program, a mixture of both
binding orientations was found, but the water was always determined
to be off, also for the poses that were in agreement with the
minor metabolite. This demonstrates on the one hand the effect of
water molecules on the prediction of docking poses, while on the
other side it shows that the presence or absence of such water molecules
is not necessarily predicted in a consistent manner [39].
CYP2D6
Fig. (1) compares different experiments for CYP2D6 based on
the percentage of poses that agree with observed metabolite formation
for a set of 65 CYP2D6 substrates [38], using different protein
models.
Before the availability of this X-ray structure, a homology
model was described based on the rabbit CYP2C5 structure [40]. In
192 Current Drug Metabolism, 2012, Vol. 13, No. 2 Oostenbrink et al.
this structure the active site was modeled in the presence of the
substrate codeine. About 60% of the substrates could be docked to
this homology model in poses that agree with the experimentally
observed metabolite formation. This value could be further increased
by adding static water molecules to the active site [38],
probably by restricting the available space in the active site. The Xray
structure of an apo form of CYP2D6 was subsequently solved
[13]. Using this structure, only 20% of the substrates could be
docked in catalytically active poses. In a series of molecular dynamics
simulations in which substrates of increasing volume were
subsequently added, the active site was expanded to accommodate
larger substrates as well. 2500 protein structures were extracted
from 10 simulations containing five different substrates and two
different initial protein conformations of Phe483 in the active site
[14]. The average quality of the SOM predictions over all these
protein conformations was 52 %.
In a subsequent experiment, water molecules were carefully
selected based on their presence during MD simulations for 8 diverse
protein structures. The docking program was used to guide
the presence or absence of these water molecules. Averaged over
the 8 protein structures, the overall percentage of binding poses in
agreement with experiment marginally increased to 56 % [41]. It
was observed that this percentage fluctuates strongly over the different
protein snapshots, indicating that small thermal fluctuations
as observed in the protein active site may have a profound effect on
the predicted binding orientations of the substrates.
In fact, it was possible to identify a single, water-free, protein
conformation for which 71% of the substrates were found to dock
in a catalytically active pose. By reducing the complete ensemble of
2500 protein conformations to three structures, an overall accuracy
of maximally 90% could theoretically be obtained, if one could
determine which protein structure is most appropriate for a given
substrate. Docking a substrate into three protein structures is usually
feasible, but the selection of the proper protein structure cannot
be made on the basis of the scoring function (unpublished results).
A simple decision tree was established, which suggests which protein
structure to use based on the molecular mass and the number of
hydrophobic atoms in the substrate. Using the decision tree, the
agreement of the predicted poses with experiment remains at an
impressive 80% [14].
Using a very similar approach for CYP3A4, similar results were
obtained [42]. When docking 16 substrates into 125 protein structures
as obtained from MD simulations, on average 26% of the
docking experiments yielded a pose in agreement with experiments.
A single protein structure could be identified in which 62% of the
compounds were correctly placed in the protein.
In summary, docking experiments on CYPs show that it is
highly sensitive to the exact protein structure that is used in the
presence or absence of water molecules. For the selection of an
appropriate protein structure for CYP2D6 from a small set of three
structures a simple decision tree was established [14], while the
proper selection of which water molecules to take into account for
any given substrate remains elusive for this versatile enzyme [41].
CYP2C9
In CYP2C9 a different approach was used to predict the binding
modes for a series of thiourea-containing inhibitors. This series of
compounds contains both an imidazole moiety and a thiourea
group, both of which are known to interact favourably with the
heme iron. As these compounds are inhibitors rather than substrates,
there is no direct indication of a proper binding mode. All
12 compounds were docked into the active site of CYP2C9 and up
to four different poses, coordinating the heme with the imidazole or
the thiourea moiety, were identified. In an effort to calculate the
Fig. (1). Percentage of compounds for which the first-ranked pose from a docking experiment agrees with the experimentally determined site of metabolism. A
set of 65 CYP2D6 substrates were docked into various protein structures obtained from homology modeling or X-ray crystallography possibly followed by
molecular dynamics simulations, and/or the addition of water molecules. The sixth and seventh bars involve one or three protein structures, where in the latter
case the optimal protein structure is selected (out of three) for every substrate. For the last bar a simple decision tree was used to select the appropriate protein
structure from the same set of three.
Malleability and Versatility From Molecular Simulations Current Drug Metabolism, 2012, Vol. 13, No. 2 193
binding affinities for these compounds, MD simulations were performed
for all docking poses and an iterative scheme using the linear
interaction energy methodology [43] was used. In short, this
scheme gives weights to the individual simulations, based on the
free energy of the ligand in the active site. By averaging the interaction
energies according to the weights, an overall free energy of
binding may be established. The weights may subsequently be interpreted
as the probability that a given pose occurs [15]. From this
we observed that 9 of the 12 compounds in this series prefer to bind
in very similar orientations, while 3 bind in alternative poses.
Moreover, for 2 of the compounds, multiple binding orientations
contribute to the total free energy of binding, suggesting that they
are equally likely to occur. This demonstrates that the active site of
CYP2C9 is quite versatile in the binding modes that it accepts and
that a dynamic equilibrium between multiple binding modes needs
to be taken into account to explain the experimental data [15]. Similar
observations were previously made based on MD simulations of
CYP2D6 [44].
ANALYSIS OF INTERACTIONS
Once reasonable binding modes of substrates or inhibitors have
been determined, it is interesting to investigate which amino acids
are most important for the interactions between the ligands and the
protein. From docking experiments this can be done by establishing
protein-ligand interaction fingerprints. One striking feature from
such analyses is the diversity of the interactions. In general, only
one or two amino acids really interact with the majority of substrates,
while many others only form interactions with less than
50% of the substrates. This has been observed for CYP1A2[39],
CYP2D6 [40] and CYP3A4 [42] and may indicate another source
of observed promiscuity. The free energy of binding to CYP1A2
was seen to correlate mostly with the non-polar protein-ligand interaction
energy, while the electrostatic contribution to the free
energy became negligible in linear interaction energy (LIE) models
[16]. Even in a hydrophobic cavity as observed in CYP1A2, the
protein is able to present a variety of suitable hydrogen bonding
partners when this is required by the ligand, therefore the shape
complementarity becomes more important to describe differences in
binding than the actual electrostatic interactions.
Phe483 in CYP2D6
As was outlined in the introduction, the strength of computational
approaches is that it allows us to investigate the flexibility
and dynamics of protein-ligand complexes at a submolecular level.
In the following, we will shift our focus to the behaviour of Phe483
in 2D6. Experimentally this residue was seen to play a role in the
recognition of R- and S-propranolol. While the wildtype protein
binds these substrates with very comparable affinity, the affinity of
the F483A mutant was stereospecifically reduced for R-propranolol
by a factor 20, or 7 .7 kJ mol-1
[45]. In molecular dynamics simulations
of CYP2D6 with various substrates bound in the active site,
the 1 torsional angle of the Phe483 sidechain was observed to
change considerably (see Fig. (2)). Over 4 ns, this dihedral angle
preferentially samples a value of 170° for R-3,4-methylenedioxy-Nethylamphetamine
(EDR), 7-methoxy-4-(aminomethyl)coumarin
(MAMC) and chlorpromazine (CHZ), while it prefers a value of
70° for R-propranolol (PPD) and tamoxifen (TMF). Note that these
are not static pictures. During several simulations, the dihedral angle
occasionally takes on different values, indicating an intrinsic
variability of this residue within the active site for which the equilibrium
may be shifted depending on the substrate bound.
The 1 torsional angle of Phe483 was restrained to 70° or 170°
in the 10 conformational ensembles that were described earlier to
perform the docking experiments in [14]. From these docking experiments
conformational preferences could be observed. Some
substrate, like e.g. a group of substrates similar to MAMC, could be
found to dock catalytically active poses significantly more often in
protein structures for which 1 of Phe483 fluctuates around 170°
[14], which agrees to the observations in Fig. (2). Fig. (3) shows the
structure of MAMC in the CYP2D6 active site. It forms a salt
bridge with Glu216, also indicated are the Heme group and Phe483.
In subsequent simulations, the unbinding of MAMC from the
active site was simulated. In a preliminary simulation, a distance
restraint was applied between the Heme Fe ion and the MAMC
molecule. This distance was gradually increased, thereby forcing
the MAMC to leave from the active site. As observed earlier, the
Fig. (2). Time series (A) and normalized distributions (B) of the x1 torsional dihedral angle of Phe483 in CYP2D6, with different substrates
bound to the active site: R-3,4-methylenedioxy-N-ethylamphetamine (EDR) in black; 7-methoxy-4-(aminomethyl)coumarin (MAMC) in red;
R-propranolol (PPD) in green; chlorpromazine (CHZ) in blue and tamoxifen (TMF) in yellow.
194 Current Drug Metabolism, 2012, Vol. 13, No. 2 Oostenbrink et al.
Fig. (3). Cytochrome P450 2D6 active site with 7-methoxy-4(aminomethyl)coumarin
(MAMC) bound. The surface of the protein is
drawn in light grey, the heme group in green sticks, MAMC in yellow and
Phe483 in magenta. The surface of the negatively charged Glu216 is colored
red.
molecule exited the protein through the so called solvent channel
[31]. Using initial coordinates taken from this pathway, Hamiltonian
Replica Exchange MD (H-REMD) simulations were performed
with 26 replicas differing in the reference value of the distance
restraint (see methods section). Every 5 ps a switch between
the replicas is attempted, thereby allowing configurations at shorter
distances to move to longer distances and vice versa. As no restrictions
with respect to the path are given, the molecule theoretically
can also sample different exit pathways in this process. However,
this was not observed in the current simulations. From the HREMD
simulations, correct ensembles at every restraining distance
were obtained which include rather diverse protein conformations.
The potential of mean force (PMF) that is calculated from this set
of 26 ensembles is shown in panel A of Fig. (4).
Even at a distance of 2.5 nm, the MAMC molecule is still interacting
with the surface of the protein, forming on average 0.77 hydrogen
bond, explaining that the PMF does not yet level off at this
distance. Up to a distance of 1.1 nm, MAMC interacts through an
average of 1.5 hydrogen bonds with the protein, of which 0.8 are
formed with Glu216 (see panel B in Fig. (4)). At larger distances,
the average number of hydrogen bonds rapidly decreases to 0.2 at
1.4 nm, corresponding to the steep increase in the PMF. Interestingly,
the ensemble at 1.5 nm contains snapshots with two distinct
orientations of the MAMC molecule. One in which the molecule
has rotated completely with respect to the orientation in Fig. (3),
such that a hydrogen bond to Glu216 may still be formed and another
orientation in which the positively charged ammonium group
is pointing towards bulk solvent. At even larger distances, alternative
interactions may be formed, leading to a plateau in the PMF,
but the average number of hydrogen bonds steadily decreases from
1 to 0.5 over the distance range 1.5 to 2.3 nm. Only at the very end
it increases again to 0.8 when MAMC diffuses slightly over the
surface of the protein. The current PMF cannot be used as such to
estimate the free energy of binding. It does not level off to a plateau
Fig. (4). A) Potential of mean force along the unbinding pathway of MAMC
from CYP2D6 as a function of the Fe - MAMC distance; B) number of
hydrogen bonds between MAMC and CYP2D6 (solid curve) and between
MAMC and Glu216 (dashed curve); C) fraction of Phe483 conformations in
which the 1 torsional angle takes a value of ~70° (between 0 and 120°).
corresponding to the unbound substrate. Moreover, corrections
would need to be estimated corresponding to the release of the distance
restraint and to the transfer of the substrate from the accessible
volume in these simulations to the standard state [33]. Still, it is
reassuring to note that the free energy change corresponding to
moving the ligand from the binding site to the distance at which all
native interactions are broken, amounts to roughly 20 kJ/mol, which
is in the same order of magnitude as an estimated experimentally
determined affinity of 27 kJ/mol [46].
Panel C in Fig. (4) represents the fraction of the time that the
Phe483 1 dihedral angle was observed to be around 70° (using
limits of 0 and 120°). It can be seen that as long as MAMC is bound
to the active site, with a MAMC to Fe distance up to 1 nm, Phe483
prefers a conformation with 1 = ~170°, similar to the one in Fig.
(3), and as observed in the simulation containing MAMC (Fig. (2)).
Once the MAMC molecule starts to move out, the sidechain of
Phe483 rotates into the active site ( 1 = ~70°), thereby forming
favourable hydrophobic interactions with the coumarin backbone of
MAMC. Once the MAMC has left the active site at a distance of
1.5 nm, this interaction is no longer possible and Phe483 returns to
its initial position ( 1 = ~170°).
CONCLUSION
Above we have summarized and described some of our own
molecular modeling work on Cytochrome P450 active sites and the
Malleability and Versatility From Molecular Simulations Current Drug Metabolism, 2012, Vol. 13, No. 2 195
interactions with substrates or inhibitors. As the most important
phase I drug metabolizing enzymes, the human Cytochromes P450
display an enormous versatility in the molecular structures of substrates.
Individual isoforms may preferentially bind specific classes
of molecules, but also within these classes, some isoforms show
remarkable levels of promiscuity. Even the relatively rigid and
well-structured active site of CYP1A2 allows for the creation of a
versatile network of hydrogen bonds, allowing the protein to accommodate
quite diverse substrates. This observation is also made
for more malleable proteins in which smaller or larger fluctuations
in the protein structure were additionally seen to facilitate a large
substrate promiscuity. In particular for CYP2D6 and CYP3A4 different
substrates prefer binding to different conformations of the
active site, in agreement with the conformational selection model.
For CYP2C9 it was shown that it is well possible that multiple
orientations of inhibitors contribute similarly to the overall binding
affinity, which may complicate predictions of binding orientations
even more.
A more detailed analysis of different conformational states of a
single amino acid in CYP2D6 reveals that the active site malleability
is not only crucial to understand how substrates may bind in
catalytically active poses, but may also play a role during the binding
and unbinding processes of inhibitors, substrates and products.
ACKNOWLEDGEMENTS
C.O. and A.d.R. gratefully acknowledge financial support of the
Vienna Science and Technology Fund (WWTF) and the European
Research Council (ERC) of the European Union (grant number
260408). J.H. acknowledges financial support by a Marie Curie
International Outgoing Fellowship within the 7th European Community
Framework Programme (Contract No. PIOF-GA-2009-
235902).
REFERENCES
[1] Ortiz de Montellano, P. R., Cytochrome P450: structure, mechanism
and biochemistry. Kluwer Academic/Plenum Publishers: New
York, 2005.
[2] Guengerich, F., P,, A malleable catalyst dominates the metabolism
of drugs. Proc. Nat. Acad. Sci. USA, 2006, 103, 13565 - 13566.
[3] Mestres, J., Structure conservation in Cytochromes P450. Proteins,
2005, 58, 596 - 609.
[4] Skopalik, J.; Anzenbacher, P.; Otyepka, M., Flexibility of human
Cytochromes P450: Molecular dynamics reveals differences between
CYPs 3A4, 2C9 and 2A6, which correlate with their substrate
preferences. J. Phys. Chem. B., 2008, 112, 8165 - 8173.
[5] Hendrychová, T.; Berka, K.; Navrátilova, V.; Anzenbacher, P.;
Otyepka, M., Dynamics and hydration of the active sites of mammalian
cytochromes P450 probed by molecular dynamics simulations.
Curr. Drug Metab., 2012, 13(2): 177-189.
[6] Ekroos, M.; Sjögren, T., Structural basis for ligand promiscuity in
cytochrome P450 3A4. Proc. Nat. Acad. Sci. USA, 2006, 103,
13782 - 13687.
[7] van Gunsteren, W. F.; Bakowies, D.; Baron, R.; Chandrasekhar, I.;
Christen, M.; Daura, X.; Gee, P.; Geerke, D.; Glättli, A.; Hünenberger,
P. H.; Kastenholz, M. A.; Oostenbrink, B. C.; Schenk, M.;
Trzesniak, D.; van der Vegt, N. F. A.; Yu, H. B., Biomolecular
modeling: goals, problems, perspectives. Angew. Chem. Intnl. Ed.
2006, 45, 4064 - 4092.
[8] Carlson, H. A., Protein flexibility and drug design: how to hit a
moving target. Curr. Opin. Chem. Biol., 2002, 6, (447 - 452).
[9] Stjernschantz, E.; Vermeulen, N. P. E.; Oostenbrink, C., Computational
prediction of drug binding and rationalisation of selectivity
towards cytochromes P450. Exp. Opin. Drug Metab. Tox., 2008, 4,
513 - 527.
[10] Rose, P. W.; Beran, B.; Bi, C. X.; Bluhm, W. F.; Dimitropoulus,
D.; Goodsell, D. S.; Prlic, A.; Quesada, M.; Quinn, G. B.; Westbrook,
J. D.; Young, J.; Yukick, B.; Zardecki, C.; Berman, H. M.;
Bourne, P. E., The RCSB protein data bank: redesighned web site
and web services. Nucl. Acids. Res., 2011, 39, D392 - D401.
[11] Sansen, S.; Yano, J. K.; Reynald, R. L.; Schoch, G. A.; Griffin, K.
J.; Stout, C. D.; Johnson, E. F., Adaptations for the oxidation of
polycyclic aromatic hydrocarbons exhibited by the structure of human
P450 1A2. J. Biol. Chem., 2007, 282, 14348 - 14355.
[12] Wester, M. R.; Yano, J. K.; Schoch, G. A.; Yang, C.; Griffin, K. J.;
Stout, C. D.; Johnson, E. F., The structure of human cytochrome
P450 2C9 complexed with flurbiprofen at 2.0-Å resolution. J. Biol.
Chem., 2004, 279, 35630 - 35637.
[13] Rowland, P.; Blaney, F. E.; Smyth, M. G.; Jones, J. J.; Leydon, V.
R.; Oxbrow, A. K.; Lewis, C. J.; Tennant, M. M.; Modi, S.; Eggleston,
D. S.; Chenery, R. J.; Bridges, A. M., Crystal structure of human
cytochrome P450 2D6. J. Biol. Chem., 2005, 281, 7614 -
7622.
[14] Hritz, J.; de Ruiter, A.; Oostenbrink, C., Impact of plasticity and
flexibility on docking results for Cytochrome P450 2D6: a combined
approach of molecular dynamics and ligand docking. J. Med.
Chem., 2008, 51, 7469 - 7477.
[15] Stjernschantz, E.; Oostenbrink, C., Improved ligand-protein binding
affinity predictions using multiple binding modes. Biophys. J.,
2010, 98, 2682 - 2691.
[16] Vasanthanathan, P.; Olsen, L.; Jorgensen, F. S.; Vermeulen, N. P.
E.; Oostenbrink, C., Computational prediction of binding affinity
for CYP1A2-ligand complexes using empirical free energy calculations.
Drug Metab. Disp., 2010, 38, 1347 - 1354.
[17] Jones, G.; Willet, P.; Glen, R. C.; Leach, A. R.; Taylor, R., Development
and validation of a genetic algorithm for flexible docking.
J. Mol. Biol., 1997, 267, 727 - 748.
[18] Eldritch, M. D.; Murray, C. W.; Auton, T. R.; Paolini, G. V.; Mee,
R. P., Empricial scoring functions 0.1. The development of a fast
empirical scoring function to estimate the binding affinity of
ligands in receptor complexes. J. Comp.-Aid. Mol. Des., 1997, 11,
425 - 445.
[19] Kirton, S. B.; Murray, C. W.; Verdonk, M. L.; Taylor, R., Prediction
of binding modes for ligands in the Cytochromes P450 and
other heme-containing proteins. Proteins, 2005, 58, 836 - 844.
[20] Verdonk, M. L.; Chessari, G.; Cole, J. C.; Hartshorn, M. J.;
Murray, C. W.; Nissink, J. W. M.; Taylor, R. D.; Taylor, R., Modeling
water molecules in protein-ligand docking using GOLD. J.
Med. Chem., 2005, 48, (20), 6504 - 6415.
[21] van Gunsteren, W. F.; Billeter, S. R.; Eising, A. A.; Hünenberger,
P. H.; Krüger, P.; Mark, A. E.; Scott, W. R. P.; Tironi, I. G., Biomolecular
simulation: The GROMOS96 manual and user guide.
Vdf Hochschulverlag AG an der ETH Zürich: Zürich, 1996.
[22] Christen, M.; Hünenberger, P. H.; Bakowies, D.; Baron, R.; Bürgi,
R.; Geerke, D.; Heinz, T. N.; Kastenholz, M. A.; Kräutler, V.; Oostenbrink,
C.; Peter, C.; Trzesniak, D.; van Gunsteren, W. F., The
GROMOS software for biomolecular simulation: GROMOS05. J.
Comput. Chem., 2005, 26, 1719 - 1751.
[23] Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; Hermans,
J., Interaction models for water in relation to protein hydration.
In Intermolecular Forces, Pullman, B., Ed. Reidel: Dordrecht,
The Netherlands, 1981; pp 331-342.
[24] Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; DiNola,
A.; Haak, J. R., Molecular-dynamics with coupling to an external
bath. J. Chem. Phys., 1984, 81, (8), 3684-3690.
[25] Ryckaert, J.-P.; Ciccotti, G.; Berendsen, H. J. C., Numerical integration
of cartesian equations of motion of a system with constraints:
Molecular dynamics of n-alkanes. J. Comput. Phys., 1977,
23, (3), 327-341.
[26] Tironi, I. G.; Sperb, R.; Smith, P. E.; van Gunsteren, W. F., A
generalized reaction field method for molecular-dynamics simulations.
J. Chem. Phys., 1995, 102, (13), 5451-5459.
[27] Heinz, T. N.; van Gunsteren, W. F.; Hünenberger, P. H., Comparison
of four methods to compute the dielectric permittivity of liquids
from molecular dynamics simulations. J. Chem. Phys., 2001,
115, (3), 1125-1136.
[28] Ludemann, S. K.; Lounnas, V.; Wade, R. C., How do substrates
enter and products leave the buried active site of cytochrome
P450cam? 1. Random expulsion molecular dynamics investigation
of ligand access channels and mechanisms. J. Mol. Biol., 2000,
303, 797 - 811.
[29] Ludemann, S. K.; Lounnas, V.; Wade, R. C., How do substrates
enter and products exit the buried active site of cytochrome
P450cam? 2. Steered molecular dynamics and adiabatic mapping of
substrate pathways. J. Mol. Biol., 2000, 303, 813 - 830.
[30] Sugita, Y.; Kitao, A.; Okamoto, Y., Multidimensional replicaexchange
method for free-energy calculations. J. Chem. Phys.,
2000, 113, (15), 6042 - 6051.
196 Current Drug Metabolism, 2012, Vol. 13, No. 2 Oostenbrink et al.
[31] Cojocaru, V.; Winn, P. J.; Wade, R. C., The ins and outs of cytochrome
P450s. Biochim. Biophys. Acta, 2007, 1170, 390 - 401.
[32] Souaille, M.; Roux, B., Extension to the weighted histogram analysis
method: combining umbrella sampling ith free energy calculations.
Comput. Phys. Comm., 2001, 135, 40 - 57.
[33] Doudou, S.; N.A., B.; Henchman, R. H., Standard free energy of
binding from a one-dimensional potential of mean force. J. Chem.
Theor. Comp., 2009, 5, 909 - 918.
[34] Cruciani, G.; Carosati, E.; De Boeck, B.; Ethirajulu, K.; Mackie,
C.; Howe, T.; Vianello, R., MetaSite: Understanding metabolism in
human cytochromes from the perspective of the chemist. J. Med.
Chem., 2005, 48, 6970 - 6979.
[35] De Graaf, C.; Vermeulen, N. P. E.; Feenstra, K. A., Cytochrome
P450 in silico: An integrative modeling approach. J. Med. Chem.,
2005, 48, (8), 2725 - 2755.
[36] Afzelius, L.; Hasselgren Arnby, C.; Broo, A.; Carlsson, L.; Isaksson,
C.; Jurva, U.; Kjellander, B.; Kolmodin, K.; Nilsson, K.;
Reaubacher, F.; Weidolf, L., State-of-the-art tools for compuational
site of metabolism predictions: comparative analysis, mechanistical
insights, and future applications. Drug Metab. Rev., 2007, 39, 61 -
86.
[37] De Graaf, C.; Pospisil, P.; Pos, W.; Folkers, G.; Vermeulen, N. P.
E., Binding mode prediction of cytochrome P450 and thymidine
kinase protein-ligand complexes by consideration of water and rescoring
in automated docking. J. Med. Chem., 2005, 48, 2308 -
2318.
[38] De Graaf, C.; Oostenbrink, C.; Keizers, P. H. J.; Van der Wijst, T.;
Jongejan, A.; Vermeulen, N. P. E., Catalytic site prediction and virtual
screening accuracy of cytochrome P450 2D6 substrates by
consideration of water and rescoring in automated docking. J. Med.
Chem., 2006, 49, 2417 - 2430.
[39] Vasanthanathan, P.; Hritz, J.; Taboureau, O.; Olsen, L.; Jorgensen,
F. S.; Vermeulen, N. P. E.; Oostenbrink, C., Virtual screening and
prediction of site of metabolism for cytochrome P450 1A2 ligands.
J. Chem. Inf. Model., 2009, 49, 43 - 52.
[40] De Graaf, C.; Oostenbrink, C.; Keizers, P. H. J.; van VugtLussenburg,
B. M. A.; van Waterschoot, R.; Tschirret-Guth, R.;
Commandeur, J. N. M.; Vermeulen, N. P. E., Molecular modelingguided
site-directed mutagenesis of cytochrome P450 2D6. Curr.
Drug Metab., 2007, 8, 59 - 77.
[41] Santos, R.; Hritz, J.; Oostenbrink, C., The role of water in molecular
docking simulations of Cytochrome P450 2D6. J. Chem. Inf.
Model. 2010, 10, 55 - 66.
[42] Teixeira, V. H.; Ribeiro, V.; Martel, P. J., Analysis of binding
modes of ligands to multiple conformations of CYP3A4. Biochim.
Biophys. Acta, 2010, 1804, 2036 - 2045.
[43] Åqvist, J.; Medina, C.; Samuelsson, J. E., New method for predicting
binding affinity in computer-aided drug design. Protein Eng.,
1994, 7, (3), 385-391.
[44] Keizers, P. H. J.; De Graaf, C.; de Kanter, F. J. J.; Oostenbrink, C.;
Feenstra, K. A.; Commandeur, J. N. M.; Vermeulen, N. P. E.,
Metabolic regio- and stereoselectivity of cytochrome P450 2D6
towards 3,4-methylenedioxy-N-alkylamphetamines: in silico predictions
and experimental validation. J. Med. Chem., 2005, 48,
6117 - 6127.
[45] De Graaf, C.; Oostenbrink, C.; Keizers, P. H. J.; van VugtLussenburg,
B. M. A.; Commandeur, J. N. M.; Vermeulen, N. P.
E., Free energies of binding of R- and S-propranolol to wildtype
and F483A mutant Cytochrome P450 2D6 from molecular dynamics
simulations. Eur. Biophys. J., 2007, 36, 589 - 599.
[46] Venhorst, J.; Onderwater, R. C. A.; Meerman, J. H. N.; Commandeur,
J. N. M.; Vermeulen, N. P. E., Influence of N-substitution of
7-methoxy-4-(aminomethyl)-coumarin on Cytochrome P450 metabolism
and selectivity. Drug Metab. Disp., 2000, 28, 1524 - 1532.
Received: October 15, 2010 Revised: March 28, 2011 Accepted: April 29, 2011
EFFICIENT INCORPORATION OF PLASTICITY AND WATER MOLECULES INTO THE MOLECULAR
DOCKING APPROACH
37
Paper 4
Byeon I-J. , Ahn J., Mitra M., Byeon C-H., Hercík K., Hritz J., Charlton L., Levin J., Gronenborn A.M. NMR
structure of human restriction factor APOBEC3A reveals substrate binding and enzyme specificity.
Nat. Commun. 2013, 4, 1890
ARTICLE
Received 29 Jan 2013 | Accepted 12 Apr 2013 | Published 21 May 2013
NMR structure of human restriction factor
APOBEC3A reveals substrate binding and
enzyme specificity
In-Ja L. Byeon1,2, Jinwoo Ahn1,2, Mithun Mitra3, Chang-Hyeock Byeon1,2, Kamil Hercı´k3,w, Jozef Hritz1,w,
Lisa M. Charlton1,2, Judith G. Levin3,* & Angela M. Gronenborn1,2,*
Human APOBEC3A is a single-stranded DNA cytidine deaminase that restricts viral pathogens
and endogenous retrotransposons, and has a role in the innate immune response.
Furthermore, its potential to act as a genomic DNA mutator has implications for a role in
carcinogenesis. A deeper understanding of APOBEC3A’s deaminase and nucleic acid-binding
properties, which is central to its biological activities, has been limited by the lack of structural
information. Here we report the nuclear magnetic resonance solution structure of APOBEC3A
and show that the critical interface for interaction with single-stranded DNA substrates
includes residues extending beyond the catalytic centre. Importantly, by monitoring deaminase
activity in real time, we find that A3A displays similar catalytic activity on APOBEC3Aspecific
TTCA- or A3G-specific CCCA-containing substrates, involving key determinants
immediately 50
of the reactive C. Our results afford novel mechanistic insights into
APOBEC3A-mediated deamination and provide the structural basis for further molecular
studies.
DOI: 10.1038/ncomms2883
1 Department of Structural Biology, University of Pittsburgh School of Medicine, 3501 Fifth Avenue, Pittsburgh, PA 15261, USA. 2 Pittsburgh Center for HIV
Protein Interactions, University of Pittsburgh School of Medicine, Pittsburgh, PA 15261, USA. 3 Section on Viral Gene Regulation, Program on Genomics of
Differentiation, Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, MD 20892-
2780, USA. w Present addresses: Fraunhofer IME – Division Molecular Biology, Biological Operating Systems: Infectious Diseases, Forckenbeckstrabe 6 ,
52074 Aachen, Germany (K.H.); CEITEC, Masaryk University, Kamenice 5, 625 00 Brno, Czech Republic (J.H.). * These authors contributed equally to this
work. Correspondence and requests for materials should be addressed to A.M.G. (email: amg100@pitt.edu).
NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications 1
& 2013 Macmillan Publishers Limited. All rights reserved.
T
he human APOBEC3 (A3) proteins are a family of
deoxycytidine deaminases that convert dC residues in
single-stranded DNA (ssDNA) to dU, and act as DNA
mutators. These proteins, which have an important role in the
innate immune response, function as host restriction factors and
display a broad range of activities against endogenous and
exogenous retroelements1–3. There are seven members in the A3
family, each having one (A3A, A3C, A3H) or two (A3B, A3D,
A3F, A3G) zinc (Zn)-binding domains with HX1EX23-24CX2-4C
motifs, where X is any amino acid4. The histidine and cysteine
residues coordinate Zn2 þ , while glutamic acid is thought to
function as a proton shuttle during the deaminase reaction5.
The single-domain A3A protein, the subject of this study,
has multiple activities. A3A degrades foreign DNA introduced
into human cells6,7 and blocks replication of exogenous
viruses such as human papilloma virus8, Rous sarcoma virus9,
parvoviruses10,11 and human T-lymphotropic virus type 1
(ref. 12). In addition, it strongly inhibits retrotransposition of
LINE-1, Alu and LTR retroelements10,13–16, which cause
insertional mutations. Unlike A3G (ref. 17), A3A is capable of
deaminating 5-methylcytosine18,19, an epigenetic marker in
genomic DNA, as well as inducing cell cycle arrest20 and
somatic hypermutation of nuclear and mitochondrial DNAs in a
dynamic interplay between A3 editing and DNA catabolism20,21.
How all these activities are regulated, however, is not fully
understood, although Tribbles 3, a human protein, was recently
reported to protect nuclear DNA from A3A-mediated
deamination22.
A3A is highly expressed in cells of the myeloid lineage, such as
monocytes and macrophages, and its expression is upregulated by
treatment with interferon-alpha10,23–26. Interestingly, silencing of
A3A in monocytes is associated with increased susceptibility to
human immunodeficiency virus (HIV)-1 infection, suggesting
that the presence of A3A may be protective against HIV-1
(ref. 23). Recent studies indicate that endogenous A3A in
macrophages restricts HIV-1 replication by reducing synthesis
of viral DNA during reverse transcription27. This result is
consistent with an independent observation that HIV-1
transcripts in interferon-alpha-treated infected macrophages
seem to be edited predominantly by A3A (ref. 28).
Although abundant information on the biological activities of
the A3 proteins has been reported, only limited structural data are
available. For example, the structure of the C-terminal domain
(CTD) of A3G, which contains the catalytic centre for its
deaminase activity17,29,30, was solved by nuclear magnetic
resonance (NMR)31–33 and X-ray crystallography34,35, but the
structure of full-length A3G has been more elusive. While this
manuscript was in preparation, the X-ray structure of another
single-domain A3 protein, A3C, was reported36.
Given human A3A’s function as an inhibitor of retroviruses and
retroelements with significant effects on cellular activities, the
availability of an atomic, three-dimensional structure clearly is of
significant importance. Here we report the NMR solution structure
of human A3A, define the interface that is critical for its interaction
with single-stranded oligonucleotide substrates and characterize its
catalytic activity. Detailed analysis of A3A binding to nucleic acids
and real-time monitoring of the deamination reaction by NMR
allow us to propose a mechanism for substrate selection and
specificity. These studies provide the structural basis for a deeper
understanding of A3A’s biological activities and broaden our
knowledge of the molecular properties of the A3 proteins.
Results
Biochemical characterization of purified A3A. To ensure that
the purified, recombinant wild-type A3A (199 aa), (containing a
C-terminal His6-tag (LEHHHHHH)), was enzymatically active, a
uracil DNA glycosylase (UDG)-dependent gel-based deaminase
assay was performed (Fig. 1a,b). The catalytic activity of this
tagged protein was evaluated using a fluorescently labelled 40-nt
Substrate
(40 nt)
Product
(22 nt)
1 2 3 4 5 6 7 8 9 10 11
(A3A)0
0 200 400 600 800 1,000
50
100
150
[A3A] (nM)
%Deaminaseproduct
40-nt ssDNA
A3A-bound
DNA
Free DNA
40-nt ssRNA
Free RNA
[A3A] (μM)
A3A-bound
RNA
[A3A] (μM)12010090800 10 25 50 60 70 0 1 5 100755040302010
2011 12 13 14 15 16 17 18 1910987654321
Figure 1 | Deaminase activity and ssDNA and ssRNA binding of A3A. (a) Deamination of a 40-nt ssDNA as a function of A3A concentration measured in
a UDG-dependent assay. Reactions were performed as described in Methods with increasing concentrations of A3A. Lanes: 1, no A3A; 2, 20 nM; 3, 40 nM;
4, 60 nM; 5, 80 nM; 6, 100 nM; 7, 200 nM; 8, 400 nM; 9, 600 nM; 10, 800 nM; 11, 1000 nM, and relative amounts of substrate and product were assessed
by gel electrophoresis. A representative gel (of three independent assays) was chosen for the figure. (b) Quantification of the relative amounts of
deaminase product versus A3A concentration from gel analysis as shown in (a). The error bars represent the s.d. for three independent measurements.
(c,d) Binding of A3A to 40-nt ssDNA (c) or 40-nt ssRNA (d) evaluated by electrophoretic mobility shift assay (EMSA; details in Methods). The positions
of the A3A-bound as well as free DNA or RNA are indicated. A3A concentrations are listed under each lane. In each case, a representative gel (of five
independent assays) was chosen for the figure.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883
2 NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications
& 2013 Macmillan Publishers Limited. All rights reserved.
ssDNA substrate, containing the TTCA deaminase recognition
site, in the presence of increasing concentrations of protein,
ranging from 20 to 1000 nM. Over 50% of the substrate was
converted to the deaminated product with as little as 20 nM A3A,
and complete conversion was seen with 200 nM A3A. These
results demonstrate that the recombinant A3A protein is highly
active as a cytidine deaminase.
Binding of A3A to ssDNA and ssRNA was evaluated in
electrophoretic mobility shift assay experiments with 32P-labelled
40-nt oligonucleotides and varying concentrations of A3A
(Fig. 1c,d). Complexes are seen at A3A concentrations of Z50 mM
with ssDNA (Fig. 1c, lane 4) and Z20mM with ssRNA (Fig. 1d,
lane 15). To achieve B50% complexation, twofold more A3A
(80mM) was required for ssDNA (Fig. 1c, lane 7), compared with
ssRNA (40mM) (Fig. 1d, lane 17), suggesting that A3A has a
somewhat higher binding affinity for ssRNA than that for ssDNA.
Estimated Kd values of B80mM for the A3A–ssDNA complex
agree well with values derived from NMR titration data (see below).
NMR sample preparation and behaviour in solution. Fulllength
A3A was monomeric and soluble up to B0.2 mM at pH
6.5, 200 mM NaCl, and the 1H–15N heteronuclear single quantum
coherence (HSQC) spectrum exhibited well dispersed, narrow
resonances (Fig. 2). Nearly complete backbone and side-chain
assignments were obtained using unlabelled, uniformly 15N-, 13C/
15N- and 2H/13C/15N-labelled A3A with specific-protonations
(see Methods) for three different conditions: (1) B0.2 mM A3A
at pH 6.5, 200 mM NaCl (Fig. 2); (2) B0.2 mM A3A at pH 6.9;
and (3) B0.3 mM A3A at pH 8.1, all in 25 mM sodium phosphate
buffer. As essentially identical nuclear O¨ verhauser effect (NOE)
patterns and chemical shifts were observed for the different
conditions (only the amide resonances exhibited small, pHdependent
chemical shift differences), the structure was assumed
to be unaffected by these different conditions. A mutant A3A,
L63N/C64S/C171Q, was also used to help resolve ambiguous
assignments (see Methods).
A3A structure. The structure of full-length A3A was calculated
on the basis of 3279 NMR-derived experimental constraints. The
final model satisfies all experimental constraints, displays excellent
covalent geometry and the 30-conformer ensemble exhibits
atomic r.m.s. deviations of 0.60±0.05 and 1.18±0.05 Å with
respect to the mean coordinate positions for the backbone (N, Ca,
C0
) and all heavy atoms, respectively (Table 1). A stereoview of
the ensemble of conformers as well as a ribbon representation of
the lowest energy structure from the ensemble are depicted in
Fig. 3a,b, respectively.
Overall, the A3A structure consists of six helices surrounding a
central b sheet of five strands (Fig. 3a,b), common to all APOBEC
proteins whose structures are known; that is, A3G-CTD, A3C and
A2 (Supplementary Table S1). The structured region of A3A is
limited to residues 10–194 (Fig. 3a,b), as very intense, sharp
amide resonances, exhibiting random-coil chemical shifts (Fig. 2),
are present for the N-terminal nine (1–9) and final five (195–199)
residues in the A3A sequence as well as the His6-tag residues,
indicating a substantial degree of flexibility. In addition to the
termini, the loop connecting b2’ and a2 (loop 3, residues 57–70;
Fig. 3a) is also highly plastic and undergoes motions on an
0
10
20
30
40
33 34 35 36 37 38
12 10 8 6
130
125
120
115
110
105
T154
F102
L55
Y155
F95
I96 T93
T19
G178
E36
S187
V37
Y35
R91
H119
C34
A127
V46
L122
L33 E38
R39
R123
E72 S103
G170
G108 G147
G105
G65 G53
G198
G43
G25
G8
G188
G27
T118
S99
L84H182
Q169
D163 F54
L190
D180
N42
Q58
S97
V150 D14
K47
L40 H207
I1 7
K137
R189
Q149
E109
W162
A148
Q174
Q115
S4
F18
A146
F66
R74
L78
V120
A192
C106
L179
V79
N24
I26
H168
V166
D133
L140
Y132
R28
E116
T44 H160
A6
R121
H70
A126 W94
A185
C171
Y136
H16
S183
N57
L194
S7
L76
T164
S20
S81
H56 C64
W98
L73
W94ε
W176
F165
A71
E157
R128
V92
F125
W176ε
W162ε
Y130
D85
Q83
N23
W98ε
W104ε
Q169ε
N24δ
1H (p.p.m.)
15N(p.p.m.)
R52
A139
N117
I124
D41
D49
Q141
Y32
I152
M13
V110
D145
Y90
M153
A3
D167
D131
A112L200M48
Q169ε
D156
L186
F173
D177
W104
C161
I193
K30
N196δ
N24δ
9.0 8.8 8.6 8.4 8.2 8.0
123
122
121
120
119
118 H182
A107 S45D180
S151
R111
D145
D177
R69
Q197 E201
R10
L135
K60
H206
L63
Q88
K159
N199
T31
K30
D49
E181
Q184
L200
L12
L186
0
F173
R144
L82
E138
I89
M153
F75
I129
L114 A87
M48
N61
C101
Y90
I152
A112F18
Q149
W162
E109
C171
L190
L84 R191 D77
M142
N196
Q195
F158
F113
F22
Elution time (min)
Molecularmass(kDa)
L62
N199δ
N21
Figure 2 | A3A NMR assignments. 600 MHz 1H–15N HSQC NMR spectrum of 0.17 mM 13C/15N-labelled A3A in 25 mM sodium phosphate, 200 mM
NaCl, pH 6.5, 25 °C. Assignments are indicated by residue name and number. An expansion of the boxed region is provided in the lower left corner.
Size-exclusion chromatography/multi-angle light scattering data are shown in the inset in the upper left corner, with the elution profile shown with black
circles and the estimated molecular masses across the peak with blue triangles.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883 ARTICLE
NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications 3
& 2013 Macmillan Publishers Limited. All rights reserved.
intermediate (m-ms) timescale, as many amide resonances exhibit
severe line broadening, and are of low intensity (Q58, N61, L62)
or entirely missing (A59, Y67 and G68) in the 1H–15N HSQC
spectrum (Fig. 2). This is suggestive of multiple conformations
undergoing chemical exchange.
Comparison with other APOBEC protein structures. We carried
out a detailed comparison of the present A3A solution
structure with other available structures of APOBEC proteins
(Supplementary Table S1). Among these structures, the present
A3A NMR structure is most similar to the X-ray structure of the
A3G-CTD quintuple mutant (residues 191–384, PDB: 3IR2)35
(Fig. 3c,d), exhibiting an average pairwise backbone atomic r.m.s.
difference of 1.86 Å. The next closest structures are the X-ray
structures of wild-type A3C, another single-domain human A3
protein (1–190, PDB: 3VOW)36, and wild-type A3G-CTD
(residues 197–380, PDB: 3E1U/3IQS)34, both exhibiting average
backbone r.m.s. differences of B2.3–2.4 Å. The NMR solution
structures of monomeric A2 proteins (murine A2, PDB: 2RPZ
and a CS-HM Rosetta model for human A2 (ref. 37)) are also
similar to A3A (atomic r.m.s. differences of B3.4 Å). In the
previously solved A3G-CTD structures, some local differences
were reported31–34. Our current A3A structure clearly possesses
an interrupted b2 strand (b2-bulge-b20
; Fig. 3b and
Supplementary Fig. S1), as well as an N-terminal a-helix (a1,
residues 15–21).
A more detailed comparison was carried out between the
current A3A and the most similar A3G-CTD structures
(A3G191-384-2K3A; Fig. 3c,d and Supplementary Table S1).
We focused on the regions around the active site where the A3A
and A3G sequences differ. A two amino-acid (W104 and G105)
insertion, located between two Zn-coordinating cysteines (C101
and C106) in the active site of A3A, is the most prominent
sequence change. Despite this insertion, the positions of the
catalytic site residues H70 (H257 in A3G), E72 (E259), C101
(C288) and C106 (C291) are very similar in both structures
(Fig. 3c). The insertion, however, distorts the N-terminal end of
helix a3, with W104 bulging out, causing positioning of the
backbone and/or side chain atoms of S103 and G105 in A3A
equivalent to those of F289 and S290 in A3G. This structural
adjustment places the hydrophobic F102 and W104 side chains at
the protein surface, next to the active site. In contrast, A3G
possesses only one hydrophobic residue (F289) in this region.
Loop 7 is located near the active site (Fig. 3), and its C-terminal
half in A3G is known to have an important role in substrate
selection (TC or CC)38. In A3A, loop 7 (127–135) adds additional
hydrophobic residues (Y132, P134 and L135), whereas in A3G
two of the equivalent residues (D317 and R320) are polar, and the
proline is replaced by glycine, resulting in conformational
differences between the A3A and A3G loop 7 (Fig. 3c). Note
that among the human A3 proteins, the polar characteristic of
this region is unique for A3G (Fig. 3e).
A3A interacts with dCTP and dUTP. Although A3A is a wellknown
cytidine deaminase, the determinants of substrate specificity
(that is, preferential recognition of TC (refs 7,10,39)) have
not been characterized in detail. As an initial approach, we
investigated binding of A3A to dCTP, several ssDNA
oligonucleotide substrates and/or their dU-containing products
by NMR.
dUTP (Fig. 4a) and dCTP (Fig. 4e) binding was monitored by
1H–15N HSQC spectroscopy, and affected regions were mapped
onto the A3A structure (insets). Perturbations of amide
resonances of the active site amino acids H70, E72, C101 and
C106 as well as other surrounding ones (K30, L55, H56, N57,
Q58, K60, N61, L63, C64, H70, A71, W98, S99 and Y132) were
very similar, indicating that both the substrate and the product
bind to the active site. Several of the amino acids whose
resonances were affected reside in loop 3, with N61, L63 and C64
located distal to the active site, indicating that a remote
conformational change was induced by dCTP/dUTP binding or
that a particular preferred conformation was selected from the
flexible ensemble in the substrate-free A3A state. Titration curves
of selected perturbed resonances upon dCTP (Fig. 4e) and dUTP
(Fig. 4a) binding were used to extract Kd values of 536±72 mM
and 578±115 mM, respectively (Table 2). Note that as up to 30%
of the dCTP was converted to dUTP during the titration, the
value for dCTP should be regarded as an approximation.
A3A binds ssDNA substrates via an extended surface. The
interaction of 50
-ATTTCATTT-30
(B0.1–3 mM) with A3A
(B0.15 mM) under our NMR conditions resulted in conversion
to 50
-ATTTUATTT-30
within o1 min, rendering it impossible to
measure a true Kd value for 50
-ATTTCATTT-30
. However, given
the similarity between dUTP and dCTP binding to A3A (Fig. 4a
versus Fig. 4e), it seemed reasonable to assume that 50
-ATT
TUATTT-30
and 50
-ATTTCATTT-30
would also have similar
binding properties. The Kd value for the ATTTUATTT product
was 58±8 mM (Fig. 4b, Table 2), and its interaction surface on the
A3A structure (Fig. 4b, inset) covers a larger surface, extending
beyond the catalytic site.
This extended interface involves an area to the left side of
the active site, opposite of loop 3 (where residues are coloured in
hot pink (Dd40.050 p.p.m.) or cyan (Dd between 0.028 and
0.050 p.p.m.) in the ribbon diagram). Interestingly, the twoTable
1 | Statistics for the final 30 conformer ensemble of
A3A.
Number of NOE distance constraints
Intra-residue (i–j ¼ 0) 1114
Sequential (|i–j| ¼ 1) 655
Medium range (2r|i–j|r4) 309
Long range (|i-j|Z5) 749
Total 2827
Number of hydrogen bond constraints 152
Number of dihedral angle constraints
f 151
c 149
Total 300
Structural quality
Violations*
Distances constraints (Å) 0.016±0.002
Dihedral angles constraints (°) 0.299±0.037
Deviation from idealized covalent geometry
Bond lengths (Å) 0.001±0.000
Bond angles (°) 0.398±0.006
Improper torsions (°) 0.222±0.005
Average r.m.s.d. of atomic coordinates (Å)w
Backbone heavy atoms 0.60±0.05
All heavy atoms 1.18±0.05
Ramachandran plot analysis (%)z
Most favourable region 73.7±2.3
Additional allowed regions 22.5±2.6
Generously allowed regions 2.9±1.1
Disallowed regions 0.9±0.6
*No individual member of the ensemble exhibited distance violations 40.5 Å or dihedral angle
violations 45°.
wAverage r.m.s.d. of atomic coordinates for residues (11–57 and 70–194) with respect to the
mean structure. A3A regions (1–9, 58–69 and 195–199) were excluded from the statistics
because they exhibit a flexible, random-coil conformation.
zStatistics were calculated using PROCHECK for full-length A3A (residues 1–199).
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883
4 NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications
& 2013 Macmillan Publishers Limited. All rights reserved.
residue insertion (W104/G105), unique to A3A, is part of the
interface: the W104 backbone amide (Dd ¼ 0.122 p.p.m.) and the
W104 e1 side chain resonance (Dd ¼ 0.079 p.p.m.), as well as the
G105 (Dd ¼ 0.066 p.p.m.) and S103 (Dd ¼ 0.072 p.p.m.) amide
resonances exhibit the largest perturbations. In addition, loop 7
and helix a4 are also part of the extended interface, as effects on
the amide resonances of residues D133 (Dd ¼ 0.112 p.p.m.), L135
(Dd ¼ 0.102 p.p.m.) and E138 (Dd ¼ 0.079 p.p.m.) were observed.
Additional binding studies of A3A using a 15-nt oligonucleotide,
50
-ATTATTTUATTTATT-30
yielded essentially the same binding
site (Fig. 4c, inset) and affinity (Kd ¼ 57±11 mM; Fig. 4c, Table 2),
as found with the 9-mer.
Although smaller perturbations were observed for some additional
resonances, these changes were non-saturable and most
likely reflect non-specific binding (for example, the I17 resonance
marked by a dashed line in Fig. 4c). The non-specific binding site
of the 15-mer oligonucleotide mapped onto the A3A structure
involves residues 12–19 (a1), 36–37, 40–42 (b1), 44–49 (b2),
53–56 (b20
) and 179–186 (a6) (Supplementary Fig. S2). To
validate the binding data, the interaction of an A3A catalytic site
mutant, E72Q, with a 15-nt substrate 5-0
(ATTATTTCAT
TTATT)-30
was evaluated. The mutant exhibited the same affinity
(Kd ¼ 55±11 mM, data not shown) as wild-type A3A with the
50
-ATTATTTUATTTATT-30
product, confirming that binding of
substrate and product occurs with essentially identical affinities.
Further titration experiments using smaller oligonucleotides
(50
-ATTT(C/U)A-30
, Fig. 4d; 50
-TCATTT-30
, Fig. 4f) delineated a
binding site similar to the one observed with the nona-nucleotide.
This implies that three nucleotides, (TCA), constitute the
essential moiety for binding to A3A. The affinities of the
hexanucleotides, however, were approximately threefold
weaker (KdB190 mM) than the one measured for 50
-ATTT(C/U)
ATTT-30
(Kd ¼ 58±8 mM; Table 2), suggesting a stabilizing effect
of the flanking nucleotides in the complex.
Interaction of A3A with a 9-nt substrate, 50
-AAACCCA
AA-30
, containing the A3G deaminase-specific recognition site2,
maps to the same surface (Fig. 4g, inset) as 50
-ATTT(C/U)AT
TT-30
(Fig. 4b, inset). This suggests that both dT and dC at the
À 1 and À 2 positions, that is, pyrimidine bases, can interact
with the extended interface next to the catalytic site of A3A.
However, slightly weaker binding (Kd ¼ 91±15 mM, Fig. 4g,
Table 2) was observed for 50
-AAACCCAAA-30
compared with
50
-ATTTCATTT-30
. Replacement of the central CCC in 50
-AAAC
CCAAA-30
by CCA (50
-AAACCAAAA-30
, Fig. 4h) or CAA
(50
-AAACAAAAA-30
, Table 2) resulted in identical (Kd ¼
94±11 mM) or slightly weaker (Kd ¼ 161±19 mM) affinities,
respectively.
Structural model for A3A–ssDNA complexes. Model structures
of the A3A-oligonucleotide complexes were created by flexible
docking40 and selecting binding poses compatible with chemical
shift perturbations (see Methods for details). Initial docking
results for 50
-ATTTCATTT-30
and 50
-AAACCCAAA-30
indicated
that only the central region of the oligonucleotides interacts
specifically with A3A, leaving the two ends free. Therefore, we
only generated final structural models of A3A with the
pentanucleotides 50
-TTCAT-30
(Fig. 5a) and 50
-CCCAA-30
(Fig. 5b). In the A3A/TTCAT complex model, the central
reactive C occupies the deep pocket delineated by the active site
and surrounding residues T31, N57, Q58, H70, E72, W98, C101,
C106, D131 and the Zn2 þ ion. The thymidine (TÀ1) immediately
α2
α1
β1β2
α2
β3
β4
α3
α4
β5
α5
α6
β2′
b
α1
β1β2
β2′
α2
β3
β4
α3
α4
β5
α5
α6
C106
C101
H70
E72
Loop3
Loop3
W104 W104
F102 F102
F289F289
E72/E259E72/E259
H70/H257H70/H257
C106/C291 C106/C291α3
α2
C101/C288 C101/C288
S103 S103
S290S290
G105G105
Loop3 Loop3
a
NN
CC
C
N
C N
Loop7
Loop7
Loop7 Loop7
Y132Y132
D317 D317
P134 P134
L135
G319G319
L135
R320 R320
α4 α4
Loop7 Loop7
α3
Loop3Loop3
dc
e β4 loop 7 α4
A3A 120 VRLRIFAARIYDY-DPLYKEALQMLRDA 146
A3G 305 VSLCIFTARIYDD-QGRCQEGLRTLAEA 331
A3B 303 VRLRIFAARIYDY-DPLYKEALQMLRDA 329
A3C 114 VNLTIFTARLYYFQYPCYQEGLRSLSQE 141
A3D/E 310 VNLTIFTARLCYFWDTDYQEGLCSLSQE 337
A3F 297 VNLTIFTARLYYFWDTDYQEGLRSLSQE 324
A3H 102 LNLGIFASRLYYHWCKPQQKGLRLLCGS 129
Figure 3 | A3A NMR solution structure. (a) Stereoview of the final 30 conformer ensemble (N, Ca, C0
). Regions of helical and beta sheet structures are
coloured hot pink and royal blue, respectively, and the remainder of the structure in grey. (b) Ribbon representation of the lowest energy structure
of the ensemble, using the same colour scheme as in (a). Secondary structure elements are labelled and the active site residues (H70, E72, C101 and C106)
and the Zn2 þ ion are shown in ball-and-stick representation with carbon, nitrogen, oxygen, sulphur and Zn atoms in green, blue, red, yellow and brown,
respectively. (c) Stereoview of the superimposition of the active site regions of the current A3A NMR and the A3G-CTD (PDB: 3IR2) X-ray structures. The
backbone traces of A3A and A3G-CTD are coloured grey and khaki, respectively. Side chains are shown in ball-and-stick representation with carbon,
nitrogen, oxygen, sulphur and Zn atoms of A3A and A3G in green, blue, red, yellow and brown, and pale green, cyan, pink, yellow and orange, respectively.
A3A residues are labelled in bold. (d) Ribbon representation of the A3G-CTD (A3G191-384-2K3A, PDB: 3IR2) X-ray structure35. (e) Amino-acid sequences
of the loop 7 region in different A3 proteins. Large, hydrophobic residues are highlighted in yellow and the polar residues D317 and R320 in A3G are
highlighted in cyan.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883 ARTICLE
NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications 5
& 2013 Macmillan Publishers Limited. All rights reserved.
preceding the C interacts with a surface formed by residues D133,
P134, L135 (loop 7) and F102 (loop 5), whereas TÀ2 contacts
residues F102, S103, W104, and G105 (loop 5), L135 (loop 7) and
E138 (a4). The adenosine (Aþ 1) at the 30
side of the C interacts
with residues D131, Y132 and D133 (loop 7). The 5-methyl
groups of TÀ 1 and TÀ 2 are in close contact with several
hydrophobic residues, for example, F102, W104 and L135,
suggesting that this hydrophobic interaction may contribute to
tighter binding of TTCA- compared with CCCA-containing
ssDNA. The model for the A3A/CCCAA complex (Fig. 5b) is
very similar to that of A3A/TTCAT (Fig. 5a), including the
localization of the individual nucleotides.
Chemicalshiftchange(p.p.m.)
–0.12
–0.08
–0.04
0.00
0.04
0.08
0.12
0.0 0.5 1.0 1.5 2.52.0 3.0
–0.10
–0.08
–0.06
–0.04
–0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.0 0.5 1.0 1.5 2.0
[dATTTUA] (mM)
Chemicalshiftchange(p.p.m.)
0.10
–0.10
–0.08
–0.06
–0.04
–0.02
0.00
0.02
0.08
0.06
0.04
0.0 1.00.5 1.5 2.0 3.02.5
H70 1HN
G105 1HN
–0.06
–0.04
–0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
K30 1HN
T31 1HN
C64 1HN
Chemicalshiftchange(p.p.m.)
3.5
–0.06
–0.04
–0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.0 0.5 1.0 1.5 2.0 2.5 3.0 4.0 4.5 5.0
K30 1HN
G65 1HN
D131 1HN
–0.08
–0.06
–0.04
–0.02
–0.00
0.02
0.04
0.06
0.08
0.0 0.5 1.0 1.5 2.0
Chemicalshiftchange(p.p.m.)
H70 1HN
G105 1HN
[dTCATTT] (mM)
W104 1Hε1
T31 1HN
W98 1Hε1
W104 1Hε1
F66 1HN
W104 1Hε1
I17 1HN
T31 1HN
[dATTATTTUATTTATT] (mM)
Chemicalshiftchange(p.p.m.)
E72 1HN
W98 1Hε1
W104 1Hε1
G65 1HN
[dUTP] (mM) [dCTP] (mM)
Chemicalshiftchange(p.p.m.)
–0.09
–0.07
–0.05
–0.03
–0.01
0.01
0.03
0.05
0.07
0.09
0.0 0.5 1.0 1.5 2.0
Chemicalshiftchange(p.p.m.)
W98 1Hε1
[dAAACCAAAA] (mM)
N57 1HN
F66 1HN
W104 1Hε1
–0.08
–0.06
–0.04
–0.02
0.00
0.02
0.04
0.06
0.08
0.0 0.5 1.0 1.5
[dAAACCCAAA] (mM)
Chemicalshiftchange(p.p.m.)
W104 1Hε1
H70 1HN
F66 1HN
[dATTTUATTT] (mM)
Loop 3
C N
Loop3
C N
Loop 3
C N
Loop 3
C N
Loop 3
C N
Loop 3
C N
Loop 3
C N
Loop 3
C N
W104 W104
W104
W104
W104
W104
Loop 7 Loop 7
Loop 7
Loop 7
Loop 7
Loop 7
Figure 4 | Binding of A3A to mononucleotides and ssDNAs. Titration curves for representative HN resonances and binding site mapping (inset) for
binding of dUTP (a), ATTTUATTT (b), ATTATTTUATTTATT (c), ATTTUA (d), dCTP (e), TCATTT (f), AAACCCAAA (g) and AAACCAAAA (h). A3A
residues whose resonances exhibit large 1H,15N-combined chemical shift changes upon nucleotide addition are coloured red (40.050 p.p.m.) and
orange (0.028–0.050 p.p.m.). Those only affected by ssDNA binding, but not by dCTP/dUTP, are shown in dark pink (40.050 p.p.m.) and cyan
(0.028–0.050 p.p.m.). All 1H–15N HSQC spectra were recorded at 25°C using 25 mM sodium phosphate buffer, pH 6.9. 1H,15N-combined chemical shift
changes were calculated using
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Dd2
HN þ ðDdN=6Þ2
q
, with DdHN and DdN, the 1HN and 15N chemical shift differences observed for A3A before and after
adding ligands (B90% saturation).
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883
6 NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications
& 2013 Macmillan Publishers Limited. All rights reserved.
Deamination by real-time NMR. The A3A-catalyzed deamination
reaction with several substrates was followed by two-dimensional
(2D) 1H–13C HSQC (Fig. 6a) or 1D 1H (Fig. 6b–d) NMR. Using
dCTP as the substrate, we determined that slow but measurable
deaminase activity is present (Fig. 6a): at 25°C, 5.2mM dCTP was
completely converted to dUTP by 170 mM A3A in B50h. Longer
substrates, such as the 9-nt 50
-ATTTCATTT-30
, were deaminated
much more rapidly (Fig. 6b), and in such cases 1D 1H NMR
spectroscopy was used (Fig. 6b inset). The 9-nt substrate (0.98mM)
was completely deaminated by 0.196mM A3A in B1.7h, with an
initial rate of B0.8mM hÀ 1. Other substrates, a 15-mer, 50
-ATTA
TTTCATTTATT-30
, and two hexanucleotides, 50
-ATTTCA-30
and
50
-TCATTT-30
, (Supplementary Fig. S3a–c, respectively) exhibited
essentially identical rates to that observed with the nona-nucleotide.
A3A and A3G purportedly possess a difference in substrate
specificity2,6,7,10,39,41. To quantify A3A’s substrate specificity, we
also assayed the deamination of an A3G-specific substrate,
AAACÀ 2CÀ 1CAAA, by A3A. Interestingly, all three cytidines
were deaminated in a sequential 30
-50
manner (Fig. 6c), with the
third dC (C) converted with an initial rate of B0.2mM hÀ 1, only
fourfold slower than the dC in the A3A-specific 50
-ATTTC
ATTT-30
substrate. When comparing substrates with varying
numbers of cytosines, such as 50
-AAACCCAAA-30
(Fig. 6c),
50
-AAACCAAAA-30
(Fig. 6d) and 50
-AAACAAAAA-30
(Supplementary Fig. S3d), deamination of the 30
dC (C) was
about two- to four-fold faster in 50
-AAACCCAAA-30
than in the
other two substrates, indicating a preference for two pyrimidine
bases at the 50
side of the reactive C.
We also noted that the kinetics of the C-U conversions of CÀ 1
and CÀ 2 are much slower, quite distinct from that of C in the
reaction with the AAACÀ 2CÀ 1CAAA substrate. Similar behaviour
was observed with AAACÀ 2CÀ 1UAAA (Supplementary Fig. S3e),
a substrate containing a uridine instead of the most reactive C. In
contrast, when dC is followed by dA, and not by dU, for example,
the C in 50
-AAACÀ 1CAAA-30
(Fig. 6d) and 50
-AAACAAAAA-30
(Supplementary Fig. S3d), the reaction proceeds readily, albeit
somewhat slower than with AAACÀ 2CÀ 1CAAA (Fig. 6c).
Given our observation that dC and dU essentially possess the
same A3A-binding affinities and interfaces, it seems reasonable to
assume that once dC is converted to dU, competitive binding
and inhibition occurs, slowing the deamination of CÀ 1 and CÀ 2 in
AAACÀ 2CÀ 1UAAA. Therefore, optimal substrates for A3A
contain a cytidine preceded by two pyrimidine bases and
followed by an adenine; that is, (T/C)(T/C)CA.
Further studies were performed to determine the catalytic
constants for A3A deaminase activity on single cytidine-containing
substrates; that is, 50
-ATTTCATTT-30
, 50
-ATTATTTCATTTA
TT-30
, 50
-ATTTCA-30
, 50
-TCATTT-30
and 50
-AAACAAAAA-30
.
The kinetics of deamination revealed that all TCA-containing
oligonucleotide substrates, regardless of length, exhibited identical
turnover rates (kcatB60–70 minÀ 1; Table 2, Fig. 6e–g). The
hexanucleotides, however, displayed an approximately threefold
lower apparent second-order rate constant (kcat/KMB5 Â
103 MÀ 1 sÀ 1) than the one measured for the 9-nt and 15-nt
substrates (kcat/KMB1.8 Â 104 MÀ 1 sÀ 1), caused mainly by an
increase in KM. For all A3A substrates, KM and Kd values were very
similar. A large reduction (B40-fold) of kcat/KM was seen only with
a substrate that is lacking pyrimidine bases at the 50
side of the
reactive C; that is, 50
-AAACAAAAA-30
, caused by a large (13-fold)
reduction in kcat and a small (approximately threefold) increase in
KM (less-favourable binding), highlighting the importance of a
pyrimidine base at the 50
side.
Collectively, the data in Fig. 6 demonstrate that the nucleotide
context surrounding the reactive C is a major determinant of
enzymatic activity.
Discussion
Here we present the NMR solution structure of human A3A as
well as a detailed analysis of its nucleic acid interaction surface,
providing new insights into substrate selection and binding. The
A3A NMR structure very closely resembles the A3G-CTD and
A3C X-ray structures31–36 (Supplementary Table S1) and not
surprisingly, there are also similarities between the activities of
Table 2 | Dissociation and catalytic constants for A3A
interaction with ss deoxymono- and deoxyoligonucleotides.*
KM kcat/KM
Nucleotide sequences Kd (lM) (lM) kcat (minÀ 1) (MÀ 1 sÀ 1)
dCTP 536±72 600w 0.033±0.003w 0.91
dUTP 578±115
50
-dATTTCATTT-30
58±8z 66±7 71±6 1.8 Â 104
50
-dATTATTTCATTTATT-30
57±11z 62±5 66±5 1.8 Â 104
50
-dATTTCA-30
184±20z 210±17 65±6 5.2 Â 103
50
-dTCATTT-30
194±18z 219±29 55±5 4.2 Â 103
50
-dAAACCCAAA-30
91±15z 100w 20±3y 3.3 Â 103
50
-dAAACCAAAA-30
94±11z 100w 13±4y 2.2 Â 103
50
-dAAACAAAAA-30
161±19z 192±32 5.5±1.1 4.7 Â 102
*Dissociation constants (Kd) and catalytic constants (Michaelis–Menten) derived from NMR
data at 25 °C (pH 6.9), as described in Methods.
wAs accurate measurements of KM are precluded when more than one dC is present, the KM
values of these substrates are assumed to be similar to the Kd values and are set to 100 (Kd
values rounded up to the nearest hundred). This assumption is based on data obtained with
substrates containing a single dC.
zThe Kd values for these dC-containing oligonucleotides reflect values for their dU-containing
products, as these substrates are rapidly deaminated under the NMR experimental conditions
(A3AB0.1 mM).
ykcat values are derived from the initial rate in a sample with 1000 mM substrate and 0.197 mM
A3A.
T–1
A+1
Loop 3
W104G105
E138
L135
D133
C
E72
H70
N57
F102
Loop 7
α3
α4
T–2
T–1
A+1
Loop 3
W104G105
E138
L135
D133
C
E72
H70
N57
F102
Loop 7
α3
α4
T–2
C–1
A+1
Loop 3
W104G105
E138
L135
D133
C
E72
H70
N57
F102
Loop 7
α3
α4
C–2
C–1
A+1
Loop 3
W104G105
E138
L135
D133
C
E72
H70
N57
F102
Loop 7
α3
α4
C–2
α2α2
α2α2
Figure 5 | Model of A3A complexed with TTCAT or CCCAA. Stereoviews
of the A3A backbone structure and the 5-nt ssDNAs (TTCAT (a) and
CCCAA (b)), as well as interacting residues (from Fig. 4b or Fig. 4g) in balland-stick
representation, with carbon, nitrogen, oxygen, phosphorus and Zn
atoms in green, blue, red, gold and brown, respectively. The 30
-end
thymidine (a) and adenosine (b), which appear to be random, are omitted
for clarity.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883 ARTICLE
NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications 7
& 2013 Macmillan Publishers Limited. All rights reserved.
A3A and the A3G-CTD. Using NMR, we show that A3A binds
TTCA- or CCCA-containing single-stranded oligonucleotides
(Z9 nt) with Kd values ranging from 50–100 mM (Table 2), in
excellent agreement with binding affinity data (B80 mM)
estimated from electrophoretic mobility shift assay (Fig. 1c) and
with fluorescence depolarization results obtained by Love et al.39
Intriguingly, A3A binds ssDNA much more weakly (B1000fold)
than A3G, for which Kd values ranging from 50 to 240 nM
have been reported17,39,42–44. However, like A3A, Kd values
(200–450 mM) were obtained for the single-domain A3G-CTD
protein31,32, implying that the tighter binding of full-length A3G
is associated with its double domain structure. Indeed, the A3GN-terminal
domain contains numerous positively charged amino
acids that contribute to high efficiency binding to ssDNA17,29,
whereas both A3A and the A3G-CTD are slightly acidic and are
unable to interact in this manner.
Importantly, titration of A3A with diverse ssDNA substrates by
NMR made it possible to distinguish specific binding from nonspecific
binding (Fig. 4c). Our results showed that A3A can bind
both TTCA- and CCCA-containing oligonucleotides, using the
same five A3A contacts, namely the active site, loop 3, loop 5
including the exposed dipeptide W104-G105, loop 7 and helix a4.
Surprisingly, although the W104NeH side chain resonance
experiences the largest chemical shift changes upon substrate
binding (Fig. 4), W104 mutations do not seem to influence or
abrogate A3A’s catalytic activity16,18.
Note that A3A oligonucleotide-binding regions are highly
localized and clustered around the active site, in stark contrast to
0.0
0.2
0.4
0.6
0.8
1.0
0.0 0.5 1.0 1.5 2.0
0.0
0.2
0.4
0.6
0.8
1.0
0 5 10 15 20 25
0.0
0.2
0.4
0.6
0.8
1.0
0 5 10 15 20 25
[dCTP](mM)
0
1
2
3
4
5
0 10 20 30 40 50 60
6.2 6.1 6.0 5.9
106
104
102
100
98
dCTP
dUTP
2.7 h
106
104
102
100
98
6.2 6.1 6.0 5.9
dCTP
dUTP
7.9 h
6.2 6.1 6.0 5.9
dCTP
dUTP
16.8 h
6.2 6.1 6.0 5.9
dUTP
47.9 h
1
H (p.p.m.) 1
H (p.p.m.)
13
C(p.p.m.)13
C(p.p.m.)
[AAAC-2C-1CAAA](mM)
Time (h)
[AAAC-1CAAAA](mM)
Time (h)
[ATTTCATTT](mM)
Time (h)
5.85.9 p.p.m
DC dU
0.0 min
5.7 min
44.9 min
84.0 min
98.9 min
C
C-1
C
C-1
C-2
Time (h)
0
10
20
30
40
50
60
70
80
90
0
10
20
30
40
50
60
70
80
90
0
10
20
30
40
50
60
70
80
90
0
100
200
300
400
500
600
700
0
100
200
300
400
500
600
700
0
100
200
300
400
500
600
700
[ATTTCATTT] (μM)
Initialrate(μMh–1
)
[ATTATTTCATTTATT] (μM)
Initialrate(μMh–1
)
[ATTTCA] (μM)
Initialrate(μMh–1
)
y=87.1x/(53.7+x); R 2
=0.998
[A3A]=20 nM
y=92.8x/(45.7+x); R2
=0.994
[A3A]=20 nM
y=75.2x/(183+x); R2
=0.999
[A3A]=20 nM
y=16.7x/(66.3+x); R 2
=0.996
[A3A]=4.5 nM
y=16.0x/(62.4+x); R2
=0.998
[A3A]=4.5 nM
y=34.2x/(210+x); R2
=0.998
[A3A]=9 nM
Figure 6 | A3A-catalyzed deamination of dCTP and several ssDNA substrates monitored by real-time NMR. Concentrations of dCTP (a), 9-nt ssDNA
ATTTCATTT (b), AAACÀ 2CÀ 1CAAA (c) and AAACÀ 1CAAAA (d) versus incubation time are provided. All concentrations of unreacted substrates
(cytidine) and end products (uridine) were determined by measuring the intensities of the 13C-5-1H resonances of cytosine and uracil in the 2D 1H–13C
HSQC spectra (a; 600 MHz) or the 1H-5 resonances in 1D 1H spectra (b–d; 900 MHz), as a function of time. The A3A concentration was 0.17 mM (a) or
0.197 mM (b–d). Best fit curves are shown by a solid line. Representative 2D 1H–13C HSQC (a) or 1D 1H NMR (b) spectra acquired at the indicated times are
shown in the inset. (e–g) Kinetics of A3A-catalyzed deamination reactions for ATTTCATTT (e), ATTATTTCATTTATT (f) and ATTTCA (g); initial reaction
rates (o5%) are plotted versus substrate concentrations. Reactions were monitored by 1D 1H real-time NMR (900 MHz) at 25 °C in 25 mM sodium
phosphate buffer, pH 6.9. Two different A3A concentrations were used: (e and f), 4.5 nM (circles) and 20 nM (triangles); (g), 9.0 nM (circles) and 20 nM
(triangles). Two independent experiments were performed with 4.5 and 9 nM A3A and average values with s.d. are shown. Reactions with the higher A3A
concentration (20 nM) resulted in KM and kcat values very similar to those obtained with low A3A concentrations (4.5 or 9 nM).
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883
8 NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications
& 2013 Macmillan Publishers Limited. All rights reserved.
results in the two studies that mapped ssDNA binding to A3GCTD
(refs 31,32). Interestingly, our A3A–ssDNA complex model
(Fig. 5) suggests that the DNA bends to insert the reactive
cytidine into the active site, permitting only the immediate
neighbouring ( À 1, À 2 and þ 1) nucleotides to interact locally
near the catalytic site. Interestingly, there are reports of RNA
bending in the crystal structure of a transfer RNA adenosine
deaminase/RNA complex45 and DNA contraction during A3G
scanning of the ssDNA substrate46.
NMR real-time kinetic data for the deamination reaction of
A3A using 50
-ATTTCATTT-30
and 50
-ATTATTTCATTT-30
as
substrates yielded values of kcatB70 minÀ 1 and KMB60 mM,
which differ from the values (kcatB15 minÀ 1 and KMB230 nM)
obtained using a 43-nt TTCT-containing ssDNA18. These
differences may be related to variations in the ssDNA
sequences and experimental conditions. Interestingly, Carpenter
et al. reported that A3A is a stronger (B200-fold) deaminase
than full-length A3G (ref. 18). Indeed, although similar amounts
of deaminase product were observed in the NMR deamination
assays described here for A3A (Fig. 6b) and for A3G-CTD in
Furukawa et al.32, the amount of A3A used was 1000-fold less
than the amount of A3G-CTD.
The fact that A3A also deaminates the most 30
dC (C) in
50
-AAACCCAAA-30
(thought to be an A3G-specific substrate2)
with only an approximate fivefold reduction in kcat/KM, compared
with TTCA-containing 9-nt and 15-nt oligonucleotides (Table 2)
appears puzzling at first. However, these results are in agreement
with the deaminase specificity observed for A3A in cell-based
assays. For example, in an investigation of foreign DNA
restriction in human primary cells, A3A preferentially
deaminated TC and CC sequences in the green fluorescent
protein gene of the transfected plasmid DNA (ref. 6). In addition,
a strong bias for deamination of TC (B50–70%) and CC (B15–
25%) dinucleotide sequences was detected in nascent HIV-1
complementary DNAs isolated from infected macrophages28 and
in an in vitro HIV-1 model replication assay, performed using
purified A3A protein39. Furthermore, a preference for TC and CC
substrate recognition sites was also observed for A3A editing of
human T-lymphotropic virus type 1 (ref. 12). Thus, A3A’s active
site possesses the flexibility to accommodate and deaminate dC
residues (C) in both TC and CC dinucleotides.
Interestingly, in a study by Shinohara et al.7, it was shown that
A3A mediates genomic DNA editing in human cells, but no
editing site preference was detected. In contrast, Suspe`ne et al.21
found that A3A preferentially deaminates cytidines in TC and CC
dinucleotides in genomic DNA, when cells are exposed to UDG
inhibitor. APOBEC-mediated genomic DNA mutations have
been implicated in carcinogenesis47 and, for example, A3B was
shown to be a source of DNA mutations in breast cancer48. These
observations suggest that the strong mutagenic potential of A3A
might be detrimental to the stability of the human genome.
Thus, the dual function of A3A as a host restriction factor and
as a DNA mutator that can potentially act on genomic DNA, an
activity that may be associated with malignancies, suggests that
A3A can act as a ‘double-edged sword’. The high resolution NMR
structure of A3A presented here is a first step in aiding future
structure-function studies for addressing these seemingly diverse
A3A functions. Furthermore, the addition of the A3A structure to
the still limited list of currently known APOBEC structures
contributes to efforts towards elucidating the molecular mechanisms
of the innate immune response.
Methods
Protein expression and purification. Wild-type (Accession number
NM_145699), E72Q and L63N/C64S/C171Q mutant synthetic A3A genes with a
C-terminal His6-tag (LEHHHHHH) were inserted into the NdeI–XhoI site of the
pET21 plasmid (Novagen) for expression in Escherichia coli Rosetta 2 (DE3).
Uniform 15N- and 13C-labelling of the proteins was carried out by growth in
modified minimal medium at 18 °C, using 15NH4Cl and 13C6-glucose as the sole
nitrogen and carbon sources, respectively. Uniform 2H-, 15N- and 13C-labelling of
the proteins was achieved using 2H2O, 15NH4Cl and 13C6/2H7-glucose as deuterium,
nitrogen and carbon sources, respectively, with two different selective protonation
of the side chains of (1) Tyr/Phe/Ile residues and (2) Tyr/Phe/Trp/Ile/Val/
Leu residues, by adding 0.10–0.15 mg of 13C/15N-tyrosine, -phenylalanine and isoleucine
(for sample 1), and 13C/15N-tyrosine and -phenylalanine, unlabelled
tryptophan, 2-keto-butyrate (1,2,3,4-13C, 98%; 3,3’-2H, 98%, CIL, Andover, MA,
USA) and 2-keto-3-methyl-butyrate (1,2,3,4-13C, 99%; 3,4,4’,4’’-2H, 98%, CIL; for
sample 2), respectively. These chemicals were added to the culture 1 h before
induction with 0.4 mM isopropyl-1-thio-b-D-galactopyranoside (total induction
time ¼ 16 h). Proteins were purified over a 5-ml Hi-Trap His column (GE
Healthcare) and Hi-Load Superdex 200 (1.6 cm  60 cm) column, equilibrated in
buffer containing 25 mM Tris–HCl (pH 7.5), 50 mM NaCl, 5% glycerol, 2 mM
dithiothreitol (DTT) and 0.02% NaN3. Fractions containing A3A were further
purified over an 8-ml MONO-Q column (GE Healthcare) in 25 mM Tris-HCl
buffer (pH 8.5), 5% glycerol, 2 mM DTT and 0.02% sodium azide, employing a
linear gradient of 0–1 M NaCl. The final A3A preparations were 499% pure, as
estimated by SDS–polyacrylamide gel electrophoresis. The molecular mass of the
A3A proteins were confirmed by LC-ESI-TOF mass spectrometry (Bruker
Daltonics, Billerica, MA, USA).
Multi-angle light scattering. Size-exclusion chromatography/multi-angle light
scattering data were obtained at room temperature using an analytical Superdex
200 (S200) column with in-line multi-angle light-scattering refractive index (Wyatt
Technology, Inc., Santa Barbara, CA, USA) and ultraviolet (Agilent Technologies,
Santa Clara, CA) detectors. One hundred microlitres of 78.4 mM A3A were applied
to the S200 column pre-equilibrated and eluted with 25 mM sodium phosphate
buffer (pH 6.5), 200 mM NaCl, 0.02% sodium azide and 1 mM DTT at a flow rate
of 0.5 ml minÀ 1.
Deaminase assay using fluorescent-tagged ssDNA substrates. Deaminase
assay conditions were adapted from Iwatani et al.17 Forty microlitres reactions,
containing 180 nM of a 40-nt ssDNA (JL913, 50
-ATT ATT ATT ATT ATT ATT
ATT TCA TTT ATT TAT TTA TTT A-30
), labelled at its 50
end with Alexa Fluor
488, (Integrated DNA Technologies (IDT, Coralville, IA, USA) and varying
amounts of A3A in 10 mM Tris–HCl buffer, pH 8.0, 50 mM NaCl, 1 mM DTT,
1 mM EDTA, pH 8.0, and 10 units of E. coli UDG (New England BioLabs) were
incubated at 37 °C for 1 h. The reaction was stopped by incubation with Proteinase
K (40 mg, Ambion) at 65 °C for 20 min, followed by sequential addition of 10 ml of
1 N NaOH for 15 min at 37 °C and 10 ml of 1 N HCl. Ten microlitres aliquots of the
final mixture were subjected to electrophoresis in a 10% denaturing polyacrylamide
gel. Gels were scanned in fluorescence mode on a Typhoon 9400 Imager and the
data were quantified using ImageQuant software (GE Healthcare).
Electrophoretic mobility shift assay. Ten microlitres reactions, containing
varying amounts of A3A, 20 nM of a 50 32P-labelled 40-nt ssDNA (JL895, identical
sequence to JL913 but without the Alexa Fluor 488 label; Lofstrand Labs Ltd,
Gaithersburg, MD, USA) or 40-nt ssRNA (JL931, identical sequence to JL895,
except that U was substituted for T; IDT), were incubated with 4 U SUPERase IN
(Ambion) at 37 °C for 10 min in 50 mM Tris–HCl buffer, pH 7.0, 100 mM NaCl,
1 mM DTT, 1% Ficoll-400 and 2.5 mM EDTA. Aliquots from each reaction were
loaded onto an 8% native polyacrylamide gel in 40 mM Tris-acetate buffer, pH 8.4,
1 mM EDTA and 5% glycerol. One microlitre DNA Loading Gel Solution (Quality
Biological, Inc.), containing bromphenol blue and xylene cyanol, was added to a
control sample without A3A. Gels were run at 4 °C (5 mA) until the bromphenol
blue dye had migrated B2/3 through the gel. Radioactive products were detected
with a Typhoon 9400 Imager and were quantified using ImageQuant software.
NMR spectroscopy. All NMR spectra for the structure determination of A3A were
recorded at 25 °C on Bruker AVANCE900, AVANCE800, AVANCE700 and
AVANCE600 spectrometers, equipped with 5 mm triple resonance, Z-axis gradient
cryoprobes. The NMR samples contained unlabelled, 13C/15N- or 2H/13C/15Nlabelled
A3A with two types of selective protonations (see above). At pH 6.9 and
8.1, A3A showed higher solubility (B0.5 and B1 mM, respectively) than that at
pH 6.5 (B0.2 mM). However, at these higher protein concentrations, soluble
aggregation occurred, as evidenced by severe line broadening. We therefore performed
all of our NMR experiments at concentrations of B0.2–0.3 mM. The
sample temperature in the spectrometer was calibrated with 100% methanol.
Backbone and side chain resonance assignments were carried out using 2D 1H–15N
HSQC, 1H–13C HSQC and nuclear O¨ verhauser enhancement spectroscopy
(NOESY) and three-dimensional (3D) HNCACB, HN(CO)CACB, HNCA,
HN(CO)CA and HCCH-total correlation spectroscopy experiments49. Distance
constraints were derived from 3D simultaneous 13C- and 15N-edited NOESY
(ref. 50) and 2D NOESY experiments. All NOESY spectra were acquired at 800 or
900 MHz, using a mixing time of 100 ms (non-perdeuterated samples) or 150 ms
(perdeuterated samples). Spectra were processed with TOPSPIN 2.1 (Bruker) and
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883 ARTICLE
NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications 9
& 2013 Macmillan Publishers Limited. All rights reserved.
NMRPipe51, and analyzed using SPARKY3 (version 3.113; T.D. Goddard and D.G.
Kneller, University of California, San Francisco) and NMRView J (version 8.0.3)52.
NMR structure calculation. All NOE cross peaks were assigned from the 3D and
2D NOESY spectra using the SPARKY3 assignment tool. Structure calculations
were performed for A3A residues 1–199, using the anneal.py protocol in XPLORNIH
(ref. 53). An iterative approach with extensive, manual cross-checking of all
distance constraints against the NOESY data and the generated structures was
employed. The final number of the NMR-derived constraints were 3279, with 2827
NOE distances, 152 H-bond distances identified from NOE patterns for helices and
b-sheets, and 300 f and c backbone torsion angles from TALOS calculations54. In
addition, a Zn2 þ ion was added at a late stage in the calculations to coordinate
with H70, C101 and C106 (refs 31,32) using constraints based on the X-ray
structures of A3G-CTD (refs 34,35). Five hundred and twelve structures were
generated and the 30 lowest energy structures were selected and analyzed using
PROCHECK-NMR (ref. 55) (Table 1). All structure figures were generated with
MOLMOL (ref. 56).
A3A nucleotide-binding site mapping and titration by NMR. To monitor
nucleotide binding to A3A, aliquots of 14–100 mM mono- or oligodeoxynucleotide
stock solutions were added to 0.10–0.18 mM 13C/15N-labelled A3A. 100 mM dCTP
and dUTP stock solutions were purchased from Promega (Madison, WI, USA) and
HPLC-purified DNA oligonucleotides were obtained from IDT or Midland Co.
(Midland, TX, USA). A series of 2D 1H–15N HSQC titration spectra were acquired
and binding isotherms were obtained by plotting 1HN proton chemical shift
change versus nucleotide concentrations for 5–8 unambiguously traceable amide
resonances. Dissociation constants were calculated by non-linear best fitting of the
isotherms using KaleidaGraph (Synergy Software, Reading, PA, USA), and values
for 5–8 resonances were averaged.
Real-time studies of A3A-catalyzed deamination. A series of 2D 1H–13C HSQC
and/or 1D 1H NMR spectra were acquired as a function of time, after addition of
A3A to solutions of mono- or oligodeoxynucleotide. Concentrations of A3A and
deoxynucleotides for the different samples are provided in the figure captions. The
intensities of well-resolved 13C-5-1H resonances (volumes in 2D 1H–13C HSQC
spectra) and/or 1H-5 resonances (integrals in 1D 1H spectra) of cytosine and uracil
were used for quantification. Real-time monitoring of the A3A-catalyzed deamination
reaction by NMR permitted the extraction of initial (o5% dC-dU conversion)
rates for a series of substrate concentrations (50, 100, 200, 400 and
600 mM). kcat (Vmax/[A3A]) and KM values were obtained using the Michaelis–
Menten module in Kaleidagraph.
Molecular docking. Structures of 50
-TTCAT-30
, 50
-CCCAA-30
, 50
-ATTTCATTT-30
and 50
-AAACCCAAA-30
ssDNAs were generated by MacroMoleculeBuilder (version
2.8)57. The oligonucleotide and A3A NMR structures were converted to a mol2
file format using the Hermes programme (version 1.4) and standard Tripos atom
and bond types. Docking was performed with the flexible docking programme
GOLD version 5.1 (ref. 40), in combination with the Chemscore scoring function58.
The Zn2 þ ion was chosen as the centre point in the docking and the radius was set to
20 Å. For each nucleotide, five independent docking runs into the five lowest energy
NMR structures of A3A, each producing fifty binding poses, were performed using a
population of 100 ligands, with a maximal number of genetic algorithm operations
dependent on ligand size: 50,000 for 50
-TTCAT-30
and 50
-CCCAA-30
, and 150,000
for 50
-ATTTCATTT-30
and 50
-AAACCCAAA-30
. Docking poses (1250) for the
pentanucleotides were subjected to a two-step selection process utilizing the NMR
data. In the first step, complexes were discarded if the oligonucleotide engaged in
contacts with residues that did not exhibit experimental chemical shift changes. In
the second step, structures were selected that exhibited good agreement with the
NMR binding-site mapping data, taking the size of the chemical shift changes and
the scoring value of the binding pose into account. The final structural models for
A3A, complexed with 50
-TTCAT-30
(Fig. 5a) and 50
-CCCAA-30
(Fig. 5b), represent
B80% of all the structures that result from the selection process.
References
1. Chiu, Y. L. & Greene, W. C. The APOBEC3 cytidine deaminases: an innate
defensive network opposing exogenous retroviruses and endogenous
retroelements. Annu. Rev. Immunol. 26, 317–353 (2008).
2. Malim, M. H. APOBEC proteins and intrinsic resistance to HIV-1 infection.
Philos. Trans. R Soc. Lond. B Biol. Sci. 364, 675–687 (2009).
3. Duggal, N. K. & Emerman, M. Evolutionary conflicts between viruses and
restriction factors shape immunity. Nat. Rev. Immunol. 12, 687–695 (2012).
4. Jarmuz, A. et al. An anthropoid-specific locus of orphan C to U RNA-editing
enzymes on chromosome 22. Genomics 79, 285–296 (2002).
5. Betts, L., Xiang, S., Short, S. A., Wolfenden, R. & Carter, Jr C. W. Cytidine
deaminase. The 2.3 Å crystal structure of an enzyme: transition-state analog
complex. J. Mol. Biol. 235, 635–656 (1994).
6. Stenglein, M. D., Burns, M. B., Li, M., Lengyel, J. & Harris, R. S. APOBEC3
proteins mediate the clearance of foreign DNA from human cells. Nat. Struct.
Mol. Biol. 17, 222–229 (2010).
7. Shinohara, M. et al. APOBEC3B can impair genomic stability by inducing base
substitutions in genomic DNA in human cells. Sci. Rep. 2, 806 (2012).
8. Vartanian, J. P., Guetard, D., Henry, M. & Wain-Hobson, S. Evidence for
editing of human papillomavirus DNA by APOBEC3 in benign and
precancerous lesions. Science 320, 230–233 (2008).
9. Wiegand, H. L. & Cullen, B. R. Inhibition of alpharetrovirus replication by a
range of human APOBEC3 proteins. J. Virol. 81, 13694–13699 (2007).
10. Chen, H. et al. APOBEC3A is a potent inhibitor of adeno-associated virus and
retrotransposons. Curr. Biol. 16, 480–485 (2006).
11. Narvaiza, I. et al. Deaminase-independent inhibition of parvoviruses by the
APOBEC3A cytidine deaminase. PLoS Pathog. 5, e1000439 (2009).
12. Ooms, M., Krikoni, A., Kress, A. K., Simon, V. & Mu¨nk, C. APOBEC3A,
APOBEC3B, and APOBEC3H haplotype 2 restrict human T-lymphotropic
virus type 1. J. Virol. 86, 6097–6108 (2012).
13. Bogerd, H. P. et al. Cellular inhibitors of long interspersed element 1 and Alu
retrotransposition. Proc. Natl Acad. Sci. USA 103, 8780–8785 (2006).
14. Muckenfuss, H. et al. APOBEC3 proteins inhibit human LINE-1
retrotransposition. J. Biol. Chem. 281, 22161–22172 (2006).
15. Kinomoto, M. et al. All APOBEC3 family proteins differentially inhibit LINE-1
retrotransposition. Nucleic Acids Res. 35, 2955–2964 (2007).
16. Bulliard, Y. et al. Structure-function analyses point to a polynucleotideaccommodating
groove essential for APOBEC3A restriction activities. J. Virol.
85, 1765–1776 (2011).
17. Iwatani, Y., Takeuchi, H., Strebel, K. & Levin, J. G. Biochemical activities of
highly purified, catalytically active human APOBEC3G: correlation with
antiviral effect. J. Virol. 80, 5992–6002 (2006).
18. Carpenter, M. A. et al. Methylcytosine and normal cytosine deamination by the
foreign DNA restriction enzyme APOBEC3A. J. Biol. Chem. 287, 34801–34808
(2012).
19. Wijesinghe, P. & Bhagwat, A. S. Efficient deamination of 5-methylcytosines in
DNA by human APOBEC3A, but not by AID or APOBEC3G. Nucleic Acids
Res. 40, 9206–9217 (2012).
20. Landry, S., Narvaiza, I., Linfesty, D. C. & Weitzman, M. D. APOBEC3A can
activate the DNA damage response and cause cell-cycle arrest. EMBO Rep. 12,
444–450 (2011).
21. Suspe`ne, R. et al. Somatic hypermutation of human mitochondrial and nuclear
DNA by APOBEC3 cytidine deaminases, a pathway for DNA catabolism. Proc.
Natl Acad. Sci. USA 108, 4858–4863 (2011).
22. Aynaud, M. M. et al. Human Tribbles 3 protects nuclear DNA from cytidine
deamination by APOBEC3A. J. Biol. Chem. 287, 39182–39192 (2012).
23. Peng, G. et al. Myeloid differentiation and susceptibility to HIV-1 are linked to
APOBEC3 expression. Blood 110, 393–400 (2007).
24. Koning, F. A. et al. Defining APOBEC3 expression patterns in human tissues
and hematopoietic cell subsets. J. Virol. 83, 9474–9485 (2009).
25. Refsland, E. W. et al. Quantitative profiling of the full APOBEC3 mRNA
repertoire in lymphocytes and tissues: implications for HIV-1 restriction.
Nucleic Acids Res. 38, 4274–4284 (2010).
26. Thielen, B. K. et al. Innate immune signaling induces high levels of TC-specific
deaminase activity in primary monocyte-derived cells through expression of
APOBEC3A isoforms. J. Biol. Chem. 285, 27753–27766 (2010).
27. Berger, G. et al. APOBEC3A is a specific inhibitor of the early phases of HIV-1
infection in myeloid cells. PLoS Pathog. 7, e1002221 (2011).
28. Koning, F. A., Goujon, C., Bauby, H. & Malim, M. H. Target cell-mediated
editing of HIV-1 cDNA by APOBEC3 proteins in human macrophages.
J. Virol. 85, 13448–13452 (2011).
29. Navarro, F. et al. Complementary function of the two catalytic domains of
APOBEC3G. Virology 333, 374–386 (2005).
30. Newman, E. N. et al. Antiviral function of APOBEC3G can be dissociated from
cytidine deaminase activity. Curr. Biol. 15, 166–170 (2005).
31. Chen, K.-M. et al. Structure of the DNA deaminase domain of the HIV-1
restriction factor APOBEC3G. Nature 452, 116–119 (2008).
32. Furukawa, A. et al. Structure, interaction and real-time monitoring of the
enzymatic reaction of wild-type APOBEC3G. EMBO J. 28, 440–451 (2009).
33. Harjes, E. et al. An extended structure of the APOBEC3G catalytic domain
suggests a unique holoenzyme model. J. Mol. Biol. 389, 819–832 (2009).
34. Holden, L. G. et al. Crystal structure of the anti-viral APOBEC3G catalytic
domain and functional implications. Nature 456, 121–124 (2008).
35. Shandilya, S. M. D. et al. Crystal structure of the APOBEC3G catalytic domain
reveals potential oligomerization interfaces. Structure 18, 28–38 (2010).
36. Kitamura, S. et al. The APOBEC3C crystal structure and the interface for HIV-
1 Vif binding. Nat. Struct. Mol. Biol. 19, 1005–1010 (2012).
37. Krzysiak, T. C., Jung, J., Thompson, J., Baker, D. & Gronenborn, A. M.
APOBEC2 is a monomer in solution: implications for APOBEC3G models.
Biochemistry 51, 2008–2017 (2012).
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883
10 NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications
& 2013 Macmillan Publishers Limited. All rights reserved.
38. Bransteitter, R., Prochnow, C. & Chen, X. S. The current structural and
functional understanding of APOBEC deaminases. Cell. Mol. Life Sci. 66,
3137–3147 (2009).
39. Love, R. P., Xu, H. & Chelico, L. Biochemical analysis of hypermutation by the
deoxycytidine deaminase APOBEC3A. J. Biol. Chem. 287, 30812–30822 (2012).
40. Verdonk, M. L., Cole, J. C., Hartshorn, M. J., Murray, C. W. & Taylor, R. D.
Improved protein-ligand docking using GOLD. Proteins 52, 609–623 (2003).
41. Aguiar, R. S., Lovsin, N., Tanuri, A. & Peterlin, B. M. Vpr.A3A chimera inhibits
HIV replication. J. Biol. Chem. 283, 2518–2525 (2008).
42. Chelico, L., Pham, P., Calabrese, P. & Goodman, M. F. APOBEC3G DNA
deaminase acts processively 30
-50
on single-stranded DNA. Nat. Struct. Mol.
Biol. 13, 392–399 (2006).
43. Nowarski, R., Britan-Rosich, E., Shiloach, T. & Kotler, M. Hypermutation by
intersegmental transfer of APOBEC3G cytidine deaminase. Nat. Struct. Mol.
Biol. 15, 1059–1066 (2008).
44. Iwatani, Y. et al. Deaminase-independent inhibition of HIV-1 reverse
transcription by APOBEC3G. Nucleic Acids Res. 35, 7096–7108 (2007).
45. Losey, H. C., Ruthenburg, A. J. & Verdine, G. L. Crystal structure of
Staphylococcus aureus tRNA adenosine deaminase TadA in complex with
RNA. Nat. Struct. Mol. Biol. 13, 153–159 (2006).
46. Senavirathne, G. et al. Single-stranded DNA scanning and deamination by
APOBEC3G cytidine deaminase at single molecule resolution. J. Biol. Chem.
287, 15826–15835 (2012).
47. Roberts, S. A. et al. Clustered mutations in yeast and in human cancers can
arise from damaged long single-strand DNA regions. Mol. Cell 46, 424–435
(2012).
48. Burns, M. B. et al. APOBEC3B is an enzymatic source of mutation in breast
cancer. Nature 494, 366–370 (2013).
49. Clore, G. M. & Gronenborn, A. M. Determining the structures of large proteins
and protein complexes by NMR. Trends Biotechnol. 16, 22–34 (1998).
50. Sattler, M., Maurer, M., Schleucher, J. & Griesinger, C. A Simultaneous 15N,
1H-HSQC and 13C, 1H-HSQC with sensitivity enhancement and a
heteronuclear gradient-echo. J. Biomol. NMR 5, 97–102 (1995).
51. Delaglio, F. et al. NMRPipe: a multidimensional spectral processing system
based on UNIX pipes. J. Biomol. NMR 6, 277–293 (1995).
52. Johnson, B. A. & Blevins, R. A. NMR View - a computer-program for the
visualization and analysis of NMR data. J. Biomol. NMR 4, 603–614 (1994).
53. Brunger, A. T. et al. Crystallography & NMR system: a new software suite for
macromolecular structure determination. Acta. Crystallogr. D Biol. Crystallogr.
54(Pt 5): 905–921 (1998).
54. Cornilescu, G., Delaglio, F. & Bax, A. Protein backbone angle restraints from
searching a database for chemical shift and sequence homology. J. Biomol.
NMR 13, 289–302 (1999).
55. Laskowski, R. A., Rullmannn, J. A., MacArthur, M. W., Kaptein, R. &
Thornton, J. M. AQUA and PROCHECK-NMR: programs for checking the
quality of protein structures solved by NMR. J. Biomol. NMR 8, 477–486
(1996).
56. Koradi, R., Billeter, M. & Wu¨thrich, K. MOLMOL: a program for display and
analysis of macromolecular structures. J. Mol. Graph. 14, 29–32 (1996).
57. Flores, S. C., Sherman, M. A., Bruns, C. M., Eastman, P. & Altman, R. B. Fast
flexible modeling of RNA structure using internal coordinates. IEEE/ACM
Trans. Comput. Biol. Bioinform. 8, 1247–1257 (2011).
58. Eldridge, M. D., Murray, C. W., Auton, T. R., Paolini, G. V. & Mee, R. P.
Empirical scoring functions: I. The development of a fast empirical scoring
function to estimate the binding affinity of ligands in receptor complexes. J.
Comput. Aided Mol. Des. 11, 425–445 (1997).
Acknowledgements
We thank Maria DeLucia for help with A3A purification, Doug Bevan and Phil Greer for
computer technical support and Michael J Delk for NMR instrumental support. J.H.
acknowledges financial support by an International Outgoing Fellowship of the European
Community program ‘Support for training and career development of researchers
(Marie Curie)’, under Contract No. PIOF-GA-2009-235902. This work was supported in
part by the National Institutes of Health NIGMS Grant P50GM082251 (A.M.G.) and the
Intramural Research Program at the National Institutes of Health, Eunice Kennedy Shriver
National Institute of Child Health and Human Development (M.M., K.H. and J.G.L.).
Author contributions
The study was conceived by A.M.G. and J.G.L. The NMR experiments were designed by
A.M.G. and I.-J.L.B.; I.-J.L.B. and C.-H.B. performed experiments and data analysis;
I.-J.L.B. and A.M.G. interpreted the NMR data; J.A., L.M.C. and C.-H.B. prepared
purified A3A; J.H. built structural models of A3A, complexed with short oligonucleotides;
the biochemical assays were designed by J.G.L., M.M. and K.H.; M.M. and K.H.
performed the experiments and M.M., K.H. and J.G.L. analyzed the results; the manuscript
was written by I.-J.L.B., M.M., J.G.L. and A.M.G.
Additional information
Accession codes: The A3A atomic coordinates and NMR constraints have been
deposited in the RCSB Protein Data Bank under accession code 2m65, and the NMR
chemical shift data have been deposited in the Biological Magnetic Resonance Bank
under accession code 19108.
Supplementary Information accompanies this paper at http://www.nature.com/
naturecommunications
Competing financial interests: The authors claim no competing financial interests.
Reprints and permission information is available online at http://npg.nature.com/
reprintsandpermissions/
How to cite this article: Byeon, I.-J. L. et al. NMR structure of human restriction factor
APOBEC3A reveals substrate binding and enzyme specificity. Nat. Commun. 4:1890
doi: 10.1038/ncomms2883 (2013).
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2883 ARTICLE
NATURE COMMUNICATIONS | 4:1890 | DOI: 10.1038/ncomms2883 | www.nature.com/naturecommunications 11
& 2013 Macmillan Publishers Limited. All rights reserved.
METHODOLOGICAL DEVELOPMENTS OF ENHANCED SAMPLING METHODOLOGY H-REMD
38
4 Methodological developments of enhanced sampling
methodology H-REMD
Paper 5: Hritz J., Oostenbrink C. Hamiltonian replica exchange molecular dynamics
using soft-core interactions. J. Chem. Phys. 2008, 128, 144121
Paper 6: Hritz J., Oostenbrink C. Optimization of Replica Exchange Molecular Dynamics
by Fast Mimicking. J. Chem. Phys. 2007, 127, 204104
Paper 7: Nagy G., Oostenbrink C., Hritz J.*: Exploring the Binding Pathways of the
14-3-3 Protein: Structural and Free-Energy Profiles Revealed by Hamiltonian Replica
Exchange Molecular Dynamics with Distance Field Distance Restraints.
PLoS ONE 2017,12(7), e0180633
Two types of Hamiltonian changes within H-REMD were addressed by the applicant:
increasing softness of the selected non-bonded interactions and distance restraints applied
to the ligand with respect to its binding site.
Figure 2: Structure of GTP and 8-Br-GTP in syn conformation.
Conformational transitions between the syn and anti states occur by rotation around the glycosidic bond
(indicated). The glycosidic dihedral angle φ is defined over atoms: C4-N9-C1-O4.
METHODOLOGICAL DEVELOPMENTS OF ENHANCED SAMPLING METHODOLOGY H-REMD
39
Figure 3: Time dependence of dihedral angle around the glycosidic bond during two MD simulations
for GTP and 8-Br-GTP
Values for MD simulations starting from anti conformation are shown in black and from syn conformation
in red color.
4.1 H-REMD using soft-core interactions
High intramolecular energy barriers of complex biomolecules restrict their efficient
conformational sampling. A millisecond MD of even such simple biomolecules as GTP
does not provide sufficient conformational sampling due to the high energy barrier for
rotation around the glycosidic bond between the base and sugar part of GTP (Fig. 2).
This energy barrier is even higher for C8- analogs of GTP (Fig. 3). Therefore the applicant
has developed the novel scheme of H-REMD, where the main idea is to use softcore
interactions24 between atoms of the base and sugar part of GTP analog in the following
forms:P5
METHODOLOGICAL DEVELOPMENTS OF ENHANCED SAMPLING METHODOLOGY H-REMD
40
Figure 4: Schematic representation of the energy profile as a function of the softness of nonbonded
interactions.
Replicas at higher levels of softness convert more easily from one stable conformation to the other,
which are separated by a high energy barrier if no softness is applied.
The softness of Van der Walls and electrostatic interactions are controlled here by the
λ parameter.24 Because soft-core potentials are almost the same as regular ones at
longer distances, most of the interactions between atoms of the perturbed parts are
changed only slightly. Rather, the strong repulsion between atoms that are close in
space, which results in high energy barriers in most cases, is weakened within higher
replicas of our scheme. Their desired effect on the overall energy landscape between
𝐸𝑖𝑗
𝑣𝑑𝑤
(𝑟𝑖𝑗, 𝜆) = (
𝐶𝑖𝑗
(12)
𝐴𝑖𝑗(𝜆) + 𝑟𝑖𝑗
6 − 𝐶𝑖𝑗
(6)
)
1
𝐴𝑖𝑗(𝜆) + 𝑟𝑖𝑗
6 ; 𝐴𝑖𝑗(𝜆) =
𝐶𝑖𝑗
(12)
𝐶𝑖𝑗
(6)
𝜆2 (4.1)
𝐸𝑖𝑗
𝑒𝑙
(𝑟𝑖𝑗, 𝜆) =
𝑞𝑖 𝑞𝑗
4
1
√𝐵𝑖𝑗(𝜆) + 𝑟𝑖𝑗
2
; 𝐵𝑖𝑗(𝜆) = 𝜆2
(4.2)
METHODOLOGICAL DEVELOPMENTS OF ENHANCED SAMPLING METHODOLOGY H-REMD
41
two stable conformations (e.g. anti and syn conformation of C8-analogs of GTP) can be
seen in Fig. 4.
The advantage of this approach over other REMD schemes is the possibility to use a
relatively small number of replicas (6 for GTP) with larger local differences between
the individual Hamiltonians. A huge computational gain is also harnessed since it is
possible to apply the soft-core interaction for only a selected region and therefore to
allow for enhanced conformational sampling only in this region, with the rest of the
system still moving on the timescale of the MD. The replica with unmodified Hamiltonian
produces unbiased canonical ensembles of structures in a highly efficient way.P5
The problem with H-REMD using soft-core interactions is its setup (how many replicas
are needed, at what softness, etc.) Therefore, the optimization of REMD setup parameters
was developed.P6 This approach has general applicability for any REMD, not only
H-REMD using soft-core interactions.
4.2 Umbrella sampling
The actual binding process of ligands with the accompanying free energy differences
can be studied by methods such as umbrella sampling.25,26 Such methods are computationally
more demanding than molecular docking by several orders of magnitude as
described in the previous scientific chapter. However, they are in principle capable of
determining reliable binding affinities. Umbrella sampling attempts to overcome the
sampling problem by modifying the original Hamiltonian (unbiased) - where 𝐻 𝑢(𝑟 𝑁
)
by a bias term (𝑉𝑏(𝑟 𝑁
)) resulting in the overall (biased) Hamiltonian 𝐻 𝑏(𝑟 𝑁
). A bias, an
additional potential energy term, is applied to the system to ensure efficient sampling
along the reaction coordinate, a path connecting two end-point states separated by
an energy barrier, e.g. bound and unbound state of a ligand of our interest. MD simulation
of the biased system provides the biased (unrealistic) distribution along the reaction
coordinate. It can still be used for the calculation of the unbiased average of a property
A by the following equation:
METHODOLOGICAL DEVELOPMENTS OF ENHANCED SAMPLING METHODOLOGY H-REMD
42
where angle brackets indicate the ensemble averaging. The effect of the bias potential
to connect energetically separated regions in phase space gave rise to the name umbrella
sampling.26,27
4.3 H-REMD using distance (field)restraints
A typical problem when applying the umbrella sampling method to biomolecular systems
is that the acting bias very often “damages” the biomacromolecule. For instance,
when we applied the distance restraints of substrates within the CYP 2D6 catalytic site
to simulate the unbinding process we often observed the ligand hitting against the
closed binding channel. This motivated us to apply the distance restraint within the HREMD
scheme. Idea was that the switches in the H-REMD protocol allow for the reversible
binding and unbinding of the ligand, ensuring that all relevant pathways are
sampled. However, this may lead to unfavorable situations as an unbound ligand may
be directed towards the active site along a path that would lead through the protein,
rather than along an entrance channel. To circumvent this problem, a distance-field
was defined, using Dijkstra’s algorithm to determine the optimal path to the active site
without going through the protein as indicated in Fig. 5.19 The applicant successfully
applied this approach to the sampling of binding pathways of phosphopeptides to the
14-3-3 protein, calculating the corresponding binding affinities that showed reasonable
agreement with the available experimental data.P7
〈𝐴〉 𝑢 =
〈𝐴(𝑟 𝑁)𝑒
+
𝑉 𝑏(𝑟 𝑁)
𝑘 𝐵 𝑇 〉 𝑏
〈𝑒
+
𝑉 𝑏(𝑟 𝑁)
𝑘 𝐵 𝑇 〉 𝑏
(4.3)
METHODOLOGICAL DEVELOPMENTS OF ENHANCED SAMPLING METHODOLOGY H-REMD
43
Figure 5: Schematic representation of distance-field restraints.
A protein is represented by a green shape, with the active site opening showing the strongest distancefield
restraints. The figure is taken from19
METHODOLOGICAL DEVELOPMENTS OF ENHANCED SAMPLING METHODOLOGY H-REMD
44
Paper 5
Hritz J., Oostenbrink C. Hamiltonian replica exchange molecular dynamics using softcore
interactions. J. Chem. Phys. 2008, 128, 144121
Hamiltonian replica exchange molecular dynamics using soft-core
interactions
Jozef Hritz and Chris Oostenbrinka͒
Leiden Amsterdam Center for Drug Research (LACDR), Division of Molecular Toxicology,
Vrije Universiteit, Amsterdam NL-1081 HV, The Netherlands
͑Received 19 December 2007; accepted 5 February 2008; published online 14 April 2008͒
To overcome the problem of insufficient conformational sampling within biomolecular simulations,
we have developed a novel Hamiltonian replica exchange molecular dynamics ͑H-REMD͒ scheme
that uses soft-core interactions between those parts of the system that contribute most to high energy
barriers. The advantage of this approach over other H-REMD schemes is the possibility to use a
relatively small number of replicas with locally larger differences between the individual
Hamiltonians. Because soft-core potentials are almost the same as regular ones at longer distances,
most of the interactions between atoms of perturbed parts will only be slightly changed. Rather, the
strong repulsion between atoms that are close in space, which in many cases results in high energy
barriers, is weakened within higher replicas of our proposed scheme. In addition to the soft-core
interactions, we proposed to include multiple replicas using the same Hamiltonian/level of softness.
We have tested the new protocol on the GTP and 8-Br-GTP molecules, which are known to have
high energy barriers between the anti and syn conformation of the base with respect to the sugar
moiety. During two 25 ns MD simulations of both systems the transition from the more stable to the
less stable ͑but still experimentally observed͒ conformation is not seen at all. Also temperature
REMD over 50 replicas for 1 ns did not show any transition at room temperature. On the other hand,
more than 20 of such transitions are observed in H-REMD using six replicas ͑at three different
Hamiltonians͒ during 6.8 ns per replica for GTP and 12 replicas ͑at six different Hamiltonians͒
during 8.7 ns per replica for 8-Br-GTP. The large increase in sampling efficiency was obtained from
an optimized H-REMD scheme involving soft-core potentials, with multiple simulations using the
same level of softness. The optimization of the scheme was performed by fast mimicking ͓J. Hritz
and C. Oostenbrink, J. Chem. Phys. 127, 204104 ͑2007͔͒. © 2008 American Institute of Physics.
͓DOI: 10.1063/1.2888998͔
I. INTRODUCTION
The energy landscape for biomolecules in explicit solvent
exhibits many local free energy minima.1
While many
minima are readily sampled in a molecular dynamics ͑MD͒
simulation, some are separated by high free energy barriers.2
A molecule can be trapped in local energy minimum conformations
for times comparable to or longer than typical simulation
times that are reachable by conventional MD. In these
cases, regular MD simulations will not lead to a complete
conformational sampling of the studied system.3
One possible solution is the application of temperature
replica exchange MD ͑T-REMD͒, in which several parallel
simulations are performed at different temperatures. At regular
intervals switches of the temperature between simulations
are attempted, allowing high temperature simulations to cool
down and low temperature simulations to heat up. A gain in
efficiency should be obtained if the free energy barriers can
be easily crossed in high temperature simulations.4
T-REMD,
however, suffers from the fact that the number of replicas
needed to cover the necessary temperature range is proportional
to the square root of the number of degrees of freedom
of the system. This means that T-REMD simulations of biomolecules
in explicit solvent can be computationally very
demanding. The requirement of a large number of replicas in
T-REMD can be overcome by applying Hamiltonian REMD
͑H-REMD͒,5,6
where not the temperature but the Hamiltonian
is varied over the replicas, through a perturbation of
the original Hamiltonian. Note that T-REMD is a special
case of H-REMD in which all the terms in the Hamiltonian
are multiplied by the same scaling factor.
Hamiltonians can be perturbed in different ways. Useful
approaches are those, in which the free energy landscape of
the higher replicas allows for faster conformational conversions
compared to the unperturbed Hamiltonian at the lowest
replica.
We choose to perturb the Hamiltonian by describing selected
interactions with a soft-core potential7
and vary the
level of softness over the replicas through a softness parameter
␭. This approach allows for a diminishing of free energy
barriers at higher replicas with higher levels of softness. The
advantage of our approach is that we can specifically use
soft-core interactions for those biomolecular parts that are of
interest, e.g., parts contributing the most to the free energy
barriers between different conformations. In this study we
have also applied the novel concept of a degenerate highest
level of softness ␭max. This involves multiple ͑n͒ replicas ata͒
Electronic mail: c.oostenbrink@few.vu.nl.
THE JOURNAL OF CHEMICAL PHYSICS 128, 144121 ͑2008͒
0021-9606/2008/128͑14͒/144121/10/$23.00 © 2008 American Institute of Physics128, 144121-1
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
␭max, allowing the system to spend more time at this ␭-value
so that alternative conformations can be reached in shorter
overall simulation time. This is reminiscent of the J-walking
method8
or finite reservoir simulations,9,10
where a reservoir
of structures is pregenerated at the highest temperature or
Hamiltonian ͑Tmax/␭max͒ and then “coupled” to REMD to
get correct ensembles at lower values of T or ␭. However,
the concept of a degenerate Tmax/␭max is an integral part of
REMD requiring no precalculation. It allows for the most
efficient balance between conformational transitions at
Tmax/␭max and replica exchanges toward T0/␭0. Refinement
of the optimal n and the optimal selection of ␭-values by fast
REMD mimicking is described in our previous paper.11
We have tested the H-REMD scheme using soft-core
interactions on two biologically relevant systems: GTP and
8-Br-GTP ͑Fig. 1͒. These systems can adopt two stable conformations
by rotation around the glycosidic bond: Anti and
syn. The boundaries of these conformations are not exactly
the same for GTP and 8-Br-GTP. Based on the dihedral angle
distributions shown in the Results section, we consider GTP
to be in a syn conformation if the dihedral angle around the
glycosidic bond is within the interval ͗−25° ,150°͘. For
8-Br-GTP we use the interval ͗−35° ,160°͘ to define the syn
state. NMR studies indicate that the dominant conformation
for GTP is anti, while it is syn for 8-Br-GTP.12
Both GTP and
8-Br-GTP have high free energy barriers between the syn and
anti conformations. However, H-REMD simulations using a
few different ␭-values ͑3 for GTP and 6 for 8-Br-GTP͒ suffice
to enhance the conformational sampling enormously.
The preference of syn and anti should follow quantitatively
from the ensemble generated at the lowest ␭-value corresponding
to the unperturbed Hamiltonian of the REMD
scheme. We compare the relative population of the two states
with direct free energy calculations between them using hidden
restraints along the glycosidic dihedral angle.
II. METHODS
The Methods section is divided into four parts. Firstly
we discuss the implementation of soft-core interactions in
REMD simulations. Secondly, we discuss the REMD simulations
in more detail, and thirdly the use of hidden restraints
to calculate the free energy difference between syn and anti.
The section is concluded with a description of the exact
simulation settings.
A. Implementation of soft-core interactions
We have used the following functional form for van der
Waals and electrostatic soft-core interactions7
between atoms
i and j as a function of the interatomic distance rij
Eij
vdW
͑rij,␭͒ = ͩ C12ij
Aij͑␭͒ + rij
6 − C6ijͪ 1
Aij͑␭͒ + rij
6 , ͑1͒
Eij
el
͑rij,␭͒ =
qiqj
4␲␧
1
ͱBij͑␭͒ + rij
2
, ͑2͒
with Aij͑␭͒=␣vdW͑C12ij /C6ij͒␭2
and Bij͑␭͒=␣el␭2
. C12ij,
C6ij are the Lennard–Jones parameters for atom pair i and j,
qi and qj the partial charges of particles i and j, and ␣vdW and
␣el are the softness parameters. In the current study we used
in all simulations ␣vdW=␣el=1, and the softness of the interactions
was controlled through the parameter ␭. It can be
seen that at longer distances ͓rij ӷA͑␭͒ and rij ӷB͑␭͔͒ the
soft-core interaction approximates the interaction for normal
atoms and that they differ mostly at short distances between
the atoms ͓rij ഛA͑␭͒ or rij ഛB͑␭͔͒. Potential energy barriers
are mostly the result of a short-ranged repulsion between
atoms, which can strongly be reduced by increasing the levels
of softness. An idealized potential energy landscape for
two states separated by a high barrier when no softness is
applied, but part of the same energy valley at higher levels of
softness is sketched in Fig. 2.
In GROMOS96 ͑Ref. 13͒ as well as GROMOS05 ͑Ref. 14͒
one can use soft-core interactions in the context of a free
energy perturbation in which the interaction energy Eij is
written as a linear combination of the potential energy for
two different states A and B,
Eij͑rij,␭͒ = ͓1 − ␭͔n
EA
͑rij,␭͒ + ␭n
EB
͑rij,1 − ␭͒. ͑3͒
In cases where the Lennard–Jones parameters and partial
charges are identical in states A and B and we set ␭=0.5 and
n=1, this reduces to the functional forms of Eqs. ͑1͒ and ͑2͒.
For other ␭-values one gets a mixture of potentials at different
levels of softness. In order to be able to control the level
FIG. 1. Structure of GTP and 8-Br-GTP in syn conformation. Conformational
transitions between the syn and anti states occur by rotation around
the glycosidic bond ͑indicated͒. The glycosidic dihedral angle ͑␾͒ is defined
over atoms: C4-N9-C1Ј-O4Ј.
FIG. 2. Schematic representation of the energy profile as function of the
softness of nonbonded interactions. Replicas at higher levels of softness
convert more easily from one stable conformation to the other, which are
separated by a high energy barrier if no softness is applied.
144121-2 J. Hritz and C. Oostenbrink J. Chem. Phys. 128, 144121 ͑2008͒
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
of softness by ␭, which is convenient in a REMD setting, we
extended the implementation of free energy perturbations in
GROMOS05 to include the possibility to set individual
␭-values for Lennard–Jones interaction ͑␭vdW͒, the Lennard–
Jones softness level ␭soft
vdW
, Coulombic interactions ͑␭el͒, and
the Coulombic softness level ͑␭soft
el
͒ as a polynomial function
͑up to fourth order͒ of an overall or “global” ␭-value. For the
Lennard–Jones interaction we write the interaction energy as
Eij
vdW
͑rij,␭͒ = ͓1 − ␭vdW͑␭͔͒n
Eij
vdW,A
͑rij,␭soft
vdW
͑␭͒͒
+ ͓␭vdW͑␭͔͒n
Eij
vdW,B
͑rij,1 − ␭soft
vdW
͑␭͒͒. ͑4͒
An analogous form can be written for the Coulombic interaction
Eij
el
. For the REMD simulations of GTP and 8-Br-GTP
reported here, we used individual ␭-dependencies as follows:
␭vdW͑␭͒ = 0, ␭soft
vdW
͑␭͒ = ␭, ͑5͒
␭el͑␭͒ = 0, ␭soft
el
͑␭͒ = ␭. ͑6͒
With these settings the softness at individual replicas is easily
controlled by the global lambda ␭, while the actual interactions
are only defined by the parameters defined for
state A.
B. REMD
A REMD simulation involves M noninteracting copies
͑replicas͒ of MD of one system running in parallel at various
conditions.15
Let us mark the state of a REMD ensemble of
simulations as X=͕...,xm
i
, ... ,xn
j
, ...͖, where the indices indicate
that the ith replica is simulated at the mth condition
and the jth replica is simulated at the nth condition. x represents
the collected positions q and momenta p of all particles
͑xϵ͑q,p͒͒. The weight factor for this state of the REMD
ensemble is given by the product of Boltzmann factors for
individual noninteracting replicas,
WREMD͑X͒ = ͟k=1
M
e−␤m͑k͒Hm͑k͒͑qk,pk͒
, ͑7͒
where ␤m͑k͒=1/kBTm͑k͒ and Hm͑k͒͑qk
,pk
͒ is the Hamiltonian
of the system given by sum of the kinetic energy K͑p͒ and
the potential energy E͑q͒. kB is the Boltzmann constant. The
subscript m͑k͒ indicates the condition m at which replica k is
currently simulated. In T-REMD, conditions m differ only by
the temperature T, while in H-REMD, the condition represents
different Hamiltonians, in this case described by the
softness parameter ␭. Conditions ͑␭ or T͒ do not have to be
different for all replicas. In a degenerate ␭max/Tmax scheme,11
multiple replicas are simulated at the same ␭max/Tmax simultaneously,
i.e., m͑i͒=m͑j͒.
At regular time intervals we attempt to exchange
͑switch͒ the conditions at which replicas i, j are simulated,
X = ͕...,xm
i
, ... ,xn
j
, ... ͖ → XЈ = ͕...,xm
j
, ... ,xn
i
, ... ͖. ͑8͒
In order to maintain the proper weight of the REMD ensemble
as described in Eq. ͑7͒ a detailed balance condition
must be imposed on the exchange probability,
w͑X → XЈ͒ = min͓1,exp͑− ⌬͔͒, ͑9͒
where
⌬ = ␤m͓Hm͑qj
,pj
͒ − Hm͑qi
,pi
͔͒
− ␤n͓Hn͑qj
,pj
͒ − Hn͑qi
,pi
͔͒. ͑10͒
Sugita and Okamoto showed that the kinetic energy parts of
the Hamiltonians cancel,16
leading to
⌬ = ␤m͓E␭m
͑qj
͒ − E␭m
͑qi
͔͒ − ␤n͓E␭n
͑qj
͒ − E␭n
͑qi
͔͒. ͑11͒
In the case of H-REMD, where all replicas are run at the
same temperature, ⌬ reduces further to
⌬ = ␤͓E␭m
͑qj
͒ − E␭m
͑qi
͒ − E␭n
͑qj
͒ + E␭n
͑qi
͔͒. ͑12͒
In our study the Hamiltonian applied for individual replicas
is varied by introducing the soft-core interactions ͑1-2͒ controlled
by parameter ␭.
C. Thermodynamic integration using hidden dihedral
angle restraints
The use of hidden dihedral angle restraints17
offers an
alternative approach to calculate the free energy difference
between the anti and syn conformations of GTP/8-Br-GTP. A
hidden dihedral angle restraint around the glycosidic bond
was used to propagate the GTP/8-Br-GTP systems from one
stable conformation to the other, thereby overcoming the energy
barrier and sampling conformations on the transition
path in an efficient way. The restraining energies for the dihedral
angle ␾͑C4-N9-C1Ј-O4Ј͒ were calculated as a function
of the coupling parameter ␭ according to Ref. 16,
Vdihres
͑␾,␭͒ = 4␭͑1 − ␭͒Vrestr
dihres,AB
͑␾,␭͒, ͑13͒
where
Vrestr
dihres,AB
͑␾,␭͒ = ͭVharm
dihres,AB
͑␾,␭͒ if ͉⌬␾␭͉ ഛ ␾lin
0
Vlin
dihres,AB
͑␾,␭͒ if ͉⌬␾␭͉ Ͼ ␾lin
0
,
ͮ
͑14͒
with
⌬␾␭ = ␾ − ͑1 − ␭͒␾0,A − ␭␾0,B + 2n␲, ͑15͒
and
Vharm
dihres,AB
͑␾,␭͒ = 1/2͓͑1 − ␭͒KA
+ ␭KB
͔͑⌬␾␭͒2
, ͑16͒
Vlin
dihres,AB
͑␾,␭͒ = ͓͑1 − ␭͒KA
+ ␭KB
͔͑␨⌬␾␭ − 1/2␾lin
0
͒␾lin
0
,
͑17͒
with ␨=−1 if ⌬␾␭Ͻ−␾lin
0
and ␨=1 if ⌬␾␭Ͼ−␾lin
0
for the
linearized part of the restraint. We have used KA
=KB
=100 kJ rad−2
and ␾lin
0
=30°. For both GTP and 8-Br-GTP,
four thermodynamic integration calculations were done:
From −120° ͑anti͒ to 60° ͑syn͒ and back ͑␾0,A
=−120°,
␾0,B
=60° or ␾0,A
=60°, ␾0,B
=−120°͒, from 60° ͑syn͒ to 240°
͑anti͒ and back ͑␾k
0,A
=60°, ␾k
0,B
=240° or ␾k
0,A
=240°,
␾k
0,B
=60°͒.
As can be seen from Eq. ͑13͒ the restraint energy reduces
to zero at the beginning ͑␭=␭A=0͒ and end ͑␭=␭B
=1͒ states. Therefore an unperturbed relative free energy
144121-3 Replica exchange molecular dynamics J. Chem. Phys. 128, 144121 ͑2008͒
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
⌬GBA of state B with respect to state A can be calculated
using the thermodynamic integration method,18
⌬GBA = ͵␭A
␭B
ͳ‫ץ‬H
‫ץ‬␭
ʹ␭
d␭, ͑18͒
where H is the Hamiltonian of the system including the restraining
potential energy term, Vdihres
͑␾,␭͒, and the angular
brackets indicate an ensemble average obtained at ␭. The
integration was performed by calculating the ensemble average
͗‫ץ‬H/‫ץ‬␭͘␭ at discrete ␭ points, corresponding to angular
changes of 20° ͑⌬␭=0.111͒. In regions where the curvature
of ͗‫ץ‬H/‫ץ‬␭͘ was high, the step size was halved. The initial
sampling time for each ␭ point was 1 ns ͑after 100 ps relaxation
which started from the equilibrated conformation of the
simulation at the previous ␭ value͒. For ␭ points in which the
ensemble averages did not converge to the same value in
forward and reverse thermodynamic integration calculations,
the sampling time was extended to 6 ns.
D. MD and REMD settings
All MD and REMD simulations were conducted using
the GROMOS05 simulation package running on a linux
cluster.14
All bonds were constrained, using the SHAKE
algorithm19
with a relative geometric accuracy of 10−4
, allowing
for a time step of 2 fs used in the leapfrog integration
scheme.20
Periodic boundary conditions, with a truncated octahedral
box ͑average volume of 59.6 nm3
͒, were applied.
After a steepest descent minimization to remove bad contacts
between molecules, initial velocities were randomly assigned
from a Maxwell–Boltzmann distribution at 298 K, according
to the atomic masses. The temperature was kept constant
using weak coupling to a bath of 298 K with a relaxation
time of 0.1 ps.21
The solute molecules ͑GTP or 8-Br-GTP͒
and solvent ͑i.e., 1926 explicit SPC water molecules22
and
three Na+
counterions͒ were independently coupled to the
heat bath. The pressure was controlled using isotropic weak
coupling to atmospheric pressure with a relaxation time of
0.5 ps.21
Van der Waals and electrostatic interactions were
calculated using a triple range cutoff scheme. Interactions
within a short-range cutoff of 0.8 nm were calculated every
time step from a pair list that was generated every five steps.
At these time points, interactions between 0.8 and 1.4 nm
were also calculated and kept constant between updates. A
reaction-field contribution was added to the electrostatic interactions
and forces to account for a homogeneous medium
outside the long-range cutoff, using the relative permittivity
of SPC water ͑61͒.23
All interaction energies were calculated
according to the GROMOS force field, parameter set 53A6.24
Force field parameters used for GTP and 8-Br-GTP are listed
in the supplementary material.25
In the H-REMD simulations,
all sugar-base interactions in GTP and 8-Br-GTP are
treated using the soft-core interactions through Eqs. ͑1͒, ͑2͒,
and ͑4͒.
REMD is implemented in GROMOS05 such that only
switches between “neighboring” replicas are attempted. Replica
exchanges are attempted every 2.5 ps ͑elementary period,
telem
͒. However, not all replica pairs are considered at
the same time. After an odd number of elementary periods,
exchanges between the ͑2i+1͒th and the ͑2͑i+1͒͒th replicas
are attempted ͑we call them replica exchanges of type I͒ and
after an even number of telem
exchanges between the ͑2i͒th
and the ͑2i+1͒th replicas are attempted ͑exchanges of type
II͒. This means that the effective time between two switches
of the same type is twice the elementary switching period
͑2ϫ2.5 ps=5 ps͒. During the initial 40 telem
͑100 ps͒ no replica
exchanges were attempted in order to equilibrate the
systems at the individual replicas. The conformations were
written out every 2.5 ps, just before the replica exchange
attempt. In order to see how effective REMD is in sampling
the less stable conformations, we started all REMD runs
from the most stable conformation, i.e., in the case of GTP
from anti and in the case of 8-Br-GTP from syn conforma-
tions.
A novel concept of degenerate ␭max simulations was applied
as presented in our previous study.11
In this approach
more replicas ͑n͒ are simulated at the highest ␭-value, ␭max,
simultaneously and the time between switching attempts between
these replicas is set such that every replica at ␭max
spends a time t␭max
total
, during which the system can convert
from one conformation to the other. Usually t␭max
total
ӷtelem
because
the conversion probability rather shows a sigmoidal
than a linear time dependence, and the middle of the sigmoidal
curve corresponds to a significantly longer time than
telem
. The values of ␭max and n as well as the number of
␭-values between 0 and ␭max and their exact values were
refined by the fast mimicking approach in Ref. 11. This approach
maximizes the number of global conformational transitions
͑Ngct͒ per CPU. We defined the number of global
conformational transitions as the number of times the conformational
state changes for each particular replica when
monitored at the lowest ␭-value ͑␭0͒. At which particular
␭-value the conformational transition occurs is irrelevant for
this measure. For more details, see Ref. 11. The ␭-values at
which REMD simulations were performed; t␭max
total
and n are
summarized in Table I.
III. RESULTS
A. Standard MD simulations
Theoretically, one can obtain the populations of syn and
anti conformations for GTP/8-Br-GTP from sufficiently long
MD simulations, during which many anti↔syn transitions
occur. However, 25 ns of MD simulation of GTP and 8-BrTABLE
I. Optimal REMD setting for GTP and 8-Br-GTP.
Molecule ␭max n t␭max
total
͑ps͒ Optimal ␭-set
GTP 0.45 4 100 ͓0.0,0.25,0.45,0.45,0.45,0.45͔
8-Br-GTP 0.7 7 100 ͓0.0,0.2,0.4,0.55,0.65,0.7,0.7,0.7,0.7,0.7,0.7,0.7͔
144121-4 J. Hritz and C. Oostenbrink J. Chem. Phys. 128, 144121 ͑2008͒
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
GTP shows that the syn and anti conformational states are
separated by a quite high free energy barrier. For GTP we
observe only one transition from the syn ͑less dominant͒ to
the anti ͑more dominant͒ conformation and no transitions
from anti to syn ͓Fig. 3͑a͔͒. No transitions were observed at
all for 8-Br-GTP, neither from syn to anti nor from anti to
syn ͓Fig. 3͑b͔͒, indicating an even higher energy barrier between
the anti and syn conformations for this molecule. This
means that in order to obtain converged values of the anti
and syn populations, we would have to run standard MD for
an impractically long time. In such cases it may be very
useful to use replica exchange MD, which can sample conformational
space much more efficiently than standard MD.
B. REMD of GTP and 8-Br-GTP
All replicas within REMD were started from the same
conformation ͑anti for GTP and syn for 8-Br-GTP͒. Therefore
it takes some time to relax the complete set of REMD
simulations. It is not straightforward to determine the relaxation
time and time length of REMD simulation required to
obtain sufficient statistics of measured quantities a priori.
The REMD settings used here were chosen to maximize the
number of global conformational transitions ͑Ngct͒ per CPU
by fast mimicking.11
Monitoring of Ngct is also useful for real
REMD because when all replicas are in equilibrium, Ngct
should show a linear time dependence. This allows us to
determine the relaxation time of REMD by extrapolating the
linear part of Ngct͑t͒ to 0 transition, as well as to estimate the
length of REMD, which is needed to obtain a specific value
of Ngct.
The time evolution of Ngct for GTP and 8-Br-GTP is
shown in Fig. 4 from which the relaxation times are estimated
to be 249 telem
͑622.5 ps͒ for GTP and 303 telem
͑757.5 ps͒ for 8-Br-GTP. This figure also shows that in order
to reach at least 40 global conformational transitions, we
perform REMD of GTP for 2750 telem
͑6875 ps͒ and REMD
of 8-Br-GTP for 3500telem
͑8750 ps͒. Only the parts of the
simulations after the relaxation times were used for quantitative
analysis of these simulations.
The dihedral angle distributions of the glycosidic bond
of GTP and 8-Br-GTP at different ␭-values are shown in
Figs. 5 and 6. The populations of syn and anti conformations
obtained by integration of the distributions between bounds
indicated in these figures are summarized in Tables II ͑GTP͒
and III ͑8-Br-GTP͒ together with the corresponding free energy
difference between the syn and anti states as calculated
by
⌬Gsyn-anti = − kBT ln
͓syn͔
͓anti͔
, ͑19͒
where kB is again the Boltzmann constant and T=298 K.
Usually we are most interested for values for unperturbed
interactions ͑␭=0.0͒ which are for GTP,
͓anti͔ = 95.6 % Ϯ 0.5 % ,
͓syn͔ = 4.4 % Ϯ 0.5 % ,
FIG. 3. Time dependence of dihedral angle ͑␾͒, around the glycosidic bond
during two MD simulations ͑starting from anti and syn configurations͒ for
͑a͒ GTP and ͑b͒ 8-Br-GTP. Anti and syn conformational regions are
indicated.
FIG. 4. Time dependence of the number of global transitions ͑Ngct͒ for
REMD of GTP and 8-Br-GTP. Relaxation time calculated from the extrapolation
of the linear part is for GTP 249 telem
͑elementary period, telem
=2.5 ps͒ and for 8-Br-GTP 303 telem
.
FIG. 5. Distribution of dihedral angle around the glycosidic bond of GTP at
different levels of softness ͑␭͒. Populations were calculated using a bin
width of 10°. Anti and syn conformational regions are indicated.
144121-5 Replica exchange molecular dynamics J. Chem. Phys. 128, 144121 ͑2008͒
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
⌬Gsyn-anti
GTP
= 7.6 Ϯ 0.3 kJ mol−1
,
and for 8-Br-GTP,
͓anti͔ = 6.0 % Ϯ 1.8 % ,
͓syn͔ = 94.0 % Ϯ 1.8 % ,
⌬Gsyn-anti
8-Br-GTP
= − 6.8 Ϯ 0.9 kJ mol−1
.
Error estimates on the populations are calculated from block
averages of occurrences of the conformational states followed
by an extrapolation to infinite block length.26
C. Thermodynamic integration using a hidden dihedral
angle restraint
The free energy difference between the syn and anti conformations
can also be calculated by thermodynamic integration
using a ͑hidden͒ dihedral angle restraint around the glycosidic
bond. Theoretically, the obtained value for the free
energy difference should not depend on the pathway by
which the system is pulled from one stable conformation to
another. However, in practice, this class of methods very
often shows hysteresis resulting from inadequate relaxation
of the systems at individual ␭-values. In order to obtain reliable
values, the free energy difference for both GTP and
8-Br-GTP was calculated through two different pathways
͓−120°͑anti͒↔60°͑syn͒,60°͑syn͒↔240°͑anti͔͒ in both
the forward and reverse directions ͑Figs. 7 and 8͒. This allows
us to determine the convergence of the free energy
difference and to monitor if sampling is sufficient for each
point at the individual pathways. We note that similar values
of ⌬G calculated from two opposite directions do not automatically
mean a complete convergence. Figure 7͑b͒ shows,
for example, that although the calculated values of ⌬G are
very similar ͑−9.2Ϯ1.2 and −9.5Ϯ1.3 kJ/mol͒ there are still
four ␭-values ͑0.333; 0.5; 0.556; 0.667͒ at which the values
of ͗‫ץ‬H/‫ץ‬␭͘␭ differ more than the statistical error estimate.
The conformational sampling for these points was not converged
even after 6 ns, and the small hysteresis is obtained
as the result of a fortuitous cancellation of errors. Note that
two simulations that are sampling conformations belonging
to different narrow minima can easily result in averages that
do not converge to the same value, while both show small
statistical error estimates. Indeed, we observed for ␭-values
right before and after the very local and steep conformational
barriers that the harmonic dihedral angle restraint results in a
dihedral angle distribution with two distinct maxima, sometimes
with insufficient transitions between them to obtain
converged results. Moreover, the true maximum of the barrier
is in these cases still not sampled. From the four transition
pathways shown in Figs. 7 and 8, only the
−120°͑anti͒↔60°͑syn͒ pathway for GTP shows a small
TABLE III. Relative syn and anti conformational populations of 8-Br-GTP
at different softness levels ͑␭͒. The corresponding free energy differences
are also calculated.
␭ ͓anti͔ ͓syn͔ ⌬Gsyn-anti͑kJ mol−1
͒
0.0 0.060 0.940 −6.81
0.2 0.051 0.949 −7.24
0.4 0.187 0.813 −3.64
0.55 0.320 0.680 −1.67
0.65 0.619 0.381 1.20
0.7 0.722 0.278 2.36
FIG. 7. Thermodynamic integration using hidden dihedral angle restraints
for GTP. Error estimates for individual points result from block averaging
and extrapolating the block length to infinity ͑Ref. 26͒.
FIG. 6. Distribution of the dihedral angle around the glycosidic bond of
8-Br-GTP at different levels of softness ͑␭͒. Populations were calculated
using a bin width of 10°. Anti and syn conformational regions are indicated.
TABLE II. Relative syn and anti conformational populations of GTP at
different softness levels ͑␭͒. The corresponding free energy differences are
also calculated.
␭ ͓anti͔ ͓syn͔ ⌬Gsyn-anti͑kJ mol−1
͒
0.0 0.954 0.044 7.62
0.25 0.860 0.140 4.49
0.45 0.768 0.232 2.96
144121-6 J. Hritz and C. Oostenbrink J. Chem. Phys. 128, 144121 ͑2008͒
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
hysteresis and the best convergence of ͗‫ץ‬H/‫ץ‬␭͘␭ values in all
␭-values. The steepness and curvature along this pathway
also seems smallest. It leads to ⌬Gsyn-anti
GTP
=7.7Ϯ1.2 kJ mol−1
which is very similar to the value obtained
from the REMD simulations of GTP ͑⌬Gsyn-anti
GTP
=7.6Ϯ0.3 kJ mol−1
͒. Despite of the lack of full convergence
in the other pathways, the calculated free energy differences
for GTP ͑⌬Gsyn-anti
GTP
between 9.2 and 9.5Ϯ1.3 kJ/mol͒ and
for 8-Br-GTP ͑⌬Gsyn-anti
8-Br-GTP
between −6.4 and
−4.1Ϯ1.6 kJ/mol͒ are still within kBT ͑2.5 kJ mol−1
͒ from
the values obtained from REMD simulations ͑⌬Gsyn-anti
GTP
=7.6Ϯ0.3 kJ mol−1
, ⌬Gsyn-anti
8-Br-GTP
=−6.8Ϯ0.9 kJ mol−1
͒.
IV. DISCUSSION
Any experimental observable that depends on the molecular
structures ͑e.g., 3
J-coupling values͒ does not result
from one single conformation but from an average over
many ͑all͒ conformations that are present and over the experimental
collection time. In order to compare such experimental
values with values obtained from calculation, it is
essential to generate the proper ensemble of conformations
corresponding to the same boundary conditions ͑temperature,
pressure͒ as at which the experiment was performed. According
to the ergodic theory, such an ensemble of conformations
can be generated by sufficiently long MD simulations. However,
in many biological systems the energy barriers separating
different conformations are so high that MD of even
hundreds of nanoseconds does not produce converged conformational
ensembles.
Examples of such biological systems are GTP and its
eight-substituted analog 8-Br-GTP which both have a high
energy barrier between their syn and anti conformations.
This barrier results from nonbonded interactions only. The
dihedral angle term around the glycosidic bond for these
molecules is zero in our models ͑see the force field parameters
in the supplementary material͒.25
While anti is the more
dominant conformation for GTP, syn is the more dominant
conformation for 8-Br-GTP. Experiments indicate, however,
that the less dominant conformations are still present for
non-negligible amounts of time ͑estimates range between 5%
and 30%͒.12,27
The presented study shows for both systems
that two 25 ns MD simulations ͑one started from anti and
one from syn͒ do not show any transition from the more to
the less stable conformation.
One possible way to sample such rare events in a more
efficient way is to force the system to cross the conformational
barrier. The application of hidden restraints17
belongs
to this class of methods and was shown to converge faster to
the free energy difference between unrestrained, stable, conformations
than, e.g., umbrella sampling.28
We applied a hidden
dihedral angle restraint around the glycosidic bond to
force transitions between the anti and syn conformations.
This approach does not directly yield the proper ensemble of
conformations, but such an ensemble can be approximated
by generating separate ensembles around two conformational
minima ͑anti, syn͒ and subsequently weight these ensembles
according to the free energy difference between them. Despite
its efficiency this method still required a relatively large
number of simulations ͑14–16͒ and many of these simulations
needed to be sampled for 6 ns in order to obtain a
reasonable convergence. The total simulation time for GTP
was 78 ns per pathway and for 8-Br-GTP 84 ns per pathway.
Note that in this approach, the actual end-points of unrestrained
anti or syn conformations are not simulated, but the
free energy derivative is approximated as zero. These would
still need to be simulated in order to approximate the conformational
ensemble as indicated above. In reality, these states
suffer from an end-state problem in which the derivative
fluctuates with unrestrained motion. Extrapolations towards
␭=0 and ␭=1 show that zero is a better approximation. As is
explained in Sec. III C, the full convergence for individual
͗‫ץ‬H/‫ץ‬␭͘␭ values was obtained only for GTP along the
−120°͑anti͒↔60°͑syn͒ pathway and the calculated value of
⌬Gsyn-anti
GTP
=7.7Ϯ1.2 kJ mol−1
is in excellent agreement with
⌬Gsyn-anti
GTP
=7.6Ϯ0.3 kJ mol−1
obtained from REMD of GTP.
The ͗‫ץ‬H/‫ץ‬␭͘␭ values are reasonably ͑but not fully͒ converged
for the other three pathways shown in Figs. 7 and 8,
and the calculated ⌬Gsyn-anti
GTP
values are still within an acceptable
range of 2.5 kJ mol−1
corresponding to thermal fluctuations
when compared to values calculated from REMD.
The observed problems with the convergence of these
simulations show that approaches which force the system to
cross steep and high energy barriers are not necessarily computationally
very efficient. Another important drawback of
these methods is the requirement of a suitable reaction coordinate
along which the system is forced to cross the energy
barrier. This means that they can only be practically applied
for simpler transition pathways.
REMD does not require prior knowledge about the transition
path and can therefore be applied to more complex
systems as well. In addition, REMD directly produces the
proper ensemble of conformations. However, its efficiency
can be much lower compared to methods forcing the molFIG.
8. Thermodynamic integration using hidden dihedral angle restraints
for 8-Br-GTP. Error estimates for individual points result from block averaging
and extrapolating the block length to infinity ͑Ref. 26͒.
144121-7 Replica exchange molecular dynamics J. Chem. Phys. 128, 144121 ͑2008͒
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
ecule to cross a conformational barrier. As is explained in the
Introduction, especially T-REMD becomes very inefficient
for systems containing explicit solvent.29
H-REMD involving
a judiciously chosen perturbation of the Hamiltonian is
less dependent on the overall system size and thus more generally
applicable. On the other hand, it may not always be
trivial to determine which parts of the Hamiltonian should be
perturbed.
In the presented H-REMD scheme the Hamiltonian is
modified over replicas by applying soft-core potentials for
selected interactions. The efficiency is increased compared to
other H-REMD approaches because soft-core potentials are
very similar to unperturbed ones at longer distances. This
means that only interactions between atoms that are close in
distance ͑contributing most to the high energy barriers͒ contribute
to the energy differences ͓⌬ in Eq. ͑12͔͒. The switching
probabilities are thus mainly governed by the interactions
that really matter for enhancing the transitions over energy
barriers. This is one of the main differences from other
H-REMD schemes5,6
or, e.g., the replica exchange with solute
tempering30
in which parts of the Hamiltonians are generally
scaled rather than modified in their functional form
͑e.g., softened͒. For GTP and 8-Br-GTP only three ͑GTP͒
and six ͑8-Br-GTP͒ different levels of softness were required,
thereby significantly reducing the time needed for the
replicas to diffuse from the highest level of softness to the
unperturbed Hamiltonian. In our previous study we showed
that the overall efficiency can still be increased by simulating
multiple replicas at the highest level of softness.11
This leads
to a total of six replicas for GTP and 12 replicas for 8-BrGTP
͑Table I͒. The REMD simulations were performed such
that after an initial equilibration period at least 40 global
conformational transitions are observed. Both the equilibration
time as well as the overall simulation time were obtained
by monitoring the linear increase of the number of
global transitions ͑Ngct͒. The total lengths of REMD simulation
can then be calculated as 6ϫ6.8=40.8 ns for GTP and
12ϫ8.7=104.4 ns of 8-Br-GTP. When comparing these
numbers with other methods ͑normal MD or thermodynamic
integration with hidden restraints͒, one should keep in mind
that REMD yields accurate free energy differences and the
ensemble of relevant conformations for these molecules,
which can be used for subsequent analysis.
In order to compare the efficiency of the presented
H-REMD with standard T-REMD, we performed T-REMD
for GTP using T0=298 K and Tmax=540 K at which
anti↔syn conversions are regularly observed. 50 replicas,
separated by 4 K ͑in the range of 298–370 K͒, 5 K
͑370–450 K͒, and 6 K ͑450–540 K͒, were needed to obtain
an average switching probability of 22.5%, close to the optimal
of 20%.31
This high number of replicas already implicates
a low efficiency of T-REMD for GTP in explicit water.
Indeed, T-REMD for 1 ns per replica, i.e., a total length of
50 ns ͑as compared to 40.8 ns used in H-REMD͒ did not
reveal any global conformational transitions at all despite the
occurrence of transitions at higher temperatures. The efficiency
of T-REMD for 8-Br-GTP in explicit water can be
expected to be even worse because MD up to 1500 K does
not show regular anti↔syn conversions.
The dihedral angle distributions P␭͑␸͒ in Figs. 5 and 6
allow us to generate the potential of mean force for GTP
͑Fig. 9͒ and for 8-Br-GTP ͑Fig. 10͒ using
G␭͑␸͒ = − kBT ln͑P␭͑␸͒͒ + C␭, ͑20͒
where C␭ is a constant which is determined based on five ␸
points within the anti and syn regions such that the difference
between G␭j͑␸͒−G␭i͑␸͒ is the same as calculated by the free
energy perturbation formula,
⌬Gij͑␸͒ = − kBT ln͗e−͑Hj−Hi͒/kBT
͘i. ͑21͒
Figures 9 and 10 form the free energy landscapes on which
the H-REMD simulations are performed and can be compared
to the schematic representation in Fig. 2. Because of
the nonmonotonous development of the free energy in the
direction of ␭ for 8-Br-GTP, a three dimensional representation
was very unclear. Rather, the individual curves are presented.
The constant C␭ cancels exactly in the switching
probability ͓Eqs. ͑9͒ and ͑12͔͒ and is thus irrelevant for the
REMD simulation. Figures 9 and 10 show that the soft-core
FIG. 9. Potential of mean force along the glycosidic bond of GTP as obtained
from the populations in Fig. 5. Curves at individual ␭-values were
shifted in order to fulfill the perturbation formula ͑21͒ between different
levels of softness for anti and syn conformations.
FIG. 10. Potential of mean force along the glycosidic bond of 8-Br-GTP as
obtained from the populations in Fig. 6. Curves at individual ␭-values were
shifted in order to fulfill the perturbation formula ͑21͒ between different
levels of softness for anti and syn conformations.
144121-8 J. Hritz and C. Oostenbrink J. Chem. Phys. 128, 144121 ͑2008͒
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
interactions at ␭max allow for a sufficient number of conformational
transitions because ͑i͒ the conformational barrier is
significantly decreased and ͑ii͒ the free energy difference between
the stable conformations is reduced. This results in a
higher population of the less dominant conformation and
thus in a further increased occurrence of the conformational
transitions. The free energy profiles also clearly show that
the transitions between anti and syn conformations occur almost
exclusively through the −120°͑anti͒↔60°͑syn͒ pathway
because the barrier for the 60°͑syn͒↔240°͑syn͒ pathway
is significantly higher. Note that this is also the pathway
in the thermodynamic integration simulations with hidden
restraints that corresponds best to the REMD simulations. It
is also interesting to observe how the interatomic interactions
“lose” their local identity with increasing levels of softness.
The nonsoft GTP and 8-Br-GTP have an inverse preference
for the syn conformation ͑4.4% for GTP and 94% for 8-BrGTP͒
but as can be seen from Tables II and III the syn
populations become more similar with rising ␭-values to a
final 23.2% for GTP at ␭=0.45 and 27.8% for 8-Br-GTP at
␭=0.7. Furthermore, one may note that the position of the
free energy minima is shifted at higher ␭-values ͑most obvious
for 8-Br-GTP at ␭=0.65 and 0.7͒. This, of course, raises
the question if the same dihedral angle intervals to define the
anti and syn regions should be used for all ␭-values and if
the populations at these ␭-values in Table III should not be
revised. However, as these simulations involve unphysically
perturbed molecules, the populations at these values have no
direct meaning.
Even though GTP and 8-Br-GTP were used in this study
mostly as model systems to show the application of our
REMD simulations, there clearly is an interest in the conformational
population of these molecules. Many six-or eightsubstituted
GTP and ATP analogs are being synthesized to
act as specific inhibitors.32–34
Bookser et al. correlated the
binding affinities of such compounds to their anti/syn preference
in water as obtained by 1
H NMR. Most adenosine
analogs prefer the syn conformations, but the compounds
with the highest adenosine kinase inhibitor potency all prefer
the anti conformation. The methods described here may directly
be applied to calculate the conformational preference
prior to compound synthesis.
In the future, we think the H-REMD approach using
soft-core interactions can be an asset to study ͑biomolecular͒
systems for which enhanced conformational sampling of a
smaller part is required. Examples are loop regions of proteins
͑work in progress͒ or orientational sampling of small
molecules bound to proteins ͑see, e.g., also our previous
work involving binding free energy calculations using softcore
interactions35,36
͒. The higher softness levels in these applications
allow parts of the system to partially overlap and
destabilize hydrogen bonds. Indeed, soft-core interactions allow
us to see transitions which are not observed at high
temperatures ͑T=1000 K͒ ͑unpublished results͒. This allows
us to produce a wide variety of conformations of these parts
of the system which are treated by soft-core interactions. Our
approach may seem similar to two variants of T-REMD for
local structure refinement: Partial and local replica exchange
molecular dynamics.37
These methods increase the temperature
only in parts of the system. This will undoubtedly lead
to a heat flow to the parts of the system that are still coupled
to lower temperatures, but the assumption is made that this
will not significantly modify the correct canonical ensemble.
This is not guaranteed at all and can be different from system
to system. Our approach does not induce a heat flow through
the unperturbed system and does not require any such assumption.
Moreover, it allows for even more flexibility by
making it possible to select individual interactions ͑not only
individual atoms or atom groups͒ which are treated by softcore
interactions, and each of these interaction can subsequently
still respond to the same ␭-value by specifying different
values for ␣vdW and ␣el ͓Eqs. ͑1͒ and ͑2͔͒.
As can be seen from the dihedral angle distributions in
Figs. 5 and 6, the produced isothermic-isobaric ensemble
does not only contain conformations from stable regions but
also a few conformations from the energy barrier region. An
approach involving only free energy estimates would not
sample these conformations. We have calculated the populations
of syn and anti conformations for GTP ͓͑anti͔=0.956,
͓syn͔=0.044͒ and 8-Br-GTP ͓͑͑anti͔͒=0.060, ͓syn͔=0.940͒.
These values compare qualitatively with NMR experiments
for guanosine and 8-Br-guanosine in dimethylsulfoxide
͑DMSO͒.12,27
We stress that these experiments involve a different
molecule in a different solvent and that several assumptions
underlay the estimates of populations from the
original NMR data. However, from the experiments it is
clear that 8-Br-GTP has an inverse population preference
with respect to GTP. Calculations to compare to original
NMR data are underway. For now, we would like to point
out that the quality of the obtained populations does not only
depend on the convergence of values due to the enhanced
sampling ͑main aim of our study͒ but also on the accuracy of
the force field parameters used for GTP and 8-Br-GTP. More
relevant is therefore the quantitative comparison ͑discussed
above͒ of the calculated free energy differences by the two
described methods as these both used the same force field
parameters.
V. CONCLUSIONS
We present a novel Hamiltonian REMD scheme using
soft-core potentials, which allows for effective REMD simulations
with only a few replicas. The application of the
method was demonstrated for two realistic biomolecules,
GTP and 8-Br-GTP, in explicit water. By using soft-core interactions
we only perturb those parts of the Hamiltonian that
contribute most to a local free energy barrier. This results in
efficient H-REMD simulations using only a few different
levels of softness ͑three for GTP and six for 8-Br-GTP͒ and
thus to a fast diffusion of the replicas between the lowest and
the highest levels of softness. In order to give the systems
time to undergo conformational transitions at the highest
level of softness, we applied a degenerate highest softness
level scheme. The optimal settings of the H-REMD schemes
were obtained from an optimization procedure as described
in our previous study.11
The optimization procedure utilizes
the number of global conformational transitions. The time
144121-9 Replica exchange molecular dynamics J. Chem. Phys. 128, 144121 ͑2008͒
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
dependence of this quantity was used to determine the relaxation
and production times of H-REMD simulations. 40 global
conformational transitions ͑20 from anti to syn and 20
from syn to anti͒ were observed within 6.8 ns of H-REMD
simulation for GTP and 8.7 ns for 8-Br-GTP. No single transition
from the more to the less dominant conformation was
observed for either GTP or 8-Br-GTP within two 25 ns normal
MD simulations ͑starting from initial anti and syn conformations͒
nor from T-REMD simulations of comparable
length for GTP in explicit solvent. The obtained free energy
differences between anti and syn conformations are in quantitative
agreement with values calculated using thermodynamic
integration with hidden dihedral angle restraints
around the glycosidic bond and also in qualitative agreement
with NMR estimates. The presented H-REMD approach using
soft-core interactions allows for a softening of very specific,
selected interactions which makes it a powerful technique
to enhance the conformational sampling of smaller
molecules in explicit solvent or of flexible parts of large
biomacromolecules in explicit solvent.
ACKNOWLEDGMENTS
This work was supported by the Netherlands Genomics
Initiative in the context of Horizon Breakthrough Grant No.
050-71-043 ͑J.H.͒ and by the Netherlands Organization for
Scientific Research, VENI Grant No. 700.55.401 ͑C.O.͒.
1
C. L. Brooks III, M. Karplus, and B. M. Pettitt, Advances in Chemical
Physics ͑Wiley, New York, 1988͒, Vol. 71.
2
M. Leitgeb, C. Schroder, and S. Boresch, J. Chem. Phys. 122, 084109
͑2005͒.
3
J. P. Valleau and S. G. Whittington, in Statistical Mechanics, edited by B.
J. Berne ͑Plenum, New York, 1977͒, Chap. 4, p. 145.
4
U. H. E. Hansmann, Chem. Phys. Lett. 281, 140 ͑1997͒.
5
H. Fukunishi, O. Watanabe, and S. Takada, J. Chem. Phys. 116, 9058
͑2002͒.
6
Y. Sugita, A. Kitao, and Y. Okamoto, J. Chem. Phys. 113, 6042 ͑2000͒.
7
T. C. Beutler, A. E. Mark, R. C. van Schaik, P. R. Gerber, and W. F. van
Gunsteren, Chem. Phys. Lett. 222, 529 ͑1994͒.
8
D. D. Frantz, D. L. Freeman, and J. D. Doll, J. Chem. Phys. 93, 2769
͑1990͒.
9
H. Li, G. Li, B. A. Berg, and W. Yang, J. Chem. Phys. 125, 144902
͑2006͒.
10
A. Okur, D. R. Roe, C. Guanglei, V. Hornak, and C. Simmerling, J.
Chem. Theory Comput. 3, 557 ͑2007͒.
11
J. Hritz and C. Oostenbrink, J. Chem. Phys. 127, 204104 ͑2007͒.
12
R. Stolarski, C. E. Hagberg, and D. Shugar, Eur. J. Biochem. 138, 187
͑1984͒.
13
W. F. van Gunsteren, S. R. Billeter, A. A. Eising, P. H. Hünenberger, P.
Krüger, A. E. Mark, W. R. P. Scott, and I. G. Tironi, Biomolecular Simulation:
The GROMOS96 Manual and User Guide ͑Vdf Hochschulverlag
AG an der ETH Zürich, Zürich, 1996͒.
14
M. Christen, P. H. Hunenberger, D. Bakowies, R. Baron, R. Burgi, D. P.
Geerke, T. N. Heinz, M. A. Kastenholz, V. Krautler, C. Oostenbrink, C.
Peter, D. Trzesniak, and W. F. Van Gunsteren, J. Comput. Chem. 26,
1719 ͑2005͒.
15
R. H. Swendsen and J. S. Wang, Phys. Rev. Lett. 57, 2607 ͑1986͒.
16
Y. Sugita and Y. Okamoto, Chem. Phys. Lett. 314, 141 ͑1999͒.
17
M. Christen, A.-P. E. Kunz, and W. F. Van Gunsteren, J. Phys. Chem. B
110, 8488 ͑2006͒.
18
J. G. Kirkwood, J. Chem. Phys. 3, 300 ͑1935͒.
19
J.-P. Ryckaert, G. Ciccotti, and H. J. C. Berendsen, J. Comput. Phys. 23,
327 ͑1977͒.
20
R. W. Hockney, Methods Comput. Phys. 9, 136 ͑1970͒.
21
H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, A. DiNola, and
J. R. Haak, J. Chem. Phys. 81, 3684 ͑1984͒.
22
H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, and J. Hermans,
in Intermolecular Forces, edited by B. Pullman ͑Reidel, Dordrecht,
1981͒, p. 331.
23
I. G. Tironi, R. Sperb, P. E. Smith, and W. F. van Gunsteren, J. Chem.
Phys. 102, 5451 ͑1995͒.
24
C. Oostenbrink, A. Villa, A. E. Mark, and W. F. van Gunsteren, J. Comput.
Chem. 25, 1656 ͑2004͒.
25
See EPAPS Document No. E-JCPSA6-128-509811 for force-field parameters
of GTP and 8-Br-GTP. For more information on EPAPS, see http://
www.aip.org/pubservs/epaps.html.
26
M. P. Allen and D. J. Tildesley, Computer Simulation of Liquids ͑Clarendon,
Oxford, 1987͒.
27
H. Rosemeyer, G. Toth, B. Golankiewicz, Z. Kazimierczuk, W. Bourgeois,
U. Kretschmer, H. P. Muth, and F. Seela, J. Org. Chem. 55, 5784
͑1990͒.
28
G. M. Torrie and J. P. Valleau, J. Comput. Phys. 23, 187 ͑1977͒.
29
J. W. Pitera and W. C. Swope, Proc. Natl. Acad. Sci. U.S.A. 100, 7587
͑2003͒.
30
P. Liu, B. Kim, R. A. Friesner, and B. J. Berne, Proc. Natl. Acad. Sci.
U.S.A. 102, 13749 ͑2005͒.
31
A. Kone and D. A. Kofke, J. Chem. Phys. 122, 206101 ͑2005͒.
32
T. Läppchen, A. F. Hartog, V. A. Pinas, G. J. Koomen, and T. den Blaauwen,
Biochemistry 44, 7879 ͑2005͒.
33
F. Schwede, A. Christensen, S. Liaw, T. Hippe, R. Kopperud, B. Jastorff,
and S. O. Doskeland, Biochemistry 39, 8803 ͑2000͒.
34
B. C. Bookser, M. C. Matelich, K. Ollis, and B. G. Ugarkar, J. Med.
Chem. 48, 3389 ͑2005͒.
35
B. C. Oostenbrink, J. W. Pitera, M. M. H. Van Lipzig, J. H. N. Meerman,
and W. F. van Gunsteren, J. Med. Chem. 43, 4594 ͑2000͒.
36
C. Oostenbrink and W. F. van Gunsteren, Proteins 54, 237 ͑2004͒.
37
X. Cheng, G. Cui, V. Hornak, and C. Simmerling, J. Phys. Chem. B 109,
8220 ͑2005͒.
144121-10 J. Hritz and C. Oostenbrink J. Chem. Phys. 128, 144121 ͑2008͒
Downloaded 18 Apr 2008 to 130.37.148.58. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
METHODOLOGICAL DEVELOPMENTS OF ENHANCED SAMPLING METHODOLOGY H-REMD
45
Paper 6
Hritz J, Oostenbrink C. Optimization of Replica Exchange Molecular Dynamics by Fast
Mimicking. J. Chem. Phys. 2007, 127, 204104
Optimization of replica exchange molecular dynamics by fast mimicking
Jozef Hritz and Chris Oostenbrinka͒
Leiden Amsterdam Center for Drug Research (LACDR), Division of Molecular Toxicology,
Vrije Universiteit, Amsterdam, The Netherlands
͑Received 8 August 2007; accepted 5 September 2007; published online 27 November 2007͒
We present an approach to mimic replica exchange molecular dynamics simulations ͑REMD͒ on a
microsecond time scale within a few minutes rather than the years, which would be required for real
REMD. The speed of mimicked REMD makes it a useful tool for “testing” the efficiency of different
settings for REMD and then to select those settings, that give the highest efficiency. We present an
optimization approach with the example of Hamiltonian REMD using soft-core interactions on two
model systems, GTP and 8-Br-GTP. The optimization process using REMD mimicking is very fast.
Optimization of Hamiltonian-REMD settings of GTP in explicit water took us less than one week.
In our study we focus not only on finding the optimal distances between neighboring replicas, but
also on finding the proper placement of the highest level of softness. In addition we suggest different
REMD simulation settings at this softness level. We allow several replicas to be simulated
at the same Hamiltonian simultaneously and reduce the frequency of switching attempts between
them. This approach allows for more efficient conversions from one stable conformation to the
other. © 2007 American Institute of Physics. ͓DOI: 10.1063/1.2790427͔
I. INTRODUCTION
Since the introduction of the replica exchange method
͑REM͒ using the Monte Carlo algorithm, replica exchange
molecular dynamics ͑REMD͒ or parallel tempering simulations
in the late 1990s,1–5
there has been a steep increase in
their popularity. In these simulations schemes the conformational
sampling of a molecular system is greatly enhanced by
connecting multiple simulations that are performed at different
temperatures ͓temperature REMD ͑T-REMD͔͒ or using
different functional forms of the potential interaction energy
͓Hamiltonian REMD ͑H-REMD͔͒.6
By maintaining a detailed
balance requirement when individual simulations are
switched to a different temperature or Hamiltonian, the correct
thermodynamic ensemble will be obtained for each of
the simulation parameters. An intelligent choice of REMD
settings allows for the swift generation of canonical ensembles
of systems in which the potential energy barriers
between stable conformations are too large to be crossed
repeatedly in a normal molecular dynamics ͑MD͒ simulation.
In this paper we will introduce an approach to efficiently
optimize the settings for REMD simulation for systems with
multiple stable conformations. Settings to optimize involve
the number of simultaneous simulations ͑replicas͒ and the
optimal simulation settings for each of these simulations
͑temperature, Hamiltonian͒.
Only researchers with access to extraordinary computational
resources can afford a trial and error approach when
searching for efficient REMD settings. Several studies describe
approaches leading to efficient REMD simulations,
mostly for T-REMDs. ͓Because one MD step requires the
same amount of CPU time for any temperature or Hamiltonian,
the allocation of replicas to CPUs can be trivial ͑one
replica per CPU͒. However, in replica exchange Monte Carlo
simulations, the average wall clock time to complete one
Monte Carlo move varies with the temperature or the Hamiltonian,
and the CPU allocation becomes an important issue.7
͔
Many authors claim that the highest efficiency is obtained
when the switching probability between neighboring replicas
is constant at a value of approximately 20%.8–13
This is still
the most often used criterion in REMD applications despite
the fact that in 2004 Trebst et al. presented the feedbackoptimized
parallel tempering approach.14
It was shown that
for the optimal temperature sets the switching probabilities
between neighboring replicas are not constant but rather depend
on the temperature.14–16
A significant practical drawback
of this approach, however, is the simulation time required
to obtain the optimized settings. Especially for
biomolecular simulations, this hampers the practical applicability
of the method. The application of feedback-optimized
parallel tempering for the 36-residue villin headpiece subdomain
HP-36 required REMD simulations that covered
400 000 switching trials, which takes many years of CPU
time.16
It is probably for this reason that less applications of
the method have been described.
For this reason we have developed a set of efficient tools
for optimizing the settings of REMD ͑both T- and H-REMD͒
simulations for systems, of which multiple stable conformations
are known. By generating the appropriate conformational
ensemble of the system REMD gives insight into the
relative populations of stable conformations. We present the
practical application of the proposed optimization scheme for
two biologically relevant systems: GTP and 8-Br-GTP
͑Fig. 1͒. GTP is an important component in, e.g., cellular
signaling, while 8-Br-GTP is considered as promising anti-a͒
Electronic mail: c.oostenbrink@few.vu.nl
THE JOURNAL OF CHEMICAL PHYSICS 127, 204104 ͑2007͒
0021-9606/2007/127͑20͒/204104/13/$23.00 © 2007 American Institute of Physics127, 204104-1
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
bacterial agent. It inhibits FtsZ polymerization but does not
affect tubulin polymerization.17
GTP and 8-Br-GTP both
have two stable orientations of the base toward the sugar:
anti and syn ͑Fig. 1͒. Nuclear magnetic resonance ͑NMR͒
studies show that while GTP prefers to be in the anti conformation
͑with an estimated population of ϳ70% from NMR
experiments͒, 8-Br-GTP shows preference for the syn conformation
͑population ϳ90%͒. Our MD study shows that
there is a high energy barrier between the anti and syn conformations
for both molecules. In two MD simulations of
GTP starting from anti and syn conformation ͑each of 20 ns͒
only one single transition from syn to anti was observed
͑after ϳ1 ns͒ while no anti→syn transition was seen at all.
Analogous simulations of 8-Br-GTP did not reveal any transition
indicating an even higher potential energy barrier between
the anti and syn conformations.18
In these cases
REMD can be used to enhance the conformational sampling
and obtain the correct conformational ensemble with the
proper populations of syn and anti conformations. From this
ensemble, the free energy difference between the two states
as well as a variety of molecular properties can be calculated.
II. METHODS
A. MD settings
All MD simulations were conducted using the GROMOS05
MD simulation package running on a linux cluster.19
All bonds were constrained, using the SHAKE algorithm20
with a relative geometric accuracy of 10−4
, allowing for a
time step of 2 fs used in the leapfrog integration scheme.21
Periodic boundary conditions, with a truncated octahedral
box, were applied. After the steepest descent minimization to
remove bad contacts between molecules, the initial velocities
were randomly assigned from a Maxwell–Boltzmann distribution
at 298 K, according to the atomic masses. The temperature
was controlled using a weak coupling to a bath of
298 K with a time constant of 0.1 ps.22
The solute molecules
͑GTP or 8-Br-GTP͒ and solvent ͑i.e., explicit water molecules
and 3 Na+
counterions͒ were independently coupled to
the heat bath. The pressure was controlled using isotropic
weak coupling to atmospheric pressure with a time constant
0.5 ps.22
Van der Waals and electrostatic interactions were
calculated using a triple range cutoff scheme. Interactions
within a short-range cutoff of 0.8 nm were calculated every
time step from a pair list that was generated every five steps.
At these time points, interactions between 0.8 and 1.4 nm
were also calculated and kept constant between updates. A
reaction-field contribution was added to the electrostatic interactions
and forces to account for a homogeneous medium
outside the long-range cutoff, using the relative permittivity
of SCP water ͑61͒.23
All interaction energies were calculated
according to the GROMOS force field, parameter set 53A6.24
Force field parameters used for GTP and 8-Br-GTP are listed
in the supplementary material of Ref. 18.
In this work we will focus on the optimization of a
H-REMD approach in which the modification of the Hamiltonian
consists of a softening of selected interactions. For
this reason we have used the following form for Van der
Waals and electrostatic soft-core interactions as a function of
the interatomic distance rij:25
E␭
vdw
͑rij͒ = ͩ C12
A␭ + rij
6 − C6ͪ 1
A␭ + rij
6 , ͑1͒
E␭
el
͑rij͒ =
qiqj
4␲␧
1
ͱB␭ + rij
2
, ͑2͒
where A␭=␣vdw͑C12/C6͒␭2
and B␭=␣el␭2
. C12 and C6 are the
Lennard–Jones parameters, qi and qj are the partial charges
of particles i and j, and ␣vdw and ␣el are the softness parameters
that can be set for every pair of atoms. In the current
study we used in all simulations ␣vdw=␣el=1 and the softness
of the interactions was controlled through the ␭ parameter.
It can be seen that at longer distances the soft-core
interaction approximates the interaction for normal atoms
and that they differ mostly at short distances between atoms.
Potential energy barriers are mostly the result of short-ranged
repulsion between atoms, which can strongly be reduced at
higher levels of softness ͑Fig. 2͒. In this study, only interactions
between the nucleotide base and sugar are treated using
the soft-core interaction.
The systems were first equilibrated for 100 ps MD at
constant pressure, where position restraints were applied on
FIG. 1. Structure of GTP and 8-Br-GTP in syn conformation. Conformational
transitions between the syn and anti states occur by rotation around
the glycosidic bond ͑indicated͒.
FIG. 2. Schematic figure of the energy landscape profile as function of the
softness of nonbonded interactions. In the presented H-REMD using softcore
interactions, the replicas at higher levels of softness have a higher
probability for the conformational transition between two stable conformational
states. Still, this transition requires some time.
204104-2 J. Hritz and C. Oostenbrink J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
heavy atoms of GTP/8-Br-GTP, after which another 100 ps
of equilibration at constant pressure without any restraints
were carried out. Finally, the systems were simulated for 1 ns
at different ␭ values, where coordinates of the whole system
were recorded every 1 ps. Ten independent 1 ns MD simulations
at different levels of softness were performed for
GTP ͑␭=0.0,0.05, ... ,0.4,0.45͒ and 15 for 8-Br-GTP
͑␭=0.0,0.05, ... ,0.65,0.7͒.
B. REMD
In a REMD simulation, a number of noninteracting MD
runs ͑called replicas͒ are simulated at different conditions.
Let us label replicas as 0,1, ... ,n. After a given time
͑elementary period, telem
͒, an exchange between two neighboring
replicas is attempted, followed by another set of independent
MD simulations. In the GROMOS05 implementation
switches are first attempted between pairs
0↔1,2↔3,... ͑type I of replica exchange trials͒ and after
the next telem
of MD simulations switches are subsequently
attempted between pairs 1↔2,3↔4,... ͑type II of replica
exchange trials͒. This means that the effective time between
switching attempts of the same type is twice the elementary
period. In our study we used telem
=2.5 ps leading to an effective
period of 5 ps between identical switching attempts.
In contrast to T-REMD, where the temperature is increased
to facilitate the crossing of high energy barriers, we
are modifying the Hamiltonian within H-REMD by making
use of the soft-core interactions, Eqs. ͑1͒ and ͑2͒. This will
lead to a decrease of the potential energy barrier between
conformations and thus also facilitate the transition from one
potential energy minimum into the other ͑see Fig. 2͒. The
proper ensemble of conformations at every value of ␭ can be
obtained by applying the Metropolis criterion for the exchange
probability w␭i,␭j
between neighboring replicas running
at Hamiltonians corresponding to the parameters ␭i and
␭j,
w␭i,␭j
= min͓1,exp͑− ⌬␭i,␭j
͔͒, ͑3͒
where
⌬␭i,␭j
= ␤͓E␭j
͑q␭i
͒ − E␭i
͑q␭i
͒ + E␭i
͑q␭j
͒ − E␭j
͑q␭j
͔͒. ͑4͒
The coordinates q␭i
represent a conformation that was obtained
from a simulation at the Hamiltonian corresponding to
parameter ␭i and E␭j
is the potential energy calculated according
to the Hamiltonian corresponding to the parameter
␭j ͓Eqs. ͑1͒ and ͑2͔͒. ␤=1/kBT with kB as the Boltzmann
constant and T as the absolute temperature.
C. Concept of double/multiple replicas at the highest
lambda/temperature
The elementary period, telem
, in REMD should be long
enough to relax the energy before the next switching attempt.
Otherwise one observes many “reswitches” at the next
switching attempt of the same type after a replica switch
with low probability. Of course one can increase telem
, but
then the overall REMD efficiency decreases with decreasing
number of replica switches. Therefore, a reasonable balance
will need to be struck.
However, there are more relaxation processes that play a
role. Even at values of ␭ where the barrier has been removed
or at temperatures where it is readily crossed, the system still
needs time to move from one conformation to the other.
͑Note: Even for a very small conformational barrier the average
transition time is significantly longer than times generally
used in REMD between switching trials. This is
mainly true for transitions to the less favorable conformational
states.͒ The occurrence of syn states for a set of simulations
starting at anti under the conditions of the highest ␭
value or temperature and the occurrence of anti states for
simulations starting from a syn conformation as a function of
time, will typically be represented by a sigmoidal curve.
We define the transition time t␭max
transit
as the time at
which this sigmoidal curve reaches saturation. The height of
the curve allows us to estimate the relative populations of
syn and anti at this ␭ value and from that the transition
probabilities at times larger than t␭max
transit
, P␭max
anti→syn
͑t␭max
transit
͒, and
P␭max
syn→anti
͑t␭max
transit
͒.
The conformational transition time, t␭max
transit
, is for the majority
of systems much longer than telem
. If telem
would be
extended to t␭max
transit
, then the efficiency gain of REMD becomes
very low. In order to combine frequent switching trials
with long enough relaxation times to allow for anti↔syn
conformational transitions to occur at ␭max we introduce a
new scheme, called degenerated ␭max scheme, involving
multiple ͑n͒ replicas at ␭max ͑or at the highest temperature
for T-REMD͒.
This can be done by “degenerating” ␭max into
␭max,0,␭max,1, ... ,␭max,n−1. In the switching scheme all
␭ values are ordered as ␭0,␭1, ... ,␭max−1,
␭max,0,␭max,1, ... ,␭max,n−1, and only switches are attempted
between neighboring ␭ values. The switching frequency between
replicas at ␭max is reduced by only allowing switching
attempts after a given multiple of the elementary period ͑Fig.
3͒. We define the time t␭max
total
as the total simulation time beFIG.
3. Schematic example of real REMD with three replicas at ␭max
͑␭2,0 =␭2,1 =␭2,2 =␭max͒ and one middle ␭1 illustrating global conformational
transitions, indicated by a diamond symbol. Crosses indicate the presence of
syn conformation. Switches between replicas at ␭2,0 and ␭2,1 are attempted
only at times, t, which are a multiple of 5 telem
and between ␭2,1 and ␭2,2 if
t+telem
is a multiple of 5 telem
. It ensures that a replica that switches from ␭2,0
to ␭2,1 will spend the required time, t␭max
total
=10telem
at ␭2,1 and ␭2,2 altogether.
204104-3 Optimization of replica-exchange molecular dynamics J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
tween the switch ␭max,0→␭max,1 until the return switch
␭max,1→␭max,0. Note that the switching probability between
replicas at ␭max equals 1 because they correspond to the same
Hamiltonian and ⌬ in Eq. ͑4͒ becomes 0 by definition. The
procedure to find the optimal t␭max
total
is described in Sec. II D 2.
The switching trial frequency between all other pairs of replicas
is much higher ͑with an elementary period of 2.5 or 5
ps between two switching trials of the same type͒, which
maintains an efficient diffusion of replicas between the lowest
and highest ␭ value.
An increasing number of replicas at ␭max increases the
convergence of conformational populations at ␭max, which
subsequently leads to a faster convergence of conformational
populations at ␭max−1,␭max−2,... and finally to the faster convergence
of populations over the whole REMD system. The
REMD efficiency gain when using more replicas at ␭max
comes from the fact that multiple conformational transition
“attempts” are performed in parallel.
With n replicas at ␭max, a replica that has had sufficient
time to show conformational transition becomes available
at ␭max,0 with a period of ͑1/n−1͒t␭max
total
. This allows
for values of n up to nmax=t␭max
total
/telem
+1, although the efficiency
is limited in this case because replicas tend to stay for
several periods of t␭max
total
at ␭max rather than get an opportunity
to switch down to ␭max−1. For this reason, the maximal overall
efficiency is obtained at nϷw␭max−1,␭max,0
nmax. Especially
for complex systems the required time t␭max
total
at ␭max can be
high, so a higher number of replicas at ␭max can significantly
increase the overall REMD efficiency, almost by a factor
͑n−1͒. For the examples described in this work, a onedimensional
setup with multiple replicas at ␭max seems sufficient,
but the concept can easily be extended to having a
degenerate set of replicas at a certain ␭ value or temperature
by going to a second or third dimension of ␭ or T. See Fig. 4.
D. Searching for the optimal REMD settings
The efficiency of REMD to sample conformational
space depends mainly on the efficiency of conformational
transitions at ␭max and an efficient diffusion of replicas between
the lowest and highest ␭ values. In this section, we
describe the tools that are used to optimize the choice of ␭
values and the number of replicas at ␭max.
Section II D 1 describes the algorithm we propose for
the fast mimicking of REMD without performing actual MD.
This algorithm requires knowledge of the optimal value of
␭max, the optimal conformational conversion time at ␭max,
t␭max
total
, as well as estimates of the saturated conformational
transition probabilities, P␭max
anti→syn
, P␭max
syn→anti
. In addition, the
algorithm makes use of switching probabilities between replicas
at different ␭ values. Section II D 2 describes the selection
of ␭max, the calculation of t␭max
total
, and conversion probabilities.
Section II D 3 describes how switching
probabilities are sampled from precalculated probability distributions
and Sec. II D 4 finally describes the approach we
take to use the mimicking algorithm to obtain the most efficient
settings for REMD simulations.
1. Mimicked REMD
Mimicked REMD assumes a number of discreet stable
conformational states exists, labeled, c=0,1, ... ,Nc−1 ͑in
our case anti and syn͒. In the present case these conformations
are described as anti and syn. Instead of performing
MD we estimate the probability of a conformational transition
as well as the REMD switching probabilities and subsequently
propagate the conformational states through time.
This requires knowledge of the probability distributions of
switching probabilities ͕w␭i,␭j
ck,cl ͖ for all neighboring pairs of ␭
values: ␭i, ␭j as well as approximate probabilities of the
conformational changes at each ␭ value from one stable conformational
region to the other ones. We would like to note
that if precise conformational transition probabilities would
be known then no REMD mimicking is needed, as one could
easily derive the individual conformational populations as
well as the free energy difference between the stable conformations.
The merit of the current approach comes from the
fact that the transition probabilities can be estimated for ␭max
fairly easily, while they are ϳ0 for all other ␭ values.
As explained in Sec. II C the transition between conformational
states is a dynamic process in which the conformational
transitions depend sigmoidally on time. From these
curves at ␭max, one can estimate the probability of the transition
ci→cj, after the ͑prolonged͒ residence time,
P␭max
ci→cj͑t␭max
transit
͒. Because ␭max will be selected such that the
corresponding t␭max
total
is shortest ͑see Sec. II D 2͒ and because
for all other replicas the time between switching attempts
͑2telem
͒ is much shorter than the corresponding t␭
total
, it is
reasonable to assume that the probabilities of conformational
transitions at any ␭ value other than ␭max approach zero
͓P␭ ␭max
ci→cj ͑2telem
͒=0͔.
In our REMD mimicking algorithm we begin by assigning
starting conformational states to all replicas. This can be
done either randomly, sampled from the correct ensemble, or
biased toward one of the states.
Subsequently four steps make up the main cycle of our
algorithm:
FIG. 4. Schematic figure showing the application of degenerate ␭’s. By
limiting the number of switching attempts between replicas at the same ␭
value, the systems are allowed more time to relax toward different conformations
at this value of ␭.
204104-4 J. Hritz and C. Oostenbrink J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
͑1͒ Intralambda conformational transitions. We predict if
the conformational state changes during the elementary
period ͑in our case telem
=2.5 ps͒ at each ␭ value by the
given probabilities. As described above all conformational
transition probabilities are set to zero for all ␭
values, except for ␭max, where P␭max
ci→cj͑t␭max
total
͒
=P␭max
ci→cj͑t␭max
transit
͒ is nonzero once per number of cycles
corresponding to the t␭max
total
spent at ␭max,1, ... ,␭max,n−1.
͑2͒ Replica exchanges of type I ͑between 0↔1,2↔3,...͒.
According to the actual conformational state at each ␭
value we take the corresponding probability distributions
of switching probabilities ͕w␭i,␭j
ck,cl ͖ for all neighboring
␭ pairs of type I. A switching probability is randomly
chosen from the distribution of switching
probabilities ͕w␭i,␭j
ck,cl ͖ and a second random number determines
if the switch occurs.
͑3͒ Intralambda conformational transitions. The same as
the first step.
͑4͒ Replica exchanges of type II ͑between 1↔2,
3↔4,...͒. The same as the second step but now for the
␭ pairs of type II.
The length of the whole cycle corresponds to the double
of elementary switching attempt period ͑in our case 2
ϫ2.5 ps=5 ps͒. Sampling of a half million cycles ͑corresponding
to 106
telem
of real REMD͒ takes typically 15 min
͑depending on the exact number of replicas͒ using an unoptimized
python script.
2. Selection of ␭max
The average conformational transition time at the unperturbed
Hamiltonian ͑␭=0.0͒, t0.0
transit
is by far too long to
sample sufficient transitions reversibly. In H-REMD, the
Hamiltonian is parameterized such that the potential energy
barrier associated to the conformational transition is reduced
for higher values of ␭, thereby also reducing t␭i
transit
.
To obtain the optimal value of ␭max, we perform ten
short MD simulations ͑200 ps͒ starting from anti and 10
short MD runs starting from syn conformation at different ␭
values ͑for GTP ␭=0.4;0.45;0.5 and for 8-Br-GTP ␭
=0.6;0.65;0.7;0.8;0.9͒. ͑Note: ␭max–values lower than 0.4
respectively. 0.6 were not considered based on the results of
1 ns runs at different ␭ values used for the calculation of
switching probability distributions as described in the next
paragraph.͒
Our aim is to find the value of t␭max
total
for which we would
get the highest number of conformational transitions at ␭max
during a given length of REMD simulation ͑see also Sec.
II C͒. While the time dependency of the number of simulations
where the conformation has changed with respect to the
initial conformation has an S-curve profile, it is clear that the
most efficient t␭max
total
will be in the interval between the midpoint
of the S curve and the saturated region. Quantitatively
t␭max
total
is obtained as the time corresponding to the maximum
of the number of conformational changes divided by time.
Because these maxima occur at different times for the anti
→syn transition and the syn→anti transitions, and because
we need to have a large enough number of both types of
transitions, we select the longer time of both transitions at
one ␭max value. We select the value of ␭max for which
t␭max
total
is shortest and where enough transitions occur in
both directions. The set of ten MD runs starting from different
conformations is subsequently prolonged for the selected
␭max, in order to refine the converged values of the conformational
conversion probabilities P␭max
anti→syn
͑t␭max
transit
͒ and
P␭max
syn→anti
͑t␭max
transit
͒. Because the exact shape of the S curve is
not properly converged from only ten simulations, it is not
wise to directly read P␭max
transit
͑t␭max
total
͒ from these curves. Rather,
we set P␭max
transit
͑t␭max
total
͒=P␭max
transit
͑t␭max
transit
͒, a value which converges
also for a limited number of simulations. Using P␭max
transit
͑t␭max
transit
͒
also ensures the generating the proper conformational populations
at ␭max within the mimicked REMD.
3. Generation of probability distributions of switching
probabilities
Let us consider the system, which has several different
stable conformational sates ͑syn, anti͒, for which standard
MD simulations do not produce enough transitions due to the
conformational barriers between these states. We assume that
the conformational space within each of these stable regions
is sampled “reasonably” well by a relatively short MD simulation
͑ϳ1 ns͒. The description below follows the schematic
representation in Fig. 5.
To obtain the probability distributions of the switching
probabilities needed for the REMD mimicking we performed
a set of 1 ns MD simulations starting from different stable
conformers at different values of ␭ ͑alternatively, one could
use different temperatures͒ between 0.0 and ␭max usually at
increments of 0.05 ͑in principle one can use also REMD at
these ␭ values for this purpose͒. Every 1 ps we store one
conformational configuration frame leading to 1000 different
configurations. It is possible that during these MD simulations
conformational transitions occur. For this reason, we
collect conformations belonging to the same stable conformational
region c and the same value of ␭ at which the
simulation was performed into one set of conformations
͕q␭
c
͖,␭ = 0,0.05,0.1, ... ,␭max;
c = 0,1, ... ,Nc − 1.
For every combination of ␭ and c we obtain about 1000
different structures. The rth structural frame from this “trajectory”
is expressed as q␭
c
͑tr͒. For each of these structures
we calculate its energy not only using the Hamiltonian at
which it was simulated ͑represented by ␭i͒, but also the energy
according to the Hamiltonian corresponding to all other
␭ values ͑␭j͒, and expressed as: E␭j
͑q␭i
c
͑tr͒͒.
For each pair of configurations q␭i
ck͑tr͒; q␭j
cl ͑ts͒ we calculate
switching energy difference
⌬␭i,␭j
ck,cl ͑tr,ts͒ = ␤͓E␭j
͑q␭i
ck͑tr͒͒ − E␭i
͑q␭i
ck͑tr͒͒ + E␭i
͑q␭j
cl ͑ts͒͒
− E␭j
͑q␭j
cl ͑ts͔͒͒ ͑5͒
and the corresponding switching probability
204104-5 Optimization of replica-exchange molecular dynamics J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
w␭i,␭j
ck,cl ͑tr,ts͒ = min͓1,exp͑− ⌬␭i,␭j
ck,cl ͑tr,ts͔͒͒, ͑6͒
where ␤=1/kBT, kB is the Boltzmann constant, and T is the
absolute temperature.
The distributions of ⌬␭i,␭j
ck,cl and w␭i,␭j
ck,cl values are thus calculated
from altogether ϳ106
values, which are marked as
͕⌬␭i,␭j
ck,cl ͖ and w0.0,0.3
syn,anti
after normalization. This gives us the
probability distribution for a randomly chosen pair of configurations,
q␭i
ck, q␭j
cl with the switching probability between
them given by w␭i,␭j
ck,cl . For the REMD mimicking we found it
convenient to produce Cum_͕w␭i,␭j
ck,cl ͖, which contains the cumulants
of the probability distribution of switching prob-
abilities.
4. Searching algorithm for the parameters
of the most efficient REMD
500 000 cycles of REMD mimicking ͑corresponding to
106
telem
of real REMD, in our case 2.5 ␮s͒ takes ϳ15 min
which allows us to perform “REMD” for hundreds of different
combinations of ␭ values and numbers of replicas at ␭max
producing time sequences of the occurrence of conformational
states at different w0.0,0.3
syn,anti
=min͓1,exp͑−⌬0.0,0.3
syn,anti
͔͒ values.
From these one can decide for different criteria from
which to prefer one set of parameters over the other.
For cases where different conformational states can be
defined ͑e.g., the anti and syn states͒, we propose to measure
efficiency of REMD, through the “number of global conformational
transitions,” Ngct. We count the number of conformational
transitions for each particular replica at the lowest ␭
value, ␭0.
The schematic REMD example shown in Fig. 3, contains
three replicas at ␭max ͑␭2,0=␭2,1=␭2,2=␭max͒ and one middle
value of lambda, ␭1. The replica represented by a solid
curve, is in the anti conformation at ␭0 after 103 telem
and
then exchanges to higher ␭ values. At ␭2,1 it makes the transition
to the syn conformation ͑after 119 telem
͒ and subsequently
exchanges with replicas at lower ␭ values until it
finally returns to ␭0, but now in the different conformational
state ͑syn͒ as before. The combination of replica exchanges
and a conformational transition is counted as one global
anti→syn transition. From this moment onwards, we will
monitor for the next syn→anti global conformational transition
͑for the same replica͒, etc. From this description one can
see that for a global conformational transition to occur it is
not necessary for the replica to reach ␭max, because the conformational
transition can occur also at lower ␭ values ͑see,
e.g., in Fig. 3 the replica represented by a dotted line. At 131
telem
an anti→syn transition occurs at ␭1 leading to the global
conformational transition at 133 telem
͒ or even at ␭0
within real REMD. ͑This is not the case for the mimicked
REMD, where conformational transition can occur only at
␭max.͒ On the other hand, the replica represented by the dotted
curve is in a syn conformation at ␭0 at 102 telem
, then
went up to ␭max where it changes its conformation to anti and
later back to syn again. When it exchanges back to ␭0 at 119
telem
, this event is not counted as a global conformational
transition, because the conformations at 102 telem
and 119
telem
are the same. In the case of MD ͑as a special case of
REMD with only one replica at ␭0͒ Ngct is equal to the number
of conformational transitions within one MD run. This
allows for a direct comparison of the efficiency between very
different REMDs or MDs performed under different condi-
tions.
Our aim is to find the REMD settings which give us the
maximum value of Ngct per CPU. ͑In the following, we assume
that in real REMD simulations every replica is assigned
one CPU.͒ Having decided on the optimal value of
␭max and t␭max
total
we vary:
͑1͒ the number and placement of middle ␭’s ͑between 0.0
and ␭max͒ and
͑2͒ the number of replicas at ␭max.
Whether it is efficient to “invest” into more replicas at
␭max or not depends on the ratio of t␭max
total
relative to the average
time needed for global conformational transitions to occur.
If more time is needed for a conformational change at
␭max, it becomes more likely that an increased number of
replicas at ␭max improves the overall efficiency per CPU.
FIG. 5. Scheme describing the steps for obtaining the
probability distribution of the switching probabilities.
204104-6 J. Hritz and C. Oostenbrink J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
In cases where only a few replicas are needed one can
test various settings in a systematic manner and select the
optimal combination. In spite of the efficiency of mimicked
REMD, a larger number of required replicas will make the
optimization problem more complex. We propose three useful
approaches:
͑1͒ The number and placement of ␭ values between 0.0 and
␭max ͑middle ␭’s͒ and the number of replicas at ␭max
can be optimized independently. Middle ␭’s are essential
for the diffusion of replicas between 0.0 and ␭max,
which will be the same for any number of replicas at
␭max. The optimal setting of middle ␭’s will therefore
be independent of the number of replicas at ␭max. However,
the gain in efficiency due to optimization of the
middle ␭’s is more pronounced if the conformational
change at ␭max is not the bottleneck of the simulation.
Therefore, we propose to search for the optimal settings
with a high number of replicas at ␭max, initially 5 in the
case of 8-Br-GTP. After this initial optimization, the
number of optimal replicas at ␭max should be obtained
after which one should check if having one more
middle ␭ does not improve results even further. Once
the optimal number of replicas at ␭max has been obtained
for an optimal set of m middle ␭ values, this
nopt
͑m͒ can be used to reduce the searching possibilities
for different numbers of middle ␭ values by taking into
account the following inequalities: nopt
͑kϾm͒
Նnopt
͑m͒ and nopt
͑kϽm͒Յnopt
͑m͒.
͑2͒ Searching for the optimal settings of middle ␭’s. When
expanding or reducing the number of middle ␭’s in the
optimization process, we do not need to consider the
complete range ͑0.0, ␭max͒ for the new ␭ values. If we
denote the optimal ␭ values in a scheme with n middle
␭’s as n
␭i
opt
͑n
␭0
opt
=0.0; n
␭n+1
opt
=␭max͒, then it is most
likely that in a scheme with n+1 middle ␭’s, the optimal
␭ values are in the following range:
n+1
␭i
opt
ʦ͗n
␭i−1
opt
,n
␭i
opt
͘. Similarly, one can generally
write that for a reduction of the number of middle ␭’s,
the optimal ␭ values are restricted to
n
␭i
opt
ʦ͗n+1
␭i
opt
,n+1
␭i+1
opt
͘.
͑3͒ In many cases, already a short simulation indicates if a
certain set of settings is promising or not. In an approach
we call “continuous filtering” we disregard possible
settings on the fly, thereby only spending computer
time on the most promising settings.
We perform a mimicked REMD simulation for all potentially
relevant REMD settings for 30 000 cycles. We subsequently
disregard all settings that show less than 10% of the
͑at that point͒ maximum value of Ngct. After 40 000 cycles
this threshold is increased to 20% and by another 10% every
10 000 cycles until after 100 000 cycles only those settings
are kept that lead to 80% of the maximum value of Ngct. We
subsequently refine the search by increasing the threshold by
2.5% every 100 000 cycles and select the optimal setting
after 500 000 cycles.
III. RESULTS
Both GTP and 8-Br-GTP have two stable conformations:
anti and syn, depending on the dihedral angle around the
glycosidic bond ͑Fig. 1͒. Based on the distributions of this
dihedral angle we define for GTP to be in the syn conformation,
when its dihedral angle around the glycosidic bond is
within the interval Ͻ−25° ,150°͒, and otherwise to be in the
anti conformation. For 8-Br-GTP the syn conformational interval
is Ͻ−35° ,160°͒.18
A. Selection of ␭max
MD simulations of GTP in explicit water were performed
for 200 ps at different values of ␭ ͑0.4; 0.45; 0.5͒.
Ten simulations started from anti and ten simulations started
from syn conformations. During the MD runs conformational
transitions occur.
Figure 6͑a͒ show the number of simulations in which
GTP is in a syn conformation as function of time, when
starting from anti and when starting from syn conformation.
We obtained t␭max
total
as the larger of two times corresponding to
the maximas of the time dependence of the number of
changed conformations divided by total time. From Fig. 6͑b͒
follows: t0.4
total
=155 ps, t0.45
total
=100 ps, and t0.5
total
=135 ps. As
the optimal value of ␭max we choose 0.45 because it gives the
shortest t␭max
total
͑100 ps͒ together with a sufficient number of
conformational transitions after this time. Another advantage
of ␭max=0.45 over 0.5 is the more efficient diffusion of replicas
between ␭0 and ␭max. For ␭max=0.45, the set of ten MD
simulations starting from syn and anti was prolonged leading
FIG. 6. ͑a͒ Number of syn conformations observed in a set of ten MD
simulations of GTP as function of time at different ␭ values starting from
ten different anti and syn conformations. Runs for ␭=0.45 were prolonged
up to 400 ps in order to reach convergence. ͑b͒ Population of converted
conformations per time ͑syn→anti or anti→syn͒. For clarity, running averages
over 20 time points have been taken.
204104-7 Optimization of replica-exchange molecular dynamics J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
to estimate of the converged conformational transition probabilities
P␭=0.45
anti→syn
=0.3 and P␭=0.45
syn→anti
=0.7 ͓Fig. 6͑a͔͒.
The same kind of analysis was performed for 8-Br-GTP
using ␭=0.6;0.65;0.7;0.8;0.9 and leading to the population
time dependences shown in Figs. 7͑a͒ and 7͑b͒. We considered
as ␭max candidates ␭max=0.7 and ␭max=0.8 for which
the prolonged MD simulations were performed.
Based on these we decided to take ␭max=0.7 for 8-Br-GTP
with the corresponding t␭max
total
=100 ps and saturated transition
probabilities P␭=0.7
anti→syn
=0.2 and P␭=0.7
syn→anti
=0.8. For ␭=0.9 the
transition syn→anti is quite efficient. However, the conversion
probability from anti to syn has become very low which
means that these transitions would occur only very rarely at
␭=0.9.
B. Generation of probability distribution of switching
probabilities
To obtain the probability distributions of switching probabilities
we performed MD simulations for ͑8-Br-͒GTP at ␭’s
in the interval ͗0.0,␭max͘ with a ␭ increment of 0.05. For
each ␭ two MD simulations of 1 ns were performed, one
starting from an anti and the second from a syn conformation.
Configurations of trajectories were stored every ps leading
to 1000 frames for each trajectory. Configurations have
been separated into sets according their conformational state
͕͑q␭
anti
͖, ͕q␭
syn
͖͒.
The distributions for the switching energy differences
͕⌬␭i,␭j
ck,cl ͖ as well as for the switching probabilities ͕w␭i,␭j
ck,cl ͖ and
their cumulative values were obtained as described in Sec.
II D 3 and schematically outlined in Fig. 5. Representative
examples of the switching energy differences and switching
probability distributions are given in Figs. 8 and 9.
C. REMD mimicking
REMD mimicking as described in the Sec. II D 1 produces
as output the time evolutions of conformational states
͑anti and syn͒ at all ␭ values that are included. Because of its
speed we can run mimicked REMD for a very long time and
thus obtain converged populations of anti and syn conformations
for GTP and 8-Br-GTP. Typically, we simulate 500 000
cycles of mimicked REMD corresponding to 106
telem
in real
REMD ͑ϳ2.5 ␮s͒.
Syn populations at individual ␭ values of mimicked
REMD for GTP ͑500 000 cycles͒ with two replicas at ␭max
=0.45 combined with no or a single middle ␭ value are
shown in Table I. Theoretically, syn populations at ␭=0.0
should be the same for all mimicked REMDs. We do not
suffer here from insufficient lengths of REMD simulations,
but inaccuracies rather stem from the probability distributions
of switching probabilities. Discrepancies in the syn
FIG. 7. ͑Color͒ ͑a͒ Number of syn conformations observed in a set of ten
MD simulations of 8-Br-GTP as function of time at different ␭ values starting
from ten different anti and syn conformations. Runs for ␭=0.7 and ␭
=0.8 were prolonged up to 400 ps in order to reach convergence. ͑b͒ Population
of converted conformations per time ͑syn→anti or anti→syn͒. For
clarity, running averages over 20 time points have been taken.
FIG. 8. Probability distributions of switching energy differences for GTP:
⌬0.0,0.3
ci,cj =␤͓E0.3͑q0.0
ci ͒r −E0.0͑q0.0
ci ͒r +E0.0͑q0.3
cj ͒s −E0.3͑q0.3
cj ͒s͔, where ci, cj are different
conformational states: syn, anti and r, s are different configurations
from MD trajectories.
FIG. 9. Probability distribution of replica exchange switching probabilities
for GTP: w0.0,0.3
syn,anti
=min͓1,exp͑−⌬0.0,0.3
syn,anti
͔͒ for GTP. Notice that probability
w0.0,0.3
syn,anti
is much lower than for the opposite replica switch w0.0,0.3
anti,syn
.
204104-8 J. Hritz and C. Oostenbrink J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
populations at ␭=0.0 can be explained from an insufficient
sampling during the 1 ns MD simulation from which the
probabilities are obtained. For ␭ values that are evident outliers
compared to the other values, this simulation should
probably be extended.
Largest deviations are expected for REMDs involving
middle ␭ values, for which we multiply very small and very
high probabilities ͑i.e., 1
␭1=0.05 and 1
␭1=0.4͒.
The root-mean-square deviation of ͓syn͔ at ␭0 divided by
its average value when excluding the values obtained with
1
␭1=0.05 and 1
␭1=0.4 is calculated to be 0.068 indicating
the relative inaccuracy in the distributions of switching prob-
abilities.
The average of syn populations at ␭0 for runs with 1
␭1
between 0.1 and 0.35 amounts to ͓syn͔aver=0.048, which corresponds
to a free energy difference ⌬Gsyn-anti
GTP
=7.40 kJ mol−1
at 298 K. As shown in Ref. 18 the obtained
population of syn conformations from real H-REMD was
0.044 for GTP corresponding to a value of ⌬Gsyn-anti
GTP
=7.63 kJ mol−1
. 500 000 cycles of mimicking REMD using
the same settings as the real REMD reported there ͑␭
=0.0;0.25,0.45,0.45,0.45,0.45͒ gives ͓syn͔=0.043, corresponding
to ⌬Gsyn-anti
GTP
=7.69 kJ mol−1
. It shows that if most
of the transitions can indeed occur only at ␭max, then REMD
mimicking can produce very reasonable values of populations
of the individual conformational states.
Therefore, REMD mimicking is not only useful as a tool
for the optimization of REMD settings but can also as be
used in itself for calculating the populations of conformations.
We can see an analogy between our method and a free
energy method that makes use of alternative pathways to
calculate free energy differences more efficiently along unphysical
thermodynamic cycles.26
We want to note, however,
that mimicked REMD covers very many of such pathways
simultaneously.
D. Finding the optimal settings for REMD
In the previous paragraph we demonstrated how long
mimicked REMD can approximate syn and anti populations
for GTP and 8-Br-GTP. These simulations would correspond
to several ␮s of real REMD. In real REMD, however, we can
afford to simulate only limited time lengths ͑ϳtens ns͒,
therefore it is crucial to find settings which corresponds to
the highest possible REMD efficiency. We have used mimicked
REMD, because it allows us to attempt REMD for
hundreds of different combinations of ␭’s. From the time
sequences of conformational states at different ␭ values we
can choose the combination of ␭’s that produces the highest
number of global conformational transitions, Ngct, which will
ensure the fastest convergence of populations at ␭0.
1. GTP
Mimicked REMD ͑500 000 cycles͒ without middle replica
and two replicas at ␭max=0.45 gives 5221 global conformational
transitions, which is 5221/3=1740 global conformational
transitions per CPU. The results for mimicked
REMD with an increasing number of replicas at ␭max=0.45
are summarized in the Table II. It shows that the highest
efficiency per CPU ͑7506/4=1877͒ is obtained for three replicas
at ␭max.
In the next step we performed mimicked REMD with
different numbers of replicas at ␭max=0.45 and one middle
␭-value 1
␭1ʦ͕0.05,0.1,0.15,0.2,0.25,0.3,0.35,0.4͖ for
500 000 cycles ͓Fig. 10͑a͔͒. With two replicas at t␭max
total
=10telem
the highest values of Ngct are obtained for 1
␭1
=0.15 ͑6069͒, 1
␭1=0.2 ͑6071͒, and for 1
␭1=0.25 ͑6089͒,
which are higher than the values of Ngct obtained from
REMD without middle ␭ having the same two replicas at
␭max=0.45. It means that putting one middle 1
␭1 can increase
the efficiency of diffusion between ␭0=0.0 and ␭max=0.45.
The efficiency per CPU for mimicked REMD containing one
middle 1
␭1=0.25, however, is lower ͑1522 compared to
1877͒. Still the efficiency per CPU may be higher for REMD
containing one middle lambda in combination with a higher
number of replicas at ␭max. Figure 10 presents the efficiency
of REMD containing one middle lambda with an increasing
number of replicas at telem
. It reveals that per CPU the most
efficient set of ␭’s is ͓0.0, 0.25, 0.45, 0.45, 0.45, 0.45͔ with
12646/6=2108 global conformational transitions per CPU.
It also shows that the differences due to the placement of 1
␭1
are more pronounced for settings with a higher number of
replicas at ␭max.
Because the optimal REMD setting containing one 1
␭1 is
more efficient per CPU than a set of ␭’s without any middle
␭ value, we continued to test REMD containing two middle
␭ values. As is explained in the methods Sec. II D 4 the
optimal setting of REMD with two middle ␭’s will still have
t+telem
replicas at ␭max=0.45, because this is the optimal
TABLE I. Syn populations at ␭0, 1
␭1, ␭max,0, ␭max,1 depending on the placement of 1
␭1 from mimicked REMD of GTP with two replicas at ␭max=0.45 over
500 000 cycles.
1
␭1 None 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
͓syn͔͑␭0͒ 0.053 0.047 0.053 0.050 0.044 0.043 0.049 0.047 0.081
͓syn͔͑1
␭1͒ ¯ 0.058 0.095 0.133 0.163 0.228 0.287 0.295 0.298
͓syn͔͑␭max,0͒ 0.302 0.303 0.296 0.301 0.305 0.307 0.299 0.306 0.302
͓syn͔͑␭max,1͒ 0.300 0.301 0.295 0.299 0.303 0.304 0.298 0.304 0.300
TABLE II. Number of global conformational transitions, Ngct, during mimicked
REMD of GTP with n=2,3,4,5 replicas at ␭max=0.45 and no middle
␭ value after 500 000 cycles.
n replicas at ␭max=0.45 2 3 4 5
Ngct 5221 7506 8954 9911
Ngct/CPU 1740 1877 1791 1652
204104-9 Optimization of replica-exchange molecular dynamics J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
setting of REMD with one middle ␭ value. Table III shows
Ngct per CPU for n=4,5,6 and two additional values 2
␭1 and
2
␭2. It shows that the optimal setting for REMD in this case
is ͓0.0, 0.15, 0.3, 0.45, 0.45, 0.45, 0.45, 0.45͔. However, its
efficiency is slightly lower than for the optimal setting using
only one middle 1
␭1 ͑2092 compared 2108 of global conformational
transitions per CPU͒.
In conclusion the most effective setting predicted by
mimicked REMD is ͓0.0, 0.25, 0.45, 0.45, 0.45, 0.45͔ with
12646/6=2108 global conformational transitions per CPU.
This number corresponds to 500 000 REMD mimicking
cycles or 106
elementary switching periods of real REMD.
This means that one can expect ϳ13 global conformational
transitions per 1000 telem
of real REMD, which amounts to
2.5 ns with telem
=2.5 ps. From an extrapolation of the linear
increase of Ngct as a function of time, over ten individual
5000 cycles mimicked REMD simulations we estimate the
REMD equilibration period to be about 200 telem
͑0.5 ns in
real REMD͒. For comparison, in real REMD we reported
that Ngct increases by ϳ17 Ngct per 1000 telem
and that a
REMD equilibration of 249 telem
was required.18
2. 8-Br-GTP
8-Br-GTP has a higher conformational barrier between
the anti and syn conformations than GTP, therefore we have
found a higher value of ␭max=0.7 and the optimal REMD
will probably require more middle replicas. For this reason
we will not use a systematic search for the optimal settings
as for GTP, but rather use a more efficient approach as described
in Sec. II D 4.
We initiate the optimization of the middle ␭’s using five
replicas at ␭max=0.7 and obtain 1
␭1
opt
=0.5 as the optimal
value if only one middle value of ␭ is taken into account,
with Ngct=1079 per CPU ͑Table IV͒. When increasing the
number of middle ␭ values to two, we restricted the searching
ranges to 2
␭1ʦ͗0.05,0.5͘ and 2
␭2ʦ͗0.5,0.65͘. Similarly
the optimizations were extended to three and four
middle ␭ values. The Ngct per CPU for different numbers of
middle ␭ values in combination with five replicas at ␭max are
summarized in Table IV, where arrows indicate the steps in
the optimization. As can be seen from this table, the optimal
setting containing four middle ␭ values in combination with
five replicas at ␭max is less efficient ͑1252 global conformational
transitions per CPU͒ than the optimal three middle
␭-values setting ͑1274͒. Because this may still be caused by
too few conformational transitions at ␭max, the same optimal
settings of four middle ␭ values ͓0.2; 0.4; 0.55; 0.65͔ was
tested in combination with n=6,7,8 and indeed for nopt
͑m=4͒=7 we obtained a more efficient setting ͑1315 Ngct per
CPU͒.
Taking into account the inequalities: nopt
͑mϾ4͒
Նnopt
͑m=4͒=7 and nopt
͑mϽ4͒Յnopt
͑m=4͒=7 that are described
in Sec. II D 4, the optimization procedure
was continued by including five middle ␭ values in combination
with n=7 and subsequently with n=8. It appears that
nopt
͑m=5͒=7, but that its efficiency is lower ͑1305͒ than for
the optimal four middle ␭-values setting ͑1315 Ngct per
CPU͒, meaning that the overall optimal scheme has Յ4
middle ␭ values with nՅ7. Because the optimal three
middle ␭-values setting ͓nopt
͑m=3͒=6͔ is also not giving a
higher efficiency, the overall most efficient setting
for 8-Br-GTP has thus been determined as
4
␭i
opt
ʦ͓0.0;0.2;0.4;0.55;0.65;0.7;0.7;0.7;0.7;0.7;0.7;0.7͔.
Ngct per CPU ͑1315͒ is lower for 8-Br-GTP than for GTP
͑2108͒ indicating that REMD of 8-Br-GTP will be computationally
more demanding. Using the optimal set for 8-BrGTP
of 12 replicas we estimate ϳ16 global conformational
transitions during 1000 telem
of real REMD ͑ϳ2.5 ns͒ and
the equilibration period to be ϳ180 telem
͑based on a linear
extrapolation of time dependence of Ngct as function of time
from ten individual runs of 5000 mimicked REMD cycles͒.
IV. DISCUSSION
The present study on the GTP/8-Br-GTP two state model
systems deepens our understanding about REMD efficiency.
For simplicity, we assumed that conformational transitions
occur only at the highest lambda, ␭max in the case of
H-REMD, or Tmax for T-REMD. To obtain a different conformational
state for a given replica at ␭0, the replica has to
diffuse from ␭0 through the middle ␭ values to ␭max, where a
conformational change has to occur. The system subsequently
needs to diffuse from ␭max back to ␭0. Such process
we named a global conformational transition. The advantage
FIG. 10. Number of global conformational transitions, Ngct observed in
500 000 cycles of mimicked REMD of GTP with one middle replica 1
␭1 at
different positions and with different numbers of replicas at ␭max=0.45: ͑a͒
absolute value and ͑b͒ value per CPU.
204104-10 J. Hritz and C. Oostenbrink J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
of monitoring the Ngct compared to the number of roundtrips
between the lowest and highest temperature as proposed by
Trebst et al.14–16
comes from the fact that many roundtrips
do not necessarily ensure any conformational changes at ␭0.
For example for GTP, we have observed simulations in
which there were many roundtrips thanks to relatively high
values of w␭i,␭j
anti,anti
, while w␭i,␭j
anti,syn
were very low. In such cases
a syn conformation that occurs at ␭max will not be able to
efficiently diffuse down to ␭0. Another advantage of counting
Ngct is that if P␭ ␭max
transit
͑telem
͒ 0 it does not require a replica
to diffuse all the way up to ␭max because a conformational
transition within real REMD can already occur at any
␭ value. The disadvantage of monitoring Ngct is that defined
stable conformational states are required.
We described the approach for finding the optimal ␭max
at which conformational transitions occur in sufficiently
short time, which is, however, typically still much longer
than the elementary switching period, telem
, between switching
attempts. The occurrence of conformational transitions
shows a sigmoidal time dependency, from which a relatively
long t␭max
transit
, t␭max
total
can be estimated. The presented approach
using n multiple replicas at ␭max allows several replicas to
TABLE III. Number of global conformational transitions, Ngct per CPU for mimicked REMD of GTP using
n=4,5,6 replicas at ␭max=0.45 and two additional ␭-values 2
␭1, 2
␭2. The maximum is obtained for 2
␭1
=0.15 and 2
␭2 =0.3, at n=5.
n=4 2
␭1
2
␭2
0.1 0.15 0.2 0.25 0.3 0.35 0.4
0.05 1614 1741 1726 1727 1697 1413 1269
0.1 ¯ 1787 1807 1879 1964 1692 1622
0.15 ¯ ¯ 1802 1898 1990 1807 1707
0.2 ¯ ¯ ¯ 1839 1988 1767 1761
0.25 ¯ ¯ ¯ ¯ 1841 1701 1719
0.3 ¯ ¯ ¯ ¯ ¯ 1678 1688
0.35 ¯ ¯ ¯ ¯ ¯ ¯ 1484
n=5 2
␭1
2
␭2
0.1 0.15 0.2 0.25 0.3 0.35 0.4
0.05 1626 1814 1755 1766 1746 1446 1261
0.1 ¯ 1834 1871 1960 2030 1795 1644
0.15 ¯ ¯ 1813 1948 2092 1867 1803
0.2 ¯ ¯ ¯ 1945 2070 1845 1790
0.25 ¯ ¯ ¯ ¯ 1931 1776 1795
0.3 ¯ ¯ ¯ ¯ ¯ 1742 1740
0.35 ¯ ¯ ¯ ¯ ¯ ¯ 1470
n=6 2
␭1
2
␭2
0.1 0.15 0.2 0.25 0.3 0.35 0.4
0.05 1562 1742 1742 1709 1702 1353 1167
0.1 ¯ 1781 1827 1920 2026 1722 1524
0.15 ¯ ¯ 1812 1917 2075 1830 1715
0.2 ¯ ¯ ¯ 1947 2069 1795 1648
0.25 ¯ ¯ ¯ ¯ 1906 1739 1714
0.3 ¯ ¯ ¯ ¯ ¯ 1715 1716
0.35 ¯ ¯ ¯ ¯ ¯ ¯ 1430
TABLE IV. Number of global transitions, Ngct per CPU for the mimicked REMD of 8-Br-GTP, having optimal set of m middle lambdas m
␭i
opt
combined with
n replicas at ␭max=0.7. The maximum is obtained for m=4 and n=7. The arrows indicate the sequence of steps during the optimization procedure.
m 0 1 2 3 4 5
Set of m
␭i
opt
None ͓0.5͔ ͓0.4,0.55͔ ͓0.35,0.5,0.65͔ ͓0.2,0.4,0.55,0.65͔ ͓0.2,0.4,0.45,0.55,0.65͔
Ngct/CPU, n=5 159 → 1079 → 1263 → 1274 → 1252
↓
Ngct/CPU, n=6 1301 1264
↑ ↓
Ngct/CPU, n=7 1297 ← 1313 → 1305
↓ ↓
Ngct/CPU, n=8 1282 1266
204104-11 Optimization of replica-exchange molecular dynamics J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
spend the required time t␭max
total
at ␭max ͑in parallel͒. At the same
time it allows for frequent switching attempts between
␭0, ... ,␭i, ...␭max,0. In order to optimize the number of replicas
at ␭max as well as the number and placement of ␭ values
between ␭0 and ␭max, we present a REMD mimicking approach.
As this approach propagates the conformations based
on calculated probabilities, it is very fast and allows for
simulations that correspond to ␮s time scales in real REMD.
This approach was inspired by the following analogy
with real REMD. In real REMD “parallel” MD simulations
are halted from time to time and switch their temperatures or
Hamiltonians depending on the current energies. If we collect
the switching probabilities into probability distributions
then we can mimic the REMD simulation by sampling a
switching probability from the probability distribution and
subsequently perform the switch depending on this probability.
Although real REMD switching probabilities converge to
the same distribution, it builds up this distribution relatively
inefficiently because for every structure at a particular time
point, only one partner conformation is selected to attempt a
replica switch. In our approach, all possible pairs of structures
at a given set of ␭i, ␭j are used to estimate the distribution
of switching probabilities. In many cases quite long
REMD simulations are required to obtain a statistically
sound probability distribution of switching probabilities. In
this work we perform relatively short MD simulations at
different ␭’s ͑H-REMD͒ or temperatures ͑T-REMD͒ starting
from different conformational states and subsequently calculate
the corresponding distributions of energy differences using
different Hamiltonians or temperatures. These distributions
are then used to calculate the probability distributions
of switching probabilities. Together with estimates of the
probability of conformational transitions at ␭max it allows us
to mimic several ␮s of REMD simulations in a few minutes.
We can then study the effect of different middle ␭ sets on the
REMD efficiency, characterized by Ngct per CPU. For example,
we clearly showed that increasing the number of replicas
does not necessary increase the efficiency ͑per CPU͒ of
REMD simulations. This means that “blind” brute force applications
of REMD may be very inefficient.
A systematic search for the optimal ␭ settings that give
the highest value of Ngct per CPU is affordable for simpler
systems with a small number of replicas. However, the optimization
process for complex systems requiring many replicas
this can lead to a huge number of combinations. For such
cases set we have suggested approaches to reduce the number
of relevant combinations. The whole proposed optimization
scheme is largely automated and fully parallelized.
Once the optimal settings have been found, REMD mimicking
can make two more practical predictions for real
REMD: it can estimate: ͑1͒ the equilibration time and ͑2͒ the
required length of an REMD simulation to reach a given
number of Ngct. For GTP this later time estimate showed an
accuracy of about 20% in real REMD. This allows the user
to estimate the feasibility of the real REMD simulation and
to allocate the needed computational time in advance.
The most important advantage of the presented optimization
scheme for REMD settings compared to the feedbackoptimized
parallel tempering approach14–16
is its speed.
While feedback optimization of the 36-residue villin headpiece
subdomain HP-36 requires several years of CPU
time14–16
optimization by REMD mimicking can be performed
within one week. This is a crucial factor for REMD
simulations of complex system such as proteins. On the other
hand, our approach requires some preliminary knowledge of
stable conformational states, at least of the most dominant
ones. Starting with an incomplete set of stable conformational
states can lead to not completely optimal REMD settings.
However, real REMD using partially optimized settings
may suffice to reveal additional stable conformational
states, which can then be used to refine the REMD settings.
The whole process can be repeated iteratively until the real
REMD leads to converged populations of the individual conformations.
Note that one set of REMD generated conformations
may yield part of the simulations that are used to calculate
the probability distributions required for refining the ␭
settings.
We also showed that REMD mimicking is not only useful
for the optimization of REMD settings, but also by itself
can give us reasonable estimates of conformational populations.
For GTP we obtained an excellent agreement for the
syn population at ␭0=0.0 from mimicked REMD ͓͑syn͔aver
=4.8%͒, as compared to real REMD ͓͑syn͔=4.4%͒ presented
in our other work.18
The advantage of mimicked REMD
compared to thermodynamic cycle approaches is the fact that
global conformational transition can occur through many different
pathways, which are directly counted in a massively
parallel manner in REMD mimicking using probability distributions
of all combinations of switching probabilities. Usage
of probability distributions, instead of average values
takes into account the conformational variety of structures
belonging to the same stable conformational region.
V. CONCLUSIONS
We have presented an algorithm that mimics replica exchange
͑REMD͒ simulations by a stochastic propagation in
time of conformational states rather than explicit MD simulations.
The approach was demonstrated for Hamiltonian
REMD simulations on two model systems, GTP and 8-BrGTP,
for which two stable conformations are known, but the
potential energy barrier separating them is too high to be
crossed in normal MD. The method is, however, also readily
applicable to temperature REMD.
We have shown that mimicked REMD can be used to
serve three different purposes: ͑1͒ it can be used to obtain the
optimal set of ␭ values by allowing an extremely fast attempt
to try different REMD settings; ͑2͒ it gives the user an estimate
of the simulation length in real REMD simulations,
both for the equilibration of conformational states over the
replicas and for the time required to obtain reasonably converged
populations; and ͑3͒ it can make estimates of such
populations directly which were shown to match remarkably
well with real REMD simulations.
We are convinced that our method can contribute significantly
to deepen our understanding of the processes governing
REMD simulations. For many different applications, it
204104-12 J. Hritz and C. Oostenbrink J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
will help to design more efficient REMD simulations for
systems as are described in this work, but also for many
more complex systems.
ACKNOWLEDGMENTS
This work was supported by the Netherlands Genomics
Initiative in the context of Horizon Breakthrough Grant No.
050-71-043 ͑postdoc stay of J.H.͒.
1
R. H. Swendsen and J. S. Wang, Phys. Rev. Lett. 57, 2607 ͑1986͒.
2
M. C. Tesi, E. J. J. van Rensburg, E. Orlandini, and S. G. Whittington, J.
Stat. Phys. 82, 155 ͑1996͒.
3
K. Hukushima and K. Nemoto, J. Phys. Soc. Jpn. 65, 1604 ͑1996͒.
4
Y. Sugita and Y. Okamoto, Chem. Phys. Lett. 314, 141 ͑1999͒.
5
U. H. E. Hansmann, Chem. Phys. Lett. 281, 140 ͑1997͒.
6
Y. Sugita, A. Kitao, and Y. Okamoto, J. Chem. Phys. 113, 6042 ͑2000͒.
7
D. J. Earl and M. W. Deem, J. Phys. Chem. B 108, 6844 ͑2004͒.
8
D. A. Kofke, J. Chem. Phys. 117, 6911 ͑2002͒.
9
D. A. Kofke, J. Chem. Phys. 121, 1167 ͑2004͒.
10
A. Kone and D. A. Kofke, J. Chem. Phys. 122, 206101 ͑2005͒.
11
C. Predescu, M. Predescu, and C. Ciobanu, J. Chem. Phys. 120, 4119
͑2004͒.
12
C. Predescu, M. Predescu, and C. Ciobanu, J. Phys. Chem. B 109, 4189
͑2005͒.
13
N. Rathore, M. Chopra, and J. J. de Pablo, J. Chem. Phys. 122, 024111
͑2005͒.
14
S. Trebst, D. A. Huse, and M. Troyer, Phys. Rev. E 70, 046701 ͑2004͒.
15
H. G. Katzgraber, S. Trebst, D. A. Huse, and M. Troyer, J. Stat. Mech.:
Theory Exp. 2006, P03018 ͑2006͒.
16
S. Trebst, M. Troyer, and U. H. E. Hansmann, J. Chem. Phys. 124,
174903 ͑2006͒.
17
T. Läppchen, A. F. Hartog, V. A. Pinas, G. J. Koomen, and T. den Blaauwen,
Biochemistry 44, 7879 ͑2005͒.
18
J. Hritz and C. Oostenbrink ͑in preparation͒.
19
M. Christen, P. H. Hünenberger, D. Bakowies, R. Baron, R. Burgi, D. P.
Geerke, T. N. Heinz, M. A. Kastenholz, V. Krautler, C. Oostenbrink, C.
Peter, D. Trzesniak, and W. F. Van Gunsteren, J. Comput. Chem. 26,
1719 ͑2005͒.
20
J.-P. Ryckaert, G. Ciccotti, and H. J. C. Berendsen, J. Comput. Phys. 23,
327 ͑1977͒.
21
R. W. Hockney, Methods Comput. Phys. 9, 136 ͑1970͒.
22
H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, A. DiNola, and
J. R. Haak, J. Chem. Phys. 81, 3684 ͑1984͒.
23
I. G. Tironi, R. Sperb, P. E. Smith, and W. F. van Gunsteren, J. Chem.
Phys. 102, 5451 ͑1995͒.
24
C. Oostenbrink, A. Villa, A. E. Mark, and W. F. van Gunsteren, J. Comput.
Chem. 25, 1656 ͑2004͒.
25
T. C. Beutler, A. E. Mark, R. C. van Schaik, P. R. Gerber, and W. F. van
Gunsteren, Chem. Phys. Lett. 222, 529 ͑1994͒.
26
A. E. Mark, W. F. Van Gunsteren, and H. J. C. Berendsen, J. Chem. Phys.
94, 3808 ͑1991͒.
204104-13 Optimization of replica-exchange molecular dynamics J. Chem. Phys. 127, 204104 ͑2007͒
Downloaded 30 Mar 2011 to 130.37.129.78. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
View publication statsView publication stats
METHODOLOGICAL DEVELOPMENTS OF ENHANCED SAMPLING METHODOLOGY H-REMD
46
Paper 7
Gabor Nagy, Chris Oostenbrink, Jozef Hritz*: Exploring the Binding Pathways of the
14-3-3 Protein: Structural and Free-Energy Profiles Revealed by Hamiltonian Replica
Exchange Molecular Dynamics with Distance Field Distance Restraints. PLoS ONE
2017,12(7), e0180633
RESEARCH ARTICLE
Exploring the binding pathways of the 14-3-3ζ
protein: Structural and free-energy profiles
revealed by Hamiltonian replica exchange
molecular dynamics with distancefield
distance restraints
Gabor Nagy1¤
, Chris Oostenbrink2
, Jozef Hritz1
*
1 CEITEC-MU, Masaryk University, Brno, Czech Republic, 2 Institute for Molecular Modeling and
Simulation, University of Natural Resources and Life Sciences, Vienna, Austria
¤ Current address: Department of Theoretical and Computational Biophysics, Max Planck Institute for
Biophysical Chemistry, Go¨ttingen, Germany
* jozef.hritz@ceitec.muni.cz
Abstract
The 14-3-3 protein family performs regulatory functions in eukaryotic organisms by binding
to a large number of phosphorylated protein partners. Whilst the binding mode of the phosphopeptides
within the primary 14-3-3 binding site is well established based on the crystal
structures of their complexes, little is known about the binding process itself. We present a
computational study of the process by which phosphopeptides bind to the 14-3-3ζ protein.
Applying a novel scheme combining Hamiltonian replica exchange molecular dynamics and
distancefield restraints allowed us to map and compare the most likely phosphopeptidebinding
pathways to the 14-3-3ζ protein. The most important structural changes to the protein
and peptides involved in the binding process were identified. In order to bind phosphopeptides
to the primary interaction site, the 14-3-3ζ adopted a newly found wide-opened
conformation. Based on our findings we additionally propose a secondary interaction site on
the inner surface of the 14-3-3ζ dimer, and a direct interference on the binding process by
the flexible C-terminal tail. A minimalistic model was designed to allow for the efficient calculation
of absolute binding affinities. Binding affinities calculated from the potential of mean
force along the binding pathway are in line with the available experimental estimates for two
of the studied systems.
Introduction
14-3-3 proteins are important regulatory factors found in all kingdoms of life and are vital for
the survival of higher organisms. In mammals, the 14-3-3 family consists of seven isoforms
that can be found in large abundance within the brain. The human 14-3-3z isoform was
selected for this project because of its high biological relevance. 14-3-3 proteins function
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 1 / 30
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Nagy G, Oostenbrink C, Hritz J (2017)
Exploring the binding pathways of the 14-3-3ζ
protein: Structural and free-energy profiles revealed
by Hamiltonian replica exchange molecular
dynamics with distancefield distance restraints.
PLoS ONE 12(7): e0180633. https://doi.org/
10.1371/journal.pone.0180633
Editor: Yaakov Koby Levy, Weizmann Institute of
Science, ISRAEL
Received: September 27, 2016
Accepted: June 19, 2017
Published: July 20, 2017
Copyright: © 2017 Nagy et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: All relevant data are
within the paper and its Supporting Information
files.
Funding: This work was supported by Czech
Science Foundation, no. GF15-34684L (J.H., G.N.).
J.H. was also supported from the SoMoPro II
Programme, co-financed by the European Union
and the South Moravian Region (Czech Republic).
The paper reflects only the author’s views and the
Union is not liable for any use that may be made of
mainly as dimers, which are composed of two 28-kDa monomers that are both capable of
binding phosphorylated serine (pS) and threonine (pT) motifs in other proteins. Crystal structures
of all seven mammalian homodimers are now available and show that each monomer is
composed of nine α-helices, arranged in an antiparallel fashion. The helices form an amphipathic
groove that mediates pS and pT target binding [1]. Most 14-3-3 targets have two phosphoserine/threonine-containing
motifs with a consensus sequence RSXpSXP (mode I) or RX
[FY]XpSXP (mode II) [2], representing the optimal recognition sites for 14-3-3. Upon binding
to these sites 14-3-3 proteins induce conformational changes in their target protein, (and ‘finish
the job’) when phosphorylation alone may lack the power to drive the necessary allosteric
changes for modulating the activity of an intracellular protein. Owing to their dimeric nature,
14-3-3 proteins are capable of distinguishing between non-, single- and double- phosphorylated
binding protein partners; in this sense 14-3-3 proteins are considered to act as coupled
binary devices [3]. Recently, we have presented an approach based on experimental 31
P NMR
spectroscopy which revealed that a double-phosphorylated protein can be complexed with the
14-3-3z dimer in a much more dynamic fashion than was originally thought. In addition to
the traditionally considered single partner with two phosphorylation sites occupying the individual
binding cavities within the 14-3–3z dimer, two more major binding modes were confirmed
[4]. All that is currently known about the structural features of the phosphopeptide
binding to 14-3-3 proteins originates from the available crystal structures of 14-3-3 proteins in
apo and holo state. Comparison of the various 14-3-3 crystal structures also revealed the different
width of the main peptide binding groove, thus suggesting a dynamic opening process [5].
Here, we present the structural and energetic features of selected phosphopeptides along
their binding/unbinding pathways to/from the 14-3-3z binding site, determined by enhanced
sampling computational approaches. The main applied methodology is based on Hamiltonian
replica exchange molecular dynamics (HRE-MD) combined with distancefield (DF) distance
restraints [6], which has several advantages over the more conventional umbrella sampling
and distance restraints approaches. HRE-MD is a highly parallel perturbed molecular dynamics
(MD) technique, whereby each parallel simulation (replica) represents a discrete state along
a thermodynamic pathway. The simulation of replicas are independent of each other, however,
at given time periods the replicas can exchange their coordinates to allow the ensembles
of each replica to include conformations derived from multiple starting coordinates. The
application of HRE-MD instead of a set of individual MD simulations with different distance
restraints significantly enhances sampling efficiency. HRE-MD could be combined with regular
distance restraints, however, such an approach could lead to protein damage when coordinates
at larger Cartesian distances switch to shorter distances [7]. Applying DF restraints
avoids the protein damage whilst the ligand is pulled into the binding site by using an altered
reaction coordinate, for which the distance to the binding site is defined by the shortest sterically
possible pathway. Combining DF restraints and HRE-MD allows the simulation coordinates
of the replicas to traverse reversibly between the various states of the binding pathway
and represent conformations in structural ensembles restrained at different DF distances that
would otherwise require much longer simulation times due to kinetic barriers. The set of thermodynamic
ensembles that is generated is used to determine the structural features of the peptide-binding
pathways as well as the corresponding potential-of-mean-force (PMF) profiles
which can be used for the calculation of the absolute binding affinities.
Results and discussion
We studied the binding of phosphorylated peptides to the 14-3-3z protein by constructing
two different models of 14-3-3z, one containing a full length 14-3-3z dimer (dim) and one
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 2 / 30
the information contained therein. G.N. also
acknowledges the BioTop doctoral exchange
program, funded by the Austrian Science Fund
(FWF; grant no. W1224). C.O. gratefully
acknowledges the financial support of the Austrian
Science Foundation (FWF; grant no. I 1999-N28)
and the European Research Council (ERC; grant
no. 240608). The computational simulations were
realised in the National Supercomputing Center
IT4Innovations, which was supported by The
Ministry of Education, Youth and Sports from the
Large Infrastructures for Research, Experimental
Development and Innovations project–
LM2015070. Further computational resources
were provided by the CESNET LM2015042 and the
CERIT Scientific Cloud LM2015085, provided
under the programme "Projects of Large Research,
Development, and Innovations Infrastructures."
The funders had no role in study design, data
collection and analysis, decision to publish, or
preparation of the manuscript.
Competing interests: The authors have declared
that no competing interests exist.
including only a truncated 14-3-3z monomer (tmon) without the flexible C-terminal stretch
(tail, aa. 230–245). The proper simulation of a full length 14-3-3z dimer requires a large simulation
box in order to avoid artificial periodic effects arising from the opening motion of 14-3-
3z and the flexible C-terminal regions. Furthermore, the presence of the flexible C-terminal
stretch also significantly slowed down the convergence of calculated free energies, as is demonstrated
by the PMF calculations (see below). Four different phosphopeptide fragments are
used as models for this study which were derived from the diphosphorylated PKC-ε and
C-RAF kinase binding sites for 14-3-3z, and are referred to as the head and tail fragments of
peptides 1 and 2, respectively (p1h, p1t, p2h, p2t), based on their location in the full protein
sequence. These phosphopeptide fragments were chosen because their crystal structure bound
to 14-3-3z was available and their binding affinities were previously measured [8–10].
Phosphopeptide sequences, abbreviations and naming conventions used in this work are
given in Fig 1. The monomers of 14-3-3 proteins are horse-shoe shaped, and the first four αhelices
form the dimerization interface. The two monomers of a 14-3-3z dimer are denoted as
M1 and M2 when a distinction is necessary. The helices three, five, seven and nine of each
Fig 1. 3D representation, nomenclature and sequence of the model molecules used during the simulations. Dimer and monomer
14-3-3ζ systems (on the right) were simulated with fragments of phosphopeptide 1 and 2 (sequence shown on the left). The phosphoserine
in the sequence is highlighted in red. 14-3-3ζ monomers are represented as cartoons in green and cyan, phosphopeptides are shown in
stick representations in orange and purple. The phosphopeptide under DF restraining is labelled by an upper-case letter in the abbreviations.
https://doi.org/10.1371/journal.pone.0180633.g001
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 3 / 30
monomer form the phosphopeptide binding grooves on the inner side of the 14-3-3z dimer.
When referring to our simulations we denote our simulated complex by first defining the 14-
3-3z model (dim or tmon) followed by peptide fragment present in the simulation (either p1h,
p1t, p2h, p2t or nothing, in case of apo simulations). In the case of dimeric simulations, if two
peptide fragments are present we will denote it by p1ht or p2ht representing either peptide 1
(region 342–372 of protein kinase C-ε) or peptide 2 (region 229–264 of C-RAF kinase) fragments.
In addition, when DF distance restraints are applied to pull one of the phosphopeptide
fragments from its respective binding groove, we mark this fragment with an upper-case letter
(e.g. tmon_p2T or dim_p2hT).
Our results are divided into three sections. The first section describes phosphopeptide binding
pathways elucidated from the DF/HRE-MD simulations. We characterized the binding/
unbinding pathways of studied 14-3-3z/phosphopeptide complexes by determining the most
important protein-peptide interactions between 14-3-3 and its binding partner, and identifying
the regions most frequently populated by the restrained phosphopeptide. In the second
section we focus on the structural changes along the binding pathways for both 14-3-3z and
the phosphopeptide. Here we aim to identify the large-scale protein motions, which may be
important for the phosphopeptide and protein binding. The third section presents the free
energy changes along the binding pathways. The determined free-energy profiles allowed us to
calculate estimates for the corresponding phosphopeptide binding affinities.
1. Major binding pathways of the 14-3-3ζ protein
In this section, we have elucidated the phosphopeptide binding pathways of 14-3-3z, using
DF/HRE-MD simulations. We achieved this by gradually pulling one of the phosphopeptides
bound to the 14-3-3z from its respective binding site in a reversible HRE-MD process. During
the DF/HRE-MD simulations replicas with increasing index numbers restrained the phosphopeptide
to regions with DF distances further away from the peptide binding groove (interaction
site 1, IS1). Note that for every dimeric 14-3-3z complexed with two phosphopeptide
fragments (e.g. p1h and p1t in dim_p1Ht) one phosphopeptide fragment (p1h in case of
dim_p1Ht) per process was pulled from its monomer, whilst the other (p1t in case of
dim_p1Ht) remained in its respective monomer binding site (Fig 1). Such a setup avoids complications
arising from the two available binding sites (for p1h) within one 14-3-3z dimer.
1.1. Characterization of the phosphopeptide binding pathways. In order to characterize
the phosphopeptide binding pathways of 14-3-3z, we monitored the position and interactions
of the phosphopeptide during its binding/unbinding process. Fig 2 summarizes the results of
such an analysis for simulation dim_p1hT (S1–S7 Figs corresponds to the other 7 DF/
HRE-MD simulations). The position of the DF-restrained phosphopeptide was tracked on a
three dimensional grid through its virtual atom (as described in Methods section) near the
pSer residue during the DF/HRE-MD simulations (Fig 2A). The most occupied/preferred
positions of the phosphopeptide at various DF distances were aligned along a dominant binding/unbinding
pathway (as shown in Fig 2B). Similar dominant pathways were observed for
seven out eight simulations (as shown in S1B–S7B Figs).
The analysis of phosphopeptide-protein interaction energies (Fig 2C) shows a local minimum
around the DF distance of 3.5–6.0 nm, in addition to the global minimum of original
binding site at the DF distance of about 0.2 nm. The local interaction energy minimum cooccurred
with stable, specific interactions with certain 14-3-3z residues around amino acids
64–70 (Fig 2D). Similar characteristic interaction energy minima were also observed for all
seven simulations with a dominant pathway, including stable interactions between the phosphopeptide
and the same 14-3-3z residues (64–70, in S1D–S6D Figs), suggesting a previously
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 4 / 30
Fig 2. Pathway visualization of the DF/HRE-MD simulation dim_p1hT. A) The volume sampled by the p1t
phosphopeptide during the simulation is shown by dots around the 14-3-3ζ protein coloured based on their
position along the pathway. The most probable points in space to find the peptide in for each replica
(probability density peaks) are represented by larger spheres. B) Density peaks and a few representative
structures corresponding to the density peaks, coloured according to their position along the pathway. Replica
density peaks are connected by lines to visualize the binding pathway. The 14-3-3ζ protein in panels A-B is
shown as a surface representation, with the two monomers shown in light and dark grey. C) Average
interaction energy between 14-3-3ζ and p1t. D) Interaction map between any atom of the pulled p1t peptide
and the amino acids of the 14-3-3ζ protein, summarized for each replica (only amino acids with at least 0.1
hydrogen bond/salt bridge on average are shown). The scale indicates the average number of interactions.
https://doi.org/10.1371/journal.pone.0180633.g002
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 5 / 30
unknown secondary interaction site. To better characterize our binding pathways, we divided
the pathways into five parts; starting from the primary interaction site (IS1) within the main
phosphopeptide binding groove, followed by the early pathway (Pw1) between the two interaction
sites, the vicinity of the secondary interaction site (IS2), the late pathway (Pw2) after IS2,
and the unbound state (Unb), where the pathway becomes diffuse and no significant interactions
with 14-3-3z are observed for the phosphopeptide.
Experimental support for the existence of the IS2 site can possibly be found in previously
published 31
P NMR titration data of the doubly phosphorylated (at positions 19 and 40)
human tyrosine hydroxylase 1 peptide binding to 14-3-3z [4]. Fig 2A of the referred article
shows an “unexpected” peak labelled as pS40Ã
, which was different from pS40 in the free state
or bound to IS1. One of the possible explanations is that while pS19 is bound to IS1, pS40 can
interact with IS2—still being partially solvent exposed—resulting in the additional peak.
1.2. Comparison between phosphopeptide pathways. Comparing the binding pathways
obtained for the four studied phosphopeptide fragments (with both 14-3-3z models), we
found a remarkable similarity between their most probable pathways in 7 out of 8 simulations
(with the exception of dim_p2Ht, Fig 3A). The dominant pathways generally followed helix
three, “sliding” from IS1 in the primary binding groove to IS2, and becoming increasingly diffuse
before the final detachment of phosphopeptide and reaching the Unb state. This general
behaviour indicates the existence of one dominant pathway for all four phosphopeptides,
observed both in the dimeric and truncated monomeric simulations. Examining the electrostatic
surface potential (ESP) in multiple 14-3-3z conformations showed both IS1 and IS2
display highly positive surface potentials; and the connecting region along helix three also
exhibits a mildly positive surface potential (Fig 3B). Thus, these areas can form favourable
Fig 3. Comparison of binding pathways of different phosphopeptides. A) Binding pathways from DF/HRE-MD simulations depicted as
connected dots referring to the probability density peaks along the respective pathway (See Table 1 for more information). B) Electrostatic
surface potential (ESP) of 14-3-3ζ in blue, white and red for positive, neutral, and negative surface patches, respectively. The positively
charged main interaction site (IS1), and secondary interaction site (IS2) are connected by a positive surface along the binding pathways. A
negative surface patch (NSP) involved in the binding process is also indicated. Serine58 (S58), a phosphorylation site is located near the
binding pathway. For clarity, the 14-3-3ζ C-terminal tail is not shown.
https://doi.org/10.1371/journal.pone.0180633.g003
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 6 / 30
electrostatic interactions with a negatively charged phosphoserine side chain, which could
explain the presence of a dominant binding pathway.
In addition, based on the ESP we identified a negative surface patch (NSP) which was often
found to accommodate the positively charged side chain in position -3 or -2 relative to the
phosphorylation site of the peptide fragment. The NSP is primarily composed of residues
Y179, E180, N183, D223, N224, L227 and W228. The NSP provides an additional anchor
point on the binding pathway and also concurs with the strong selection of positive amino
acids in this position as reported by Jaffe et. al. [2].
It was previously reported [11], that phosphorylation and mutation of S58 into negative
amino acids lead to dimer dissociation and a (usually negative) change in the 14-3-3 binding
affinity, independently of the oligomeric state. S58 is located near the dimer interface of 14-3-
3z and directly next to the binding pathway between IS1 and IS2. A negatively charged residue
in position 58 could negate the mildly positive surface potential (Fig 3 and S13 Fig), and disrupt
the observed binding pathway, which is in line with the observed experimental changes.
We identified the amino acids of 14-3-3z which were important phosphopeptide interaction
partners at each of the five stages of the binding pathway. Most interactions between 14-
3-3z and the phosphopeptides were polar hydrogen bonds or salt bridges. During our simulations,
as the phosphopeptide sampled the available volumes along the dominant pathway, the
overall probability to find a particular 14-3-3 amino acid interacting with the restrained phosphopeptide
was calculated (shown in S8 Fig) as function of the replica number (restraining DF
distance). In Fig 4 we display the table of all the 14-3-3z residues found significant (over 10%
probability at a given replica) in this analysis. The residues are coloured according to the stage
in which the residue was found most significant.
In dark purple we have depicted residues found most significant while the phosphopeptide
is bound in IS1 including many residues previously identified as crucial interaction partners
(such as Lys 49, and Arg 127) based on 14-3-3 crystal structures, and residues from the negative
surface patch such as Asp 224, which mostly interacted with the positively charged residues
of the phosphopeptide. The second group of amino acids (shown in orange) were found
most prominent on the early pathway (Pw1) between IS1 and IS2. Many of these residues were
probable interaction partners already when the peptide was in IS1 (such as Glu 180 and Arg
56), whilst other residues became significant only as the phosphopeptides left the primary
binding site (e.g. Glu 113 and Asp 124).
The third group of residues were most prominent whilst the phosphopeptide visited IS2.
Most of these residues (e.g. Ser 64, Lys 68 and Lys 75) are located in the positive surface patch
at the end of helix three (marked with light green in Fig 4), near the dimerization interface.
There were also three residues (Lys 11, Ser 207 and Glu 209, shown light blue) considered as
important interaction partners only during our dimeric DF/HRE-MD simulations. In dimeric
simulations these residues from the other, 14-3-3 monomer enhanced and expanded the positive
surface patch at IS2, and prevented the phosphopeptide to diverge from a dominant pathway
for ~1.0 nm longer in the distance field during the late pathway stage (Pw2) compared to
monomeric simulations.
The final group of interacting residues, were located on the disordered C-terminal region
(aa. 231–245) of the monomer from which the peptide is being pulled, shown in brown in Fig
4. The bulk of the tail interactions were observed in the PW1, IS2 and Pw2 stage of the binding
pathway (including the salt bridges with Glu 241 and Glu 244), although some of the tail residues
could interact with the phosphopeptide fragments while they were bound to IS1 (Thr
232, Gly 234 and Ala 237, most prominently with the p2h).
Fig 4 also highlights that the occurrence of phosphopeptide—tail interactions was very high
in two of the DF/HRE-MD simulations (dim_p1hT and dim_p2Ht) and very low in the other
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 7 / 30
two cases (dim_p1hT and dim_p2hT). The strongest interactions with the 14-3-3 tail region
were observed for the simulation dim_p2Ht, which at the same time had the weakest interactions
with IS2 and did not follow the dominant phosphopeptide binding pathway observed
during the other seven DF/HRE-MD simulations (S7 Fig). These observations suggest that Cterminal
tail may disrupt the dominant peptide binding pathway.
1.3. Comparing the pathways within 14-3-3z dimeric and truncated monomeric
model. The summary of our DF/HRE-MD simulations indicates a dominant phosphopeptide
binding pathway for 14-3-3z where the phosphopeptide in the process of unbinding
“slides” from the IS1 towards IS2 along the described pathway shown in Figs 3 and 4. Here, we
summarize the differences observed between our full dimeric systems (dim) and the minimalistic,
truncated, monomeric systems (tmon) during the detailed analysis of preferred positions
and phosphopeptide interactions (shown in Figs 2–4 and S1–S8 Figs).Our minimalistic systems
lacked the neighbouring monomer and the C-terminal region of the 14-3-3z protein,
Fig 4. 14-3-3ζ residues involved in the phosphopeptide interactions. The table on the left side shows the 14-3-3ζ amino acids, which
were found important based on the interaction map analysis (Fig 2, S1–S7 Figs) of the DF/HRE-MD simulations. Entries in the various
simulations are marked as not applicable (n/a), not significant (-), significant (+), important (++), or major interaction partners (+++). The last
three amino acids in the table were interaction partners from the other monomer. The right side shows the three dimensional structure of the
14-3-3ζ protein, where the monomers are depicted in cartoon and surface representations, respectively. The amino acids are depicted in
stick representation, coloured differently for the two monomers. In the cartoon representation, peptide-interacting amino-acids are coloured
according their position in the secondary structure, where helices 3, 5, 7 and 9 are shown in red, blue, green and brown, respectively, while
the C-terminal stretch is depicted in light purple. For the surface representation, amino acids found significant only in the bound-state
replicas (IS1) are shown as dark purple, amino acids along the binding pathway are marked in orange, light green and light blue, if they
appeared prior, in, or after the secondary interaction site (IS2). See S8 Fig for details.
https://doi.org/10.1371/journal.pone.0180633.g004
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 8 / 30
which, in the case of full dimeric systems, restricted the accessible space for the peptides and
necessitated longer DF distances to reach the unbound state.
The truncated monomer pathways were more similar to each other near the main peptide
binding groove (IS1), and showed enhanced interactions with helices 7–9, which were partially
blocked by the C-terminal region in the dimers. The monomeric pathways fanned out after
leaving the IS2 around 5 nm in the DF and sampled the outer side of 14-3-3z monomer before
final detachment from the 14-3-3z surface, and showed no significant interactions above 7
nm.
Full dimeric simulations lead to a more complete picture of 14-3-3z binding pathways and
could yield additional, biologically relevant information. The pathway comparison revealed a
more localized phosphopeptide presence for all dimeric systems near the dimerization interface,
along with new interaction partners from the other monomer (K11, S207, and E209).
During dimeric HRE-MD simulations which followed the dominant pathway, phosphopeptides
occupied the secondary interaction site (IS2) for a wider range of distances than their
monomeric counterparts and had non-negligible interactions between the restrained peptide
and the 14-3-3z dimer for DF distances smaller than 9.5 nm.
The interactions between the phosphopeptide and the C-terminal tail of the restrained 14-
3-3 monomer showed a large variation between the four full dimeric simulations. In case of
the p2h peptide fragment, strong interactions with the tail seemed to prevent the phosphopeptide
to follow the dominant pathway (S7 Fig), which is a major difference compared to the
truncated simulation with the same peptide (S2 Fig). The signs of direct interference from the
C-terminal tail, and the tail-associated discrepancies in the full dimeric phosphopeptide binding
simulations necessitated further investigation, shown below.
2. Structural changes along the binding pathway
The structural ensembles generated at different replicas within the DF/HRE-MD allowed us
to analyse structural changes as function of the DF distance from IS1. The results of this analysis
allowed us to monitor the slow conformational degrees of freedom which may affect the
binding/unbinding process of 14-3-3z. In the following section we identify these large scale
motions, and compare our DF/HRE-MD simulations in the bound (IS1) and unbound (Unb)
states with unrestrained MD simulations under the same conditions and—whenever possible
—experimental observations. This analysis allowed us to explore the observed discrepancies,
assess the effect of these large scale motions on the phosphopeptide binding process during the
limited sampling of our simulations, as well as effects of the applied DF/HRE-MD restraints
on the system.
2.1. The C-terminal region. The 14-3-3z proteins contain a 15 amino acid long C-terminal
region, which is a highly flexible disordered segment of the protein (Fig 5). In our simulations
two types of models were employed: full length dimeric (dim) models, and truncated
monomeric (tmon) models lacking the C-terminal tail. The C-terminal regions in our fulllength
models visited both the inner (dimer interface formed by helices 3–4 of both monomers)
and outer (helices 7–9) side of the 14-3-3z protein surface, as well as free conformations
in which the tail was detached from the protein surface (Fig 5A). The position of the C-terminal
tail in the simulations was determined based on the distances measured between the terminal
carbonyl (of N245) and the Nz atom of K68 (in IS2), Nz of K120 (in IS1), and Cγ of D197
(on the outer side), with cut-off distances 2.5, 2.0, and 2.5 nm, respectively. The tail was considered
to be on the inner side of 14-3-3z if the terminus was within the cut-off distance of
IS1 or IS2, considered to be on the outer side if it was within the cut-off distance of D197
but not the other two residues, and was considered free otherwise. The detailed population
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 9 / 30
distributions and number of transitions between the four states are shown in S2 and S3 Tables
for HRE-MD and MD simulations, respectively.
The exchange between the three conformational sub-states of the protein tail was a process
observed, but not properly sampled on the time scales of our DF/HRE-MD simulations. Analysis
of the tail residence during the DF/HRE-MD simulations (S2 Table) revealed a strong bias
towards the conformation of the tail at the start of the DF/HRE-MD run (shown in Fig 5B).
The tail of monomer 1 (M1) at the start of the simulations was facing outwards, while the tail
of monomer 2 (M2) was facing towards the inner side of the 14-3-3 dimer (in both cases the
starting conformation was detached from the surface). During the simulations each C-terminal
region of M1 dominantly sampled free conformations and the outer surface of 14-3-3z,
while tail of M2 sampled free conformations and the inner surface. A few transitions from the
inner to the outer side were observed for the M2 tail, suggesting a slow conformational transition
and preference towards the outer surface.
Fig 5. Flexibility of the 14-3-3ζ C-terminal tail. A) Representative structures of the conformational sub-states of the C-terminal stretch
(marked in red) interacting with the inner or outer surface of 14-3-3ζ or being detached and exposed to the solvent. Populations of these
conformational states as obtained from DF/HRE-MD (in red) and MD (in black) simulations are listed under the cartoon figures. B) 14-3-3
model starting conformation, coloured according to the backbone RSMF, where blue and red represent the least and most flexible aminoacids,
respectively. C) The atom-positional root-mean-square fluctuations (RMSF) of the 14-3-3ζ backbone atoms.
https://doi.org/10.1371/journal.pone.0180633.g005
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 10 / 30
The bias of the preferred C-terminal tail conformation also explains the differences
observed in the phosphopeptide interaction analysis. During the simulations dim_p1Ht
and dim_p2hT the peptide fragment was pulled from M1, and since the C-terminal tail was
mostly located on the outer side of the 14-3-3 dimer, no or very little interaction between
the tail and the phosphopeptide was observed (Fig 4). On the other hand, for the simulations
dim_p1hT and dim_p2Ht the peptide was pulled from M2 and the tail was close to the
inner surface and the phosphopeptide binding pathway, allowing for strong tail-peptide
interactions.
The explanation for the difference between the p2h binding pathways in tmon_p2H (S2
Fig) and dim_p2Ht (S7 Fig) is that the C-terminal tail partially buried and neutralized the positive
surface potential of IS2 in dim_p2Ht. This prevented the p2h fragment in dim_p2Ht from
following the dominant pathway, which consequently explored less likely, alternative binding/
unbinding pathways. In the case of dim_p1hT (Fig 2), IS2 was more exposed, and the Cterminal
tail formed salt bridges and remained attached to the phosphopeptide as it traversed
through the dominant pathway (until the p1t fragment was pulled out of reach).
We compared the tail behaviour of the DF/HRE-MD simulations to four unrestrained MD
simulations (two with and two without peptides) of 40 ns length (S3 Table). The unrestrained
simulations showed a similar behaviour (black and red percentages in Fig 5A) as the DF/
HRE-MD simulations. The tails showed a strong bias towards either the inner or outer side of
the 14-3-3z dimer, and even fewer transitions between the two were observed. However once
the C-terminal tail travelled from the inner surface to outer surface, we did not observe its
return. Consequently the classical MD simulations have shown an even stronger preference
towards the tail to be found near the outer surface. Despite this preference, the C-terminal
tails in both DF/HRE-MD and unrestrained MD simulations remained the most flexible part
of 14-3-3z (as shown in Fig 5C) with regular transitions between detached and surface bound
conformations.
The biological function of the 14-3-3 C-terminal region is not fully understood. It was suggested
previously [12] that the 14-3-3z C-terminal tail can have an auto-inhibitory effect by
binding to the phosphopeptide binding groove of its respective 14-3-3z monomer (IS1). The
C-terminal tail was not observed occupying the IS1 in any of our MD simulations, but this
occurred transiently (<5) % in the HRE-MD simulations of dim complexes. Our results suggest
that instead of occupying IS1, the tail rather interacts with amino acids on the outer surface
of its 14-3-3 monomer, or amino acids from IS2. In addition, the C-terminal region also
spent a considerable amount of time (~20%) being detached from the globular parts of 14-3-
3z, and in some of the DF/HRE-MD simulations directly interacted with the phosphopeptides
themselves during the binding process. Even though the sampling of the conformational substates
is incomplete, our simulations are more in line with NMR studies that showed high flexibility
and only transient interactions between the structured part of 14-3-3z and the C-terminal
region [13].
2.2. 14-3-3z monomer opening. Structural changes in the 14-3-3z protein monomers
leading to an opening or closing of the peptide-binding groove were previously reported using
both computational and experimental methods [5,14]. This breathing motion was observed in
our simulations as well, for both dim and tmon systems and their phosphopeptide complexes.
We measured the groove width of a 14-3-3z monomer associated with the opening process, as
the distance between the Cα atom of G53 (middle of helix 3) and L191 (middle of helix 8) (Fig
6A). Based on the measurement of the groove width distributions, we defined three sub-states
(closed, open, wide-open) of the 14-3-3z monomer, depicted in Fig 6. We considered the 14-3-
3z monomer closed, if the groove width was below 2.5 nm, wide-open above 2.9 nm and open
in between.
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 11 / 30
Fig 6. Changes in the 14-3-3ζ monomer groove width. A) Models of 14-3-3ζ at different levels of opening (the green, blue and orange colours
mark closed, open and wide-open states, respectively), the last 3 helices are marked with a darker colour. B) Representative replicas of five
different stages along the binding pathway (with the replica ID number shown in brackets) from DF/HRE-MD simulation dim_p1hT. The panel
shows groove width distributions whilst the phosphopeptide moves from the binding site (IS1) to the unbound state (Unb). C) Average groove
width distributions of 14-3-3ζ monomers as obtained from DF/HRE-MD bound (IS1) and unbound (Unb) states, and unrestrained MD simulations
in holo (bound) and apo (unbound) states.
https://doi.org/10.1371/journal.pone.0180633.g006
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 12 / 30
We monitored the groove width distribution as the function of phosphopeptide location
along the binding/unbinding pathway and found that the breathing motion in the restrained
monomer was changed for all DF/HRE-MD simulations. Fig 6B presents the distribution of
the groove widths for replicas representing the five stages along the binding pathway of the
simulation dim_p1hT (replicas 2, 6, 20, 32 and 48 for IS1, Pw1, IS2, Pw2 and Unb, respectively).
When the phosphopeptide was further away from the primary binding site (in the IS2,
Pw2 or Unb stage) the groove width distributions were similar, centered on the open state
(~2.6 nm), with a smaller probability to visit both the closed and wide-open states. When the
phosphopeptide was bound to IS1, however, the interactions with residues from helices three,
five and seven resulted in narrower groove width distributions which were also shifted towards
the closed state. A similar observation was reported by Hu et. al. [14] (See S13 Fig). Interestingly,
shortly after the phosphopeptide left IS1 (typically between replicas 6–11, Pw1) very
wide grove width distributions were observed, with a high probability of the monomer from
which the peptide was pulled adopting a wide-open conformation. The wide-open conformation
in Pw1 is present for all our 14-3-3z/phosphopeptide models and is not dominant for any
other part of the binding/unbinding pathway. These results (to our knowledge not reported
before) indicate that the wide-open conformation of 14-3-3z may be necessary for the phosphopeptide
ligand to detach from or enter into the primary binding site at IS1.
The distributions of the groove width for apo (unbound) and holo (bound) 14-3-3z in unrestrained
MD simulations were also compared in Fig 6C with average groove width distributions
for IS1 (bound) and Unb (unbound) DF/HRE-MD replicas. As the figure shows, the
range of groove width distributions for the DF/HRE-MD and unrestrained MD simulations
were very similar, and a shift towards the closed state was also observed between the bound
and unbound monomers. The similarity between groove width distributions suggest, that the
presence of a phosphopeptide strongly affected the breathing motion of 14-3-3z monomers in
our simulations, but the applied DF restraints did not perturb this conformational degree of
freedom directly.
2.3. Inter-monomer twist within 14-3-3z homo-dimer. Apart from the internal motion
of the monomers, motion of monomers relative to each other was also observed in our (dim)
simulations. This motion was monitored by measuring the dihedral angle of the Cα atoms of
the residues L43(M1)-A54(M1)-A54(M2)-L43(M2) where M1 and M2 indicate different
monomers within the 14-3-3z dimer (Fig 7A).
During DF/HRE-MD simulations an inter-monomer twist of 175 ± 15˚ was observed, with
little variation with respect to the pulled phosphopeptide fragment or its DF restraint position
(shown in Fig 7B). The similar distributions of the inter-monomer twist angles suggest that
this conformational degree of freedom is not involved in the phosphopeptide binding process.
We speculate the inter-monomers twist may still be important for protein-protein binding
processes, as it can help 14-3-3 to better adapt the binding interface.
In some preliminary regular MD simulations at lower salt content, the inter-monomers
twist angle was significantly reduced (see Methods section and S9 Fig). However, the intermonomer
twist angles in the DF/HRE-MD simulations were consistent with the twist angles
observed in the available crystal structures of 14-3-3z dimers (~180 ± 3˚ calculated from the
pdb entries 2WHO and 1A4O) and yielded a better agreement with SAXS measurements
(S14 Fig).
2.4. Secondary structure changes of phosphopeptides along the binding pathway.
The secondary structure of the 14-3-3z dimer remained mostly unchanged during our DF/
HRE-MD and classical MD simulations, except for the occasional shortening of helix nine,
and smaller conformational changes within the loops and disordered tails (e.g shown in
S10 Fig). Secondary structure analyses of phosphopeptides for individual replicas of the
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 13 / 30
DF/HRE-MD along the binding pathway were also performed using the DISICL algorithm.
The secondary structure (SS) of phosphopeptides along their binding pathway was summarized
for all eight DF/HRE-MD simulations in S11 and S12 Figs. In Fig 8A we present the
DISICL profile of the p1h phosphopeptide along the binding pathway during the simulation
dim_p1Ht. The phosphopeptide in IS1 (replicas 1–4) adopts an extended conformation, dominated
by the β-cap (BC, brown) and polyproline-like (PP, maroon) classes (~60% of all amino
acids in the ensemble). After the phosphopeptide leaves IS1 it adopts less extended conformations,
the PP and BC classes become less characteristic, and the population of the helix-cap
(HC, blue) class rises slightly.
The analysis of the other phosphopeptides revealed a similar highly extended backbone
structure for all four phosphopeptide fragments in the bound state, with a dominance of the
polyproline-like (PP) and β-cap (BC) classes. This is also in agreement with the conformations
observed in the crystal structures and unrestrained MD simulations (a representative conformation
is shown in Fig 8B). Similar trends were also observed for later stages of the eight DF/
HRE-MD simulations. The presence of helical classes, mainly the helix-cap, π-helical (PIH,
cyan), and α-helical (ALH, green) population was gradually increased to 5–20% as the peptide
reached the unbound state, whilst the population of the extended classes (PC and BC)
decreased to ~30%). The population of the secondary structure classes in the 100 ns MD simulations
of the free phosphopeptide fragments (e.g shown in Fig 8C), also showed an increased
helical preference (~20%) compared to the extended 14-3-3z-bound peptide fragments. In
these simulations, the average population of π-helix is ~10%, significantly higher than what
was expected for a random coil peptide (~0.4%). We found that this conformation was stabilized
by intra-molecular interactions with the pSer sidechain.
Despite the common trends, phosphopeptide-specific conformational changes were also
observed during the DF/HREMD simulations, with the largest conformational changes in secondary
structure occurring between replicas 5–12 (Pw1, DF: 0.9–2.2 nm) and 20–25 (IS2, DF:
4.4–5.5 nm) along the binding pathway. The simulation dim_p2Ht showed an unusually high
π-helical content, similar to the free peptide simulations. Note that dim_p2Ht did not follow
the dominant binding pathway. The distorted π-helical conformation of the backbone in this
Fig 7. 14-3-3ζ inter-monomer twist. A) The top view of the protein dimer, with the twist angle between the monomers displayed in dark
blue. B) Probability distribution of inter-monomers twist angles averaged over all DF/HRE-MD simulations in the IS1 (bound) and Unb
(unbound) stages, and for the whole pathway.
https://doi.org/10.1371/journal.pone.0180633.g007
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 14 / 30
simulation increased (to ~10%) gradually while the phosphopeptide gets unbound, stabilised
by an intra-molecular hydrogen-bond bridge between the pSer 5 side-chain and residues His 2
and Arg 3. These interactions appear early in the pathway, when the phosphopeptide is prevented
to follow the positive surface patch of the 14-3-3 dimer, due to the intervention of the
C-terminal tail.
3. Free-energy changes along the binding pathway
3.1. PMF profiles and binding affinity determination. The HRE-MD simulations were
used to determine the Helmholtz free energy profiles (Awham
(l)) along the binding pathway of
the four phosphopeptide fragments bound to both dim and tmon models of 14-3-3z. It is important
to emphasise that the DF/HRE-MD methodology greatly enhances the sampling of phosphopeptide
binding/unbinding but it does not enhance the sampling of slow conformational
Fig 8. Phosphopeptide p1h secondary structure. Changes in the backbone structure are shown during DF/HRE-MD simulation
dim_p1Ht (panel A) and unrestrained MD (panels B and C) simulations, analysed by the DISICL algorithm. The change in the average
secondary structure content during the DF/HRE-MD simulation is shown in the middle of panel A, whilst the most dominant conformations in
bound and unbound states are tabulated on the left and right side, respectively. Representative conformations for the bound and unbound
states are depicted on the left and right sides, respectively, where the residues are coloured according to secondary structure classification.
Intra-molecular hydrogen bonds are depicted as dashed lines. The tables besides the depictions show the 5 most populated secondary
structure classes for the corresponding MD simulation and the appropriate (bound(IS1)/unbound) stage of the DF/HRE-MD simulation,
respectively. The most populated DISICL classes are depicted in the following colours: π-helix (PIH)–cyan, Extended β-strand (EBS)–red,
normal β-strand (NBS)–orange, polyproline-like (PP)–brown, turn type VIII (TVIII)–indigo, Gamma turn (GXT)–maroon, β-cap (BC)–gold,
helix-cap (HC)–blue, turn-cap (TC)–black. DISICL secondary structure elements are listed in S1 Table.
https://doi.org/10.1371/journal.pone.0180633.g008
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 15 / 30
transitions of the protein (e.g. C-terminal tail) to the same extent. Due to the slow transitions
between different conformations, the simulations of dimeric complexes are not fully converged
and should be considered only in a qualitative way (e.g. shape of curves Awham
(l) in Fig 9). On
the other hand, our minimalistic systems (tmon complexes) lack the C-terminal tail, sample a
Fig 9. Free energy profiles of the DF/HRE-MD simulations. A) The raw free-energy profiles, derived from
WHAM analysis, as function of simulation time per replica (every ns) for the simulation tmon_p2H. B)
Convergence of the free energy difference between the unbound and bound state of all eight DF/HRE-MD
simulations. C) Final free energy profiles (ΔAwham
) for the eight 14-3-3ζ peptide-binding simulations.
https://doi.org/10.1371/journal.pone.0180633.g009
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 16 / 30
less complex conformational space and, therefore, the convergence of free energy profiles is
much better. Fig 9A depicts the change of the free-energy profile (Awham
(l)) over simulation
time, as obtained from the weighted histogram analysis method (WHAM) for simulation
tmon_p2h. The physical meaning of the free energy profile (Awham
(l)) is that the probability
(ρ(l)) of finding the phosphopeptide at the DF distance l is proportional to exp(-Awham
(l)). The
Awham
(l) profiles for all systems were shifted by a constant such, that their minimal value is zero.
Fig 9B presents the convergence of Awham
(l) profiles as a function of DF/HRE-MD simulation
time for the 8 studied complexes.
In order to calculate binding free energies that can be compared with the experimental
binding affinities, a standard state correction has to be added to the raw WHAM profile as
described in the Methods section (Eq 5). This correction relates the volume associated with the
unbound state to the standard state volume. Table 1 shows the sampled and accessible volume
(based on the restraining distance), as well as the approximate free energy from the WHAM
calculations for every replica of simulation tmon_p2h, together with the zero-energy DF distance
(l0). Once the unbound replicas are identified, the unbound volume is calculated from
the sampled volume of the unbound replicas. The free energy of binding is calculated by subtracting
the free energy of the unbound state (Eq 4) from the free energy of the bound state
(Eq 3) and adding the standard state correction in accordance with Eq 5.
Table 2 summarises the resulting binding free energies and draws a comparison with the
available experimental binding affinities. The binding affinities of 14-3-3z to very similar peptide
fragments with identical binding recognition sequences were recently documented using
fluorescence spectroscopy and isothermal calorimetry (ITC) methods [8,15], resulting in binding
affinities in the 100–10 μM range corresponding to binding free energies of -20 to -30 kJ/
mol. Surface plasmon resonance (SPR) experiments [2,10] on identical or very similar peptides
reported binding affinities of about 100 nM, corresponding to binding free energies of -45 kJ/
mol (p2t peptide). Our computational absolute binding affinities in the range of -30 to -55 kJ/
mol correspond roughly to the available experimental data taking into consideration the fact
that no fitting parameters were used. Please note that the presented experimental binding freeenergies
were measured for longer peptides with an identical recognition sequence, thus the
free energies of binding do not need to agree completely with the ones calculated from our
simulations. The monomeric PMF profiles as well as two of the dimeric profiles (dim_p1Ht
and dim_p2hT) have a second, local free-energy minimum at 3.5–5.0 nm in the distancefield,
associated with IS2 and a free energy barrier at ~3.0 nm separating IS1 and IS2 (Fig 9C). For
the other two complexes (dim_p1hT and dim_p2Ht) neither the barrier nor the local minimum
was observed, this is likely due to interference from the C-terminal tail of 14-3-3z, which
was—in both cases—mostly located on the inner side of the 14-3-3 dimer.
3.2. Impact of C-terminal tail on binding to the IS2. Comparison of free energy profiles
between full length dimeric and truncated monomeric models of the 14-3-3z complexes
allowed us to estimate the impact of the C-terminal tail on the phosphopeptide binding. The
conformational transition of the C-terminal tail between the inner and outer side of 14-3-3z
(Fig 2, S2 and S3 Tables) was a slow collective motion, poorly sampled in the dimeric simulations.
However, two of DF/HRE-MD simulations (dim_p1Ht and dim_p2hT) thoroughly
sampled the most probable binding pathways whilst the C-terminal tail remained attached to
the outer protein surface. The other two simulations (dim_p1hT and dim_p2Ht) sampled
pathways, during which the tail was attached to IS2 (more structural details in section 2.1).
The PMF profile of simulations where the C-terminal tail was for a considerable time present
on the inner side of the 14-3-3z dimer (dim_p1hT and dim_p2Ht) does not exhibit a local
free-energy minimum associated with IS2 (Fig 9C), or a free-energy barrier which would prevent
the phosphopeptide to directly bind to IS1. In case of dim_p2Ht, the C-terminal tail was
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 17 / 30
Table 1. Details of the DF/HRE-MD simulation of tmon_p2H. The table contains quantities for every replica that are required for the free energy calculations.
The columns show the state assignment, the replica number, the sampled volume (Vsampled), replica exchange probability (Pex), reaction coordinate
value (λ), ideal distancefield distance (l0), accessible volume (V(l)) and raw free-energy according to the free-energy profile (Awham
). V(l) and Awham
were calculated
based on the restraining distance assigned to the replica. The last row contains the unbound volume (Vunb), calculated from the volume sampled by the
phosphopeptide in all Unb state replicas, and the volume of simulation box (Vbox).
State Replica Vsampled Pex λ l0 V(l) Awham
tmon_p2H Number nm3
Nm nm3
kJ/mol
IS1 (bound) 1 0.15 0.105 0.00 0.2 0.03 7
IS1 (bound) 2 0.27 0.223 0.02 0.4 0.09 5.6
IS1 (bound) 3 0.37 0.219 0.04 0.6 0.21 0.6
IS1 (bound) 4 0.44 0.222 0.06 0.9 0.60 1.7
IS1 (bound) 5 0.55 0.175 0.08 1.1 0.86 10.9
IS1/Pw1 6 0.75 0.099 0.11 1.3 0.77 16.7
IS1/Pw1 7 1.02 0.100 0.13 1.5 0.99 20.2
IS1/Pw1 8 1.38 0.130 0.15 1.7 1.21 22.9
Pw1 9 1.59 0.143 0.17 1.9 1.89 24.5
Pw1 10 1.60 0.123 0.19 2.2 2.19 28.1
Pw1 11 1.50 0.123 0.21 2.4 1.70 28.3
Pw1 12 1.78 0.118 0.23 2.6 2.02 32.2
IS2 13 2.05 0.089 0.25 2.8 2.47 35.4
IS2 14 2.65 0.102 0.27 3.0 2.99 35.2
IS2 15 2.71 0.121 0.29 3.3 5.61 33.9
IS2 16 2.66 0.130 0.32 3.5 7.00 31.8
IS2 17 2.78 0.113 0.34 3.7 5.84 30.3
IS2 18 2.72 0.098 0.36 3.9 6.93 30.3
IS2 19 3.19 0.137 0.38 4.1 10.94 29.8
IS2 20 2.78 0.170 0.40 4.4 13.14 29.5
IS2 21 3.35 0.156 0.42 4.6 10.47 30.5
IS2 22 4.12 0.120 0.44 4.8 12.05 31.1
IS2 23 4.91 0.119 0.46 5.0 13.82 31.7
IS2 24 5.61 0.150 0.48 5.2 15.69 32.6
IS2 25 6.76 0.128 0.51 5.5 27.42 34.7
IS2 26 7.37 0.114 0.53 5.7 31.08 35.7
IS2 27 7.69 0.142 0.55 5.9 22.56 35.4
Pw2 28 8.16 0.140 0.57 6.1 23.65 35.7
Pw2 29 7.67 0.132 0.59 6.3 24.40 35.7
Pw2/Unb 30 7.95 0.124 0.61 6.5 24.72 35.2
Pw2/Unb 31 9.47 0.125 0.63 6.8 24.76 36.2
Pw2/Unb 32 9.57 0.125 0.65 7.0 24.51 37
Pw2/Unb 33 9.75 0.111 0.67 7.2 23.97 37.5
Pw2/Unb 34 10.37 0.129 0.69 7.4 23.25 38.1
Unb (unbound) 35 10.69 0.138 0.72 7.6 22.52 39.1
Unb (unbound) 36 10.30 0.117 0.74 7.9 32.17 39.6
Unb (unbound) 37 9.83 0.114 0.76 8.1 30.47 39.9
Unb (unbound) 38 9.59 0.109 0.78 8.3 18.96 40.4
Unb (unbound) 39 10.17 0.112 0.80 8.5 17.80 40.7
Unb (unbound) 40 9.34 0.111 0.82 8.7 16.40 41
Unb (unbound) 41 8.84 0.107 0.84 9.0 15.00 40.9
Unb (unbound) 42 9.89 0.123 0.86 9.2 13.74 41.7
Unb (unbound) 43 9.90 0.124 0.88 9.4 12.47 41.8
(Continued)
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 18 / 30
directly interacting with IS2 and prevented interactions between the p2h peptide and IS2. In
case of dim_p1hT, IS2 was partially available for the p1t peptide, however, interactions
between p1t and the C-terminal tail disrupted interactions with IS2. On the other hand, in simulations
where the C-terminal tail was not present (all tmon simulations) or mostly present on
the outer surfaces of 14-3-3z (dim_p1Ht and dim_p2hT) the drop in free-energy profile in the
IS2 area and a free energy barrier (at ~3.0 nm separating IS1 and IS2) are present. These two
features suggest an intermediate state (when the peptide is bound IS2) along the phosphopeptide
binding pathway, which is diminished by the presence of the C-terminal tail near the
inner side of the 14-3-3z dimer. This led us to the conclusion that one of the roles of the C-terminal
tail may be to weaken the interaction between phosphopeptides and IS2 of the 14-3-3z
protein.
Conclusions
We explored the phosphopeptide binding pathways of the 14-3-3z protein through molecular
dynamics simulations of four phosphopeptide fragments derived from PKC-ε and C-RAF
kinase. The pathways were explored by a novel Hamiltonian replica exchange molecular
dynamics method with incorporated distancefield restraints (DF/HRE-MD). The eight DF/
HRE-MD simulations (4 dimeric and 4 truncated monomeric complexes) combined corresponded
to more than 6.7 μs of enhanced-sampling simulation time, allowing for the unbiased
determination of the most probable binding/unbinding pathways, the corresponding structural
changes of the phosphopeptides and 14-3-3z, as well as the PMF profiles along the binding
pathway.
The determined binding pathways were very similar for 7 out of the 8 studied complexes,
suggesting a dominant phosphopeptide pathway, which roughly followed helix 3 between the
Table 1. (Continued)
State Replica Vsampled Pex λ l0 V(l) Awham
tmon_p2H Number nm3
Nm nm3
kJ/mol
Unb (unbound) 44 9.04 0.109 0.91 9.6 11.20 42.1
Unb (unbound) 45 7.60 0.064 0.93 9.8 9.98 42.9
Unb (unbound) 46 6.96 0.090 0.96 10.2 8.82 43.7
Unb (unbound) 47 7.73 0.134 0.98 10.4 7.74 43.9
Unb (unbound) 48 6.87 0.059 1.00 10.6 6.73 45
Vunb 74.8 nm3
Vbox 626.4 nm3
https://doi.org/10.1371/journal.pone.0180633.t001
Table 2. Comparison of the experimental and calculated binding free energies. The ΔGexp shows the
experimental free energies calculated from dissociation constants [8–10]. ΔAbind(mon) shows the binding free
energy calculated from the tmon DF/HRE-MD simulations (20ns/replica).
Simulated ΔGexp ΔAbind(mon)
System kJ/mol kJ/mol
tmon_p1H -28.8 -30.9
tmon_p1T -23 * -49.1
tmon_p2H -21.6 -47.8
tmon_p2T -27.4 / -44.2 -52.9
*: weak binding, estimate based on detection limit
https://doi.org/10.1371/journal.pone.0180633.t002
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 19 / 30
primary binding site (IS1) and a newly identified secondary interaction site (IS2), localized in
the second half of helix 3 (residues 60–70). We found that the flexible C-terminal tail of 14-3-
3z may interact with both IS2 and the phosphopeptide in our simulations. When the C-terminal
region interfered with the phosphopeptide binding pathway the interactions between the
phosphopeptide and 14-3-3z IS2 were changed significantly, and this change was also reflected
in the corresponding PMF profiles.
We confirmed previous findings suggesting that 14-3-3 monomers in complex with a phosphopeptide
are shifted towards a more closed conformation as compared to the apo state. Our
DF/HRE-MD simulations revealed that the 14-3-3z monomer adopts a wide-opened conformation
when a phosphopeptide is to enter or leave IS1. The phosphopeptide secondary structure
during the DF/HRE-MD simulations also changed from an extended conformation at IS1
into a less ordered structure with ~20% helical content, with a high probability of π-helical
conformations in the unbound state. Sequence specific rearrangements in the peptide structure
were detected during the Pw1 and IS2 stages, followed by the gradual increase of helical
content as the phosphopeptides detached from 14-3-3z.
The DF/HRE-MD simulations allowed effective pathway sampling of the phosphopeptide
fragments within the 14-3-3z protein. Application of distancefield restraints prevented the 14-
3-3z protein damage that was observed in cases when regular distance-restraints were applied.
While the full length 14-3-3z WT dimer models complexed with the four phosphopeptides
proved to be useful for exploring the structural properties, their size and slow convergence
prevented effective free-energy calculation for these systems. Therefore minimalistic models
based on a truncated 14-3-3z monomer were designed, which proved to be more efficient in
calculating the potential-of-mean-force (PMF) profiles for the binding of phosphopeptide/14-
3-3z complexes. The binding free energies derived from the calculated PMF profiles of minimalistic
14-3-3z models show a reasonable agreement with known experimental binding affinities
between similar PKC-ε and C-RAF kinase fragments and the 14-3-3z protein.
Taken together, these findings deepen our understanding about the binding phenomena of
phosphopeptides to 14-3-3z and the obtained results likely have a wider applicability for other
human isoforms considering their high sequence identity.
Methods
All molecular dynamics (MD) and Hamiltonian replica exchange MD (HRE-MD) simulations
were performed using the GROMOS11 software package [16]. The structure preparation and
analysis of the simulation trajectories was based on the GROMOS++ analysis package [17].
The PyMol molecular graphics system version 1.5.0.3 [18] was used for visualization and calculation
of the electrostatic surface potentials based on an adaptive Poisson-Boltzmann solver
[19] as implemented in the PyMol software. Changes in the protein and peptide secondary
structure were followed by the DISICL algorithm [20].
Generated starting coordinates
The starting coordinates of the presented 14-3-3z models were generated based on the same
crystal structure (pdb code: 2WH0) in order to prevent structural changes due to different protein
starting structures. The crystal structure 2WH0 represents the dimeric 14-3-3z in complex
with protein kinase C ε peptide fragments [8] (p1h, p1t). The structure of the C-RAF proto
oncogene peptide fragments [9] (p2h, p2t) bound to 14-3-3z was taken from the structure of
their co-crystal (pdb code: 4FJ3), after aligning the 14-3-3z dimer structures with 2WH0. The
missing parts of the 14-3-3z protein, including the C-terminal tail, were added using Modeller
version 9v8 [21]. The phosphopeptide structure was energy minimized in vacuum to avoid
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 20 / 30
clashes with the 2WH0 protein atoms. The 14-3-3z dimer model without ligands was constructed
based on the 2WH0 crystal structure where all peptide atoms were removed. The
sequence of the complete peptides, as well as the head and tail fragments are indicated in Fig 1.
The head and tail peptide fragments were chosen on the basis that they contain the consensus
binding motif and the phosphorylation site in the middle of the sequence, consist of 8 amino
acids for all peptide fragments, and have a total charge of -1.
Minimalistic (truncated monomeric) models were constructed from the dimer complexes
by extracting the corresponding 14-3-3z monomer (along with its phosphopeptide) and truncating
the last 15 amino acids (C-terminal stretch).
The prepared PDB coordinate files were transformed into GROMOS configuration files
compatible with the GROMOS 54a7 force field [22], with phosphorylation parameters taken
from the Vienna PTM 54a7 extension [23]. The structures were then energy minimised, solvated
in pre-equilibrated SPC water [24] to fill a rectangular box with a minimal solute-to-wall
distance of 2.0, 2.0, and 2.5 nm in the X, Y and Z dimensions, respectively, to provide sufficient
space to pull the phosphopeptide fragments out of the binding site. The solute models were
rotated in such a way that the largest dimension of the solute complex was oriented along the
Z axis. Sodium and chloride ions were added to all simulation boxes to neutralise the total
charge and provide a NaCl concentration of 0.15 or 0.25 M for electrostatic screening. The
systems of the monomers typically contained 57 000 atoms, whilst the dimer complexes
amounted to roughly 121 000 atoms. After solvation, the simulation boxes were energy minimised,
then heated up from 60 to 298 K while gradually reducing the position restraints on
protein and peptide atoms (initial force constant of 2.5 x 104
kJ/mol/nm2
) in 5 discrete steps of
20 ps each, and subsequently equilibrated for 60 ps without position restraints.
Molecular dynamics simulations
Equilibration and all simulations were performed under periodic boundary conditions, using
a 2-fs time step and the SHAKE algorithm [25] to constrain bond lengths and H–O–H bond
angles. The weak-coupling algorithm [26] was used to maintain a stable temperature (300 K)
and pressure (101 kPa) when required, with relaxation times of 0.1 and 0.5 ps respectively. For
long-range interactions the reaction field method [27] was used with a 1.4 nm cut-off and 61
as dielectric permittivity [28]. During all of the production simulations roto-translational constraints
[29] were kept on all protein atoms to prevent tumbling of the complexes in the rectangular
simulation box.
During our simulations SHAKE errors occurred frequently at planar amide groups found
in peptide bonds of the polypeptide chain and in the side chain amide groups with a delocalised
character, such as arginines, glutamines and asparagines. The backbone N-H bond of
arginine 18 appeared particularly regularly due to local structural strain. To treat this frequently
occurring numerical problem, we increased the mass of this particular hydrogen by a
factor of 5, in both our MD and HRE-MD production runs. MD simulations of all 14-3-3z
dimer, monomer and complex systems were performed for 40 ns, and all phosphopeptide
related MD simulations for 100 ns.
Preliminary simulations proved to be quite sensitive to the NaCl concentration. Our initial
intention was to run all simulations at 0.15 M NaCl (physiological) concentration. This corresponds
to the salt concentration at which most of experimental binding affinities were measured.
However, preliminary simulations of the dimeric systems at 0.15 M showed significant
instabilities in terms of inter-monomer twist angles, defined as the dihedral angle of the Cα
atoms of the residues L43(M1)-A54(M1)-A54(M2)-L43(M2) where M1 and M2 indicate different
monomers within the 14-3-3z dimer (Fig 7A). These problems were not observed at
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 21 / 30
higher, 0.25 M NaCl concentration (S9A Fig). The comparison of experimental and calculated
SAXS profiles of dim systems (S14 Fig) indicated that low values of inter-monomer twist
(often observed at 0.15 M NaCl) do not agree with solution SAXS data. Therefore, we decided
to perform all dimeric (dim) simulations (presented in this study) at 0.25 M NaCl and all tmon
simulations at 0.15 M NaCl where inter-monomer twist instability cannot occur.
Distancefield replica-exchange simulations
DF distance restraints and HRE-MD were applied as they are implemented in the GROMOS11
version 1.3.0 [16]. To compute the binding free energies for the phosphopeptide fragments
(p1h, p1t, p2h, p2t), a reaction coordinate was defined as the DF distance between a virtual
atom at the binding site of the corresponding 14-3-3z monomer and a virtual atom at the
phosphorylation site of the peptide. The virtual atoms for all peptides were defined by the centre
of mass of the Cβ carbon atoms of the phosphorylated serine and its two neighbouring
amino acids. The virtual atom for the binding site was defined as the centre of mass of the Cα
atoms in residues N50, A192 and the amide hydrogen of N224 of the appropriate monomer.
Simulations were started along the reaction coordinates to pull one of the peptides out of its
respective binding pocket. This was done using a harmonic distancefield [6] distance restraint
in 20 discrete steps with a minimal-energy distance ranging from 0.2 to 10.6 nm, a force constant
of 2500 kJ/mol/nm2
and a simulation length of 250 ps at each step. The final configurations
at each step were used to start 100 ps of HRE-MD [30] equilibration. During the
HRE-MD equilibration with 48 replicas (Table 1) exchange events were prohibited and position
restraints (25 kJ/mol/nm2
) were applied to the protein atoms. After this equilibration, the
final coordinates were then used to start the DF/HRE-MD production simulations, where replica
exchanges were permitted every 10 ps and position restraints on the proteins were replaced
with roto-translational constraints to maintain the protein orientation. Each of the four
HRE-MD simulations of dim complexes was run for 15 ns; and of tmon complexes for 20 ns,
using 48 replicas to cover a DF distance of ~10 nm from the IS1. The average interaction
energy, and hydrogen bonds were monitored between peptide and protein atoms at each replica
to determine the minimal distance for the unbound state (Fig 2C).
HRE-MD production simulations for each peptide fragment uniformly used a harmonic
DF restraining potential with a force constant of 350 kJ/mol/nm2
. The harmonic potential was
linearised after deviations larger than 2 nm in order to avoid large forces over longer distances.
The DF for the distance calculations was updated each 100 steps with a grid spacing and protein
cut-off of 0.2 nm, and a single smoothing step to decrease repulsion at the protein surface
[6]. HRE-MD simulations were kept at a constant temperature of 300 K by the weak coupling
algorithm and were performed at a constant volume. The combined simulation time length of
the DF/HRE-MD comes to a total of ~6.7 μs of enhanced sampling to study the 14-3-3z peptide
binding.
Phosphopeptide pathway assignment
The phosphopeptide binding-pathways obtained from DF/HRE-MD simulations were divided
into 5 parts (IS1, Pw1, IS2, Pw2, Unb) during the analysis. The replicas of the DF/HRE-MD
simulations were assigned to IS1 or IS2 based on three factors:
1. the average protein-phosphopeptide interaction energy of the replica had to be near a local
minimum.
2. interactions observed with 14-3-3z residues in the corresponding interaction sites (more
details below).
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 22 / 30
3. the most probable locations to find the phosphopeptide for the replica had to be in the
proximity of the interaction site.
An interaction partner was considered stable in the interaction analysis if on average at
least 0.1 interaction was present in the particular replica. Interaction partners for IS1 were
K49, R127, Y128 towards the pS of the peptide. Interaction partners for IS2 were K68 towards
the pS and any interaction involving S64-G70. Proximity from an interaction site was determined
by measuring the distance between density peak maxima of the phospopeptide virtual
atom for that particular replica (as shown in Fig 8) and Cα of K49 and K68 for IS1 and IS2,
respectively. The density peak was considered proximal within 1.5 nm. The replicas were
assigned to unbound state (Unb) if the interaction energy was close to zero and no stable
hydrogen-bonds were detected between the phosphopeptide and 14-3-3z. Replicas that were
fulfilling only part of the criteria were assigned to Pw1 and Pw2 instead. An example of assignment
is show in Table 1, and replica density peaks are coloured according to their assignment
in Fig 2, S1–S7 Figs.
Free energy calculations
The free energy profile was calculated from the replica exchange simulations, using a weighted
histogram analysis method (WHAM) [31]. The WHAM is an iterative method that uses the
probability distribution of biased ensembles ðrðbÞ
i Þ and reconstructs the unbiased probability
distribution (ρ) by fitting the free-energy of a given number of windows (Nw) along the reaction
coordinate. Using the DF distance distributions of the replicas—biased by the harmonic
DF restraining potentials (Bj)–the probability distribution along the peptide binding pathway
is calculated iteratively according to Eqs 1 and 2:
rðlÞ ¼
PNw
i¼1 ni rðbÞ
i ðlÞ
PNw
j¼1 nje
À ðBjðlÞÀ Awham
j
Þ=RT
ð1Þ
eÀ Alwham
j
=RT
¼
R
rðlÞ eÀ BjðlÞ=RT
dl ð2Þ
Where ρ(l) is the probability of finding the peptide at the DF distance l, Awham
j is the Helmholtz
free energy at the DF window j, ni is the number of data points in replica i, R is the ideal gas
constant, and T is the temperature. The resulting free-energy profile(Awham
(l)) was used to calculate
the free energy of the bound (Abound) and unbound (Aunb) states, by integrating the profile
over the DF distances associated with the two states, as shown in Eqs 3 and 4, respectively:
Abound ¼ À RT ln
R
bound
eÀ AwhamðlÞ=RT
dl ð3Þ
Aunb ¼ À RT ln
R
unb
eÀ AwhamðlÞ=RT
dl ð4Þ
Because of the shape of Awham
(l) in the bound area the Abound value is quite insensitive on the
particular choice of boundaries of the integral in Eq 3. On the other hand, Aunb depends significantly
on the particular choice of the unb region boundaries in Eq 4. This dependence is
mainly due to entropic contributions from the increased available volume of the unbound
state, and is compensated by the standard state correction (ΔAstd) in the final calculation of the
binding free energy (ΔAbind) by Eq (5):
DAbind ¼ Abound À Aunb þ DAstd; DAstd ¼ À RT ln
Vunb
V0
ð5Þ
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 23 / 30
where V0 is the standard state volume (1.66 nm3
) corresponding to a 1 M concentration. The
physical meaning of Eq 5 is that the binding free energy (ΔAbind) corresponds to the freeenergy
difference of bringing a ligand from a bound to an unbound state and subsequently
from the unbound to the standard state volume [32]. ΔAbind can be directly compared with
experimental values calculated from dissociation constants. The unbound volume (Vunb) was
calculated based on the number of grid-points visited by the phosphopeptide virtual atom in
any replica of unbound state (Table 1).
To calculate the free-energy profiles of the peptide binding processes, we used the DF distributions
of the DF distances collected over the 48 replicas of the HRE-MD simulations and 100
DF widows for the iterative WHAM process, which was continued for 10 000 steps or until the
energy change was less than 10−5
kJ/mol. Calculations of the available volume were approximated
from the DF grid based on the number of gridpoints assigned to a given distance. The
Fig 10. Graphical representation of pathway mapping in simulations. A) Distancefield grid, where grid points are coloured according to
distancefield (DF) distance. B) Direct distance (as a black dashed line), and DF distance (along the coloured grid points) between the ligand
and the protein binding side. Grid points located within the protein (which should be avoided) are shown in purple. C-E) Grid points sampled
during a replica exchange simulation, coloured according to their position along the pathway. Panels C and D are showing grid points at
different levels of relative probability ðPr
i Þ per replica; with all visited grid points in Panel C, and only the often-visited grid points per replica in
Panel D. Panel E shows the derived peptide-binding pathway, with one peak maximum per replica. The surface of the protein is shown in
grey.
https://doi.org/10.1371/journal.pone.0180633.g010
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 24 / 30
volumes sampled for the replica exchange simulations were calculated based on a similar grid
with a mesh size identical to DF grids (0.2 nm) where the movements of the peptide virtual
atom were tracked. By recording which gridpoint was the closest to the peptide virtual atom
position at each frame (after every 5 ps of simulation), we could determine the probability density
of the given gridpoint (Pi). To find the conformations most often visited by the peptide,
the relative probability density ðPr
i Þ of the gridpoint i was calculated according to Eq 6:
Pr
i ¼
Pi À Pmin
Pmax À Pmin
Pi ¼
Oi
N
ð6Þ
Where Pmax and Pmin are the probabilities of visiting the least and most often visited gridpoints,
Oi is the number of times the gridpoint was visited and N is the number of simulation
frames used in the calculation (an example is shown in Fig 10).
Supporting information
S1 Table. DISICL secondary structure elements. Structure elements and abbreviations are
listed below. For a detailed description of the DISICL classes see [20].
(DOCX)
S2 Table. C-terminal tail distributions in DF/HRE-MD simulations. The table displays the
name of the simulation, the identity of the monomer (mon), the probability to find the C-terminal
tail near the primary binding site (IS1), secondary binding site (IS2), on the outer protein
surface (out), and free in solution (sol) over the entire length of the simulation, along the
total number of transitions between the listed sub-states (observed transitions) Note that the
phosphopeptide fragments p1h and p2t were binding to M1, whilst the p1t and p2h were binding
to M2.
(DOCX)
S3 Table. C-terminal tail distribution in unrestrained MD simulations. The table displays
the name of the simulation, the identity of the monomer (mon), the probability to find the Cterminal
tail near the primary binding site (IS1), secondary binding site (IS2), on the outer
protein surface (out), and free in solution (sol) over the entire length of the simulation, along
with the total number of transitions between the listed sub-states (Observed transitions) Note
that simulations were started with the M1 tail close to the outer surface and the M2 tail close to
the inner surface (near IS2). Dim-2 denotes dimeric 14-3-3z- simulation started from an alternative
set of starting coordinates.
(DOCX)
S1 Fig. Pathway visualization of the simulation tmon_p1H. A) The volume sampled by the
p1h phosphopeptide during the simulation is shown by dots around the 14-3-3z protein coloured
based on their position along the pathway. The most probable points in space to find the peptide
in for each replica (probability density peaks) are represented by larger spheres. B) Density peaks
and a few representative structures corresponding to the density peaks, coloured according to
their position along the pathway. Replica density peaks are connected by lines to visualize the
binding pathway. The 14-3-3z protein in panels A-B is shown as a surface representation, with
the two monomers shown in light and dark grey. C) Average interaction energy between 14-3-3z
and the p1h. D) Interaction map between any atom of the pulled p1h peptide and the amino
acids of the protein, summarized for each replica (only amino acids with at least 0.1 hydrogen
bond/salt bridge on average are shown). The scale indicates the number of interactions.
(TIF)
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 25 / 30
S2 Fig. Pathway visualization of the simulation tmon_p2H. A) The volume sampled by the
p2h phosphopeptide during the simulation is shown by dots around the 14-3-3z protein coloured
based on their position along the pathway. The most probable points in space to find the
peptide in for each replica (probability density peaks) are represented by larger spheres. B)
Density peaks and a few representative structures corresponding to the density peaks, coloured
according to their position along the pathway. Replica density peaks are connected by lines to
visualize the binding pathway. The 14-3-3z protein in panels A-B is shown as a surface representation,
with the two monomers shown in light and dark grey. C) Average interaction energy
between 14-3-3z and the p2h. D) Interaction map between any atom of the pulled p2h peptide
and the amino acids of the protein, summarized for each replica (only amino acids with at
least 0.1 hydrogen bond/salt bridge on average are shown). The scale indicates the number of
interactions.
(TIF)
S3 Fig. Pathway visualization of the simulation tmon_p1T. A) The volume sampled by the
p1t phosphopeptide during the simulation is shown by dots around the 14-3-3z protein coloured
based on their position along the pathway. The most probable points in space to find the
peptide in for each replica (probability density peaks) are represented by larger spheres. B)
Density peaks and a few representative structures corresponding to the density peaks, coloured
according to their position along the pathway. Replica density peaks are connected by lines to
visualize the binding pathway. The 14-3-3z protein in panels A-B is shown as a surface representation,
with the two monomers shown in light and dark grey. C) Average interaction energy
between 14-3-3z and the p1t. D) Interaction map between any atom of the pulled p1t peptide
and the amino acids of the protein, summarized for each replica (only amino acids with at
least 0.1 hydrogen bond/salt bridge on average are shown). The scale indicates the number of
interactions.
(TIF)
S4 Fig. Pathway visualization of the simulation tmon_p2T. A) The volume sampled by the
p2t phosphopeptide during the simulation is shown by dots around the 14-3-3z protein coloured
based on their position along the pathway. The most probable points in space to find the
peptide in for each replica (probability density peaks) are represented by larger spheres. B)
Density peaks and a few representative structures corresponding to the density peaks, coloured
according to their position along the pathway. Replica density peaks are connected by lines to
visualize the binding pathway. The 14-3-3z protein in panels A-B is shown as a surface representation,
with the two monomers shown in light and dark grey. C) Average interaction energy
between 14-3-3z and the p2t. D) Interaction map between any atom of the pulled p2t peptide
and the amino acids of the protein, summarized for each replica (only amino acids with at
least 0.1 hydrogen bond/salt bridge on average are shown). The scale indicates the number of
interactions.
(TIF)
S5 Fig. Pathway visualization of the simulation dim_p1Ht. A) The volume sampled by the
p1h phosphopeptide during the simulation is shown by dots around the 14-3-3z protein coloured
based on their position along the pathway. The most probable points in space to find the
peptide in for each replica (probability density peaks) are represented by larger spheres. B)
Density peaks and a few representative structures corresponding to the density peaks, coloured
according to their position along the pathway. Replica density peaks are connected by lines to
visualize the binding pathway. The 14-3-3z protein in panels A-B is shown as a surface representation,
with the two monomers shown in light and dark grey. C) Average interaction energy
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 26 / 30
between 14-3-3z and the p1h. D) Interaction map between any atom of the pulled p1h peptide
and the amino acids of the protein, summarized for each replica (only amino acids with at
least 0.1 hydrogen bond/salt bridge on average are shown). The scale indicates the number of
interactions.
(TIF)
S6 Fig. Pathway visualization of the simulation dim_p2hT. A) The volume sampled by the p2t
phosphopeptide during the simulation is shown by dots around the 14-3-3z protein coloured
based on their position along the pathway. The most probable points in space to find the peptide
in for each replica (probability density peaks) are represented by larger spheres. B) Density peaks
and a few representative structures corresponding to the density peaks, coloured according to
their position along the pathway. Replica density peaks are connected by lines to visualize the
binding pathway. The 14-3-3z protein in panels A-B is shown as a surface representation, with
the two monomers shown in light and dark grey. C) Average interaction energy between 14-3-
3z and the p2t. D) Interaction map between any atom of the pulled p2t peptide and the amino
acids of the protein, summarized for each replica (only amino acids with at least 0.1 hydrogen
bond/salt bridge on average are shown). The scale indicates the number of interactions.
(TIF)
S7 Fig. Pathway visualization of the simulation dim_p2Ht. A) The volume sampled by the
p2h phosphopeptide during the simulation is shown by dots around the 14-3-3z protein coloured
based on their position along the pathway. The most probable points in space to find
the peptide in for each replica (probability density peaks) are represented by larger spheres. B)
Density peaks and a few representative structures corresponding to the density peaks, coloured
according to their position along the pathway. Replica density peaks are connected by lines to
visualize the binding pathway. The 14-3-3z protein in panels A-B is shown as a surface representation,
with the two monomers shown in light and dark grey. C) Average interaction energy
between 14-3-3z and the p2h. D) Interaction map between any atom of the pulled p2h peptide
and the amino acids of the protein, summarized for each replica (only amino acids with at
least 0.1 hydrogen bond/salt bridge on average are shown). The scale indicates the number of
interactions. Note that dim_p2Ht did not follow the general pathway observed for other DF/
HRE-MD simulations, and IS2 residues were not significant interaction partners.
(TIF)
S8 Fig. Interaction map between the perturbed phosphopeptides and amino acids of 14-3-
3z. The map shows an average over the 8 HRE-MD simulations, where the average number of
hydrogen bonds observed during the simulations are shown as the function of replica number.
Replica 1–3 refers to the bound state in interaction site1 (IS1), and larger replica numbers correspond
to higher DF distance from IS1.
(TIF)
S9 Fig. The salt concentration dependence of the 14-3-3z inter-monomer twist. A) The
drift of the monomer twist angle during MD simulations. B) Atom-positional root-meansquare
deviation (RMSD) from the original conformation (derived from the crystal structure)
during the various simulations. The structural deviation calculated for the individual monomers
of simulation X are marked as X_M1 and X_M2, respectively.
(TIF)
S10 Fig. 14-3-3z secondary structure analysis. DISICL backbone analysis shows a stable secondary
structure during the MD simulation dim. The left side of the Figure shows a schematic
representation of the 14-3-3z helices. The most populated DISICL classes are depicted in the
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 27 / 30
following colours: α-helix (ALH)–green, π-helix (PIH)–cyan, helix-cap (HC)–blue, turn-cap
(TC)–black, polyproline-like (PP)–brown, turn type I (TI)–magenta, Turn type II (TII)–purple.
For a description of abbreviations of DISICL secondary structure elements see S1 Table.
(TIF)
S11 Fig. Pathway dependence of phosphopeptide secondary structure. The population of
secondary structure elements is dependent on the replica ID (and DF distance) within the dim
DF/HRE-MD simulations. The change of the backbone secondary structure content was analysed
by the DISICL algorithm. For a description of abbreviations of DISICL secondary structure
elements see S1 Table.
(TIF)
S12 Fig. Pathway dependence of phosphopeptide secondary structure. The population of
secondary structure elements is dependent on the replica ID (and DF distance) within the
tmon DF/HRE-MD simulations. The change of the backbone secondary structure content was
analysed by the DISICL algorithm. For a description of abbreviations of DISICL secondary
structure elements see S1 Table.
(TIF)
S13 Fig. The effect of phosphorylation of Ser58 on the electrostatic surface potential (ESP)
of 14-3-3z. The IS1 and IS2 in A) 14-3-3z WT are connected by a mildly positive surface
potential patch in contrary to B) 14-3-3z phosphorylated at S58, where the connection is disrupted
by the negatively charged phosphoserine. Blue, white and red colour represents the positive,
neutral, and negative surface patches, respectively. For the clarity, the 14-3-3z C-terminal
tails are not shown.
(TIF)
S14 Fig. SAXS comparison. Small angle X-ray scattering calculated from the crystal structure
(cryst, pdb code 2WHO), the IS1 and UnB stages dim_p1hT, and the unrestrained MD simulations
with a low intermonomer twist angles (dim_0.15 dim_p1ht_0.15). The_SAXS curves
were calculated using the software Crysol (1), and compared to the experimental SAXS curve
(exp) after ensemble averaging. The full length 14-3-3z protein sample used for SAXS measurements
was expressed and purified as described in Hritz et al.[4]. SAXS data were collected
on the beamline BM29 BioSAXS ESFR in Grenoble, France. The concentrations of the 14-3-
3-WT during the measurement were 1.18; 2.35 and 4.70 mg/ml in the 50 mM Tris buffer,
pH = 8.0. The data were recorded at 20.12˚C using the pixel 1M PILATUS detector at a sample-detector
distance of 2.867 m, and a wavelength (λ) of 0.099 nm, covering the range of
momentum transfer 0.025 nm−1
< s < 5 nm−1
(s = 4π sin(θ)/λ, where 2θ is the scattering
angle). No radiation damage was observed during the data collection.The data were processed
using standard procedures with the PRIMUS software (2). Solvent contributions (buffer backgrounds
collected before and after the protein sample) were averaged and subtracted from the
associated protein sample. A slight concentration dependency was noticeable. Therefore, the
scattering curves collected at different concentrations were used to obtain a final zero concentration
scattering curve through extrapolation according to the guidelines provided by (3).
(TIF)
Acknowledgments
We kindly acknowledge Dr. Anita de Ruiter, for support on the calculation of WHAM profiles,
Vojteˇch Zapletal for SAXS measurements, Erik Zˇupa for help with 14-3-3z sample purification
and Jamie McDonald for critical reading of the manuscript.
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 28 / 30
Author Contributions
Conceptualization: Gabor Nagy, Chris Oostenbrink, Jozef Hritz.
Data curation: Gabor Nagy.
Formal analysis: Gabor Nagy.
Funding acquisition: Chris Oostenbrink, Jozef Hritz.
Investigation: Gabor Nagy.
Methodology: Gabor Nagy, Chris Oostenbrink, Jozef Hritz.
Project administration: Jozef Hritz.
Resources: Jozef Hritz.
Supervision: Jozef Hritz.
Validation: Gabor Nagy, Jozef Hritz.
Visualization: Gabor Nagy.
Writing – original draft: Gabor Nagy, Chris Oostenbrink, Jozef Hritz.
Writing – review & editing: Gabor Nagy, Chris Oostenbrink, Jozef Hritz.
References
1. Gardino AK, Smerdon SJ, Yaffe MB. Structural determinants of 14-3-3 binding specificities and regulation
of subcellular localization of 14-3-3-ligand complexes: A comparison of the X-ray crystal structures
of all human 14-3-3 isoforms. Semin Cancer Biol. 2006; 16: 173–182. https://doi.org/10.1016/j.
semcancer.2006.03.007 PMID: 16678437
2. Yaffe MB, Rittinger K, Volinia S, Caron PR, Aitken A, Leffers H, et al. The Structural Basis for 14-3-3:
Phosphopeptide Binding Specificity. Cell. 1997; 91: 961–971. https://doi.org/10.1016/S0092-8674(00)
80487-0 PMID: 9428519
3. Johnson C, Crowther S, Stafford MJ, Campbell DG, Toth R, MacKintosh C. Bioinformatic and experimental
survey of 14-3-3-binding sites. Biochem J. 2010; 427: 69–78. https://doi.org/10.1042/
BJ20091834 PMID: 20141511
4. Hritz J, Byeon I-JL, Krzysiak T, Martinez A, Sklenar V, Gronenborn AM. Dissection of Binding between
a Phosphorylated Tyrosine Hydroxylase Peptide and 14-3-3 zeta: A Complex Story Elucidated by
NMR. Biophys J. 2014; 107: 2185–2194. https://doi.org/10.1016/j.bpj.2014.08.039 PMID: 25418103
5. Yang X, Lee WH, Sobott F, Papagrigoriou E, Robinson CV, Grossmann JG, et al. Structural basis for
protein—protein interactions in the 14-3-3 protein family. Proceedings of the National Academy of Sciences.
2006; 103: 17237–17242.
6. de Ruiter A, Oostenbrink C. Protein-Ligand Binding from Distancefield Distances and Hamiltonian Replica
Exchange Simulations. J Chem Theory Comput. 2013; 9: 883–892. https://doi.org/10.1021/
ct300967a PMID: 26588732
7. Oostenbrink C, de Ruiter A, Hritz J, Vermeulen N. Malleability and Versatility of Cytochrome P450
Active Sites Studied by Molecular Simulations. Curr Drug Metab. 2012; 13: 190–196. PMID: 22208533
8. Kostelecky B, Saurin AT, Purkiss A, Parker PJ, McDonald NQ. Recognition of an intra-chain tandem
14-3-3 binding site within PKCε. EMBO reports. 2009; 10: 983–989. https://doi.org/10.1038/embor.
2009.150 PMID: 19662078
9. Molzan M, Kasper S, Roeglin L, Skwarczynska M, Sassa T, Inoue T, et al. Stabilization of Physical
RAF/14-3-3 Interaction by Cotylenin A as Treatment Strategy for RAS Mutant Cancers. ACS Chem
Biol. 2013; 8: 1869–1875. https://doi.org/10.1021/cb4003464 PMID: 23808890
10. Muslin AJ, Tanner JW, Allen PM, Shaw AS. Interaction of 14-3-3 with signaling proteins is mediated by
the recognition of phosphoserine. Cell. 1996; 84: 889–897. PMID: 8601312
11. Truong AB, Masters SC, Yang HZ, Fu HA. Role of the 14-3-3 C-terminal loop in ligand interaction. Proteins.
2002; 49: 321–325. https://doi.org/10.1002/prot.10210 PMID: 12360521
12. Williams DM, Ecroyd H, Goodwin KL, Dai H, Fu H, Woodcock JM, et al. NMR spectroscopy of 14-3-3ζ
reveals a flexible C-terminal extension: differentiation of the chaperone and phosphoserine-binding
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 29 / 30
activities of 14-3-3ζ. Biochemical Journal. 2011; 437: 493–503. https://doi.org/10.1042/BJ20102178
PMID: 21554249
13. Hu G, Li H, Liu J-Y, Wang J. Insight into Conformational Change for 14-3-3σ Protein by Molecular
Dynamics Simulation. International Journal of Molecular Sciences. 2014; 15: 2794–2810. https://doi.
org/10.3390/ijms15022794 PMID: 24552877
14. Sluchanko NN, Gusev NB. Oligomeric structure of 14-3-3 protein: What do we know about monomers?
FEBS Letters. 2012; 586: 4249–4256. https://doi.org/10.1016/j.febslet.2012.10.048 PMID: 23159940
15. Molzan M, Ottmann C. Synergistic Binding of the Phosphorylated S233- and S259-Binding Sites of CRAF
to One 14-3-3ζ Dimer. Journal of Molecular Biology. 2012; 423: 486–495. https://doi.org/10.1016/j.
jmb.2012.08.009 PMID: 22922483
16. Schmid N, Christ CD, Christen M, Eichenberger AP, van Gunsteren WF. Architecture, implementation
and parallelisation of the GROMOS software for biomolecular simulation. Comput Phys Commun.
2012; 183: 890–903. https://doi.org/10.1016/j.cpc.2011.12.014
17. Eichenberger AP, Allison JR, Dolenc J, Geerke DP, Horta BAC, Meier K, et al. GROMOS plus plus Software
for the Analysis of Biomolecular Simulation Trajectories. J Chem Theory Comput. 2011; 7: 3379–
3390. https://doi.org/10.1021/ct2003622 PMID: 26598168
18. Schro¨dinger, LLC. The PyMOL Molecular Graphics System.
19. Baker NA, Sept D, Joseph S, Holst MJ, McCammon JA. Electrostatics of nanosystems: application to
microtubules and the ribosome. Proceedings of the National Academy of Sciences. 2001; 98: 10037–
10041.
20. Nagy G, Oostenbrink C. Dihedral-Based Segment Identification and Classification of Biopolymers I:
Proteins. J Chem Inf Model. 2014; 54: 266–277. https://doi.org/10.1021/ci400541d PMID: 24364820
21. Sali A, Blundell T. Comparative Protein Modeling by Satisfaction of Spatial Restraints. J Mol Biol. 1993;
234: 779–815. https://doi.org/10.1006/jmbi.1993.1626 PMID: 8254673
22. Schmid N, Eichenberger AP, Choutko A, Riniker S, Winger M, Mark AE, et al. Definition and testing of
the GROMOS force-field versions 54A7 and 54B7. European Biophysics Journal. 2011; 40: 843–856.
https://doi.org/10.1007/s00249-011-0700-9 PMID: 21533652
23. Petrov D, Margreitter C, Grandits M, Oostenbrink C, Zagrovic B. A Systematic Framework for Molecular
Dynamics Simulations of Protein Post-Translational Modifications. Wei G, editor. PLoS Computational
Biology. 2013; 9: e1003154. https://doi.org/10.1371/journal.pcbi.1003154 PMID: 23874192
24. Berendsen HJC, Postma JPM, Van Gunsteren WF, Hermans J. Interaction models for water in relation
to protein hydration. Intermolecular Forces. 1981; 11: 331–338.
25. Ryckaert J, Ciccotti G, Berendsen H. Numerical-Integration of Cartesian Equations of Motion of a System
with Constraints—Molecular-Dynamics of N-Alkanes. J Comput Phys. 1977; 23: 327–341. https://
doi.org/10.1016/0021-9991(77)90098-5
26. Berendsen H, Postma J, Van Gunsteren W, Dinola A, Haak J. Molecular-Dynamics with Coupling to an
External Bath. J Chem Phys. 1984; 81: 3684–3690. https://doi.org/10.1063/1.448118
27. Tironi I, Sperb R, Smith P, Van Gunsteren W. A generalized reaction field method for molecular-dynamics
simulations. J Chem Phys. 1995; 102: 5451–5459. https://doi.org/10.1063/1.469273
28. Heinz T, van Gunsteren W, Hunenberger P. Comparison of four methods to compute the dielectric permittivity
of liquids from molecular dynamics simulations. J Chem Phys. 2001; 115: 1125–1136. https://
doi.org/10.1063/1.1379764
29. Amadei A, Chillemi G, Ceruso MA, Grottesi A, Di Nola A. Molecular dynamics simulations with constrained
roto-translational motions: Theoretical basis and statistical mechanical consistency. J Chem
Phys. 2000; 112: 9–23. https://doi.org/10.1063/1.480557
30. Sugita Y, Kitao A, Okamoto Y. Multidimensional replica-exchange method for free-energy calculations.
J Chem Phys. 2000; 113: 6042–6051. https://doi.org/10.1063/1.1308516
31. Kumar S, Rosenberg JM, Bouzida D, Swendsen RH, Kollman PA. The weighted histogram analysis
method for free-energy calculations on biomolecules. I. The method. Journal of computational chemistry.
1992; 13: 1011–1021.
32. Doudou S, Burton NA, Henchman RH. Standard free energy of binding from a one-dimensional potential
of mean force. J Chem Theory Comput. 2009; 5: 909–918. https://doi.org/10.1021/ct8002354 PMID:
26609600
The binding pathways of the 14-3-3ζ protein
PLOS ONE | https://doi.org/10.1371/journal.pone.0180633 July 20, 2017 30 / 30
ALCHEMICAL (RELATIVE) FREE ENERGY CALCULATIONS
47
5 Alchemical (relative) Free energy calculations
Paper 8: Hritz, J.; Lappchen T.; Oostenbrink, C. Calculations of binding affinity
between C8-substituted GTP analogs and the bacterial cell-division protein FtsZ.
Eur.Biophys. J. 2010, 39, 1573-1580
Paper 9: Hritz, J.; Oostenbrink, C. Efficient free energy calculations for compounds
with multiple stable conformations separated by high energy barriers.
J. Phys. Chem. B 2009, 113, 12711-12720
Paper 10: Jandova, Z.; Trosanova, Z.; Weisova, V.; Oostenbrink, C.*, Hritz, J.*:Free energy
calculations on the stability of the 14-3-3 protein. BBA - Proteins and Proteomics,
2018, 1866, 442-450.
The previous chapter describes the challenges of achieving sufficient sampling of conformational
space. Applications also included calculations of binding free energies by
applying H-REMD using distance-field distance restraints. Although these approaches
enhance sampling along the binding pathways by several orders of magnitude with respect
to traditional MD – they are computationally still very demanding.
However, in many cases binding free energy differences between the similar type of
ligands need to be calculated, i.e. relative rather than absolute binding affinities. This
is a typical scenario faced by pharma companies where there is a lead compound regarding
a particular protein target and in the following steps, the chemical substitutions
are sought that should lead to an increase in binding affinity. For this type of case,
instead of calculating individual (absolute) binding affinities, it is possible to calculate
the difference between them by calculating the alchemical free energy difference in the
following way:
The ∆𝐆 𝐘𝐗 in the free or bound state means the free energy difference between two
chemically different compounds in the corresponding environment. Rarely, can it be
calculated from a single simulation of one of two compounds. More often it is calculated
∆∆𝐺 𝑏𝑖𝑛𝑑 = ∆𝐺 𝑏𝑖𝑛𝑑(𝑌) − ∆𝐺 𝑏𝑖𝑛𝑑(𝑋) = ∆𝐺 𝑌𝑋(𝑓𝑟𝑒𝑒) − ∆𝐺 𝑌𝑋(𝑏𝑜𝑢𝑛𝑑) (5.1)
ALCHEMICAL (RELATIVE) FREE ENERGY CALCULATIONS
48
Figure 6: Thermodynamic cycle used for the computation of the difference in binding free energies
between compound X and Y into the same protein target (P).
through several intermediate steps –e.g. by alchemical change/perturbation of one that
is a direct consequence of thermodynamic cycle in Fig. 6.28
It is usually done by a set of MD simulations at different Hamiltonians, defined as, e.g.:
H(λi) = (1- λi)HX + λiHY controlled by the λi parameter, linking the Hamiltonians HX representing
compound X and HY representing compound Y. The free energy difference is
then calculated by the free energy perturbation formula:29
where kBT is the Boltzmann constant, multiplied by the absolute temperature. The bottleneck
of traditional free energy calculation methods is achieving sufficient conformational
sampling for systems with high energy barriers.
5.1 Efficient calculations of relative free energies
High energy barriers within biomolecules restrict the efficiency of MD conformational
sampling and naturally also prohibit the use of traditional free energy calculation
∆𝐺 𝑌𝑋 = ∑ ∆𝐺𝜆𝑖
1
𝜆𝑖=0
; ∆𝐺𝜆𝑖
= −𝑘 𝐵 𝑇 ln 〈𝑒
−
𝐻(𝜆𝑖)−𝐻(𝜆𝑖+1)
𝑘 𝐵 𝑇 〉 𝐻(𝜆𝑖)
(5.2)
ALCHEMICAL (RELATIVE) FREE ENERGY CALCULATIONS
49
schemes such as alchemical free energy calculation.30–32 The applicant developed the
enhanced sampling – one-step perturbation method (ES-OS) to tackle the calculation
of free energy differences between similar C8-substituted GTP analogs in a highly efficient
way.P8 A single MD simulation of a judiciously chosen reference state (using
two sets of soft-core interactions) was sufficient to determine conformational
distributions of chemically similar compounds and the free energy differences
between them. The ES-OS method was applied to a set of five biologically relevant 8substituted
GTP analogs having high energy barriers between the anti and syn conformations
of the guanine with respect to the ribose part.P9 The reliability of ES-OS was
verified by comparing the results to Hamiltonian replica exchange simulations of GTP
and 8-Br-GTP and the experimentally determined 3J(C4,H1’) coupling constant for GMP
in water. Additional simulations in vacuum, octanol and in the binding site of the FtsZ
protein allowed us to calculate differences in the solvation free energies, lipophilicities
(log P) and relative binding affinities for FtsZ.P8,P9 The applicant also derived the theoretical
basis describing how free energy contributions from individual conformational
regions could be calculated and have their relationship with the overall
free energy derived, leading to a set of multi-conformational free energy formulas.P9
The equation for the solvation free energy, i.e. relative free energy difference between
compounds A and B in aqueous (aq) with respect to the vacuum (v) environment
was derived as:
where [i] stands for the population of particular conformation state of particular compound
(e.g. A) in a given environment. For 8-substituted GTP, the conformational states
can be [anti] and [syn]. The usefulness of this equation for the community lies in the
fact that when faced with compounds with a high intramolecular energy barrier one
does not have to cross it a sufficient number of times in the free energy perturbation
∆∆𝐺 𝐴𝐵(𝑠𝑜𝑙𝑣) = −𝑘 𝐵 𝑇 ln (∑
[𝑖](𝑎𝑞)
𝐴
[𝑖](𝑣)
𝐵
[𝑖](𝑣)
𝐴 𝑒
−
∆∆𝐺 𝐴𝐵
𝑖 (𝑠𝑜𝑙𝑣)
𝑘 𝐵 𝑇
𝑛
𝑖
)
(5.3)
ALCHEMICAL (RELATIVE) FREE ENERGY CALCULATIONS
50
calculations. The calculations of ∆∆𝐺 𝐴𝐵
𝑖
(𝑠𝑜𝑙𝑣) for individual conformational states has
fast convergence because there is no need to cross the high energy barrier.
The calculation of relative free energy differences by alchemical perturbation has general
validity and thus can be applied not only for the presented analogs of organic compounds
but also for the predicting of the impact of site-directed mutagenesis on the
overall thermodynamics of the protein. We presented such practical application for the
particular mutations of 14-3-3 protein.P10 The predicted thermodynamic stability
changes were also compared with the experimental data. P10
ALCHEMICAL (RELATIVE) FREE ENERGY CALCULATIONS
51
Paper 8
Hritz, J.; Lappchen T.; Oostenbrink, C. Calculations of binding affinity between C8-substituted
GTP analogs and the bacterial cell-division protein FtsZ.
Eur.Biophys. J. 2010, 39, 1573-1580
ORIGINAL PAPER
Calculations of binding affinity between C8-substituted GTP
analogs and the bacterial cell-division protein FtsZ
Jozef Hritz • Tilman La¨ppchen • Chris Oostenbrink
Received: 1 April 2010 / Revised: 21 May 2010 / Accepted: 26 May 2010 / Published online: 18 June 2010
Ó The Author(s) 2010. This article is published with open access at Springerlink.com
Abstract The FtsZ protein is a self-polymerizing GTPase
that plays a central role in bacterial cell division. Several
C8-substituted GTP analogs are known to inhibit the polymerization
of FtsZ by competing for the same binding site as
its endogenous activating ligand GTP. Free energy calculations
of the relative binding affinities to FtsZ for a set of
five C8-substituted GTP analogs were performed. The calculated
values agree well with the available experimental
data, and the main contribution to the free energy differences
is determined to be the conformational restriction of
the ligands. The dihedral angle distributions around the
glycosidic bond of these compounds in water are known to
vary considerably depending on the physicochemical
properties of the substituent at C8. However, within the FtsZ
protein, this substitution has a negligible influence on the
dihedral angle distributions, which fall within the narrow
range of -140° to -90° for all investigated compounds. The
corresponding ensemble average of the coupling constants
3
J(C4,H10
) is calculated to be 2.95 ± 0.1 Hz. The contribution
of the conformational selection of the GTP analogs
upon binding was quantified from the corresponding populations.
The obtained restraining free energy values follow
the same trend as the relative binding affinities to FtsZ,
indicating their dominant contribution.
Keywords GTP analogs Á FtsZ Á
One-step free energy perturbation Á
Conformational selection Á Restraining free energy Á
Ensemble average Á Heteronuclear coupling constant
Introduction
The cell division protein FtsZ is considered a promising
antibacterial target (Vollmer 2006; Huang et al. 2007;
Paradis-Bleau et al. 2007; Lock and Harry 2008; Kapoor
and Panda 2009), and the recent discovery of a small
synthetic FtsZ inhibitor with potent in vitro and in vivo
bactericidal activity against multidrug-resistant Staphylococcus
aureus suggests that these high expectations are
justified (Haydon et al. 2008; Czaplewski et al. 2009).
In the presence of guanosine 50
-triphosphate (GTP),
FtsZ assembles into a variety of polymeric structures, the
nature of which is very much dependent on the exact
experimental conditions employed (reviewed by Adams
and Errington 2009). Linear protofilaments and protofilament
bundles arising from lateral association are among the
more frequently studied polymeric species of FtsZ. Polymerization
of FtsZ activates its GTPase activity by insertion
of acidic residues from the synergy loop into the
nucleotide binding pocket of the preceding monomer in the
protofilament (Oliva et al. 2004). Despite considerable
J. Hritz Á C. Oostenbrink (&)
Leiden-Amsterdam Center for Drug Research,
Section of Molecular Toxicology,
Department of Chemistry and Pharmacochemistry,
Vrije Universiteit, De Boelelaan 1083,
1081 HV Amsterdam, The Netherlands
e-mail: c.oostenbrink@few.vu.nl
J. Hritz
e-mail: jozef.hritz@gmail.com
T. La¨ppchen
Department of Biomolecular Engineering,
Philips Research, High Tech Campus 11,
M/S WBC02 P263, 5656 AE Eindhoven, The Netherlands
e-mail: tilman.lappchen@philips.com
C. Oostenbrink
Institute of Molecular Modeling and Simulation,
University of Natural Resources and Applied Life Sciences,
Muthgasse 18, 1190 Vienna, Austria
123
Eur Biophys J (2010) 39:1573–1580
DOI 10.1007/s00249-010-0614-y
efforts, FtsZ polymer dynamics, the associated GTPase
reaction kinetics, and the modulation of both by pH, nature
and concentration of cations, GTP, guanosine 50
-diphosphate
(GDP), FtsZ, and certainly regulation by accessory
proteins is still not fully understood at a molecular level
(Lo¨we and Amos 1998; Michie and Lowe 2006; Mendieta
et al. 2009). In particular, the relationship between GTP
hydrolysis and FtsZ polymer dynamics remains controversial.
While earlier studies suggested direct exchange of
nucleotide in protofilaments (Romberg and Mitchison
2004; Tadros et al. 2006), recent data show that terminal
FtsZ subunit exchange is independent of nucleotide state
and faster than GTP hydrolysis, supporting the hypothesis
that nucleotide exchange occurs only on recycling terminal
subunits (Chen and Erickson 2009). In addition, the previously
accepted view that the GTP-bound form of the FtsZ
protofilament is intrinsically straight while the GDP-bound
form is curved has recently been challenged by the finding
that FtsZ structures in various crystal forms and nucleotide
states did not show evidence of a conformational switch
in the FtsZ monomer involving domain movement (Oliva
et al. 2007), although it should be taken into account that
the crystal structures might not be representative for the
GTP- or GDP-bound state but in fact could correspond to a
transition state. Strikingly, even 8-morpholino-GTP, one of
a series of C8-substituted GTP analogs acting as competitive
inhibitors of GTP-driven FtsZ polymerization and
GTP hydrolysis, was found to bind to Aquifex aeolicus
FtsZ in essentially the same way as GDP without inducing
any significant conformational changes in the protein
(La¨ppchen et al. 2008). Until now, the molecular basis of
the observed inhibitory action of the investigated
C8-substituted GTP derivatives has not been completely
resolved. Although the C8-morpholino substituent protrudes
from the surface of the monomer, the currently
available FtsZ protofilament structures (Oliva et al. 2004,
2007) suggest that the inhibitory action cannot be simply
attributed to direct steric clashes between the C8 substituent
and the next FtsZ monomer in a growing protofilament.
It is important to note, however, that stabilization
of intersubunit contacts and the rate of GTPase activity are
also dependent on the presence of divalent and monovalent
cations and pH (Mendieta et al. 2009), suggesting that the
C8-substituted GTP derivatives might act by interfering
with vital hydrogen-bonding interactions via rearrangement
of water molecules and cations in the active site.
In a series of C8-substituted GTP analogs, inhibitory
potencies were found to correlate with the corresponding
binding affinities to the FtsZ monomer and with the
Sterimol parameters of their C8 substituents (La¨ppchen
et al. 2008). Intrigued by this observation, we set out to
rationalize these results in terms of binding free energies.
C8-substituted GTP analogs with two stable conformations
(anti, syn) separated by high energy barriers belong to a
challenging class of compounds for binding affinity calculations
(Hritz and Oostenbrink 2007, 2008). Recently we
have developed the enhanced sampling one-step free
energy perturbation method (ES-OS) that allows for efficient
free energy calculations for GTP analogs in explicit
solvent based on sufficient sampling of both relevant
conformations (Hritz and Oostenbrink 2009).
This paper presents calculations of relative free energies
of binding to the FtsZ protein for a set of five C8-substituted
GTP analogs in which H8 is replaced by halogen
atoms or a methyl group (Fig. 1). Molecular docking
simulations of the C8-substituted GTP analogs well
reproduced the phosphate and ribose groups of the molecules,
while the base was found to be in both the syn and
the anti conformation (La¨ppchen 2007). The recent highresolution
crystal structure of the Aquifex aeolicus FtsZ
protein, co-crystallized with 8-morpholino-GTP, shows
that also compounds with a bulky substituent at the C8
position bind to the protein in the anti conformation
(La¨ppchen et al. 2008), even though in solution the syn
conformation is expected to be dominant (Davies 1978;
Stolarski et al. 1984; Cho and Evans 1991). This observation
significantly simplifies the free energy calculations
of the compounds in the binding site of the FtsZ protein,
because the simulation can be restricted to a single ligand
conformation and no high energy barriers need to be
crossed. For this reason we calculate the free energy difference
between the various compounds bound to the FtsZ
protein using the one-step (OS) perturbation method (Liu
et al. 1996; Oostenbrink and van Gunsteren 2005). The
corresponding values in solution (where both anti and syn
conformations contribute) were calculated earlier using
enhanced sampling OS (ES-OS) (Hritz and Oostenbrink
2009).
The computationally predicted values are compared
with the available experimental binding affinities to
nucleotide-free Methanococcus jannaschii FtsZ protein
(La¨ppchen et al. 2008). The main contributions to the
Fig. 1 Structure of C8-substituted analogs of GTP in syn conformation,
where X = H, F, Cl, Br, CH3. Conformational transitions
between the syn and anti conformations occur by rotation around the
glycosidic bond indicated by the arrow. The glycosidic dihedral angle
(v) is defined over atoms: C4-N9-C10
-O40
1574 Eur Biophys J (2010) 39:1573–1580
123
relative binding free energies are analyzed and provide an
explanation for the empirically observed correlation
between Sterimol parameters of C8 substituents and binding
affinities (La¨ppchen et al. 2008).
Materials and methods
One-step free energy perturbation (OS)
(Liu et al. 1996)
The aim of OS is to efficiently determine the free energy
differences between chemically similar compounds from a
single molecular dynamics (MD) simulation of a designed
reference compound, S. The free energy between the
reference compound (S) and the real compounds (R),
can be calculated from a simulation using the reference
Hamiltonian and applying Zwanzig’s perturbation formula
(Zwanzig 1954):
DGSR ¼ GR À GS ¼ ÀkBT ln e
À HRðq;pÞÀHSðq;pÞð Þ
kBT
( )
S
ð1Þ
where kB is the Boltzmann constant, T is absolute temperature,
and HS and HR are the Hamiltonians for the (soft)
reference compound and one of the real compounds,
respectively. The angular brackets indicate an ensemble
average over the positions, q, and momenta, p, obtained
from a simulation of the reference state.
The ensemble average Ah iR
of a property A for the real
compound (R) can be estimated by reweighting the individual
values corresponding to the particular configuration
(qi, pi) for the reference compound, AS
i
by a Boltzmann
factor pSR
i :
Ah iR
¼
X
i
pSR
i AS
i ; with pSR
i ¼
e
À HRðqi;piÞÀHSðqi;piÞð Þ
kBT
P
i
e
À HRðqi;piÞÀHSðqi;piÞð Þ
kBT
: ð2Þ
Simulation setup
The crystal structure of Aquifex aeolicus FtsZ complexed
with 8-morpholino-GTP (La¨ppchen et al. 2008) was
downloaded from the protein databank (Berman et al.
2003) (www.pdb.org; PDB ID: 2R75, chain B). The morpholino
substituent was replaced by a single bromine (Br)
atom. Molecular dynamics (MD) simulations of the FtsZ
protein in complex with 8-Br-GTP were performed using
the GROMOS05 simulation package (Christen et al. 2005)
in combination with the GROMOS 53A6 force field
(Oostenbrink et al. 2004). Force field parameters for five
C8-substituted GTP analogs are available in the supplementary
material of (Hritz and Oostenbrink 2009). The
magnesium cation and all 283 crystallographic water
oxygens were kept and position restrained during initial
equilibration steps.
Rectangular periodic boundary conditions were used
with an additional 24,544 water molecules; 13 of them
were replaced by 13 sodium cations in order to electroneutralize
the whole system (note: crystallographic waters
were not considered for replacement by sodium cations).
The system finally contained 24,814 explicit simple point
charge (SPC) water molecules (Berendsen et al. 1981). All
bonds were constrained, using the SHAKE algorithm
(Ryckaert et al. 1977), with relative geometric accuracy of
10-4
, allowing for a time step of 2 fs in the leapfrog
integration scheme (Hockney 1970). After a steepestdescent
minimization to remove bad contacts between
molecules, initial velocities were randomly assigned from a
Maxwell–Boltzmann distribution at 298 K, according to
the atomic masses. The temperature was kept constant
using weak coupling (Berendsen et al. 1984) to a bath of
298 K with a relaxation time of 0.1 ps. The solute molecule
and solvent were independently coupled to the heat
bath. The pressure was controlled using isotropic weak
coupling to atmospheric pressure (Berendsen et al. 1984)
with a relaxation time of 0.5 ps. van der Waals and electrostatic
interactions were calculated using a triple range
cutoff scheme. Interactions within a short-range cutoff of
0.8 nm were calculated every time step from a pair list that
was generated every five steps. At these time points,
interactions between 0.8 and 1.4 nm were also calculated
and kept constant between updates. A reaction-field contribution
was added to the electrostatic interactions and
forces to account for a homogeneous medium outside the
long-range cutoff, using the relative permittivity (61) of
SPC water (Tironi et al. 1995). Selected interactions were
calculated using a soft-core van der Waals and electrostatic
interaction between atoms i and j (Beutler et al. 1994):
Evdw
ij ðrij; kvdwÞ ¼
C12ij
AijðkvdwÞ þ r6
ij
À C6ij
!
1
AijðkvdwÞ þ r6
ij
;
ð3Þ
Eel
ij ðrij; kelÞ ¼
qiqj
4pe
1
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
BijðkelÞ þ r2
ij
q ; ð4Þ
with rij being the interatomic distance; AijðkvdwÞ ¼
avdw
C12ij
C6ij
k2
vdw and BijðkelÞ ¼ aelk2
el. C12ij and C6ij are the
Lennard–Jones parameters for atom pair i and j, qi and
qj are the partial charges of particles i and j, and avdw and
ael are the softness constants. In the current study we
used in all simulations avdwk2
vdw ¼ aelk2
el ¼ 0:3775, the value
empirically known to work well in standard OS (Scha¨fer
et al. 1999; Oostenbrink and van Gunsteren 2004). It can
be seen that at longer distances [rij ) A(kvdw) and
rij ) B(kel)] the soft-core interaction approximates the
Eur Biophys J (2010) 39:1573–1580 1575
123
interaction for normal atoms and that they differ mostly at
short distances between the atoms [rij B A(kvdw) or
rij B B(kel)]. The conformational space of the C8-substituted
GTP analogs was adequately represented within the
FtsZ protein by the reference state 8-soft_Br-GTP. Here, a
bromine substituent was placed at position 8, for which all
nonbonded interactions with the rest of the system
(including protein and solvent) were evaluated as soft-core
interactions (Eq. 3, 4). A single MD simulation of the
reference state FtsZ:8-soft_Br-GTP was performed for
2 ns, and system coordinates were saved every 0.2 ps.
Results
Figure 2 presents the normalized dihedral angle (v) distributions
for the five C8-substituted analogs of GTP in
complex with the FtsZ protein as calculated by reweighting
the probabilities of individual configurations of the MD
trajectory of the reference state (8-soft_Br-GTP) using Eq.
(2). The distributions indicate that all five studied GTP
analogs occupy a very similar conformational range,
v [ [-140°, -90°], when bound to FtsZ. This range is
only about half of the anti conformational range observed
for these compounds free in solution. The rest of the anti
range is strongly prohibited by steric repulsion of Phe175
and hydrogen bonding with Asp179 in the FtsZ active site.
The different glycosidic dihedral angle distributions in the
bound and free state have a direct influence on the 3
J
coupling constant values, calculated as ensemble averages
3
JðC4; H10
Þ
 R
using Eq. 2. 3
JðC4; H10
ÞS
i values for the
individual configurations of the reference compound (S)
were calculated using the Karplus equation (Karplus 1959):
3
JðC4; H10
ÞS
i ¼ A cos2
ðvi þ 120
Þ þ B cosðvi þ 120
Þ þ C
ð5Þ
with the Karplus coefficients, A = 4.4 Hz, B = -1.4 Hz,
and C = 0.1 Hz (Trantirek et al. 2002). The 3
J values for
both states are listed in the caption of Fig. 2. While there is
a large difference between the calculated 3
JðC4,H10
Þ
 R
values of individual compounds in water, the values are
almost identical when bound to the FtsZ protein and fall in
the very narrow range of 2.95 ± 0.1 Hz. A separate simulation
of real 8-Br-GTP bound to the protein yielded an
average 3
J-value of 3.01 Hz.
Free energy differences of the five compounds relative
to the reference 8-soft_Br-GTP bound to FtsZ as calculated
by OS are listed on the left side of the thermodynamic
cycle presented in Fig. 3. The free energies relative to GTP
(DGOS
GTP;RðFtsZÞ) obtained by cycle closure (right side) are
listed in the second column of Table 1. Relative differences
in binding affinities, DDGcalc
GTP;RðbindÞ, are calculated
using Eq. 6.
DDGcalc
GTP;RðbindÞ ¼ DGOS
GTP;RðFtsZÞ À DGESÀOS
GTP;R ðaqÞ; ð6Þ
where DGESÀOS
GTP;R ðaqÞ are the free energy values relative to
GTP in a water environment as calculated by the ES-OS
method in our previous study (Hritz and Oostenbrink
2009). It is important to note that, despite the different soft
reference states used in water and for the FtsZ bound state,
the relative free energy differences between real compounds
remain valid for the calculation of relative binding
Fig. 2 Normalized dihedral
angle (v) distributions for C8substituted
analogs of GTP as
calculated by ES-OS in water
environment [solid lines (Hritz
and Oostenbrink 2009)] and by
OS in bound state to FtsZ
protein (dashed lines).
Ensemble averages of
3
JðC4; H10
Þ
 
values in Hz are
listed in the legend above
individual line symbols. Anti
and syn conformational ranges
are indicated by dotted lines.
Note that the dihedral angle in
the protein simulations remains
in the range ½À140
; À90
Š:
1576 Eur Biophys J (2010) 39:1573–1580
123
affinities. The computationally predicted values are compared
with the last column in Table 1, which lists the only
available experimental binding affinities, to Methanococcus
jannaschii FtsZ (La¨ppchen et al. 2008). We do not
expect that the binding affinities will deviate significantly
between M. jannaschi FtsZ and A. aeolicus FtsZ, since
X-ray structures of FtsZ from both species show very similar
monomer interfaces, which comprise the nucleotide binding
pocket (Oliva et al. 2007). Moreover, for both proteins it was
observed that different ligands do not lead to different conformation
of the binding site (Oliva et al. 2007).
Discussion
The calculated relative binding affinities of C8-substituted
GTP analogs to the FtsZ protein compare well to the
experimental values, with root-mean-square error of
2.7 kJ mol-1
for 8-Cl-GTP, 8-Br-GTP, and 8-CH3-GTP.
Note that no empirical parameters, other than the force
field to calculate the interactions, were used to obtain these
values. Table 1 nicely illustrates that the relative free
energies in both environments are equally important for the
final free energies. The bromine and methyl substituent are
both predicted to be much weaker binders with respect to
chloride, by roughly *6.5 kJ mol-1
. However, while the
dominant contribution to this difference comes from the
water environment for 8-Br-GTP, it comes from the bound
state for 8-Me-GTP. It is also interesting to note that the
GTP analog that is predicted to have the highest affinity is
8-F-GTP, resulting from similar contributions in both
environments. No experimental data is available for this
compound, as difficulties concerning its synthesis were
only recently resolved (Liu et al. 2006; Ghosh et al. 2007).
It is usually considered that the anti conformation corresponds
to a low and the syn conformation to a high value
of the coupling constant, 3
J(C4,H1’) (Stolarski et al. 1984;
Cho and Evans 1991; Ippel et al. 1996; Trantirek et al.
2002). Therefore it may seem surprising that the ensemble
average 3
JðC4; H10
Þ
 
of GTP in water (to which the syn
conformation contributes with *3%) is calculated to be
lower (2.4 Hz) than the value obtained for the bound state
of GTP (2.95 Hz) in which only the anti conformation is
observed (Fig. 2). This finding follows directly from the
fact that within the FtsZ binding site the v dihedral angle is
restricted within tighter bounds for all five C8-substituted
GTP analogs as compared with in aqueous solution.
The paradigm of conformational selection describes the
binding process between ligand and protein by taking
multiple conformations of the protein into account. It states
that the correct conformation is selected from the complete
ensemble of possible conformations, which are all populated
to a given extent (Carlson 2002). Here, we apply this
model to the GTP analogs and quantify the contribution of
Fig. 3 Thermodynamic cycles used to calculate the free energy
difference between compounds in complex with FtsZ protein. The
free energy differences between the reference compound (8-soft_BrGTP)
and the real compounds were calculated by OS using the
perturbation formula (Zwanzig 1954). The free energies relative to
GTP were obtained by cycle closure
Table 1 Comparison of binding affinities to FtsZ protein for C8substituted
GTP analogs with respect to GTP as obtained from
computational simulations, DDGcalc
GTP;RðbindÞ, and corresponding
experimental values, DDGexp
GTP;RðbindÞ to Methanococcus jannaschii
FtsZ (La¨ppchen et al. 2005, 2008)
DGOS
GTP;RðFtsZÞ (kJ mol-1
) DGESÀOS
GTP;R ðaqÞ (kJ mol-1
) DDGcalc
GTP;RðbindÞ (kJ mol-1
) DDGexp
GTP;RðbindÞ (kJ mol-1
)
GTP 0 0 0 0
8-F-GTP 14.6 ± 1.1 15.0 ± 1.4 -0.8 ± 2.5 –
8-Cl-GTP 16.5 ± 1.0 12.8 ± 1.3 3.7 ± 2.3 8.0 ± 4.4
8-Br-GTP 16.0 ± 1.0 5.9 ± 1.3 10.1 ± 2.3 9.2 ± 1.8
8-CH3-GTP 23.0 ± 1.0 12.6 ± 1.4 10.5 ± 2.4 *8.7a
Values of free energy differences in complex with FtsZ relative to GTP (DGOS
GTP;RðFtsZÞ) were derived from the thermodynamic cycles shown in
Fig. 3 based on the calculated values by OS free energy perturbation calculations. Relative free energy values in explicit water, DGESÀOS
GTP;R ðaqÞ,
were calculated by the ES-OS method (Hritz and Oostenbrink 2009)
a
Value for DDGexp
GTP;8ÀCH3ÀGTPðbindÞ was estimated from an experimental 50% inhibition concentration (IC50) value of GTPase activity
Eur Biophys J (2010) 39:1573–1580 1577
123
the conformational restriction to the relative binding
affinities. For this, we calculate the free energy that is
needed to restrict the GTP analogs from their unbound state
to a conformation that is possible in the protein, using
Eq. 7.
DGrest
R ðaqÞ ¼ ÀkT ln À140
; À90
h i½ ŠR
ðaqÞ; ð7Þ
where the unitless populations À140
; À90
h i½ ŠR
ðaqÞ were
obtained from a simple integration over the selected range
of a normalized dihedral angle distribution of compound R
in water (solid lines in Fig. 2). The relative values with
respect to GTP in water, DDGrest
GTP;RðaqÞ, are listed in the
fourth column of Table 2. It is interesting to note that these
follow the same trend as the DDGcalc
GTP;RðbindÞ values
(Table 1, fourth column). It seems that the conformational
restriction of the GTP analogs accounts for roughly 65% of
the difference in affinity for the FtsZ protein. We emphasize
that, while the calculation of DDGcalc
GTP;RðbindÞ requires
extensive simulations in water and in the binding site of the
FtsZ protein, the restraining free energy values were calculated
from a simple integration of dihedral angle distributions
obtained from the water simulation only. Our
results strongly indicate that the previously reported correlation
between the binding affinities and Sterimol B1
parameter of the substituents (La¨ppchen et al. 2008) follows
from the differences in the restraining free energies
of the GTP analogs, which in turn stem from a different
syn–anti balance of the compounds, when free in solution.
In agreement with the experimental observations, all
compounds adopt roughly the same conformation when
bound to the protein, while the various substituents lead to
different conformational ensembles when the compounds
are free in solution. The conformational restriction upon
binding to the FtsZ protein accounts for roughly 65% of the
differences in binding affinity. As the binding affinity
seems to be for a large part dependent on the conformational
ensemble of the studied C8-substituted GTP analogs
in water, a low specificity may be expected against other
GTPases in which the binding site restricts the conformational
freedom of C8-substituted GTP analogs in a similar
manner. The inhibitory activities for such proteins will
presumably display the same trend. A notable exception is
tubulin, the eukaryotic homolog of FtsZ, where GTP analogs
with small C8 substituents promoted assembly more
than GTP itself.
Conclusions
Relative free energy differences of five C8-substituted GTP
analogs in complex with the bacterial cell-division protein
FtsZ were calculated using the one-step perturbation (OS)
method. Combined with previous values for the water
environment as obtained from enhanced sampling OS we
calculated the relative binding free energies for these
compounds. The results are in good agreement with the
available experimental binding affinities. The dihedral
angle distributions within the FtsZ binding site are much
narrower as compared with those obtained in water. This
results in significantly different ensemble averages of the
3
J coupling constants.
The contribution of conformational selection for the
C8-substituted GTP analogs was quantified by calculating
the restraining free energy in water that is needed to restrain
the dihedral angle to the conformational range that is
accessible within the binding site of the FtsZ protein. The
restraining free energies follow the same trend as the binding
free energies, accounting for about 65% of the differences
in affinity. This suggests low specificity towards the FtsZ
protein, because the same trend can be expected for any
GTPases in which the binding site restricts the conformational
freedom of C8-substituted GTP analogs in a similar
manner. Our results also suggest an explanation for the
empirically observed correlation between the Sterimol
parameters and the binding affinity to the FtsZ protein.
Acknowledgments We gratefully acknowledge financial support
from the Netherlands Organization for Scientific Research (NWO),
Horizon Breakthrough grant 935.18.018 (J.H.) and VENI grant
700.55.401 (C.O.).
Open Access This article is distributed under the terms of the
Creative Commons Attribution Noncommercial License which permits
any noncommercial use, distribution, and reproduction in any
medium, provided the original author(s) and source are credited.
References
Adams DW, Errington J (2009) Bacterial cell division: assembly,
maintenance and disassembly of the Z-ring. Nat Rev Microbiol
7:642–653
Table 2 The free energy of restraining the conformation of C8substituted
GTP analogs to a conformational range v 2 ½À140
; À90
Š
in water
hÀ140
; À90
i½ ŠR
ðaqÞ DGrest
R ðaqÞ
(kJ mol
-1
)
DDGrest
GTP;RðaqÞ
(kJ mol
-1
)
GTP 50% 1.7 0.0
8-F-GTP 48% 1.8 0.1
8-Cl-GTP 18% 4.2 2.5
8-Br-GTP 3.3% 8.4 6.7
8-CH3-GTP 4.0% 8.0 6.2
The free energy values DGrest
R ðaqÞ are calculated from the populations
hÀ140
; À90
i½ ŠR
ðaqÞ. A comparison of the relative free energies with
respect to GTP, DDGrest
GTP;RðaqÞ, indicates a significant contribution to
the total relative binding affinity, DDGcalc
GTP;RðbindÞ (Table 1)
1578 Eur Biophys J (2010) 39:1573–1580
123
Berendsen HJC, Postma JPM, van Gunsteren WF, Hermans J (1981)
Interaction models for water in relation to protein hydration. In:
Pullman B (ed) Intermolecular forces. Reidel, Dordrecht,
pp 331–342
Berendsen HJC, Postma JPM, van Gunsteren WF, DiNola A, Haak JR
(1984) Molecular-dynamics with coupling to an external bath.
J Chem Phys 81:3684–3690
Berman HM, Henrick K, Nakamura H (2003) Announcing the
worldwide Protein Data Bank. Nat Struct Biol 10:980
Beutler TC, Mark AE, van Schaik RC, Gerber PR, van Gunsteren WF
(1994) Avoiding singularities and numerical instabilities in free
energy calculations based on molecular simulations. Chem Phys
Lett 222:529–539
Carlson HA (2002) Protein flexibility and drug discovery: how to hit a
moving target. Curr Opin Chem Biol 6:447–452
Chen YD, Erickson HP (2009) FtsZ filament dynamics at steady state:
subunit exchange with and without nucleotide exchange.
Biochemistry 48:6664–6673
Cho BP, Evans FE (1991) Correlation between nmr spectral
parameters of nucleosides and its implication to the conformation
about the glycosyl bond. Biochem Biophys Res Commun
180:273–278
Christen M, Hunenberger PH, Bakowies D, Baron R, Burgi R, Geerke
DP, Heinz TN, Kastenholz MA, Krautler V, Oostenbrink C,
Peter C, Trzesniak D, Van Gunsteren WF (2005) The GROMOS
software for biomolecular simulation: GROMOS05. J Comput
Chem 26:1719–1751
Czaplewski LG, Collins I, Boyd EA, Brown D, East SP, Gardiner M,
Fletcher R, Haydon DJ, Henstock V, Ingram P, Jones C, Noula
C, Kennison L, Rockley C, Rose V, Thomaides-Brears HB, Ure
R, Whittaker M, Stokes NR (2009) Antibacterial alkoxybenzamide
inhibitors of the essential bacterial cell division protein
FtsZ. Bioorg Med Chem Lett 19:524–527
Davies DB (1978) Conformations of nucleosides and nucleotides.
Progress Nucl Magn Res Spectr 12:135–225
Ghosh A, Lagisetty P, Zajc B (2007) Direct synthesis of 8-fluoro
purine nucleosides via metalation-fluorination. J Org Chem
72:8222–8226
Haydon DJ, Stokes NR, Ure R, Galbraith G, Bennett JM, Brown DR,
Baker PJ, Barynin VV, Rice DW, Sedelnikova SE, Heal JR,
Sheridan JM, Aiwale ST, Chauhan PK, Srivastava A, Taneja A,
Errington J, Czaplewski LG (2008) An inhibitor of FtsZ with
potent and selective anti-staphylococcal activity. Science
321:1673–1675
Hockney RW (1970) The potential calculations and some applications.
Meth Comput Phys 9:136–211
Hritz J, Oostenbrink C (2007) Optimization of replica exchange
molecular dynamics by fast mimicking. J Chem Phys
127:204104
Hritz J, Oostenbrink C (2008) Hamiltonian replica exchange molecular
dynamics using soft-core interactions. J Chem Phys
128:144121
Hritz J, Oostenbrink C (2009) Efficient free energy calculations for
compounds with multiple stable conformations separated by high
energy barriers. J Phys Chem B 113:12711–12720
Huang Q, Tonge PJ, Slayden RA, Kirikae J, Ojima I (2007) FtsZ:
a novel target for tuberculosis drug discovery. Curr Top Med
Chem 7:527–543
Ippel JH, Wijmenga SS, de Jong R, Heus HA, Hilbers CW, de Vroom
E, van der Marel GA, van Boom JH (1996) Heteronuclear scalar
couplings in the bases and sugar rings of nucleic acids: their
determination and application in assignment and conformational
analysis. Magn Reson Chem 34:S156–S176
Kapoor S, Panda D (2009) Targeting FtsZ for antibacterial therapy:
a promising avenue. Exp Opin Therap Targets 13:1037–1051
Karplus M (1959) Contact electron-spin coupling of nuclear magnetic
moments. J Chem Phys 30:11–15
La¨ppchen T (2007) Synthesis of GTP analogues and evaluation of
their effect on the antibiotic target FtsZ and its eukaryotic
homologue tubulin. PhD thesis, University of Amsterdam
(available online http://dare.uva.nl/en/record/211017); Chapter 3
La¨ppchen T, Hartog AF, Pinas VA, Koomen GJ, den Blaauwen T
(2005) GTP analogue inhibits polymerization and GTPase
activity of the bacterial protein FtsZ without affecting its
eukaryotic homologue tubulin. Biochemistry 44:7879–7884
La¨ppchen T, Pinas VA, Hartog AF, Koomen GJ, Schaffner-Barbero
C, Andreu JM, Trambaiolo D, Lo¨we J, Juhem A, Popov AV, den
Blaauwen T (2008) Probing FtsZ and tubulin with C8-substituted
GTP analogs reveals differences in their nucleotide binding sites.
Chem Biol 15:189–199
Liu HY, Mark AE, van Gunsteren WF (1996) Estimating the relative
free energy of different molecular states with respect to a single
reference state. J Phys Chem 100:9485–9494
Liu J, Barrio J, Satyamurthy N (2006) Kinetics and mechanism of the
defluorination of 8-fluoropurine nucleosides in basic and acidic
media. J Fluor Chem 127:1175–1187
Lock RL, Harry EJ (2008) Cell-division inhibitors: new insights for
future antibiotics. Nat Rev Drug Discov 7:324–338
Lo¨we J, Amos LA (1998) Crystal structure of the bacterial celldivision
protein FtsZ. Nature 391:203–206
Mendieta J, Rico AI, Lopez-Vinas E, Vicente M, Mingorance J,
Gomez-Puertas P (2009) Structural and functional model for
ionic (K?/Na?) and pH dependence of GTPase activity and
polymerization of FtsZ, the prokaryotic ortholog of Tubulin.
J Mol Biol 390:17–25
Michie KA, Lowe J (2006) Dynamic filaments of the bacterial
cytoskeleton. Ann Rev Biochem 75:467–492
Oliva MA, Cordell SC, Lo¨we J (2004) Structural insights into FtsZ
protofilament formation. Nat Struct Mol Biol 11:1243–1250
Oliva MA, Trambaiolo D, Lo¨we J (2007) Structural insights into the
conformational variability of FtsZ. J Mol Biol 273:1229–1242
Oostenbrink C, van Gunsteren WF (2004) Free energies of binding of
polychlorinated biphenyls to the estrogen receptor from a single
simulation. Proteins 54:237–246
Oostenbrink C, van Gunsteren WF (2005) Free energies of ligand
binding for structurally diverse compounds. Proc Natl Acad Sci
U S A 102:6750–6754
Oostenbrink C, Villa A, Mark AE, van Gunsteren WF (2004) A
biomolecular force field based on the free enthalpy of hydration
and solvation: the GROMOS force-field parameter sets 53A5
and 53A6. J Comput Chem 25:1656–1676
Paradis-Bleau C, Beaumont M, Sanschagrin F, Voyer N, Levesque
RC (2007) Parallel solid synthesis of inhibitors of the essential
cell division FtsZ enzyme as a new potential class of antibacterials.
Bioorg Med Chem 15:1330–1340
Romberg L, Mitchison TJ (2004) Rate-limiting guanosine 50
-triphosphate
hydrolysis during nucleotide turnover by FtsZ, a prokaryotic
tubulin homologue involved in bacterial cell division.
Biochemistry 43:282–288
Ryckaert J-P, Ciccotti G, Berendsen HJC (1977) Numerical integration
of cartesian equations of motion of a system with
constraints: molecular dynamics of n-alkanes. J Comput Phys
23:327–341
Scha¨fer H, van Gunsteren WF, Mark AE (1999) Estimating relative
free energies from a single ensemble: hydration free energies.
J Comput Chem 20:1604–1617
Stolarski R, Hagberg CE, Shugar D (1984) Studies on the dynamic
syn-anti equilibrium in purine nucleosides and nucleotides with
the aid of 1H and 13C NMR spectroscopy. Eur J Biochem
138:187–192
Eur Biophys J (2010) 39:1573–1580 1579
123
Tadros M, Gonzalez JM, Rivas G, Vicente M, Mingorance J (2006)
Activation of the E. coli cell division protein FtsZ by low affinity
interaction with monovalent cations. FEBS Lett 580:4941–4946
Tironi IG, Sperb R, Smith PE, van Gunsteren WF (1995) A
generalized reaction field method for molecular-dynamics simulations.
J Chem Phys 102:5451–5459
Trantirek L, Stefl R, Masse JE, Feigon J, Sklenar V (2002)
Determination of the glycosidic torsion angles in uniformly
C-13-labeled nucleic acids from vicinal coupling constants
3 J(C2/4–H10
) and 3 J(C6/8–H10
). J Biomol NMR 23:1–12
Vollmer W (2006) The prokaryotic cytoskeleton: a putative target for
inhibitors and antibiotics? Appl Microbiol Biotech 73:37–47
Zwanzig RW (1954) High-temperature equation of state by a
perturbation method. I. Nonpolar gases. J Chem Phys 22:1420–
1426
1580 Eur Biophys J (2010) 39:1573–1580
123
ALCHEMICAL (RELATIVE) FREE ENERGY CALCULATIONS
52
Paper 9
Hritz, J.; Oostenbrink, C. Efficient free energy calculations for compounds with multiple
stable conformations separated by high energy barriers.
J. Phys. Chem. B 2009, 113, 12711-12720
Efficient Free Energy Calculations for Compounds with Multiple Stable Conformations
Separated by High Energy Barriers
Jozef Hritz and Chris Oostenbrink*
Leiden Amsterdam Center for Drug Research, DiVision of Molecular Toxicology, VU UniVersity Amsterdam,
The Netherlands
ReceiVed: April 1, 2009; ReVised Manuscript ReceiVed: August 4, 2009
Compounds with high intramolecular energy barriers represent challenging targets for free energy calculations
because of the difficulty to obtain sufficient conformational sampling. Existing approaches are therefore
computationally very demanding, thus preventing practical applications for such compounds. We present an
enhanced sampling-one step perturbation method (ES-OS) to tackle this problem in a highly efficient way. A
single molecular dynamics simulation of a judiciously chosen reference state (using two sets of soft-core
interactions) is sufficient to determine conformational distributions of chemically similar compounds and the
free energy differences between them. The ES-OS method is applied to a set of five biologically relevant
8-substituted GTP analogs having high energy barriers between the anti and the syn conformations of the
base with respect to the ribose part. The reliability of ES-OS is verified by comparing the results to Hamiltonian
replica exchange simulations of GTP and 8-Br-GTP and the experimentally determined 3
J(C4,H1′) coupling
constant for GMP in water. Additional simulations in vacuum and octanol allow us to calculate differences
in the solvation free energies and in lipophilicities (log P). Free energy contributions from individual
conformational regions are also calculated, and their relationship with the overall free energy is derived leading
to a set of multiconformational free energy formulas. These relationships are of general applicability and can
be used in free energy calculations for a more diverse set of compounds.
Introduction
The calculation of free-energy differences from molecular
simulations using free energy perturbation (FEP) or thermodynamic
integration (TI) is becoming more and more popular, not
in the least thanks to increasing computer power and more
accurate force fields.1-5
In these approaches, the free energy
difference is typically calculated from a set of simulations,
connecting the two compounds or states between which the free
energy is calculated. Convergence becomes difficult when
applied to compounds that have a high energy barrier between
multiple relevant conformations.6-9
In this manuscript, a method
is presented that efficiently calculates the free energy difference
between multiple conformations of several molecules and
simultaneously yields the free energy difference between these
compounds. The problem of multiple conformations in free
energy calculations was addressed earlier by the conformational
space decomposition approach and the confine-and-release
method.6,10,11
In these approaches, the free energy difference
between two compounds is calculated from the populations of
individual conformation for both the compounds and the free
energy difference between the compounds in specified conformations.
In this work, we derive multiconformational free energy
formulas which, in contrast to the conformational space
decomposition approach, require the conformational populations
of only one compound. This is highly advantageous, because
calculating the conformational populations by applying an
external force to overcome high energy barriers is known to be
slowly converging.12,13
If the occupancy of the conformational states changes during
the process for which the free energy (∆A) is calculated,
sufficient conformational sampling is of utmost importance.
8-substituted GTP analogs (Figure 1) represent a group of
biologically relevant compounds to which this applies. The base
of GTP can adopt two stable conformations with respect to the
sugar moiety by rotation around the glycosidic bond, χ. The
preference for the anti (χ ≈ -130°) or syn (χ ≈ 60°)
conformation is strongly dependent on the substituent at the
C8 atom of the base. In a previous study, we showed that a
high energy barrier between these two conformations in both
GTP and 8-Br-GTP prevents sufficient conformational sampling
by common computational techniques such as molecular
dynamics (MD). No single anti f syn transition is seen within
two independent 25 ns MD simulations of GTP or 8-Br-GTP
in water at room temperature preventing the determination of
their relative populations.13
There are two possible transition
pathways between the anti and the syn conformations, the one
with the lowest energy barrier corresponding to the region with
χ ≈ -20°. We determined that the height of this barrier for
GTP is ∼15 kJ·mol-1
with respect to the anti and ∼7 kJ·mol-1
with respect to the syn conformation of GTP in water. In the* Corresponding author. E-mail: c.oostenbrink@few.vu.nl.
Figure 1. Structure of 8-substituted analogs of GTP in syn conformation,
where X ) H, F, Cl, Br, CH3. Conformational transitions between
the syn and anti conformations occur by rotation around the glycosidic
bond indicated by the arrow. The glycosidic dihedral angle (χ) is defined
over atoms: C4-N9-C1′-O4′.
J. Phys. Chem. B 2009, 113, 12711–12720 12711
10.1021/jp902968m CCC: $40.75  2009 American Chemical Society
Published on Web 09/01/2009
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
case of 8-Br-GTP, it is higher than 25 kJ·mol-1
with respect to
both anti/syn conformations which prevents adequate sampling.
The consequence of these high intramolecular barriers is the
fact that during 25 ns of standard MD starting from both anti
and syn conformations the only observed transition was for GTP,
from syn to anti. In the same paper, we show a very poor
convergence when calculating the free energy difference
between the anti and the syn conformations using techniques
with an external force to overcome the energy barrier. Significantly
better results were obtained when, instead of forcing to
overcome a high energy barrier, the barrier itself was decreased
by applying soft-core interactions between the base and the sugar
atoms of GTP/8-Br-GTP as part of a Hamiltonian replica
exchange MD (H-REMD) scheme. This technique revealed the
inverse preference for the anti and syn conformations of GTP
and 8-Br-GTP, in agreement with NMR data.14
High intramolecular
energy barriers within the molecules not only hamper
the determination of their conformational ensembles but also
make it difficult to calculate the free energy differences between
them. Bitetti-Putzer et al. proposed to perturb the Hamiltonian
within generalized ensembles between two compounds with
inverse conformational populations in order to enhance the
conformational sampling and to calculate the free energy
difference between them.15
Unfortunately, such an approach does
not work for 8-substituted GTP analogs because they all contain
the energy barrier in the same conformational region and, for
all linear combinations of the Hamiltonians of individual ligands,
the barrier remains high. This problem can be overcome by
applying “dual topology alchemical REMD”.16
In this REMD
scheme, every replica contains coordinates for both the startingand
the end-state of a perturbation, while a coupling parameter
λ determines how the Hamiltionians for these states are mixed
in every replica. The authors point out that a soft-core potential17
may also be used at intermediate values of λ, potentially
lowering barriers. In ref 13, we suggested an alternative
approach to combine H-REMD between compounds with
variations in softness. By using a single topology approach, in
which only the softness is modified at intermediate replica’s to
obtain sufficient conformational transitions, the free energy
difference between compounds and the conformational preferences
could be obtained. In order to obtain sufficient conformational
transitions, it is not needed to decouple the system
completely from its surroundings, but a high enough level of
softness is sufficient. GTP and 8-Br-GTP could be connected
by a number of replica’s in which the sugar-base interaction
is made soft, such that conformational transitions become
possible. This approach would suffer from large computational
requirements as about 20 replicas would be needed. In addition,
we note that finding the optimal settings of such a scheme is
very demanding.18
In order to calculate the free energy difference
between any pair of GTP analogs by any of the mentioned
H-REMD schemes, we would have to perform such an extensive
H-REMD simulation for each of these pairs.
In contrast, this paper presents a very efficient enhanced
sampling-one step perturbation (ES-OS) approach, combining
the construction of an artificial reference Hamiltonian allowing
for enhanced conformational sampling and simultaneous application
of the one-step perturbation method.19,20
In the onestep
perturbation method, all interactions involving atoms which
are chemically modified within a series of compounds are
described using soft-core interactions. For many systems, such
as the 8-substituted GTP analogs, this is not sufficient to enhance
the conformational sampling. Therefore, we extend the set of
interactions treated using a soft-core by the interactions between
the atom pairs that are responsible for the high energy barrier
(base-sugar in case of 8-substituted GTP analogs in water),
thereby allowing for much faster conformational transitions.13
The set of 8-substituted analogs of GTP was chosen not only
out of methodological interest but also because of the biological
relevance of these compounds, which are considered as promising
antibacterial agents.21,22
8-Br-guanosine (with an enforced
syn conformation) was experimentally used to probe the syn
conformation in a G quartet,23
UNCG motifs,24
and leaddependent
ribozyme.25
Many studies involving 8-substituted
analogs of cyclic adenosine and guanosine are reviewed in ref
26. Much less is known about the biological effects of 8-Fguanosines,
probably because of the fact that these became
synthetically feasible only very recently.27,28
This paper is organized as follows; first ES-OS is described
and applied to a set of 8-substituted analogs of GTP in vacuum,
water, and 1-octanol. From this, we obtain free energy differences
between the compounds and the dihedral angle distributions
around the glycosidic bond for each of them. The method
is tested by comparing the distributions of the dihedral angle
and energies for GTP and 8-Br-GTP as calculated by ES-OS to
the values obtained from extensive H-REMD simulations using
soft-core interactions.13
From the free energy differences in
different media, we subsequently calculated differences in log
P and solvation free energies between the different compounds.
Differences in free energies between compounds, solvation free
energies, and lipophilicities were calculated over the complete
conformational space as well as over ensembles for the anti
and syn conformations individually. Theoretical relationships
are derived showing how these individual conformational terms
contribute to the overall values and how these equations can
be used for free energy calculations for more diverse compounds.
Methods
Enhanced Sampling-One Step Perturbation (ES-OS). The
aim of ES-OS is to determine simultaneously the free energy
differences between multiple conformations separated by high
energy barriers and between chemically similar compounds over
the complete conformational space. We show its application
for a set of five 8-substituted GTP analogs in which H8 is
replaced by halogen atoms or a methyl group (Figure 1).
The ES-OS approach consists of two steps; the first is the
(1) construction of an (artificial) reference compound. Typically,
all interactions involving the atom or group of atoms that are
different in the series of compounds are treated by soft-core
interactions (see below). For practical reasons, we usually select
the largest atom from the set; in our application, we used a soft
Br atom at position 8 of the base. Often, such a modification of
the Hamiltonian is insufficient to overcome high intramolecular
energy barriers. Therefore, additional interactions are altered.
Here, soft-core interactions were additionally applied between
atoms of the base and the sugar in the reference molecule, as
described in ref 13. (2) The second step is the projection to the
Hamiltonian of the real compounds. The conformational sampling
is strongly affected by the modifications to the Hamiltonian
in the previous step. Following the idea of the one-step
perturbation method,20
the free energy between the reference
compound (S) and the real compounds (R), can be calculated
from a simulation using the reference Hamiltonian and applying
Zwanzigs perturbation formula:29
∆GSR ) GR - GS ) -kBT ln〈e-(HR(q,p)-HS(q,p))/kBT
〉S
(1)
12712 J. Phys. Chem. B, Vol. 113, No. 38, 2009 Hritz and Oostenbrink
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
where kB is the Boltzmann constant, T is the absolute temperature,
HS and HR are the Hamiltonian for the (soft) reference compound
and one of the real compounds, respectively. The angular brackets
indicate an ensemble average over the positions, q and momenta
p obtained from a simulation of the reference state.
Any particular configuration (qi,pi) can be reweighted for the
real compounds by calculating the normalized Boltzmann factor:
The ES-OS approach has general applicability for any set of
compounds. The high efficiency comes from the fact that the
conformational characteristics for all compounds in the studied
set as well as the free energy differences between them can be
calculated from a single MD simulation of the reference
compound. If a judiciously chosen reference state is used, which
samples relevant conformations for the real compounds often
enough, eqs 1 and 2 may be calculated with sufficient accuracy
to obtain relevant results for the real compounds.19,20,30,31
Simulation Setup. All MD simulations and most of the
analyses were conducted using the GROMOS05 simulation
package running on a linux cluster.32
All bonds were constrained,
using the SHAKE algorithm33
with a relative geometric accuracy
of 10-4
, allowing for a time step of 2 fs used in the leapfrog
integration scheme.34
Periodic boundary conditions with a
truncated octahedral box were applied. System coordinates were
saved every 1 ps, and simulations of the reference compound
were performed for 50 ns in water, 200 ns in vacuum, and 100
ns in octanol, depending on the convergence of the calculated
free energy values and the speed of the simulations. Simulation
of the reference compound in vacuum was much faster than
simulation in a box containing 1926 SPC water molecules35
or
324 flexible octanol molecules, respectively. After a steepest
descent minimization to remove bad contacts between molecules,
initial velocities were randomly assigned from a
Maxwell-Boltzmann distribution at 298 K, according to the
atomic masses. The temperature was kept constant using weak
coupling to a bath of 298 K with a relaxation time of 0.1 ps.36
The solute molecule (reference compound 8-Br-GTP using soft
core interactions) and solvent were independently coupled to
the heat bath. The pressure was controlled using isotropic weak
coupling to atmospheric pressure with a relaxation time of 0.5
ps.36
van der Waals and electrostatic interactions were calculated
using a triple range cutoff scheme. Interactions within a shortrange
cutoff of 0.8 nm were calculated every time step from a
pair-list that was generated every five steps. At these time points,
interactions between 0.8 and 1.4 nm were also calculated and
kept constant between updates. A reaction-field contribution was
added to the electrostatic interactions and forces to account for
a homogeneous medium outside the long-range cutoff, using
the relative permittivity of SPC water (61),37
octanol (10),38
and
vacuum (1, no reaction field). All interaction energies were
calculated according to the GROMOS force field, parameter
set 53A6.39
Force field parameters for all compounds are listed
in Supporting Information. Note that different partial charges
of the phosphate groups in different environments result in an
overall neutral reference compound in vacuum and octanol while
it has a net charge of -3e in water which was electroneutralized
by the addition of 3 Na+
counterions. The functional form of
the soft-core van der Waals and electrostatic interactions
between atoms i and j was the following:17
with rij the interatomic distance; Aij(λVdw) ) RVdwλVdw
2
(C12ij)/(C6ij),
and Bij(λel) ) Relλel
2
. C12ij, C6ij are the Lennard-Jones parameters
for atom pair i and j; qi and qj are the partial charges of particles
i and j; RVdw and Rel are the softness constants. In the current
study, we used in all simulations RVdw ) Rel ) 1, and the softness
of the interactions was controlled through the parameters
λVdw, λel. It can be seen that at longer distances (rij . A(λVdw)
and rij . B(λel)) the soft-core interaction approximates the
interaction for normal atoms and that they differ mostly at short
distances between the atoms (rij e A(λVdw) or rij e B(λel)).
Potential energy barriers are mostly the result of a short-ranged
repulsion between atoms, which can strongly be reduced by
increasing the levels of softness, thereby allowing us to make
very specific, local modifications to the Hamiltonian.
The procedure for finding the optimal softness of selected
interactions, such that enhanced conformational sampling is
observed, is described in our previous paper.18
In water, it led
to softness parameters λvdw ) 0.7 and λel ) 0.7 for all nonbonded
interactions between the base and the sugar atoms of 8-Br-GTP.
In the ES-OS application for 8-substituted analogs of GTP, the
set of soft-core interactions was further extended to all nonbonded
interactions between bromine and the rest of the system
(including water) using the same softness parameters. In
vacuum, the procedure led to the same interactions with the
same softness parameters (λvdw ) 0.7; λel ) 0.7) extended by
soft-core interactions between the base and the phosphates with
λvdw ) 0 and λel ) 0.22. The optimal softness parameters
obtained for GTP in octanol were refined as λvdw ) 0.7 and λel
) 0.6 for interactions between base and sugar atoms as well as
between bromine and the rest of system (including octanol).
For interactions between the base and the phosphates, softness
parameters λvdw ) 0 and λel ) 0.27 were used. We stress that,
in order to calculate the free energy of transfer between two
media, the reference molecules do not need to be identical. As
long as the end-states of the perturbation (the real molecules)
are identical in the different media, they can theoretically be
connected through any arbitrary reference molecule in order to
calculate the free energy difference between the real molecules.
The most efficient reference molecule can be different in
different media.
Results
During 50 ns of MD simulation of the reference state in water
∼150 anti T syn transitions occurred ensuring sufficient
convergence of the χ dihedral angle distribution (orange curve
in Figure 2) and a precise determination of the conformational
populations, [anti](aq)
S
) 41.5% ( 3.0%, [syn](aq)
S
) 58.5% (
3.0%, and the free energy between them ∆GS
anti,syn
(aq) ) -kBT
ln([syn](aq)
S
/[anti](aq)
S
) ) -0.85 ( 0.31 kJ·mol-1
. Similar values
were obtained in vacuum ([anti](V)
S
) 40.8% ( 1.1%, [syn](V)
S
)
59.2% ( 1.1%, ∆GS
anti,syn
(V) ) -0.92 ( 0.11 kJ·mol-1
) and
in octanol ([anti](o)
S
) 49.3 ( 4.5%, [syn](o)
S
) 50.7% ( 4.5%,
∆GS
anti,syn
(o) ) -0.069 ( 0.45 kJ·mol-1
). Error estimates on
ensemble averages are calculated from block averages followed
by an extrapolation to infinite block length.40
Populations
pi
R
(qi, pi) )
e-(HR(qi,pi)-HS(qi,pi))/kBT
∑
i
e-(HR(qi,pi)-HS(qi,pi))/kBT
(2)
Eij
Vdw
(rij, λVdw) )
( C12ij
Aij(λVdw) + rij
6
- C6ij
) 1
Aij(λVdw) + rij
6
(3)
Eij
el
(rij, λel) )
qiqj
4πε
1
√Bij(λel) + rij
2
(4)
Efficient Free Energy Calculations J. Phys. Chem. B, Vol. 113, No. 38, 2009 12713
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
(including their error estimates) are calculated from the ensemble
average of occurrences of the conformational states. From the
simulations in the reference state, the free energy differences
to the real compounds, ∆GSR were calculated, see Table 1. By
selecting from the ensemble only structures that are in anti and
syn conformation, free energy differences can be obtained for
the molecules in these conformations ∆GSR
anti
and ∆GSR
syn
. From
these values and the free energy difference ∆GS
anti,syn
for the
reference state, we can calculate the free energy difference
between the anti and the syn conformations for the real
compounds, ∆GR
anti,syn
(see below) and subsequently the relative
populations [anti]R
and [syn]R
. All of these values are summarized
in Table 1. From H-REMD calculations, we have earlier
calculated ∆GGTP
anti,syn
) 7.6 ( 0.3 kJ·mol-1
and ∆G8-Br-GTP
anti,syn
)
-6.8 ( 0.9 kJ·mol-1
,13
which fit remarkably well with the data
in Table 1.
The reliability of the free energy differences between the
compounds will be mostly dependent on the overlap in phase
space sampled by the reference and real compounds (Figure
2). The sampling efficiency can be calculated as Seff ) 2N∆He∆G/
N, being twice the number of conformations in which the change
in Hamiltonian in eq 1 is less than the overall change in free
energy, divided by the total number of conformations sampled
for the reference state.41
The most efficient sampling will occur
with a symmetric distribution of energies around ∆H ) ∆G,
leading to Seff ) 1. Sampling efficiencies were calculated to be
between 3.4% (for 8-Br-GTP in water) and 6.1% (for 8-F-GTP
in water) indicating that about 2000 relevant conformations are
sampled for each of the real compounds.
In Figure 3, we compare distributions of nonbonded energies
between (A) the base and water and (B) the base and the sugar
of GTP and 8-Br-GTP as obtained from ES-OS and from
previous extensive H-REMD simulations.13
The distributions
for the ES-OS approach were obtained by reweighting the
probabilities of individual MD structures using eq 2. A good
agreement can be observed. Figure 2 presents a similar
comparison in terms of the χ dihedral angle distribution for GTP
and 8-Br-GTP. The distributions in water obtained from ESOS
(solid curves) show good agreement with the same distributions
as obtained from H-REMD simulations (dashed curves).
The solid curves in Figure 2 and Figure 3 are less smooth than
the dashed ones, which is due to the fact that only few (∼50)
MD structures have a significant (>0.1) probability for the real
compounds. We note that the [anti](aq)
R
and [syn](aq)
R
populations
show faster convergence. A reasonably accurate number can
already be obtained from 10 ns of ES-OS simulation.
3
J(C4,H1′) values for individual frames of the reference
compounds were calculated using the Karplus equation:
where the most recent Karplus coefficients parametrized for
guanosine containing systems were applied: A ) 4.4 Hz, B )
-1.4 Hz, C ) 0.1 Hz.42
Distributions of 3
J(C4,H1′) were generated in the same way
as for the glycosidic dihedral angle and subsequently used for
the calculation of the average values 〈3
J(C4,H1′)〉, listed in the
last column of Table 1. These values are directly comparable
with experimental NMR data. Schwalbe et al. reported for
different magnetization transfer delays, four values of 3
J(C4,H1′)
for 5′-GMP in D2O: 2.5 Hz, 3.5 Hz, 2.6 Hz, and 2.6 Hz.43
Our
value for GTP in H2O as obtained from ES-OS, 〈3
J(C4,H1′)〉
) 2.42 Hz is in very good agreement with these data.
Unfortunately, we did not find experimental NMR 3
J(C4,H1′)
data for other molecules in our set of 8-substituted GTP analogs.
Figure 4 shows the χ dihedral angle distributions for all real
compounds in Table 1. The values for [anti]R
and [syn]R
that
can be obtained by integrating these distributions correspond
within numerical precision to the values obtained from the free
energy differences in Table 1. In order to analyze the free energy
differences under the restriction of a given conformation, it is
useful to use the same definition of the anti and syn regions
(anti: χ∈<-180°,-15°)∪<135°,180°>; syn: <χ∈<-15°,135°)).
In reality, the distributions are shifted within these regions for
different compounds in different environments (Figure 4A (in
water), Figure 4B (in vacuum), Figure 4C (in octanol)). The
population of the anti conformation is lower in vacuum as
compared with the water environment for all studied compounds.
This is most probably due to hydrogen bonding and electrostatic
interactions between the phosphate and the NH2 group of the
base which are more pronounced in vacuum than in water. A
more detailed comparison of these effects requires additional
simulations of guanosine in water and vacuum which is beyond
the scope of this work. By using the thermodynamic cycles in
Figure 5, the free energy differences between the compounds
(Figure 5A) or between syn and anti states of the real compounds
can be calculated (Figure 5B).
From the free energy differences between individual compounds
and the soft reference state in different environments,
we calculate the differences in solvation free energies and
lipophilicities by applying the following equations:
The calculations were performed for the complete conformational
space (Table 2A) and for the anti (Table 2B) or syn
(Table 2C) conformations. We note that, since we have
simulated the reference molecules in different phosphate protonation
states in the different media, a contribution for the
deprotonation of the phosphate tail should be included in the
Figure 2. Normalized dihedral angle (χ) distributions for GTP and
8-Br-GTP calculated by ES-OS and H-REMD and for the reference
state (8-Br-GTP using soft-core interactions) as obtained from MD.
Anti and syn conformational regions are indicated by dotted lines.
3
J(C4,H1′) ) A cos2
(χ + 120◦
) + B cos(χ + 120◦
) + C
(5)
∆∆GSR(solV) ) ∆GSR(aq) - ∆GSR(V) (6)
∆∆GSR(o/w) ) ∆GSR(aq) - ∆GSR(o) (7)
∆log Po/w
)
∆∆G(o/w)
2.3kBT
(8)
12714 J. Phys. Chem. B, Vol. 113, No. 38, 2009 Hritz and Oostenbrink
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
calculations. However, we assume this contribution to be
constant for the various compounds and conformations, such
that it cancels in the relative free energies.
For comparison, ∆log Po/w
was also calculated using a widely
used atomic contribution model44
based on the two-dimensional
(2D) structure of molecules as implemented in the program
MOE.45
These values are listed in the last column of Table 2.
Discussion
We want to emphasize that all data for the set of five
8-substituted GTP analogs as summarized in Tables 1 and 2
and Figures 4 and 5 were refined from only three MD
simulations of the reference compound in different media. This
demonstrates the big potential of the ES-OS approach to
efficiently characterize chemically similar compounds with high
intramolecular energy barriers. An obvious application is a lead
optimization process during which one searches for the chemical
modification which has certain desired physicochemical properties.
We observe relatively rough dihedral angle distributions
mostly in regions of conformational space that are rarely
sampled by the reference compound and for compounds showing
a large free energy difference with respect to the reference
compound. However, even rough distributions provide very
good overview for all compounds and therefore allow for an
efficient identification of the ones having the desired properties.
For the compounds of interest, additional MD simulations can
be performed within the local minima and smooth dihedral angle
distributions can be obtained by reweighting the distributions
for each of these simulations with the population obtained with
ES-OS. We recall that the relative populations converge much
faster in ES-OS than the distributions. In the present application,
this can be applied to 3
J value distributions and the resulting
average value. The drawback obviously is the need for additional
MD simulations.
The unphysical reference compound described here allows
for enhanced conformational sampling either using ES-OS or
TABLE 1: Free Energy Differences for 8-Substituted GTP Analogs with Respect to the Reference State (8-Br-GTP Using
Soft-Core Interactions) As Calculated by ES-OS and Free Energy Differences between Their Anti and Syn Conformations in
(A) Water, (B) Vacuum, and (C) Octanola
(A)
In water ∆GSR(aq)[kJ·mol-1
] ∆GSR
anti
(aq)[kJ·mol-1
] ∆GSR
syn
(aq)[kJ·mol-1
] ∆GR
anti,syn
(aq)[kJ·mol-1
] [anti](aq)
R
[syn](aq)
R
〈3
J(C4,H1′)〉[Hz]
GTP 6.82 ( 0.87 4.72 ( 0.85 13.97 ( 0.62 8.40 ( 1.78 96.7% 3.3% 2.42 ( 0.60
8-F-GTP 21.85 ( 0.51 20.35 ( 0.62 24.04 ( 1.00 2.84 ( 1.93 75.9% 24.1% 3.33 ( 0.61
8-Cl-GTP 19.57 ( 0.38 19.93 ( 0.71 19.33 ( 0.53 -1.45 ( 1.55 35.8% 64.2% 4.49 ( 0.69
8-Br-GTP 12.67 ( 0.47 17.71 ( 0.76 11.48 ( 0.53 -7.08 ( 1.60 5.4% 94.6% 5.46 ( 1.00
8-CH3-GTP 19.37 ( 0.54 23.15 ( 0.85 18.27 ( 0.61 -5.73 ( 1.77 9.0% 91.0% 5.33 ( 1.12
(B)
In vacuum ∆GSR (V)[kJ·mol-1
] ∆GSR
anti
(V)[kJ·mol-1
] ∆GSR
syn
(V)[kJ·mol-1
] ∆GRanti,syn(V)[kJ·mol-1
] [anti](V)
R
[syn](V)
R
〈3
J(C4,H1′)〉[Hz]
GTP -22.47 ( 0.70 -24.28 ( 0.79 -19.09 ( 0.67 4.27 ( 1.57 84.9% 15.1% 1.86 ( 0.47
8-F-GTP -13.19 ( 0.33 -12.62 ( 0.52 -13.51 ( 0.48 -1.81 ( 1.11 32.5% 67.5% 4.23 ( 0.71
8-Cl-GTP -17.03 ( 0.44 -13.40 ( 0.81 -18.08 ( 0.49 -5.60 ( 1.41 9.4% 90.6% 4.69 ( 0.57
8-Br-GTP -20.69 ( 0.45 -12.71 ( 0.75 -21.95 ( 0.46 -10.16 ( 1.32 1.6% 98.4% 4.96 ( 0.74
8-CH3-GTP -18.14 ( 0.50 -9.88 ( 0.73 -19.41 ( 0.51 -10.45 ( 1.35 1.4% 98.6% 5.10 ( 0.94
(C)
In octanol ∆GSR (o)[kJ·mol-1
] ∆GSR
anti
(o)[kJ·mol-1
] ∆GSR
syn
(o)[kJ·mol-1
] ∆GRanti,syn(o)[kJ·mol-1
] [anti](o)
R
[syn](o)
R
〈3
J(C4,H1′)〉[Hz]
GTP 2.12 ( 0.86 0.42 ( 0.95 10.26 ( 0.54 9.77 ( 1.94 98.1% 1.9% 1.91 ( 0.42
8-F-GTP 7.42 ( 0.71 6.26 ( 0.90 9.55 ( 0.70 3.22 ( 2.05 78.6% 21.4% 3.24 ( 0.63
8-Cl-GTP 1.21 ( 0.43 1.04 ( 0.77 1.38 ( 0.40 0.27 ( 1.62 52.7% 47.3% 3.96 ( 0.77
8-Br-GTP -7.06 ( 0.37 -4.10 ( 0.79 -8.35 ( 0.35 -4.32 ( 1.59 14.9% 85.1% 4.99 ( 0.83
8-CH3-GTP 3.83 ( 0.47 6.59 ( 0.75 2.59 ( 0.50 -4.07 ( 1.7 16.2% 83.8% 5.01 ( 0.98
a
Values for ∆GRanti,syn (5th column) were derived from the thermodynamic cycles shown in Figure 5B.
Figure 3. Normalized energy distributions for GTP and 8-Br-GTP
calculated by ES-OS and H-REMD and for the reference state (8-BrGTP
using soft-core interactions) as obtained from MD. Nonbonded
energies between (A) base and water and (B) base and sugar, which
are modified in the reference state.
Efficient Free Energy Calculations J. Phys. Chem. B, Vol. 113, No. 38, 2009 12715
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
H-REMD. In this application, the interactions that were softened
(between sugar and base) were determined mostly by chemical
intuition. Determining the optimal reference molecule may be
the most crucial and least straightforward factor of this method,
which will be the topic of continuing research. Now that a
reference molecule for GTP is determined, it can be applied
further in any guanosine containing systems, such as RNA loops.
This would be useful not only for a structural refinement
including the populations of conformations with guanosine in
the anti or syn conformation but also for a consistent interpretation
of NMR relaxation rates depending on the chemical
shielding anisotropy (CSA). Recently, Ferner et al. presented a
significant dependence of 13
C CSA on the dihedral angle around
the glycosidic bond (-133 ppm for guanine in anti and -122
in syn). Only by taking the different conformations into account
could they obtain fully consistent data for the guanine residue.46
Two applications that require the accurate calculation of free
energy differences between 8-substituted GTP analogs in
different media are the solvation free energy and lipophilicity
differences, which are summarized in Table 2. The free energy
differences in water that are summarized in Figure 5A and in
the second column of Table 1A will prove to be useful in other
applications as well, such as the calculation of differences in
binding affinities of 8-substituted GTPs toward biomacromolecules.
Despite the fact that many proteins will bind these
compounds only in one conformation, it is important to note
that for the unbound compounds both conformations (anti/syn)
are contributing to the final affinity.
We emphasize that values calculated for complete conformational
space (Table 2A) differ significantly from the
same values calculated only considering the anti (Table 2B)
or syn (Table 2C) conformation for several compounds. The
reason is that the free energy differences for individual
conformations are different and their contributions to the
overall values depend on the conformational populations.
Both of these effects are completely ignored by any 2D
prediction method of solvation and lipophilicity44
(often used
as descriptor in e.g. medicinal chemistry). Therefore, 2D
predictions in principle can not be expected to lead to accurate
predictions for compounds with different conformational
states. Please note in Table 1 that the value of ∆GSR for the
complete conformational space is always in between the
values for the individual conformations ∆GSR
anti
and ∆GSR
syn
. This
is not necessarily true for the relative free energy differences
between different environments, ∆∆GSR. For example, differences
in the solvation free energy of 8-CH3-GTP and GTP
for individual conformations are ∆∆GGTP,8-CH3-GTP
anti
(solV) )
4.03 ( 3.05 kJ · mol-1
and ∆∆GGTP,8-CH3-GTP
syn
(solV) ) 4.62
( 2.41 kJ · mol-1
while the value for the complete ensemble
is ∆∆GGTP,8-CH3-GTP (solV) ) 8.22 ( 2.61 kJ · mol-1
(Table
2).
In order to understand how ∆G and ∆∆G values for
individual conformations contribute to values for the complete
conformational space, we derive explicit relationships between
them. The free energy for compound A is defined as
where ZA is the isothermal-isobaric partition function of
compound A:
Figure 4. Normalized dihedral angle (χ) distributions for 8-substituted
analogs of GTP as calculated by ES-OS and for the reference state
(8-Br-GTP using soft-core interactions) as obtained from MD in (A)
water, (B) vacuum, (C) octanol. Anti and syn conformational regions
are indicated by dotted lines.
Figure 5. Thermodynamic cycles used to calculate the free energy
difference between compounds (A) and between different conformations
(B) in water. The free energy differences between the reference
compound (Soft_Ref) and the real compounds where calculated by ESOS.
All other values were obtained by cycle closure.
GA ) -kBT ln ZA (9)
12716 J. Phys. Chem. B, Vol. 113, No. 38, 2009 Hritz and Oostenbrink
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
Considering that the complete conformational space of
compound A is divided into nA
distinct conformations (iA
), the
partition function can be written as sum of partition functions
of individual subspaces, or conformations:
which defines GA
iA
as the free energy of compound A, confined
to conformation iA
.6
The relative population of this conformation
is defined as
Within this framework, Straatsma and McCammon address
the free energy difference between compound A and compound
B in the presence of multiple conformations by writing6
The last term in this equation, the free energy difference
between compound B in conformation iB and compound A in
conformation iA, is typically obtained by standard TI or FEP
calculations, during which the compounds occupy only one
conformational state. The first two terms represent the free
energy difference between sampling the complete conformational
space and sampling only region iA. These terms are
directly linked to the population of the conformation, GA - GA
iA
) kBT ln [iA
]A
. Straatsma and McCammon point out that a free
energy calculation in which the compounds occupy only one
conformation should be corrected by these terms, both for
compound A and for compound B.6
However, in many cases
obtaining an accurate estimate of the relative populations is
much more difficult than obtaining a free energy difference
between compounds confined to well-defined local minima.
Also, in the current application, we have precise population
values only for the reference compound. We therefore derive
formulas that only depend on populations of one compound in
order to evaluate ∆GAB.
Using eqs 9-11, we can write
in which we multiply the individual terms by unity in the form
(1)/(nA
)∑iAnA
(ZA
iA
)/(ZA
iA
) to obtain
and using eq 12 then yields
TABLE 2: Differences in Solvation Free Energies and Lipophilicities for 8-Substited GTP Analogs with Respect to the
Reference State (8-Br-GTP Using Soft-Core Interactions; 2nd, 4th Column) and with Respect to GTP (3rd, 5th Column) As
Calculated by ES-OS over (A) the Complete Conformational Space, (B) the Anti Conformation, and (C) the Syn Conformationa
(A)
∆∆GGTP,R(solv)[kJ·mol-1
] ∆∆GGTP,R(solv)[kJ·mol-1
] ∆∆GSR(o/w)[kJ·mol-1
] ∆∆GGTP,R(o/w)[kJ·mol-1
] ∆log PES-OS
o/w
∆log P2D
o/w
GTP 29.29 ( 1.57 0 4.70 ( 1.73 0 0 0
8-F-GTP 35.04 ( 0.84 5.75 ( 2.41 14.43 ( 1.22 9.73 ( 2.95 1.71 ( 0.52 0.58
8-CL-GTP 36.6 ( 0.82 7.31 ( 2.39 18.36 ( 0.81 13.66 ( 2.54 2.40 ( 0.45 2.73
8-BR-GTP 33.36 ( 0.92 4.07 ( 2.49 19.73 ( 0.84 15.03 ( 2.57 2.64 ( 0.45 3.19
8-CH3-GTP 37.51 ( 1.04 8.22 ( 2.61 15.44 ( 1.01 10.74 ( 2.74 1.89 ( 0.48 1.29
(B)
anti ∆∆GSR
anti
(solV)[kJ·mol-1
] ∆∆GGTP,R
anti
(solV)[kJ·mol-1
] ∆∆GSR
anti
(o/w)[kJ·mol-1
] ∆∆GGTP.R
anti
(o/w)[kJ·mol-1
] ∆log PES-OS
o/w,anti
∆log P2D
o/w,anti
GTP 29.00 ( 1.64 0 4.30 ( 1.80 0 0 0
8-F-GTP 32.97 ( 1.14 3.97 ( 2.61 14.09 ( 1.52 9.79 ( 3.32 1.72 ( 0.58 0.58
8-CL-GTP 33.33 ( 1.52 4.33 ( 2.99 18.89 ( 1.48 14.59 ( 3.28 2.56 ( 0.58 2.73
8-BR-GTP 30.42 ( 1.51 1.42 ( 2.98 21.81 ( 1.55 17.51 ( 3.35 3.07 ( 0.59 3.19
8-CH3-GTP 33.03 ( 1.58 4.03 ( 3.05 16.56 ( 1.60 12.26 ( 3.40 2.15 ( 0.60 1.29
(C)
syn ∆∆GSR
syn
(solV) [kJ·mol-1
] ∆∆GGTP,R
syn
(solV) [kJ·mol-1
] ∆∆GSR
syn
(o/w) [kJ·mol-1
] ∆∆GGTP,R
syn
(o/w) [kJ·mol-1
] ∆log PES-OS
o/w,syn
∆log P2D
o/w,syn
GTP 33.06 ( 1.29 0 3.71 ( 1.16 0 0 0
8-F-GTP 37.55 ( 1.48 4.49 ( 2.77 14.49 ( 1.7 10.78 ( 2.86 1.89 ( 0.50 0.58
8-CL-GTP 37.41 ( 1.02 4.35 ( 2.31 17.95 ( 0.93 14.24 ( 2.09 2.50 ( 0.37 2.73
8-BR-GTP 33.43 ( 0.99 0.37 ( 2.28 19.83 ( 0.88 16.12 ( 2.04 2.83 ( 0.36 3.19
8-CH3-GTP 37.68 ( 1.12 4.62 ( 2.41 15.68 ( 1.11 11.97 ( 2.27 2.10 ( 0.40 1.29
a
Values of ∆log PES-OS
o/w
with respect to GTP calculated by ES-OS (6th column) are compared with values calculated in MOE using an
atomic contribution model (7th column).44
ZA )
1
h3N
N!
Be-(HA+pV)/kBT
(dV dpN
dqN
) ) e-GA/kBT
(10)
ZA ) ∑
iA
nA
ZA
iA
) ∑
iA
nA
e-GA
iA
/kBT
(11)
[iA
]A
)
ZA
iA
ZA
(12)
∆GAB ) GB - GA ) (GB - GB
iB
) - (GA - GA
iA
) +
(GB
iB
- GA
iA
) (13)
∆GAB ) GB - GA ) -kBTln
ZB
ZA
) -kBTln
(∑
iB
nB
ZB
iB
ZA
)(14)
∆GAB ) -kBTln
(∑
iB
nB
ZB
iB
ZA
1
nA ∑
iA
nA
ZA
iA
ZA
iA
)
) -kBTln
(1
nA ∑
iB
nB
∑
iA
nA
ZB
iB
ZA
ZA
iA
ZA
iA
) (15)
Efficient Free Energy Calculations J. Phys. Chem. B, Vol. 113, No. 38, 2009 12717
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
In many practical applications, the different compounds
occupy the same conformations. Here, we consider the same
definition of the anti and syn conformations for all compounds.
By imposing iA
) iB
) i and multiplying the individual terms
in eq 14 by unity in the form (ZA
i
)/(ZA
i
), eq 16 reduces to
It can be seen from eq 17 that the value of ∆GAB is always
in between the maximum and minimum of the individual
∆GAB
i
values. The requirement that ∑i
n
[i]A
) 1 ensures that ∆GAB
) ∆GAB
i
if all individual ∆GAB
i
values are equal. Application of
eq 17 to the values for ∆GSR
anti
and ∆GSR
syn
yields estimates that
correspond up to the second decimal to the values of ∆GSR in
Table 1.
We also derive similar relationships for relative free energy
differences, such as the ones reported for differences in solvation
or lipophilicity. The relative solvation free energy ∆∆GAB(solV)
can be obtained using eq 6:
In the same notation as above, we can express ∆∆GAB(solV)
in terms of the differences in free energies over the individual
conformations in water (∆GAB
iA
(aq)) and in vacuum (∆GAB
iA
(V))
using eq 16:
Note that conformation iA
may be defined differently in
different media, but it only makes sense to consider it for
compounds in the same medium. Therefore, we will indicate
the solvation state by using (aq) or (V) only once for every
symbol. GB
iB
(V) then represents the free energy of compound B
in vacuum, confined to conformation iB
as it is defined in
vacuum. Again, this equation becomes simpler if the definition
of the conformational states is the same for both compounds
(still allowing for different conformations in different media).
That is, if iA
(aq) ) iB
(aq) ) i(aq) and iA
(V) ) iB
(V) ) i(V), then
we obtain the following from eq 17:
An alternative relationship between the absolute and the
relative free energy differences can be obtained from eqs 11
and 18 and multiplying the individual terms by unity in the
form (ZA
i
(aq)ZB
j
(V))/(ZA
i
(aq)ZB
j
(V)):
Using eq 9 and 12, we can write this as
Equation 22 is similar to eq 17 in the sense that the value of
∆∆GAB(solV) will be in between the maximum and minimum
of the individual (∆GAB
i
(aq) - ∆GAB
j
(V)) values. Again, the
normalization condition ∑i(aq)
n(aq)
∑j(V)
n(V)
[i](aq)
A
[j](V)
B
) ∑i(aq)
n(aq)
[i](aq)
A
∑j(V)
n(V)
[j](V)
B
) 1 ensures that if all differences (∆GAB
i
(aq) - ∆GAB
j
(V))
are equal, they will also be equal to the relative solvation
free energy difference over the complete conformational
space (∆∆GAB(solV)).
In case the individual conformations are the same for
compounds A and B in both media iA
(aq) ) iB
(aq) ) iA
(V) )
iB
(V) ) i, we can write eq 18 as
Multiplication of the individual terms by unity in the form
(ZA
i
(V)ZA
i
(aq)ZB
i
(V))/(ZA
i
(V)ZA
i
(aq)ZB
i
(V)) and applying eqs 6, 9, and
12 then yields
Note that in most cases ∑i
n
([i](aq)
A
[i](V)
B
)/([i](V)
A
) will not add up
to 1. Therefore, even if all individual ∆∆GAB
i
(solV) are equal,
eq 24 does not necessarily lead to the same value for the overall
∆GAB ) -kBTln
(1
nA ∑
iB
nB
∑
iA
nA
[iA
]A
e-(GB
iB
-GA
iA
)/kBT
) (16)
∆GAB ) -kBTln( ∑
i
n
[i]A
e-(GB
i -GA
i )/kBT
)
) -kBTln( ∑
i
n
[i]A
e-∆GAB
i /kBT
) (17)
∆∆GAB(solV) ) ∆GAB(aq) - ∆GAB(V)
) -kBTln
ZB(aq)ZA(V)
ZA(aq)ZB(V)
(18)
∆∆GAB(solV) )
-kBTln
( 1
nA
(V)
∑
iB(V)
nB(V)
∑
iA(V)
nA(V)
[iA
](V)
A
e-(GB
iB
(V)-GA
iA
(V))/kBT
1
nA
(aq)
∑
iB(aq)
nB(aq)
∑
iA(aq)
nA(aq)
[iA
](aq)
A
e-(GB
iB
(aq)-GA
iA
(aq))/kBT)(19)
∆∆GAB(solV) ) -kBTln
(∑
i(V)
n(V)
[i](V)
A
e-∆GAB
i (V)/kBT
∑
i(aq)
n(aq)
[i](aq)
A
e-∆GAB
i (aq)/kBT)(20)
∆∆GAB(solV) ) -kBTln
∑
i(aq)
n(aq)
ZB
i
(aq) ∑
j(V)
n(V)
ZA
j
(V)
ZA(aq)ZB(V)
) -kBTln
(∑
i(aq)
n(aq)
∑
j(V)
n(V)
ZB
i
(aq)ZA
j
(V)
ZA(aq)ZB(V)
ZA
i
(aq)ZB
j
(V)
ZA
i
(aq)ZB
j
(V))(21)
∆∆GAB(solV) )
-kBTln( ∑
i(aq)
n(aq)
∑
j(V)
n(V)
[i](aq)
A
[j](V)
B
e-(∆GAB
i (aq)-∆GAB
j (V))/kBT
) (22)
∆∆GAB(solV) ) -kBTln
ZA(V) ∑
i
n
ZB
i
(aq)
ZA(aq)ZB(V)
) -kBTln(∑
i
n
ZA(V)ZB
i
(aq)
ZA(aq)ZB(V)) (23)
∆∆GAB(solV) ) -kBTln
(∑
i
n
[i](aq)
A
[i](V)
B
[i](V)
A
e-∆∆GAB
i (solV)/kBT
)(24)
12718 J. Phys. Chem. B, Vol. 113, No. 38, 2009 Hritz and Oostenbrink
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
∆∆GAB(solV). This also explains why ∆∆GAB (solV) can be
larger than both ∆∆GAB
anti
(solV) and ∆∆GAB
syn
(solV) as was observed
for compound 8-CH3-GTP relative to GTP (Table 2).
Excellent agreement is observed between eqs 20, 22, and 24
and the differences in solvation free energies and lipophilicities
listed in Table 2. The set of eqs 16, 17, 19, 20, 22, and 24
relates the absolute and relative free energy differences for
individual conformational states to values for the complete
ensemble. We will refer to them as the multiconformational free
energy formulas. Equations 17 and 24 are graphically represented
in Figure 6 for the free energy differences in water and
vacuum resulting in solvation free energies. Of course, the
multiconformational free energy formulas have general validity
for any environment. When comparing free energy differences
in water and octanol, differences in lipophilicities can be
calculated, and when comparing a water and a protein environment,
differences in binding affinities are obtained. In this
context, the different conformational states iA
(protein), iB
(protein)
in the multiconformational free energy formulas may also refer
to different relevant binding modes.
It is important to understand that here we calculated the free
energy differences over the complete conformational space by
two independent approaches. The first one, was a direct
calculation by ES-OS using a reference state which samples
the complete conformational space sufficiently. The second
approach builds on ideas of configuration space decomposition6
uses of the multiconformational free energy formulas which
require only free energy differences for individual conformations
and conformational populations of one compound (for eqs 16,
17, 19, and 20) in certain environments. While the first approach
is more efficient, it is applicable only to sets of very similar
compounds. On the other hand, the second approach has a much
wider range of applicability. For example, the calculation of
free energy differences between GTP analogs with larger and
more diversely substituted groups cannot be addressed by OS
perturbation approaches.47
A large change in the Hamiltonian
requires more steps of perturbation (e.g., within FEP29
or TI48
)
between individual compounds, and for each of these steps,
sufficient conformational sampling is required. One possible
solution by using H-REMD is discussed in the introduction.
On the other hand, the multiconformational free energy formulas
allow for a much easier solution, in which only the conformational
populations for one selected compound (soft or real) and
the free energy differences for individual conformational states
are needed. The populations can be obtained from a specifically
designed soft reference state or by using enhanced sampling
methods like REMD,49,50
umbrella sampling,51
local elevation,52
hidden restraints,12
and so forth. The free energy differences
within one conformational space are generally converging very
fast because there is no need to cross high energy barriers.
Conclusions
This paper presents the enhanced sampling-one step (ES-OS)
method, which characterizes a set of chemically similar
compounds with high intramolecular energy barriers structurally
and energetically. The method was applied to a set of five
8-substituted GTP analogs. An unphysical reference state was
constructed based on 8-Br-GTP where the Br atom was treated
as a “soft atom”; that is, all nonbonded interactions involving
Br are treated as soft-core interactions. This allowed us to
perturb the reference state into the real states of 8-substituents
of GTP. Another set of soft-core interactions was applied
between the atoms of the base and the ribose allowing the
reference state to cross high energy barriers. The final step of
ES-OS is a one-step perturbation to the real compounds using
Zwanzigs formula, thus obtaining free energy differences
between the reference and the real compounds. Combining free
energy differences between 8-substituted GTPs in water,
vacuum, and octanol enabled us to calculate differences in
solvation and lipophilicity. Using the probability of individual
structures of the MD simulation, we constructed distributions
of the dihedral angle around the glycosidic bond, the 3
J(C4,H1′)
values, and the base-water and base-sugar nonbonded energy.
The comparison of these distributions for GTP and 8-Br-GTP
in water shows a good agreement with the ones calculated by
H-REMD using soft-core interactions. Also, the ensemble
average 〈3
J(C4,H1′)〉 of GTP in water is in very good agreement
with the experimental NMR value for GMP.
All free energy differences were calculated over the complete
conformational space as well as separately for the anti and syn
conformations. In analogy to the conformational space decomposition
approach, the relationship between them was derived,
which deepens our understanding of how free energy differences
from individual conformations contribute to the overall values.
The relationships have a significant practical potential also for
more diverse compounds because the calculation of free energy
differences within one conformation is in many cases much
simpler as there is no need to cross high energy barriers.
Acknowledgment. We gratefully acknowledge financial
support from The Netherlands Organization for Scientific
Research (NWO), VENI Grant 700.55.401 (CO) and Horizon
Breakthrough Grant 935.18.018 (JH).
Supporting Information Available: Force field parameters
for 8-substituted analogs of GTP. This material is available free
of charge via the Internet at http://pubs.acs.org.
References and Notes
(1) Simonson, T.; Archontis, G.; Karplus, M. Acc. Chem. Res. 2002,
35, 430.
(2) Brandsdal, B. O.; Osterberg, F.; Almlof, M.; Feierberg, I.; Luzhkov,
V. B.; Aqvist, J. AdV. Protein Chem. 2003, 66, 123.
(3) Raha, K.; Kenneth, M.; Merz, J. Ann. Rep. Comput. Chem. 2005,
1, 113.
(4) Rodinger, T.; Pomes, R. Curr. Opin. Struct. Biol. 2005, 15, 164.
(5) Shirts, M. R.; Mobley, D. L.; Chodera, J. D. Ann. Rep. Comput.
Chem. 2007, 3, 41.
(6) Straatsma, T. P.; McCammon, J. A. J. Chem. Phys. 1989, 90, 3300.
(7) Straatsma, T. P.; McCammon, J. A. J. Chem. Phys. 1989, 91, 3631.
(8) Tobias, D. J.; Brooks, C. L.; Fleischman, S. H. Chem. Phys. Lett.
1989, 156, 256.
(9) Hermans, J.; Yun, R. H.; Anderson, A. G. J. Comp. Phys. 1992,
13, 429.
(10) Leitgeb, M.; Schroder, C.; Boresch, S. J. Chem. Phys. 2005, 122,
084109.
Figure 6. Graphical representation of multiconformational free energy
formulas (eqs 17 and 24) relating absolute and relative free energy
differences for compounds A and B in conformations i and j in aqueous
environment and vacuum.
Efficient Free Energy Calculations J. Phys. Chem. B, Vol. 113, No. 38, 2009 12719
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
(11) Mobley, D. L.; Chodera, J. D.; Dill, K. A. J. Chem. Theory &
Comput. 2007, 3, 1231.
(12) Christen, M.; Kunz, A.-P. E.; Van Gunsteren, W. F. J. Phys. Chem.
B 2006, 110, 8488.
(13) Hritz, J.; Oostenbrink, C. J. Chem. Phys. 2008, 128, 144121.
(14) Stolarski, R.; Hagberg, C. E.; Shugar, D. Eur. J. Biochem. 1984,
138, 187.
(15) Bitetti-Putzer, R.; Yang, W.; Karplus, M. Chem. Phys. Lett. 2003,
377, 633.
(16) Min, D. H.; Li, H. Z.; Li, G. H.; Bitetti-Putzer, R.; Yang, W.
J. Chem. Phys. 2007, 126, 144109.
(17) Beutler, T. C.; Mark, A. E.; van Schaik, R. C.; Gerber, P. R.; van
Gunsteren, W. F. Chem. Phys. Lett. 1994, 222, 529.
(18) Hritz, J.; Oostenbrink, C. J. Chem. Phys. 2007, 127, 204104.
(19) Liu, H. Y.; Mark, A. E.; van Gunsteren, W. F. J. Phys. Chem.
1996, 100, 9485.
(20) Oostenbrink, C.; van Gunsteren, W. F. Proc. Natl. Acad. Sci. U.S.A.
2005, 102, 6750.
(21) La¨pp chen, T.; Hartog, A. F.; Pinas, V. A.; Koomen, G. J.; den
Blaauwen, T. Biochemistry 2005, 44, 7879.
(22) Lappchen, T.; Pinas, V. A.; Hartog, A. F.; Koomen, G. J.; SchaffnerBarbero,
C.; Andreu, J. M.; Trambaiolo, D.; Lowe, J.; Juhem, A.; Popov,
A. V.; den Blaauwen, T. Chem. Biol. 2008, 15, 1.
(23) Dias, E.; Battiste, J. L.; Williamson, J. R. J. Am. Chem. Soc. 1994,
116, 4479.
(24) Proctor, D.; Kierzek, E.; Kierzek, R.; Bevilacqua, P. J. Am. Chem.
Soc. 2003, 125, 2390.
(25) Yajima, R.; Proctor, D. J.; Kierzek, R.; Kierzek, E.; Bevilacqua,
P. C. Chem. & Biol. 2007, 14, 23.
(26) Schwede, F.; Maronde, E.; Genieser, H. G.; Jastorff, B. Pharmacol.
& Ther. 2000, 87, 199.
(27) Ghosh, A.; Lagisetty, P.; Zajc, B. J. Org. Chem. 2007, 72, 8222.
(28) Liu, J.; Barrio, J.; Satyamurthy, N. J. Fluor. Chem. 2006, 127, 1175.
(29) Zwanzig, R. W. J. Chem. Phys. 1954, 22, 1420.
(30) Mark, A. E.; Xu, Y. W.; Liu, H. Y.; van Gunsteren, W. F. Acta
Biochimica Polonica 1995, 42, 525.
(31) Oostenbrink, C.; van Gunsteren, W. F. Chem.sEur. J., in press.
(32) Christen, M.; Hunenberger, P. H.; Bakowies, D.; Baron, R.; Burgi,
R.; Geerke, D. P.; Heinz, T. N.; Kastenholz, M. A.; Krautler, V.;
Oostenbrink, C.; Peter, C.; Trzesniak, D.; Van Gunsteren, W. F. J. Comput.
Chem. 2005, 26, 1719.
(33) Ryckaert, J.-P.; Ciccotti, G.; Berendsen, H. J. C. J. Comput. Phys.
1977, 23, 327.
(34) Hockney, R. W. Meth. Comput. Phys. 1970, 9, 136.
(35) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.;
Hermans, J. Interaction models for water in relation to protein hydration.
In Intermolecular Forces; Pullman, B., Ed.; Reidel: Dordrecht, The
Netherlands, 1981; pp 331.
(36) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; DiNola,
A.; Haak, J. R. J. Chem. Phys. 1984, 81, 3684.
(37) Tironi, I. G.; Sperb, R.; Smith, P. E.; van Gunsteren, W. F. J. Chem.
Phys. 1995, 102, 5451.
(38) Komooka, H. Bull. Chem. Soc. Jpn. 1972, 45, 1696.
(39) Oostenbrink, C.; Villa, A.; Mark, A. E.; van Gunsteren, W. F.
J. Comput. Chem. 2004, 25, 1656.
(40) Allen, M. P.; Tildesley, D. J. Computer simulation of liquids;
Clarendon Press: Oxford, 1987.
(41) Scha¨fer, H.; van Gunsteren, W. F.; Mark, A. E. J. Comput. Chem.
1999, 20, 1604.
(42) Trantirek, L.; Stefl, R.; Masse, J. E.; Feigon, J.; Sklenar, V.
J. Biomol. NMR 2002, 23, 1.
(43) Schwalbe, H.; Marino, J. P.; King, G. C.; Wechselberger, R.;
Bermel, W.; Griesinger, C. J. Biomol. NMR 1994, 4, 631.
(44) Wildman, S. A.; Crippen, G. M. J. Chem. Inf. Comput. Sci. 1999,
39, 868.
(45) Molecular Operating Environment, v., Chemical Computing Group,
Montreal, Canada.
(46) Ferner, J.; Villa, A.; Duchardt, E.; Widjajakusuma, E.; Wohnert,
J.; Stock, G.; Schwalbe, H. Nucleic Acids Res. 2008, 36, 1928.
(47) Oostenbrink, C.; van Gunsteren, W. F. J. Comput. Chem. 2003,
24, 1730.
(48) Kirkwood, J. G. J. Chem. Phys. 1935, 3, 300.
(49) Hansmann, U. H. E. Chem. Phys. Lett. 1997, 281, 140.
(50) Sugita, Y.; Okamoto, Y. Chem. Phys. Lett. 1999, 314, 141.
(51) Torrie, G. M.; Valleau, J. P. J. Comput. Phys. 1977, 23, 187.
(52) Huber, P.; Torda, A. E.; Van Gunsteren, W. F. J. Comput.-Aided
Mol. Design 1994, 8, 695.
JP902968M
12720 J. Phys. Chem. B, Vol. 113, No. 38, 2009 Hritz and Oostenbrink
DownloadedbyAMSTERDAMVRIJEUNIVonSeptember17,2009|http://pubs.acs.org
PublicationDate(Web):September1,2009|doi:10.1021/jp902968m
ALCHEMICAL (RELATIVE) FREE ENERGY CALCULATIONS
53
Paper 10
Jandova, Z.; Trosanova, Z.; Weisova, V.; Oostenbrink, C.*, Hritz, J.*:Free energy calculations
on the stability of the 14-3-3 protein. BBA - Proteins and Proteomics, 2018,
1866, 442-450.
Contents lists available at ScienceDirect
BBA - Proteins and Proteomics
journal homepage: www.elsevier.com/locate/bbapap
Free energy calculations on the stability of the 14-3-3ζ protein
Zuzana Jandovaa
, Zuzana Trosanovab
, Veronika Weisovab
, Chris Oostenbrinka,⁎
, Jozef Hritzb,⁎
a
Institute of Molecular Modeling and Simulation, University of Natural Resources and Life Sciences, Vienna, Austria
b
CEITEC-MU, Masaryk University, Kamenice 753/5, Bohunice, Brno, Czech Republic
A R T I C L E I N F O
Keywords:
14-3-3 protein
Protein stability
Molecular dynamics simulation
Differential scanning calorimetry
Free energy calculation
Thermodynamic integration
A B S T R A C T
Mutations of cysteine are often introduced to e.g. avoid formation of non-physiological inter-molecular disulfide
bridges in in-vitro experiments, or to maintain specificity in labeling experiments. Alanine or serine is typically
preferred, which usually do not alter the overall protein stability, when the original cysteine was surface exposed.
However, selecting the optimal mutation for cysteines in the hydrophobic core of the protein is more
challenging. In this work, the stability of selected Cys mutants of 14-3-3ζ was predicted by free-energy calculations
and the obtained data were compared with experimentally determined stabilities. Both the computational
predictions as well as the experimental validation point at a significant destabilization of mutants C94A and
C94S. This destabilization could be attributed to the formation of hydrophobic cavities and a polar solvation of a
hydrophilic side chain. A L12E, M78K double mutant was further studied in terms of its reduced dimerization
propensity. In contrast to naïve expectations, this double mutant did not lead to the formation of strong salt
bridges, which was rationalized in terms of a preferred solvation of the ionic species. Again, experiments agreed
with the calculations by confirming the monomerization of the double mutants. Overall, the simulation data is in
good agreement with experiments and offers additional insight into the stability and dimerization of this important
family of regulatory proteins.
1. Introduction
The 14-3-3 proteins form a family of acidic proteins with molecular
weight of approximately 30 kD [1] ubiquitous in all eukaryotic organisms.
There are seven highly conserved isoforms of the 14-3-3 family in
mammals, two in yeast and fifteen in plants [2–4]. The 14-3-3 proteins
function mostly as hetero and homo dimers [5,6], with each of the
monomers consisting of nine antiparallel helices. The complete structure
of the protein is shown in Fig. 1. The first four helices, starting from
the N-terminus, create a hydrophobic dimer interface and the floor of
the binding channel. The next 5 helices create the amphipathic interior
of the peptide-binding channel [7,8]. Residues aligning the channel
belong to the most highly conserved residues among all isoforms, while
the sequence of the surface and flexible C-terminal residues is more
diverse and responsible for isoform-substrate specificity [1,5,9].
14-3-3 proteins play a key role in regulating intracellular signaling
pathways, common cellular processes, primary metabolism, antiapoptotic
pathways and cellular proliferation [10,11]. They act as
adapters for > 200 binding partners by either allosteric modification of
enzymes or controlling the assembly of protein complexes [12–16].
Among the best known 14-3-3 binding partners are tyrosine hydroxylase,
Raf-1 (RAF proto-oncogene ser/thr-protein kinase), BCR (B-cell
receptor), PKC (protein kinase C), MLK (mixed lineage ser/thr kinase),
Tau protein or p53 (cellular tumor antigen p53) [16–18]. Their protein
binding partners usually contain intrinsically disordered regions, typically
containing motifs involving phosphoserine or phosphothreonine
with a consensus sequence RSXpSXP (mode I) or RX[FY]XpSXP (mode
II) [19]. Enhanced sampling molecular dynamics (MD) simulations of
the phosphopeptides binding to 14-3-3ζ showed the presence of a
dominant binding pathway, roughly following helix 3. In the process of
phosphopeptide binding, 14-3-3ζ was observed to adopt a so-called
“wide-opened” conformation that is not usually present in the apo or
holo states [20]. 14-3-3 dimers contain two binding sites and are
therefore able to accommodate also doubly-phosphorylated binding
partners. In addition to the traditionally considered single partner with
two phosphorylation sites occupying the individual binding cavities
within the 14-3-3ζ dimer [21,22], two more major binding modes were
confirmed by 31
P NMR titration experiments [23].
In this paper, we focus on computational predictions of the stability
of the 14-3-3ζ isoform upon mutations, using MD simulations and
thermodynamic integration (TI) to compute free-energy differences.
Our computational predictions are compared to experimentally determined
melting temperatures (TM) as obtained from differential
scanning calorimetry and to native gel electrophoresis experiments, as
https://doi.org/10.1016/j.bbapap.2017.11.012
Received 29 August 2017; Received in revised form 31 October 2017; Accepted 25 November 2017
⁎
Corresponding authors.
E-mail addresses: chris.oostenbrink@boku.ac.at (C. Oostenbrink), jozef.hritz@ceitec.muni.cz (J. Hritz).
BBA - Proteins and Proteomics 1866 (2018) 442–450
Available online 05 December 2017
1570-9639/ © 2017 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/).
T
indirect measures of protein stability and dimerization [24]. While the
thermodynamic stability as computed from our simulations is not expected
to quantitatively correlate with TM, both are measures of stability
of the protein [25]. A comparison of homologous proteins from
mesophillic and thermophilic organisms e.g. reveals that for the majority
of cases, the melting temperature is increased as the result of an
increase of the unfolding free energy at the entire temperature range
[26].
Mutations of cysteine may be introduced for multiple practical
reasons. Cysteine contains a reactive sulfhydryl group, allowing for the
formation of disulfide bridges. However, cysteines that are not involved
in intramolecular disulfide bridges may be prone to unwanted oxidation
or crosslinking under experimental conditions. On the other hand,
various experimental methods make use of specific labels that are anchored
to the protein (e.g. chromopohoric or spin labels) to gain insight
into the function and dynamics of proteins. To maintain the specificity
in labeling, other cysteines, naturally occurring in the sequence need to
be replaced by other amino acids preferably without altering the overall
protein stability. This is commonly done by a replacement by alanine or
serine. While mutating the surface exposed single cysteines for alanines
or serines usually does not alter the overall protein stability, it is much
more complicated when mutating the cysteins in the hydrophobic core
of the protein. The 14-3-3ζ protein contains three cysteines, of which
two (Cys25 and Cys189) are located on the protein surface and their
mutagenesis for the Ala has negligible impact on the protein stability
[23]. The third, Cys94, is located in the central helix H4 and buried in a
hydrophobic pocket, therefore its mutation influences the overall stability
of the protein. Here, we investigate the possibility to propose
protein mutants by MD simulations that would sustain or improve
protein stability upon cysteine mutation and verify our results by experimental
melting temperatures. Apart from the most common choices
C94A and C94S, we have further considered three more hydrophobic
mutants, C94I, C94L and C94V, based on the location of the residue in
the hydrophobic core of the protein.
Further, we investigate a double mutation of Leu12 and Met78,
which are located at the dimer interface and influence protein dimerization.
In the 14-3-3ζ homodimer several salt bridges as well as buried
polar and hydrophobic residues are engaged in the formation of a stable
dimer interface [8]. The double mutant Leu12Glu/Met78Lys was
identified to destabilise this interface and thereby hinder dimer formation
while at the same time it does not change the overall charge of
the 14-3-3ζ monomeric unit. As the mutations are in a position that
would allow for stable salt bridges, a destabilization of the dimer interface
seems somewhat counter intuitive. Here, our aim is to rationalize
the experimental results by molecular simulations.
For the calculation of free energy differences, we make use of
thermodynamic cycles. The one we use for the mutations of Cys94 can
be seen in Fig. 2. In this figure, the dimeric 14-3-3ζ is indicated by two
roughly L-shaped figures with the wildtype residue (Cys94) indicated
with a triangle. A hypothetical unfolded protein is indicated by a wavy
bar at the bottom of the cycle. Mutations are implied by the white
squares.
Considering that the Gibbs free energy is a state function [27] the
value around the cycle is equal to zero. This allows us to write:
∆∆ = ∆ − ∆ = ∆ − ∆
= ∆∆
G D G G D G D G D
G D
( ) 2 ( ) ( ) ( )
( )
mut mut
u
mut
f
fold
WT
fold
mut
fold (1)
where ΔΔGmut(D) represents the relative free energy difference of mutation
of one aminoacid into another, which can be computed as the
difference of the mutation free energy in the dimer, ΔGmut
f
(D) and in
the unfolded state ΔGmut
u
. The same quantity can be written as
ΔΔGfold(D), which is the relative free energy of folding and is the difference
of the folding free energy of the wildtype, ΔGfold
WT
(D), and of
the mutant, ΔGfold
mut
(D). The addition (D) explicitly refers to the dimeric
state. Current computational resources do not allow us to simulate
the full process of protein folding and unfolding to determine
ΔΔGfold(D) directly. However, we are able to estimate this value from
free energy perturbation in a folded and unfolded state [28,29]. Selection
of the right representation of the unfolded state is crucial for
free energy calculations [30]. We have decided to use a tripeptide with
its central residue being mutated and two neighbouring residues resembling
the ones in protein context, as was previously shown to agree
well with experimental results [31]. For the folded state, we simulated
both the monomer and the dimer to obtain a better understanding of
the impact of dimerization on the stability of the protein.
Fig. 1. 14-3-3ζ protein with helices H1-H9 marked in one
monomer and the mutated residues L12, M78 and C94 indicated
in stick representation. The unstructured C-terminus
(residues 229–245) was removed because of its high
flexibility.
Fig. 2. Thermodynamic cycle for the mutation of Cys94. The upper arrow corresponds to
the free energy of mutation in the dimer, the lower arrow in the unfolded state. Grey
triangles represent WT Cys94, white squares the mutants.
Z. Jandova et al. BBA - Proteins and Proteomics 1866 (2018) 442–450
443
2. Methods
2.1. Experimental
2.1.1. Cloning, expression, and purification of 14-3-3ζ variants
A codon-optimized gene for 14-3-3ζ (GenScript, Piscataway
Township, NJ) was inserted into pET15b and two mutations C25A and
C189A were inserted. Due to the fact that these mutations have negligible
impact on the protein stability (TM is slightly increased from
59.5 °C to the 60.3 °C) and avoid the formation of any intermolecular
artificial disulfide bonds, we used this construct as a 14-3-3ζ pseudoWT.
The 14-3-3ζ gene with the C25A and C189A mutations was
therefore used as the “parental” construct for all other studied mutations,
which were introduced by using the QuikChange Lightning SiteDirected
Mutagenesis Kit (Agilent Technologies, Santa Clara, CA, USA).
The DNA sequence for all constructs was confirmed by sequencing
(Macrogene, Seoul, Rep. of Korea). All 14-3-3ζ variants were expressed
and purified by procedure described in Hritz et al. [23]. Final purity of
the prepared protein variants was verified by MALDI-TOF-MS spectro-
scopy.
2.1.2. Differential scanning calorimetry (DSC)
Melting temperatures (TM) were obtained from DSC measurements
carried out at a VP-DSC instrument (MicroCal Inc., Northampton, MA)
at a heating scan rate of 1 °C per minute from 25 to 80 °C. Samples of
14-3-3ζ C94 mutants were prepared at 1 mg/ml concentration and
dialyzed against a degassed 20 mM sodium phosphate buffer (pH 6.0).
DSC data were analyzed using the Microcal Origin 7.0 software
(MicroCal Inc., Northampton, MA).
2.1.3. Native polyacrylamide gel electrophoresis (Native-PAGE)
10 μl of 2 μM protein sample were mixed with 10 μl of 2× loading
buffer containing 2% glycerol, 2% Bromophenol Blue, and 160 mM
TRIS-HCl (pH = 6.8). Samples were then loaded into a freshly prepared
12.5% native minigel. Electrophoresis was conducted for approximately
200 min using constant voltage (95 V) in a native electrophoretic buffer
(2.5 mM TRIS-HCl, 19.2 mM Glycine, pH 8.3). The apparatus was
cooled down on ice during whole procedure in order to prevent the
thermal denaturation of the studied proteins. Finally, the gel was
stained by Coomassie Brilliant Blue R-250 (AppliChem, Darmstadt,
Germany).
2.2. MD simulations
As starting configuration for molecular dynamics simulations we
used an apo 14-3-3ζ crystal structure (PDB ID:1A4O [8]) at a 2.8 Å
resolution. Missing loops and residues were remodelled from a holo
crystal structure (PDB ID:4HKC [32]). C-terminal residues 229–245
were removed from the structure to avoid interactions of the highly
flexible and unstructured C-terminus with the rest of the protein. In
preliminary simulations, a high strain in the backbone of Arg18 was
observed, leading to failures of the SHAKE algorithm to constrain bond
lengths. It appeared that the ϕ and ψ angles of Glu17 were at the very
bottom of the left-handed helical configuration in the Ramachandran
plot (ϕ = 73°, ψ = −16°), leading to strain in the loop. By turning ψ of
A16 and ϕ of E17 by approximately 180°, the strain was released and no
further SHAKE failures were observed. All arginines, lysines and cysteines
were protonated, while aspartates and glutamates were deprotonated.
His164 was protonated on Nε, which was based on better hydrogen
bonding possibilities in its surrounding. All MD simulations
were carried out using the GROMOS11 software simulation package
[33], employing the 54a8 forcefield [34]. Proteins were energy-minimized
in vacuum using the steepest-descent algorithm and subsequently
solvated in a rectangular, periodic and pre-equilibrated box of
single point charge (SPC) water [35]. Minimum solute to box-wall
distances were set to 2, 2, and 2.5 nm for dimer and 1, 1 and 2.5 nm for
monomer in the x,y and z-dimensions, respectively, after alignment of
the largest intramolecular distance with the z-axis. This led to systems
containing about 50 to 60 thousand atoms for the monomer and 110 to
117 thousand atoms for the dimer. Another minimization in water was
performed using the steepest descent algorithm. To achieve electroneutrality
of the system with monomer and dimer 8 and 16 sodium ions
were added to the system, respectively. Because previous simulations
on very similar systems were shown to have increased structural stability
at higher salt concentrations [20], simulations were additionally
performed at a concentration of 300 mM NaCl. To achieve this, additional
94–127 sodium and chloride ions were added to the box of
monomer and 209–219 sodium and chloride ions were added to the box
of dimer. For the equilibration, the following protocol was used: initial
velocities were randomly assigned according to a Maxwell–Boltzmann
distribution at 60 K. All solute atoms were positionally restrained with a
harmonic potential using a force constant of 2.5 × 104
kJ/mol nm−2
.
In each of the four subsequent 20 ps MD simulations, the force constant
of the positional restraints was reduced by one order of magnitude and
the temperature was increased by 60 K. Subsequently, the positional
restraints were removed and rototranslational constraints were introduced
on all solute atoms [36]. The last step of equilibration was
performed at a constant pressure of 1 atm for 300 ps. After equilibration
production runs of 60 ns were performed with constant number of
particles, constant temperature (300 K) and constant pressure (1 atm).
To sustain a constant temperature, we used the weak-coupling thermostat
[37] with a coupling time of 0.1 ps. The pressure was maintained
using a weak coupling barostat with a coupling time of 0.5 ps
and an isothermal compressibility of 4.575 × 10−4
kJ−1
mol nm−3
.
Solute and solvent were coupled to separate temperature baths. Implementation
of the SHAKE algorithm [38] to constrain bond lengths of
solute and solvent to their optimal values allowed for a 2-fs time-step.
Nonbonded interactions were calculated using a triple range scheme.
Interactions within a short-range cutoff of 0.8 nm were calculated at
every time step from a pair list that was updated every fifth step. At
these points, interactions between 0.8 and 1.4 nm were also calculated
explicitly and kept constant between updates. A reaction field [39]
contribution was added to the electrostatic interactions and forces to
account for a homogenous medium outside the long-range cutoff using
a relative dielectric constant of 61 as appropriate for the SPC water
model [40]. Coordinate and energy trajectories were stored every 0.5 ps
for subsequent analysis. To study the overall stability of the WT protein,
a total of eight independent simulations were performed: Two independent
simulations of both monomer and dimer in both 0 and
300 mM NaCl, further noted as MD1 and MD2.
2.3. Thermodynamic integration (TI)
Using thermodynamic integration, the free energy difference between
two states A and B can be computed via multiple discrete intermediate
steps using a coupling parameter λ. If λ is equal to 0, the
system is in state A and, on the other hand the system is in state B when
λ is equal to 0. At each intermediate step the average of the derivative
of the Hamiltionian with respect to λ is calculated. By integrating over
the derivatives along the path we obtain the total free energy difference
(ΔGBA).
∫=
∂
∂
G
H λ
λ
dλΔ
( )
BA
λ
0
1
(2)
TI calculations for perturbations of selected amino acid were initially
performed at 11 evenly spaced λ steps. Starting from initial
structures taken after above mentioned equilibration with low salt
content, 20 ps of equilibration at each λ value were followed by 1 ns of
production run. If necessary, the simulations were prolonged to maximally
5 ns per λ point or additional λ points were added to decrease the
overall error estimate below 3 kJ/mol. For perturbed atoms we used
Z. Jandova et al. BBA - Proteins and Proteomics 1866 (2018) 442–450
444
soft-core parameters of 0.5 for the van der Waals and 0.5 nm2
for
electrostatic interactions [41]. A single topology approach was used, by
introducing dummy atoms where necessary. Dummy atoms have all
nonbonded interactions set to zero while the bonded interactions and
the masses of individual atoms remain the same as for real atoms. This
way we can selectively convert real atoms into dummy atoms and vice
versa to switch between particular residues. Fig. 3 shows the complete
list of structural formulas of perturbed residues, including the dummy
atoms, for both sets of mutations.
3. Results and discussion
3.1. Overall protein stability
First, we investigated the overall stability of 14-3-3ζ as WT protein.
We performed plain MD simulations of monomers and dimers in 0 and
300 mM NaCl for 60 ns. For each of the simulation two independent
simulations were performed, noted as MD1 and MD2. The atom-positional
root-mean-square deviation of Cα atoms per monomer increased
up to a maximum of 1 nm after 30 ns of the dimer simulation MD1.
Time series of the root-mean-square deviations of all simulations can be
found in the Supplementary Material (Fig. S1). The root-mean-square
fluctuations of Cα atoms (Fig. 4) as well as the secondary structure
analysis (Fig. 5) shows, that the monomer simulations are characterized
by a lower stability, especially in the first three helices H1 to H3 (residues
1–70), than the dimer simulations. In contrast, the last three
helices (residues 162–228) seem to be less stable in simulation of dimers,
which is in agreement with previous studies [20].
For the secondary structure analysis we used the DSSP algorithm
(Define Secondary Structure of Proteins) [42]. Fig. 5 shows the average
secondary structure propensity per residue for all simulations, i.e. MD1
and MD2 in 0 and 300 mM NaCl for the monomer simulations and
averaged over both monomers for the dimer simulations. In the dimer
simulation, these helices keep their α-helical secondary structure for
almost the entire simulation, whereas in the monomer simulation we
observe unfolding of the first three helices. Exactly these three helices
are necessary for maintaining the dimer interface. After monomer separation,
the interface residues become unstable which, after some
time, results in a disruption of the secondary structure.
Apart from the changes in the local secondary structure of the
protein a global opening and closing of the binding channel occurs. As
previously described [20,43], the distance between the third and eighth
helix fluctuates significantly. Fig. 6 shows the distance between Cα
carbons of Gly53 and Leu191. This distance was previously suggested to
describe the openness of 14-3-3ζ. While the opening distance of
monomers in the dimer simulations seems to yield different distributions
for every simulation, all four monomer simulations consistently
show the maximum in the distance distribution at 2.8 nm. In particular
the simulations of dimers at a low salt concentration seem to be more
diverse in the openness of the binding channel.
Furthermore, similarly to the previous work of Nagy et al. [20], an
inter-monomer twist of 14-3-3ζ was observed. It means that one of the
monomers is rotating with respect to the other by up to 80°. The time
series of a dihedral angle showing this twist, represented by the dihedral
angle Leu43(M1)-Ala54(M1)-Ala54(M2)-Leu43(M2), where M1
and M2 represents monomer 1 and 2 respectively, can be found in the
Supplementary material (Fig. S2). Our analysis shows, that a large
variability is observed in simulations with salt concentration of 0 and
300 mM NaCl, with intermonomer twists ranging between 15 and 80°.
3.2. Mutation of Cys94
Cysteine 94 is located in a hydrophobic environment in the middle
of helix 3. We have perturbed this residue into 5 different amino acids:
Alanine, Isoleucine, Leucine, Valine and Serine. The structural formulas
and details of these mutations are shown in Fig. 3. The free energy
differences for TI perturbations in Ile-Cys-Asn tripeptide as well as in
protein were evaluated based of the thermodynamic cycle, shown in
Fig. 2. Perturbation in the protein was carried out in both, monomer
and dimer. Fig. S3 shows the TI curves of the perturbations in tripeptide,
monomer and dimer for all five mutations. Perturbation free energies
for tripeptide (ΔGmut
u
) as well as monomer (ΔGmut
f
(M)) were
Fig. 3. Structural formulas of perturbed residues. D stands for dummy atoms.
Z. Jandova et al. BBA - Proteins and Proteomics 1866 (2018) 442–450
445
multiplied by two, to be comparable with the double perturbation in
dimer (ΔGmut
f
(D)). Table 1 summarizes the free energies obtained from
the simulations. The free energies as obtained for the monomers agree
very well with the stability predicted for the dimers. Both sets of simulations
suggest that the Ala and Ser mutations lead to a loss of stability
(ΔΔGmut
f
> 0) while the Val mutant shows comparable stability
as the wildtype and Leu and Ile increase the stability at the simulated
temperature. For comparison, the experimental melting temperatures
as determined by DSC (experimental details described in the Methods)
are also included in Table 1.
While ΔΔGmut
f
is a measure of stability at a given thermodynamic
state (here at a temperature of 300 K), the melting temperature depends
on the behavior of the protein at a range of different temperatures.
Therefore, and because the thermal unfolding of the 14-3-3 constructs
was observed to be fully irreversible, ΔΔGmut
f
and TM cannot be expected
to correlate exactly. Still, both quantities are indicative of protein
stability. In fact, a common strategy by which thermophilic proteins
increase the melting temperature is to decrease the free energy of
folding over the entire temperature range [26]. Arguably, a single point
mutation is more likely to crease the folding free energy at the entire
temperature range, than to have a significant effect on the protein heat
capacity, leading to alternative temperature dependencies.
Our simulation data agrees with the experimental stability estimates
by identifying Ser and Ala as the two least stable mutants, with
ΔΔGmut(D) of 54 kJ/mol and 18 kJ/mol, respectively. However, the
simulations predict Ile and Leu to be more stable than the WT, while the
Fig. 5. Secondary structure propensities as calculated by the DSSP program [42] for 14-3-3ζ monomer and dimer simulations, averaged over all plain WT simulations.
Fig. 4. RMSF of Cα atoms of dimers and monomers. First monomer of the dimer in red, second in black. NaCl indicates the simulations that were performed at 300 mM NaCl
concentration.
Z. Jandova et al. BBA - Proteins and Proteomics 1866 (2018) 442–450
446
experimentally determined TM is slightly lower than for the wildtype.
The large instability of Ala and Ser is striking, as these are typically the
first mutations suggested to replace Cys in experiments. Both simulations
and the experimentally determined melting temperatures show
that these particular mutations are rather unfortunate choices to replace
Cys94 in 14-3-3ζ.
Our rationalization of the destabilizing effect of the C94A and C94S
mutants is that these mutations disrupt the hydrophobic core in that
region [44]. Considering that position 94 is located in a very hydrophobic
region among three helices, any disturbance of this hydrophobicity
might lead to a decrease of protein stability. Detailed snapshots
from the last step of simulations are shown in Fig. 7. When a
hydrophobic cysteine residue [45] is replaced by a hydrophilic serine,
there are only a few possibilities for serine to create hydrogen bonds
within its surrounding. Hydrogen bonding between the sidechain
oxygen of Ser94 and the backbone oxygen of Leu90 was observed in
45% of simulation time in the last simulation of the TI simulations
(λ = 1.00). Interestingly, in one monomer we could see a hydrogen
bond between a water molecule and Ser94. While hydrogen bonding is
energetically favourable, introducing water molecules into a hydrophobic
core might lead to structural reorganization and destabilization.
On the other hand, after insertion of alanine into a densely packed core,
a voluminous hydrophobic cavity is created. It has been previously
shown, that substitution of bulky hydrophobic amino acids by smaller,
cavity-creating ones is unfavourable and leads to a loss of stability [46].
The other branched hydrophobic aminoacids, on the contrary, fill the
cavity and contribute favourably to the hydrophobic interactions in the
core and, thus, stabilise the protein.
3.3. Mutation of Leu12 and Met78
3.3.1. Experimental
Under normal circumstances, without any posttranslational modifications
the monomer/dimer equilibrium of the 14-3-3ζ isoform is
Table 1
Predicted free-energy differences, ΔG, for monomer, dimer and tripeptide simulation and their differences ΔΔG in kJ/mol. TM corresponds to the experimentally determined melting
temperatures in °C.
2ΔGmut
f
(M) ΔGmut
f
(D) 2ΔGmut
u
2ΔΔGmut
f
(M) ΔΔGmut
f
(D) TM
[kJ/mol] [°C]
Ala 55 ± 1 59 ± 1 41 ± 2 14 ± 2.0 18 ± 2 54.0
Ile 24 ± 3 29 ± 4 51 ± 3 −26 ± 4 −22 ± 5 55.6
Leu 21 ± 2 23 ± 2 34 ± 2 −13 ± 3 −11 ± 2 57.0
Ser 2 ± 1 9 ± 1 −46 ± 1 47 ± 1 55 ± 1 52.5
Val 56 ± 1 61 ± 2 60 ± 1 −4 ± 2 1 ± 2 55.0
Cys (WT) 0 0 0 0 0 60.3
Fig. 6. Monomer opening: Distances between Cα carbons of Glu53 and Leu191. Different colours in dimer graphs correspond to monomer 1 and monomer 2. NaCl indicates the
simulations that were performed at 300 mM NaCl concentration.
Z. Jandova et al. BBA - Proteins and Proteomics 1866 (2018) 442–450
447
strongly shifted towards the dimer [48]. A variety of 14-3-3 dimer-incapable
forms were reported over last two decades and they are nicely
summarized in Table 1 of the review of Sluchanko et al. [48] The listed
mutations shifting this equilibrium towards monomer change the
overall charge of 14-3-3 monomer with respect to the WT. Here, we
designed a double mutation Leu12Glu and Met78Lys which does not
change the overall charge of the 14-3-3ζ monomer and at the same time
is almost exclusively in monomeric state at a μM concentration range.
Fig. 8B shows a native PAGE gel of WT and the double mutant at 1 μM
concentration. The L12E/M78K double mutant is propagating faster
through the PAGE than WT. Considering that both of these proteins
have the same overall charge per monomer our interpretation is that
the L12E/M78K double mutant is in the monomeric form in contrast to
the dimeric form of the 14-3-3ζ WT at the given conditions.
3.3.2. Simulation
To rationalize the experimental results computationally, we have
performed in silico mutations L12E and M78K in three different systems.
One of them was, again, a tripeptide representing the unfolded
state within a protein context, Lys11-Leu12-Ala13 and Gln77-Met78Ala79.
The other two were in dimer and monomer simulations. In
contrast to the mutations applied to Cys94, the individual mutations
involve full charge changes. The L12E mutation leads to negative
charges, while the M78K mutations leads to a positive charge (see
Fig. 3). Thus, the overall charge of the system stays neutral and there is
no need for additional corrections [49]. The total scheme of the resulting
thermodynamic cycles can be seen in Fig. 8A. Here, we compute
the relative dimerization free energies, ΔΔGdim as the difference between
the dimerization free energies of the mutant and the wildtype,
which is the same as the difference in the mutation free energy in the
mutant and in the monomers,
= ∆ − ∆ = ∆ − ∆ΔΔG G G G D G M( ) 2 ( )dim dim
mut
dim
WT
mut
f
mut
f
(3)
Similarly to the Cys94 mutations, we can also define the relative
folding free energy of the monomers as
= ∆ − ∆ = ∆ − ∆ΔΔG M G M G M G M G( ) ( ) ( ) ( )fold fold
mut
fold
WT
mut
f
mut
u
(4)
where in this case the mutational free energy of the unfolded state is the
sum of two independent simulations of the tripeptides
= +ΔG ΔG L E ΔG M K( 12 ) ( 78 )mut
u
mut
u
mut
u
(5)
The folding free energy of the complete dimers is subsequently related
to the previous properties as
= ∆ +ΔΔG D ΔG ΔΔG M( ) 2 ( )fold dim fold (6)
Table 2 shows the resulting free energy differences and TI curves for
the mutations in all three systems can be found in Fig. S4. The freeFig.
8. A) Thermodynamic cycle of mutations at the 14-3-
3ζ dimeric interface. WT Leu12 and Met78 are in grey cylinder
and circle and Glu12 and Lys78 in white pentagon
and parallelogram. B) Native PAGE of 14-3-3ζ (pseudo) WT
and the L12E/M78K double mutant at 1 μM concentration
showing the higher oligomeric state of WT.
Fig. 7. Snapshots of the most the WT and the two most destabilizing mutants of Cys94.
The cavities inside of the protein as computed by Pymol [47] are shown in pink.
Z. Jandova et al. BBA - Proteins and Proteomics 1866 (2018) 442–450
448
energy calculations show that the difference of free energy between the
mutant and the WT is 90 kJ/mol for dimer, i.e. 45 kJ/mol per monomer
in dimer. Despite the fact, that these values are unusually high, they
show that the 14-3-3ζ WT is much more stable in its dimeric form than
the L12E-M78K double mutant. Our predictions agree with the experimental
assay, where at 1 μM concentration no trace of a monomeric
state was found for the WT, while the monomeric state was dominant
for the L12K/M78E mutant (Fig. 8B).
The exact role of salt bridges in stabilising of protein complexes has
been a source of dispute for a long time [50]. The driving forces for salt
bridge formation are primarily energetically favourable Coulombic
charge-charge interactions between two ionized amino acids. In order
to build a stable salt bridge, however, two additional factors play a key
role: the orientation of interacting sidechains and the local context
[51]. The sidechain orientation is crucial for keeping charged moieties
in close vicinity, which is necessary for electrostatic interactions. This
might, however, lead to an unfavourable loss in entropy. If the local
context is rich in other possible interaction partners, the probability of
salt bridge formation might be decreased as well. Therefore, due to the
immobilization and desolvation of the interacting residues, creating a
salt bridge in solvent may be unfavourable. The time series of the distance
between Leu12 and Met78 throughout the TI simulation resembles
this case (Fig. S5). In the beginning of the simulation, the residue
distance varies between 0.2 and 1.0 nm. In this phase, these
neutral residues interact via hydrophobic interactions. After introducing
small charges to the perturbing residues, the residues start to
create a stable salt bridge with a length of about 0.3 nm. In the last
perturbation step, nevertheless, the distance increases again. This may
be ascribed to the high charge of both residues and preferential interaction
of Lys78 with solvent. A snapshot of the final state can be seen in
Fig. 9A. The simulation of the double mutant L12E, M78K was prolonged
up to 20 ns. In monomer 1, the salt bridge was formed for the
first 3 ns, then lost for a host time and reformed for another 4 ns, before
it was lost completely. In monomer 2, the salt bridge was immediately
lost, only to be formed transiently again after 15 ns (Fig. 9B). This indicates
that, while the salt bridge could be formed theoretically, the
residues prefer to interact favourably with the solvent.
4. Conclusion
Two types of mutation of the human regulatory protein 14-3-3ζ
were studied by free energy calculations and experimental stability
measurements. Both computational and experimental approaches
agreed that the common Ala and Ser replacements of Cys94 leads to
considerable protein destabilization. The folding free energies were
significantly reduced, as was confirmed by a lower melting temperature.
These findings could be rationalized at a molecular level by the
hydrophobic nature of the particular position and the formation of
small cavities upon mutation to Ala.
Similarly, free energy calculations on a double mutant (L12K/
M78E) that was found to show a significantly reduced tendency to dimerize
were performed and found to be in agreement with the experimental
findings. The putative salt bridge that was introduced in the
double mutant was only observed for a limited amount of time, while
the ionic species of the sidechains, in particular Lys78, are more favourably
solvated individually in the individual monomers.
Overall, our work nicely demonstrates the strength of molecular
simulations and free energy calculations as a complementary tool to
experimental observation. The simulations offer explanations at a molecular
level that are not easily accessible experimentally. This type of
computational tools can be also used for the more efficient rational
design of mutations having desired physico-chemical properties.
Transparency document
The Transparency document associated with this article can be
found, in online version.
Acknowledgments
Financial support of the Austrian Science Fund (grant numbers
W1224 and I 1999-N28) is gratefully acknowledged. J.H., Z.T. and V.W.
acknowledge the financial support of the Czech Science Foundation
(GF15-34684L). CIISB research infrastructure project LM2015043,
funded by Ministry of Education, Youth and Sports of the Czech
Republic (MEYS CR), is gratefully acknowledged for partial financial
support of the measurements at the Proteomics and Biomolecular
Interactions Core Facilities, CEITEC – Masaryk University.
Fig. 9. A) A snapshot from the simulation at the last perturbation
step (λ = 1.00), corresponding to the double
mutant, K78 is solvated. B) Distance between NZ in K78
and CD in Glu12 from prolonged simulations at λ = 1.00.
Table 2
Free energy of mutation in different systems, and their differences,
ΔΔG, in kJ/mol.
ΔG [kJ/mol]
ΔGmut
f
(D) −1169 ± 9
2ΔGmut
f
(M) −1259 ± 11
2ΔGmut
u
−1296 ± 8
2ΔΔGfold(M) 37 ± 14
2ΔΔGdim 90 ± 14
2ΔΔGfold(D) 127 ± 12
Z. Jandova et al. BBA - Proteins and Proteomics 1866 (2018) 442–450
449
References
[1] Aitken A, Collinge DB, van Heusden BPH, Isobe T, Roseboom PH, Rosenfeld G, Soll
J. 14-3-3 proteins: a highly conserved, widespread family of eukaryotic proteins.
Trends Biochem. Sci., 17: 498–501.
[2] W. Wang, D.C. Shakes, Molecular evolution of the 14-3-3 protein family, J. Mol.
Evol. 43 (1996) 384–398.
[3] M. Rosenquist, P. Sehnke, R.J. Ferl, M. Sommarin, C. Larsson, Evolution of the 14-3-
3 protein family: does the large number of isoforms in multicellular organisms reflect
functional specificity? J. Mol. Evol. 51 (2000) 446–458.
[4] A. Aitken, 14-3-3 proteins: a historic overview, Semin. Cancer Biol. 16 (2006)
162–172.
[5] D.H. Jones, S. Ley, A. Aitken, Isoforms of 14-3-3 protein can form homo- and
heterodimers in vivo and in vitro: implications for function as adapter proteins,
FEBS Lett. 368 (1995) 55–58.
[6] G. Messaritou, S. Grammenoudi, E.M.C. Skoulakis, Dimerization is essential for 14-
3-3 zeta stability and function in vivo, J. Biol. Chem. 285 (2010) 1692–1700.
[7] X.W. Yang, W.H. Lee, F. Sobott, E. Papagrigoriou, C.V. Robinson, J.G. Grossmann,
M. Sundstrom, et al., Structural basis for protein-protein interactions in the 14-3-3
protein family, Proc. Natl. Acad. Sci. U. S. A. 103 (2006) 17237–17242.
[8] D. Liu, J. Bienkowska, C. Petosa, R.J. Collier, H. Fu, R. Liddington, Crystal structure
of the zeta isoform of the 14-3-3 protein, Nature 376 (1995) 191–194.
[9] A.K. Gardino, S.J. Smerdon, M.B. Yaffe, Structural determinants of 14-3-3 binding
specificities and regulation of subcellular localization of 14-3-3-ligand complexes: a
comparison of the X-ray crystal structures of all human 14-3-3 isoforms, Semin.
Cancer Biol. 16 (2006) 173–182.
[10] P.D. Burbelo, A. Hall, 14-3-3 proteins. Hot numbers in signal transduction, Curr.
Biol. 5 (1995) 95–96.
[11] M.P. Rubio, K.M. Geraghty, B.H.C. Wong, N.T. Wood, D.G. Campbell, N. Morrice,
C. Mackintosh, 14-3-3-affinity purification of over 200 human phosphoproteins
reveals new links to regulation of cellular metabolism, proliferation and trafficking,
Biochem. J. 379 (2004) 395–408.
[12] L.N. Johnson, D. Barford, The effects of phosphorylation on the structure and
function of proteins, Annu. Rev. Biophys. Biomol. Struct. 22 (1993) 199–232.
[13] M.J. Hubbard, P. Cohen, On target with a new mechanism for the regulation of
protein phosphorylation, Trends Biochem. Sci. 18 (1993) 172–177.
[14] T. Pawson, J.D. Scott, Signaling through scaffold, anchoring, and adaptor proteins,
Science 278 (1997) 2075–2080.
[15] T. Pawson, M. Raina, P. Nash, Interaction domains: from simple binding events to
complex cellular behavior, FEBS Lett. 513 (2002) 2–10.
[16] C. Mackintosh, Dynamic interactions between 14-3-3 proteins and phosphoproteins
regulate diverse cellular processes, Biochem. J. 381 (2004) 329–342.
[17] K. Nagata, A. Puls, C. Futter, P. Aspenstrom, E. Schaefer, T. Nakata, N. Hirokawa,
et al., The MAP kinase kinase kinase MLK2 co-localizes with activated JNK along
microtubules and associates with kinesin superfamily motor KIF3, EMBO J. 17
(1998) 149–158.
[18] G. Tzivion, Y.H. Shen, J. Zhu, 14-3-3 proteins; bringing new definitions to scaffolding,
Oncogene 20 (2001) 6331–6338.
[19] M.B. Yaffe, K. Rittinger, S. Volinia, P.R. Caron, A. Aitken, H. Leffers, S.J. Gamblin,
et al., The structural basis for 14-3-3:phosphopeptide binding specificity, Cell 91
(1997) 961–971.
[20] G. Nagy, C. Oostenbrink, J. Hritz, Exploring the binding pathways of the 14-3-3zeta
protein: structural and free-energy profiles revealed by Hamiltonian replica exchange
molecular dynamics with distancefield distance restraints, PLoS One 12
(2017) e0180633.
[21] M.B. Yaffe, How do 14-3-3 proteins work?— gatekeeper phosphorylation and the
molecular anvil hypothesis, FEBS Lett. 513 (2002) 53–57.
[22] C. Johnson, S. Crowther, M.J. Stafford, D.G. Campbell, R. Toth, C. MacKintosh,
Bioinformatic and experimental survey of 14-3-3-binding sites, Biochem. J. 427
(2010) 69–78.
[23] J. Hritz, L. Byeon Ij, T. Krzysiak, A. Martinez, V. Sklenar, A.M. Gronenborn,
Dissection of binding between a phosphorylated tyrosine hydroxylase peptide and
14-3-3ζ: a complex story elucidated by NMR, Biophys. J. (2014) 2185–2194.
[24] D.C. Rees, A.D. Robertson, Some thermodynamic implications for the thermostability
of proteins, Protein Sci. 10 (2001) 1187–1194.
[25] W.J. Becktel, J.A. Schellman, Protein stability curves, Biopolymers 26 (1987)
1859–1877.
[26] A. Razvi, J.M. Scholtz, Lessons in stability from thermophilic proteins, Protein Soc.
15 (2006) 1569–1578.
[27] R. Cohen, Tom, J.G. Frey, B. Holström, IUPAC, Quantities, Units and Symbols in
Physical Chemistry, 2nd ed., Blackwell Scientific Publications, Oxford, 1993.
[28] Y.Y. Shi, A.E. Mark, C.X. Wang, F. Huang, H.J. Berendsen, W.F. van Gunsteren, Can
the stability of protein mutants be predicted by free energy calculations? Protein
Eng. 6 (1993) 289–295.
[29] C.D. Christ, A.E. Mark, W.F. van Gunsteren, Feature article basic ingredients of free
energy calculations: a review, J. Comput. Chem. 31 (2010) 1569–1582.
[30] Y. Sugita, A. Kitao, Dependence of protein stability on the structure of the denatured
state: free energy calculations of I56V mutation in human lysozyme, Biophys.
J. 75 (1998) 2178–2187.
[31] A.P. Eichenberger, W.F. van Gunsteren, S. Riniker, L. von Ziegler, N. Hansen, The
key to predicting the stability of protein mutants lies in an accurate description and
proper configurational sampling of the folded and denatured states, Biochim.
Biophys. Acta Gen. Subj. 2015 (1850) 983–995.
[32] R. Bonet, I. Vakonakis, I.D. Campbell, Characterization of 14-3-3-zeta interactions
with integrin tails, J. Mol. Biol. 425 (2013) 3060–3072.
[33] N. Schmid, C.D. Christ, M. Christen, A.P. Eichenberger, W.F. van Gunsteren,
Architecture, implementation and parallelisation of the GROMOS software for
biomolecular simulation, Comput. Phys. Commun. 183 (2012) 890–903.
[34] M.M. Reif, M. Winger, C. Oostenbrink, Testing of the GROMOS force-field parameter
set 54A8: structural properties of electrolyte solutions, lipid bilayers, and
proteins, J. Chem. Theory Comput. 9 (2013) 1247–1264.
[35] H.J.C. Berendsen, J.P.M. Postma, W.F. van Gunsteren, J. Hermans, Interaction
models for water in relation to protein hydration, in: B. Pullman (Ed.),
Intermolecular Forces: Proceedings of the Fourteenth Jerusalem Symposium on
Quantum Chemistry and Biochemistry Held in Jerusalem, Israel, April 13–16, 1981,
Springer Netherlands, Dordrecht, 1981, pp. 331–342.
[36] A. Amadei, G. Chillemi, M.A. Ceruso, A. Grottesi, A. Di Nola, Molecular dynamics
simulations with constrained roto-translational motions: theoretical basis and statistical
mechanical consistency, J. Chem. Phys. 112 (2000) 9–23.
[37] H.J.C. Berendsen, J.P.M. Postma, W.F. Vangunsteren, A. Dinola, J.R. Haak,
Molecular dynamics with coupling to an external bath, J. Chem. Phys. 81 (1984)
3684–3690.
[38] J.P. Ryckaert, G. Ciccotti, H.J.C. Berendsen, Numerical integration of cartesian
equations of motion of a system with constraints - molecular dynamics of N-alkanes,
J. Comput. Phys. 23 (1977) 327–341.
[39] I.G. Tironi, R. Sperb, P.E. Smith, W.F. van Gunsteren, A generalized reaction field
method for molecular dynamics simulations, J. Chem. Phys. 102 (1995)
5451–5459.
[40] T.N. Heinz, W.F. van Gunsteren, P.H. Hünenberger, Comparison of four methods to
compute the dielectric permittivity of liquids from molecular dynamics simulations,
J. Chem. Phys. 115 (2001) 1125–1136.
[41] T.C. Beutler, A.E. Mark, R.C. van Schaik, P.R. Gerber, W.F. van Gunsteren, Avoiding
singularities and numerical instabilities in free energy calculations based on molecular
simulations, Chem. Phys. Lett. 222 (1994) 529–539.
[42] W. Kabsch, C. Sander, Dictionary of protein secondary structure: pattern recognition
of hydrogen-bonded and geometrical features, Biopolymers 22 (1983)
2577–2637.
[43] G. Hu, H. Li, J.Y. Liu, J. Wang, Insight into conformational change for 14-3-3sigma
protein by molecular dynamics simulation, Int. J. Mol. Sci. 15 (2014) 2794–2810.
[44] K.A. Dill, Dominant forces in protein folding, Biochemistry 29 (1990) 7133–7155.
[45] N. Nagano, M. Ota, K. Nishikawa, Strong hydrophobic nature of cysteine residues in
proteins, FEBS Lett. 458 (1999) 69–71.
[46] A.E. Eriksson, W.A. Baase, X.J. Zhang, D.W. Heinz, M. Blaber, E.P. Baldwin,
B.W. Matthews, Response of a protein structure to cavity-creating mutations and its
relation to the hydrophobic effect, Science 255 (1992) 178–183.
[47] L. Schrödinger, The PyMOL Molecular Graphics System. Version~1, 8 ed., (2015).
[48] N.N. Sluchanko, N.B. Gusev, Oligomeric structure of 14-3-3 protein: what do we
know about monomers? FEBS Lett. 586 (2012) 4249–4256.
[49] M.M. Reif, C. Oostenbrink, Net charge changes in the calculation of relative ligandbinding
free energies via classical atomistic molecular dynamics simulation, J.
Comput. Chem. 35 (2014) 227–243.
[50] P. Strop, S.L. Mayo, Contribution of surface salt bridges to protein stability,
Biochemistry 39 (2000) 1251–1255.
[51] G.I. Makhatadze, V.V. Loladze, D.N. Ermolenko, X. Chen, S.T. Thomas, Contribution
of surface salt bridges to protein stability: guidelines for protein engineering, J. Mol.
Biol. 327 (2003) 1135–1148.
Z. Jandova et al. BBA - Proteins and Proteomics 1866 (2018) 442–450
450
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
54
6 Structural and interaction properties of intrinsically
disordered proteins (IDPs) determined by experimental
NMR and computational studies
Paper 11: Hritz J.; Byeon I-J.; Krzysiak T.; Martinez A.; Sklenář V.; Gronenborn A.M.
Dissection of binding between a phosphorylated tyrosine hydroxylase peptide and
14-3-3ζ: a complex story elucidated by NMR. Biophys. J. 2014, 107, 2185-2194
Paper 12: Louša, P.; Nedozrálová, H.; Župa, E.; Nováček, J.; Hritz, J.*: Phosphorylation
of the regulatory domain of human tyrosine hydroxylase 1 monitored using nonuniformly
sampled NMR. Biophysical Chemistry 2017, 223, 25-29
Paper 13: Zapletal, V.; Mládek, A.; Melková, K.; Louša, P.; Nomilner, E.; Jaseňáková,
Z.; Kubáň, V.; Makovická, M.; Laníková, A.; Žídek L.; Hritz, J.* Choice of force field for
proteins containing structured and intrinsically disordered regions. Biophys. J. 2020,
118, 1621–1633
Paper 14: Pavlíková Přecechtělová, J.; Mládek, A.; Zapletal, V.; Hritz, J. Quantum
Chemical Calculations of NMR Chemical Shifts in Phosphorylated Intrinsically Disordered
Proteins, JCTC 2019, 15, 5642-5658
Paper 15: Jansen, S.;Melková, K.; Trošanová, Z.; Hanáková, K.; Zachrdla, M.; Nováček,
J.; Župa, E.; Zdráhal, Z.; Hritz, J.*; Žídek, L.*: Quantitative Mapping of MAP2c
Phosphorylation and 14-3-3ζ Binding Sites Reveals Key Differences Between MAP2c
and Tau. J. Biol. Chem. 2017, 292, 6715-6727
6.1 Intrinsically disordered proteins (IDPs)
The structure-function paradigm is probably mentioned in every structural biology or
biochemistry textbook. It is important to pay attention to the cases in which proteins
seem to violate this quite general rule. Certain proteins do not adopt a unique 3D structure
but they still play a well-defined physiological role.33,34 Such proteins are usually
referred to as intrinsically disordered proteins in the literature. Since the automated
analysis of sequences predicted that one-third of the eukaryotic proteins would contain
disordered regions longer than 30 amino acids, IDPs have recently attracted significant
attention.35,36 However, the biophysical characterization of IDPs is much more
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
55
difficult than in the case of well-structured proteins. The lack of a unique structure represents
a fundamental problem for crystallographic and cryoEM studies. Rather, computational
simulations and nuclear magnetic resonance (NMR) play a crucial role in the
atomic-resolution studies of IDPs.
However, even for these techniques, the characterization of IDPs is far from trivial:
Computational approaches suffer from insufficient sampling, in a manner that is much
more severe than for structured proteins, because of the necessity to sample a much
wider conformational space. The resonance frequencies observed in NMR spectra are
time- and ensemble-averaged. Because IDPs take up multiple rapidly inter-converting
conformations, the averaging makes the distribution of measured frequencies more
uniform and thus more difficult to distinguish. Peak overlap in spectra of large or highly
repetitive IDPs is often so severe that it is impossible to assign frequencies to individual
atoms using standard approaches. The experimental data need to be interpreted differently
than results obtained for well-ordered proteins because the obtained values
are averages of large ensembles of very diverse protein conformations. For example,
the propensity of individual molecules to form transient secondary structures is estimated,
instead of a determination of well-defined secondary structure elements in
terms of atomic coordinates.
6.2 Structural ensembles of IDPs and IDRs verified by NMR data
MD generally can generate the canonical ensemble of any system if sufficient sampling
is achieved. More details about the sampling aspect are discussed in the second scientific
chapter. However, there is another important aspect – the quality of force-field
(FF) parameters that has not been discussed yet because there is a wide range of biomolecular
FFs that work quite well for globular proteins. However, recent publications
indicate that many of them fail when simulating IDPs and new sets of FF parameter
sets were designed for this class of proteins.37 At the same time, these FF parameters
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
56
have not been providing reliable data for globular proteins. This introduces several
complications when trying to simulate proteins containing both globular parts as well
as disordered regions.
This issue was addressed by the applicant for three different proteins: the δ subunit of
RNA polymerase (δRNAP), the regulatory domain of human tyrosine hydroxylase 1
(RD-hTH1) and the selected region of microtubule-associated protein 2 c (MAP2c).P13
The structural characterization of the δ subunit- RNAP was complicated mainly by the
physical properties of its C-terminal domain. This region of the sequence is extremely
acidic and highly repetitive, which was demonstrated to be necessary for regulating
the RNAP affinity towards nucleic acids. To proceed towards structural characterization
of the native, full-length protein, we applied and further developed NMR techniques
based on nonuniform sampling (NUS), which allowed us to overcome a severe
overlap of peaks of the disordered C-terminal domain in the NMR spectra.
RD-hTH1 has a similar architecture, i.e. region 1-69 is unstructured and the rest (70-
169) is structured. In the case of MAP2c, region 159-254 was chosen, containing transient
helical elements. MD simulations in the microsecond range were generated for
those proteins by applying a variety of popular biomolecular force fields. Primary NMR
data were used to verify generated MD trajectories by using Amber99, Charmm22 and
Charmm36 force-field parameters in combination with TIP3P and TIP4P-D explicit water
models.37,38 We noticed distinct differences in the predicted cross-relaxation
steady-state NOE relaxation data, exemplified in Fig. 7. The dynamic behavior of the
disordered region is very unrealistic in all simulations using TIP3P (original or modified)
when compared to the TIP4P-D water model. We observed the same pattern also
for other protein systems.P13 This is an interesting observation, as the TIP4P-D water
model was not parameterized using NMR relaxation data. Although simulations with
TIP4P-D water model yield a quite realistic steady-state nuclear Overhauser effect
(ssNOE) profile, there is still space for improvement (e.g. region 90-130 in Fig. 9). This
motivated us to validate simulations of IDRs, in addition to chemical shifts (CSs),
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
57
Figure 7: NMR relaxation data for δ subunit of RNA polymerase.
Comparison of predicted and measured dipolar cross-relaxation NMR rates (15N-1H steady-state NOE)
for δ subunit of RNA polymerase from Bacillus subtilis. Experimental data (presented by black vertical
lines) were collected on 600MHz Bruker Avance III NMR spectrometer. Other colors present computational
data by using three different force-fields in combination with TIP3P and TIP4P-D water models.
residual dipolar coupling (RDCs), paramagnetic relaxation enhancement (PREs) and
small angular X-ray scattering (SAXS) data and also check the dynamic performance
with respect to the NMR relaxation data.P13 Fig. 9 shows that the majority of biomolecular
FFs parameters result in unrealistic predictions. However, this data allows us to
detect those that are the most reliable for the given class of IDPs.
From NMR experimental data - chemical shifts (CS) of proteins are mostly used for the
verification of structural data. These procedures use predicted CS data for the given
protein structures by using the empirical databases and the related prediction procedures.
There are however cases, such as post-translational modifications of proteins
(e.g. phosphorylation), when such procedures are currently not available because of an
insufficient number of experimental NMR data. In such cases, one can use QM predictions
of CS. Previously QM predictions of NMR CSs were shown for small organic molecules
or selected parts of the globular proteins. The applicant presented the
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
58
procedure by which such an approach can be applied even for the IDP region hTH1 in
the explicit water solvent.P14
Unstructured domains are challenging for structural characterization and for free-energy
calculations as they should be described by ensembles of structures rather than
by a single structure. The group of prof. Zidek has recently developed nonuniform sampling
NMR experiments (NUS) which are suitable to address intrinsically disordered
proteins.39,40 This methodology was successfully applied for the δ-subunit of RNA polymerase.41
Different computational approaches will be tested to generate structural
ensembles of the hTH1 unstructured region and the δ-subunit of RNA polymerase. Ensemble
averaged data will be compared with experimentally measured NMR properties:
chemical shifts, NOEs, paramagnetic relaxation enhancement and relaxation data.
Figure 8: Binding scheme of dp_hTH1_50 phosphorylated on positions 19 and 40 with the
14-3-3ζ dimer, represented by a letter P.
Subscript and superscript next to P indicate which phosphorylation site (non, 19 or 40) is bound to the
14 3 3ζ. The whole scheme can be described by only three parameters Kd1, Kd2 and Keq when assuming
independent binding of individual phosphoserines to different binding sites with 14 3 3ζ. This assumption
does not hold for the cooperative binding and different values of equilibrium constants that
have to be assigned to the individual equilibriums.
P19-40
Æ
P19-40
19-40
PÆ
Æ
P19
40
P19-40
40-19
P40-19
Æ
P40-19
40-19
2*Kd1&
2*Kd2&
Keq*Kd2/Kd1&Keq&
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
59
6.3 NMR investigation of IDP/IDRs binding properties
The NMR methodology is well known as one of the structural biology methods but it is
often forgotten that it can also be used very well for following different kinds of
changes within the protein. In the case of the studied IDPs/IDRs we applied NMR to
follow the phosphorylation kinetic profiles of RD-hTH1P12 and MAP2c.P15 Subsequently
NMR was applied for the determination of the interaction patches of those phosphorylated
proteins towards their known binding partner 14-3-3ζ. In the case of multiple
binding patches to the 14-3-3 proteins having two different binding sites determining
the complete binding scheme is far from trivial (Fig. 8). The applicant used 31P 1D NMR
spectroscopy to determine a whole variety of existing binding complexes from their
disordered fragments (region 1-50 of hTH1) containing single or double phosphorylation
sites.P11 We identified an interaction between singly S40-phosphorylated hTH1-
50 and 14-3-3ζ that exhibits one order of magnitude weaker affinity than the singly
S19-phosphorylated hTH1-50. Detailed analysis of the NMR data, giving information
about individual phosphorylation sites, revealed that binding of 14-3-3ζ to the doubly
phosphorylated peptide occurs in a complex fashion and three major distinct forms of
the 14-3-3ζ complex with doubly phosphorylated hTH1-50 peptide were identified.P11
To the best of our knowledge, this is the first study to show the existence of more
than the single complex form of 14-3-3 with bound phosphorylated peptide.
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
60
Paper 11
Hritz J.; Byeon I-J.; Krzysiak T.; Martinez A.; Sklenář V.; Gronenborn A.M. Dissection of
binding between a phosphorylated tyrosine hydroxylase peptide and 14-3-3ζ: a complex
story elucidated by NMR. Biophys. J. 2014, 107, 2185-2194
Article
Dissection of Binding between a Phosphorylated Tyrosine Hydroxylase
Peptide and 14-3-3z: A Complex Story Elucidated by NMR
Jozef Hritz,1,2
In-Ja L. Byeon,1
Troy Krzysiak,1
Aurora Martinez,3
Vladimir Sklenar,2
and Angela M. Gronenborn1,*
1
Department of Structural Biology, University of Pittsburgh, School of Medicine, Pittsburgh, Pennsylvania; 2
Department of Structural Biology,
Central European Institute of Technology, Masaryk University, Brno, Czech Republic; and 3
Department of Biomedicine, University of Bergen,
Bergen, Norway
ABSTRACT Human tyrosine hydroxylase activity is regulated by phosphorylation of its N-terminus and by an interaction with
the modulator 14-3-3 proteins. We investigated the binding of singly or doubly phosphorylated and thiophosphorylated peptides,
comprising the first 50 amino acids of human tyrosine hydroxylase, isoform 1 (hTH1), that contain the critical interaction domain,
to 14-3-3z, by 31
P NMR. Single phosphorylation at S19 generates a high affinity 14-3-3z binding epitope, whereas singly S40phosphorylated
peptide interacts with 14-3-3z one order-of-magnitude weaker than the S19-phosphorylated peptide. Analysis of
the binding data revealed that the 14-3-3z dimer and the S19- and S40-doubly phosphorylated peptide interact in multiple ways,
with three major complexes formed: 1), a single peptide bound to a 14-3-3z dimer via the S19 phosphate with the S40 phosphate
occupying the other binding site; 2), a single peptide bound to a 14-3-3z dimer via the S19 phosphorous with the S40 free in
solution; or 3), a 14-3-3z dimer with two peptides bound via the S19 phosphorous to each binding site. Our system and data
provide information as to the possible mechanisms by which 14-3-3 can engage binding partners that possess two phosphorylation
sites on flexible tails. Whether these will be realized in any particular interacting pair will naturally depend on the details of
each system.
INTRODUCTION
Members of the 14-3-3 protein family are important modulators
of several key signaling pathways that regulate critical
biological activities, including progression through the cell
cycle, growth, proliferation, and apoptosis (1–3). Extensive
proteomics studies have revealed that the 14-3-3 family targets
>200 distinct partners, ~0.6% of the entire cellular proteome
(4,5).
In humans, seven different 14-3-3 isoforms (b, g, ε, h, s,
t, and z) are found, each encoded by a distinct gene, that
function as homo- or heterodimers of ~30-kDa proteins
(6,7). Crystal structures of all seven homodimeric mammalian
proteins are available. The fold of the monomeric unit
contains nine a-helices, arranged in an antiparallel fashion.
Within the dimeric structure, two amphipathic grooves, one
per monomer, constitute the binding pockets in which
the conserved KRRY residues-Lys49
, Arg56
, Arg127
, and
Tyr128
in 14-3-3z-are positioned for the recognition of
a phosphorylated serine or threonine target (8). Three
different phosphoserine (pS)- or phosphothreonine-containing
sequence motifs can be recognized by 14-3-3 proteins:
RSXp(S/T)XP (motif I), RXXXp(S/T)XP (motif II), and a
carboxyl-terminal p(S/T)XCOOH motif (motif III), where
X is a nonproline amino acid (9–11). Some 14-3-3 family
members have also been reported to bind nonphosphorylated
peptides as well as motifs that do not strictly display
the above canonical sequences (6,11).
Tyrosine hydroxylase (TH) and tryptophan hydroxylase
were the first proteins that were identified as binding partners
for 14-3-3 proteins (12). TH catalyzes the tetrahydrobiopterin-dependent
hydroxylation of L-tyrosine to L-3,4-dihydroxyphenylalanine,
which is the rate-limiting step in the
biosynthesis of dopamine and other catecholamines (13).
In humans, several isoforms of TH are expressed, with isoform
1 (hTH1; 497 residues, NP-000351.2) being the best
characterized (14–16). hTH1 can be phosphorylated at T8
and/or three serine residues (S19, S31, and S40) by several
protein kinases (17) on its highly dynamic 50-residue N-terminal
region that precedes the regulatory ACT domain, characteristic
of the aromatic amino-acid hydroxylase family.
14-3-3z and other isoforms have been reported to bind
to S19 phosphorylated TH (18–20). Although not strictly
exhibiting the canonical binding motifs, recent crystallographic
analysis of the complex between hTH1 S19-phosphorylated
N-terminal peptide (1–43 residues) and 14-3-3g
revealed that the phosphate interacts with the KRRY motif
in the 14-3-3g binding grove, and the region surrounding
S19 engages in similar interactions as seen for peptides containing
canonical motifs I and II (21). In addition, phosphorylation
of hTH1 on S40 stimulates TH activity (17), and
induces binding to the yeast 14-3-3 proteins BMH1 and
BMH2 and to sheep brain 14-3-3 proteins, but not to 14-3-
3z (18,19). On the other hand, 14-3-3z has been reported
Submitted June 30, 2014, and accepted for publication August 28, 2014.
*Correspondence: amg100@pitt.edu
Editor: David Eliezer.
Ó 2014 by the Biophysical Society
0006-3495/14/11/2185/10 $2.00 http://dx.doi.org/10.1016/j.bpj.2014.08.039
Biophysical Journal Volume 107 November 2014 2185–2194 2185
to interact with doubly S19- and S40-phosphorylated hTH1
(18). In a recent study (22) we have shown that the doubly
phosphorylated hTH1 binds 14-3-3g with similar stoichiometry
and affinity as the singly phosphorylated enzyme on
Ser19
. However, 14-3-3g interfered with the phosphorylation
of TH1-pS19 on Ser40
, supporting that Ser40
becomes
inaccessible in the hTH1:14-3-3g complex, although Ser40
does not significantly contribute to the binding of 14-3-3g.
Therefore the question of whether S40 phosphorylated
hTH1 can be bound by 14-3-3 proteins has not been settled
unequivocally.
The 14-3-3 proteins are thought to function by inducing a
conformational change in the target proteins, thereby modulating
the target proteins’ activities. In this way, 14-3-3 family
members impart additional control onto key cellular
enzymes, in addition to common enzymatic control mechanisms
such as feedback loops or product inhibition. The
contemporary prevalent model for the binding of doubly
phosphorylated peptides by 14-3-3 suggests that one phosphorylated
residue of the target protein engages 14-3-3 first,
linking the two proteins together, followed by binding of the
second phosphorylated residue, to effect the conformational
change (23). The notion that binding of 14-3-3 to two phosphorylation
sites is required to induce the activity altering
conformational change, has led some authors to consider
14-3-3 as a binary computer (4,24).
We explored this model of 14-3-3 modulation using 31
P
NMR in a simplified peptide system. We investigated the
binding of hTH1 peptides (region 1–50) with a single phosphorylation
at S19 (pS19) or S40 (pS40) to 14-3-3z and determined
individual binding affinities. We found that 14-3-3z
can interact with the hTH1 peptide through either phosphorylated-S19
or -S40, although the binding affinity of 14-3-3z
for the pS40 peptide is significantly lower than that for the
pS19 peptide. Furthermore, binding of 14-3-3z to the doubly
phosphorylated hTH1 peptide exhibits a complex binding
profile involving three major complexes. Our results suggest
that when considering 14-3-3 interactions with multiply phosphorylated
targets in general, and hTH1 in particular, avidity
considerations should be taken into account.
METHODS
Cloning, expression, and purification of hTH1
peptides and 14-3-3z
The coding region for residues 1–50 of human tyrosine hydroxylase
(hTH1–50) (Fig. 1 A) was inserted into a modified pET32a vector, in frame
with N-terminal thioredoxin and 6ÂHis tags, followed by a tobacco etch virus
(TEV) cleavage site. A codon-optimized gene for 14-3-3z (GenScript,
Piscataway Township, NJ) was inserted into pET15b with Cys25
and
Cys189
codons converted to Ala using the QuikChange Site-Directed Mutagenesis
Kit (Stratagene, La Jolla, CA). The DNA sequence for all constructs
was confirmed by sequencing (Genewiz, South Plainfield, NJ).
Proteins were produced in Escherichia coli BL21(DE3) cells that were
grown at 37
C in LB media to an OD600 of 0.8 and induced with 0.5 mM
IPTG for 18 h at 18
C. Cells were harvested, resuspended in 50 mM TBS
(50 mM Tris, 150 mM NaCl, pH 8.0) buffer, and lysed with a microfluidizer
(Microfluidics, Westwood, MA). The cell lysate was clarified by centrifugation
(21,000g for 60 min at 4
C), and the supernatant was loaded onto a
HisTrap column (GE Healthcare, Piscataway, NJ). Proteins were eluted using
a linear imidazole gradient (0.05–1 M in a total volume of 15 mL).
Fractions containing tagged hTH1 peptide were treated with TEV (1 mg/
20 mg of peptide) at 4
C overnight, and applied to a Vydac reverse-phase
C4 column (Grace Davison Discovery Sciences, Deerfield, IL). hTH1–50
was eluted from the reverse phase column using a linear gradient of 10–
40% acetonitrile in 50 mL. Peptide-containing samples were further purified
by gel filtration through a Superdex75 26/60 column (GE Healthcare)
equilibrated in 20 mM Tris buffer, pH 7.5. The Trp-containing three aminoacid
cloning artifact at the N-terminus of the hTH1–50 peptide (Fig. 1 A)
allowed for detection during purification and concentration determination
at 280 nm.
Fractions containing 14-3-3z were dialyzed into TBS buffer and treated
with TEVovernight. Undigested protein, 6ÂHis tag cleavage products, and
TEV protease were removed by passing the sample over the HisTrap column
for a second time and the flow-through containing untagged 14-3-3z
was collected. The sample was further purified over anion exchange
(HiTrap Q HP; GE Healthcare) followed by gel filtration on a Superdex75
26/60 column (GE Healthcare), equilibrated in 20 mM sodium phosphate
buffer, pH 7.0.
The Cys25
Ala/Cys189
Ala 14-3-3z variant exhibited essentially the same
stability as wild-type 14-3-3z protein, as evidenced by identical differential
scanning calorimetry unfolding profiles.
Phosphorylation of hTH1–50
The serine residue at position 40 of the hTH1–50 peptide was phosphorylated
using the catalytic subunit of cAMP-dependent protein kinase A
(PKA; New England BioLabs, Ipswich, MA), whereas Ser19
was phosphorylated
using p38 regulated/activated protein kinase (PRAK; obtained from
the protein phosphorylation unit, University of Dundee, Dundee, Scotland).
For phosphorylation at S19, the reaction mixture, containing hTH1–50
S40A peptide in 50 mM Tris buffer, pH 7.5, 1 mM ATP, 10 mM MgCl2,
and PRAK kinase (50 mg for 10 mg of hTH1–50 S40A), was incubated
for three days at 30
C. For phosphorylation at S40, hTH1–50 was incubated
with PKA (0.5 mg for 15 mg hTH1–50) in 50 mM Tris buffer, pH
7.5,10 mM MgCl2, 200 mM ATP at 30
C for 2 h. Reaction products were
purified by gel filtration using a Superdex30 16/60 column (GE Healthcare)
followed by anion exchange (HiTrap Q HP; GE Healthcare).
Doubly phosphorylated peptide was prepared using S40 phosphorylated
hTH1–50 peptide (pS40-hTH1–50) and PRAK kinase, as described above.
Approximately 70% of pS40-hTH1–50 was converted to the doubly phosphorylated
product (pS19pS40-hTH1–50), which was purified as described
above. For thiophosphorylation, adenosine 50
-[g-thio]triphosphate (ATPgS)
was added to the PKA reaction mixture, instead of ATP. In all cases, the correct
masses and purity of the final products were verified by matrix-assisted
laser desorption/ionization-time of flight mass spectrometry.
31
P NMR titrations
All 31
P NMR spectra were recorded at 37
C on an AVANCE 600 MHz
NMR spectrometer equipped with a broadband frequency probe (Bruker,
Billerica, MA). Samples containing 14-3-3z and singly or doubly phosphorylated
hTH1–50 peptides were prepared in 20 mM sodium phosphate, pH
7.0, at a constant peptide concentration (0.83 or 0.71 mM) with molar ratios
of 14-3-3z to peptide of 0:1–8:1.
Modeling of binding scenarios
Binding schemes for the different complexes between 14-3-3z and singly or
doubly phosphorylated peptides are presented in Fig. 1 C and Fig. 2 B,
Biophysical Journal 107(9) 2185–2194
2186 Hritz et al.
respectively. Equations for individual binding equilibria and equations
describing mass conservation are provided in Fig. S1 and Fig. S2 in the Supporting
Material. The equations were solved numerically using the software
MATLAB (The MathWorks, Natick, MA) and results yield populations of
the individual species as a function of the 14-3-3z to peptide molar ratios, r.
Three different sets of solutions were obtained, only one of which was physically
meaningful giventhatpopulationscannotbenegative for any r value R0.
RESULTS
Binding of singly phosphorylated hTH1 peptides
to 14-3-3z
Binding between 14-3-3z and the phosphorylated N-terminus
of hTH1 was probed by 31
P NMR. hTH1 peptides, containing
the first 50 residues (hTH1–50), were prepared and
each serine in the peptide was phosphorylated individually,
generating pS19-hTH1–50 and pS40-hTH1–50. PKA selectively
and specifically phosphorylated only Ser40
, but PRAK
phosphorylated both Ser19
and Ser40
(J. Hritz and A.M. Gronenborn,
unpublished results). Therefore, we introduced the
S40A change, permitting the preparation of singly phosphorylated
pS19-hTH1–50 peptide. Titration of 14-3-3z
into a solution containing pS19-hTH1–50 revealed the system
to be in slow exchange on the NMR chemical shift timescale.
Thus, resonances for both free and the 14-3-3z-bound
peptide can be observed separately (Fig. 1 B), and the relative
fractions of the free and bound peptide can be determined
from resonance intensities (25,26). A dissociation
constant (Kd1) of 0.15 5 0.03 mM for the 14-3-3z/pS19hTH1–50
complex was derived, using
Kd1 ¼
½pS19free
Š Ã
Â
14 À 3 À 3zfree
Ã
½pS19boundŠ
¼
cpep à AðpS19free
Þ Ã fr À AðpS19bound
Þg
AðpS19boundÞ
; (1)
A
B
C
D
FIGURE 1 Binding of singly phosphorylated
hTH1–50 peptides to 14-3-3z. (A) Amino-acid
sequence of the hTH1–50 peptide with S19 and
S40 highlighted (boldface). (B) 31
P spectra of
pS19-hTH1–50 at 0.83 mM peptide concentration
and molar ratios of 14-3-3z to peptide (r) 0.5:1
(black) and 1:1 (gray). (C) Schematic depiction
of pS19-hTH1–50 (denoted by 19–40) binding to
the 14-3-3z homodimer (free protein dimer is depicted
by P[
[, where Ø indicates an empty phosphoserine
binding site). (Circled) Bound pS19
residues. (D) 31
P spectra of pS40-hTH1–50 at
0.71 mM peptide concentration and molar ratios
(r) of 14-3-3z to peptide of 0:1 (blue), 0.5:1
(red), 1:1 (dark green), 2:1 (purple), 4:1 (orange),
and 8:1 (brown). (Inset) Chemical shift changes
upon 14-3-3z binding (Dd) are plotted as a function
of molar ratio (r). The curve obtained by fitting Dd
to the quadratic equation shown in Eq. 2 is shown.
To see this figure in color, go online.
Biophysical Journal 107(9) 2185–2194
Binding of Phospho-TH Peptides to 14-3-3 2187
where cpep, A(pS19free
), and A(pS19bound
) comprise the total
peptide concentration, and the free and bound fractions of
the pS19-hTH1–50 peptide, respectively, with r the molar
ratio of 14-3-3z to peptide. For both singly phosphorylated
peptides, dissociation constants are reported for the 14-3-3z
monomer, assuming that binding to a single site in the 14-3-
3z homodimer is not influenced by binding to the second
site.
To analyze our binding data, we generated a scheme for
the interaction between pS19-hTH1–50 and the 14-3-3z
dimer (Fig. 1 C). A pS19-hTH1–50 peptide (represented
by 19–40 in Fig. 1 C) can bind to the free 14-3-3z dimer
(depicted by P[
[ in Fig. 1 C) at two different sites, with
the dissociation constant being one-half of the observed
Kd1; the binding of a second pS19-hTH1–50 peptide to
the 14-3-3z dimer in which one site is already occupied
A
B
C
D
FIGURE 2 Binding of doubly phosphorylated
peptide to 14-3-3z. (A) 31
P spectra of pS19pS40hTH1–50
at 0.83 mM and the molar ratio of 14-
3-3z to the peptide (r) was 0:1 (blue), 0.5:1 (red),
1:1 (dark green), 2:1 (purple), 3:1 (yellow), 4:1 (orange),
or 5:1 (light green). (B) Binding scheme of
pS19pS40-hTH1–50 (denoted by 19–40) with
respect to the 14-3-3z homodimer (P[
[). Dissociation
constants Kd1 and Kd2 correspond to the values
determined for the singly phosphorylated peptides.
(Blue arrows) Equilibria for which slow exchange
on the 31
P chemical shift scale is observed; (red
arrows) equilibria for which fast exchange on the
31
P chemical shift scale is observed. (Circled)
Bound phosphorous groups. Phosphorous groups
contributing intensity to the same peak in the
NMR spectra are shown in the same color. (C)
Equations governing the relative peak intensities
(denoted as ‘‘A’’), based on the scheme depicted
in panel B; each phosphopeptide contributes to
one or more of the five peaks. The specific phosphoserine
that contributes to intensity of a particular
peak is colored in the same color as in panel
B. (D) Populations of phosphoserine species that
contribute to the experimentally observed peak intensities
(x) for different 14-3-3z to peptide molar
ratios, r. (Continuous lines) For comparison, the
theoretically predicted populations, based on the
binding model in panel B using Kd1 ¼ 0.15 mM,
Kd2 ¼ 1.0 mM, and Keq ¼ 1.5, are plotted. To see
this figure in color, go online.
Biophysical Journal 107(9) 2185–2194
2188 Hritz et al.
(depicted by P/
19À40 in Fig. 1 C) can only occur to the unoccupied
site, and dissociation from a doubly occupied
14-3-3z dimer (depicted by P19À40
19À40 in Fig. 1 C) can
occur from either site, resulting in 2*Kd1. The equations
describing these two equilibria, taking mass conservation
of the peptide and 14-3-3z into account, are
presented in Fig. S1 in the Supporting Material. Based
on this model, in a 1:1 mixture of pS19-hTH1–50 and
14-3-3z (monomer concentration) with a Kd1 of
0.15 mM, 34.4% of the peptide would be in the free state,
22.6% of peptide in the P/
19À40 bound form, and 43% in
the P19À40
19À40 bound form.
In contrast to the result with the pS19 peptide, titration of
14-3-3z into a solution containing the pS40-hTH1–50 peptide
revealed fast exchange on the NMR chemical shift
scale, resulting in a single resonance at the weighted averaged
chemical shifts of both the free and bound resonances
(Fig. 1 D). By fitting the chemical shift change (Dd) to Eq. 2,
a Kd2 value of 1.0 5 0.1 mM and a maximum chemical shift
difference between free and bound peptide (Ddmax) of 0.94
p.p.m. were determined for the binding of 14-3-3z to pS40-
hTH1–50:
Dd ¼
Ddmax
2cpep

Kd þ cpep þ rcpep
À
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiÀ
Kd þ cpep þ rcpep
Á2
À 4c2
pepr
q 
: (2)
These values allowed us to calculate the 31
P bound chemical
shift, d(pS40bound
), which was 3.26 p.p.m. (Fig. 1 D).
Interestingly, the bound 31
P chemical shifts for both
singly phosphorylated hTH1–50 peptides are essentially
identical (3.26 p.p.m., Fig. 1, B and D), irrespective of
the different neighboring amino acids around the phosphorylated
Ser in the hTH1–50 peptide. This suggests that the
binding of the phosphorylated residues to the identical
binding pockets of 14-3-3z that are lined by residues
Lys49
, Arg56
, Arg127
, and Tyr128
is the major determinant
of the chemical shift, irrespective of the local peptide
sequence. We speculate that the side chain of Tyr128
most
likely causes the upfield shift of the pS19 and pS40 31
P
resonances.
Binding of doubly phosphorylated hTH1 peptide
to 14-3-3z
Binding between doubly phosphorylated proteins and 14-3-
3 has previously been characterized using various methods,
including isothermal titration calorimetry and fluorescence
polarization (19,27,28). Although these methods provide
overall binding properties, such as Kd or stoichiometry,
important details for a system involving a homodimer and
two binding motifs on the target are lacking. For example,
answers to the question of how individual phosphorylated
amino acids interact with 14-3-3, and what kind of complexes
are possible, are still incomplete. 31
P NMR is ideally
suited to provide such information, since individual 31
P
signals can be monitored throughout the course of titration
experiments between various interacting partners. We
therefore prepared doubly phosphorylated hTH1–50 peptide
(pS19pS40-hTH1–50) and monitored 14-3-3z binding to
this peptide by 31
P NMR. Our data reveal that a number
of different bound species are present during the course
of the titration (Fig. 2 A). The spectrum of free
pS19pS40-hTH1–50 (blue spectrum in Fig. 2 A; r ¼ 0) contains
two well-separated resonances, at 3.76 p.p.m. and at
4.18 p.p.m. for the phosphorous atoms on pSer19
(pS19free
)
and pSer40
(pS40free
), respectively. At the end of the titration
(light green trace in Fig. 2 A; r ¼ 5), the pS19pS40hTH1–50
peptide is saturated with 14-3-3z, and resonances
at 3.26 p.p.m. and 3.43 p.p.m. for the phosphorous atoms on
pSer19
(pS19bound
) and pSer40
(pS40c
), respectively, are
observed. The bound resonances are broader than those
of the free peptide, reflecting the slower rotational correlation
time of the peptide/14-3-3z complex (molecular mass
~62-67 kDa).
The chemical shift of 3.43 p.p.m. is different from the
completely saturated position (3.26 p.p.m.) obtained with
a peptide that contains only phosphorylated Ser40
(pS40hTH1–50,
Fig. 1 D), indicating that this bound resonance
position contains a contribution from the free pS40 resonance
(due to fast exchange on the chemical shift scale).
This signal, therefore, is designated pS40c
, instead of
pS40bound
. During the titration, the intensities of both the
pS19free
and pS40free
resonances decrease to the same degree
with increasing 14-3-3z concentration, but only the
bound pS19 31
P resonance at 3.26 p.p.m. increases in intensity
at the same rate (Fig. 2 A, e.g., red spectrum). The
bound pS40 31
P resonance at 3.43 p.p.m. is only visible
for excess 14-3-3z over peptide (e.g., purple, yellow, orange,
and light-green spectra). These data suggest that
for excess peptide and little 14-3-3z (r < 1), the phosphorous
group on Ser19
(strong binder) is the one predominantly
bound by 14-3-3z, and two pS19pS40-hTH1–50
peptides are likely bound by one 14-3-3z dimer. For conditions
where 14-3-3z is in excess (r > 1), both phosphorous
groups on a single peptide can be bound by one 14-3-3z
dimer.
The complete binding scenario, considering six possible
complexes for the interaction between doubly phosphorylated
peptide (free peptide is depicted as 19–40) and 14-
3-3z (free 14-3-3z dimer is depicted as P[
[ ), is presented
in Fig. 2 B. Based on this interaction scenario and taking
into account that resonances can be in slow or fast exchange
on the chemical shift scale (marked by blue and
red arrows, respectively), five different resonances are expected
to be observable in the 31
P NMR spectra (see
below). Phosphorous groups that exhibit identical resonance
frequencies, thus contributing intensity to the same
Biophysical Journal 107(9) 2185–2194
Binding of Phospho-TH Peptides to 14-3-3 2189
peak, are labeled in the same colors in Fig. 2, B and C. The
derived binding scheme is based on two assumptions:
1. The bound 31
P chemical shifts of pS19 and pS40 are
identical when saturated with 14-3-3z, irrespective of
whether the second binding site on the 14-3-3z dimer
is occupied or not; and
2. The individual phosphorous resonances in pS19pS40hTH1–50
exhibit the same NMR exchange regime as
those for the singly phosphorylated peptides (Fig. 1, B
and D, i.e., the pS19 resonance is in slow exchange
and that of pS40 is in fast exchange on the 31
P NMR shift
scale).
Note that the slow or fast exchange regimes apply not only
to the interaction between free peptide (19–40) and the one
site occupied 14-3-3z dimer complex ðP/
19À40 or P/
40À19Þ but
also to various additional 14-3-3z complexes (Fig. 2 B). For
example, the pS40 phosphorous resonance of P40
19
j, the complex
in which both sites on the protein dimer engage the two
phosphate groups on 19‒40, is in fast exchange with that of
P/
19À40, the complex in which one site on the dimer engages
the pS19 phosphate group of the peptide, since very weak
binding of the phosphate group of pS40 is involved. On
the other hand, P40
19
j and P/
40À19 are in slow exchange, since
the exchange involves tight binding of the phosphate group
of pS19. In agreement with the above binding scheme, five
distinct peaks are observed in the 31
P spectra during the
titration (Fig. 2 A).
Multiplecomplexescontribute tothe intensityofeachphosphorous
resonance. All peptides with the pS19 phosphate
group free of 14-3-3z protein (19–40, P/
40À19; P40À19
19À40; P40À19
40À19 )
contribute to the peak labeled ‘‘pS19free
’’ (19 colored green in
Fig. 2, B and C). Free peptide (19‒40) as well as any peptides
bound to 14-3-3z via the pS40 phosphate group
ðP/
40À19; P40À19
19À40; P40À19
40À19Þ (40 colored orange in Fig. 2, B and
C) will contribute to the peak labeled ‘‘pS40free
’’ in Fig. 2 A.
Given that the same peptide can contribute intensity to two
different resonances (pS19free
and pS40free
), the total intensity
of the two resonances is expected to be similar, as is indeed
observed in the 31
P spectra (Fig. 2 A).
Any complexes of 14-3-3z that contain peptide bound via
the phosphate group on pS19 (P/
19À40, P19À40
19À40, P40
19
j, and
P40À19
19À40; 19 colored blue in Fig. 2, B and C) add intensity
to the peak labeled ‘‘pS19bound
.’’ The pS40c
peak arises
from the pS40 phosphate groups in the P40
19
j and P/
19À40 complexes
(40 colored purple in Fig. 2, B and C), which are in
fast exchange. The contribution from the free pS40 phosphate
group in P/
19À40 causes the ‘‘pS40c
’’ resonance to
slightly shift toward the free position (4.18 p.p.m.) and,
therefore, it is observed at 3.43 p.p.m., slightly downfield
from the bound position (3.26 p.p.m.) of the singly phosphorylated
pS40 peptide (Fig. 1 D). In contrast, the position
of the pS19bound
peak is at 3.27 p.p.m., identical to the
bound position of the singly phosphorylated pS19 peptide
(Fig. 1 B). The reason for this is that the phosphate groups
that contribute to the pS19bound
peak are in slow exchange,
thus no weighted chemical shift averaging comes into play.
Finally, the free pS40 phosphate groups in the P19À40
19À40 and
P19À40
40À19 complexes (40 colored brown in Fig. 2, B and C), denoted
pS40*, reside on the peptide that is bound to 14-3-3z
via pS19, but they are not buried in the 14-3-3z binding site
and are free in solution, surrounded by solvent. Note that
pS40* is observed as a distinct peak at 3.93 p.p.m., in
slow exchange with the pS40 phosphate resonance (4.18
p.p.m.) of the free peptide (Fig. 2 A), because the peptides
in these complexes are bound to 14-3-3z via the pS19 phosphate
group.
The relationships listed in Fig. 2 C imply that the total
normalized area (denoted as ‘‘A’’) of all peaks in Fig. 2 A
corresponds to 2, because the peptide contains two phosphate
groups:
À
A
À
pS19free
Á
þ A
À
pS40free
Á
þ A
À
pS19bound
Á
þ AðpS40c
Þ
þ AðpS40Ã
Þ
Á
à cpep ¼ 2
À
½19 À 40Š þ
Â
P/
40À19
Ã
þ
Â
P/
19À40
Ã
þ
Â
P40
19
j
Ã
þ 2
Â
P19À40
19À40
Ã
þ 2
Â
P19À40
40À19
Ã
þ 2
Â
P40À19
40À19
ÃÁ
¼ 2 Ã cpep:
(3)
Assignment of phosphorous resonances by
selective thiophosphorylation
The 31
P resonance assignment of the pS19 and pS40 in
the pS19pS40-hTH1–50 peptide (Fig. 2 A) was initially
obtained using the assignment of singly phosphorylated
hTH1–50 peptides (Fig. 1, B and D). We confirmed
the assignment using a hTH1–50 peptide that was phosphorylated
on Ser19
and thiophosphorylated on Ser40
(pS19tpS40-hTH1–50). Thiophosphorylation causes a large
downfield shift of the 31
P resonance from ~4.2 p.p.m. (blue
spectra in Fig. 1 D and Fig. 2 A) to 43.95 p.p.m. (blue
spectra in Fig. 3, A and B). Thus, in the spectrum of the
pS19tpS40-hTH1–50 peptide (Fig. 3 B), both resonances
are clearly separated.
A thiophosphate group is assumed to functionally mimic a
phosphate group, given its close chemical structure. To
confirm that no gross changes to 14-3-3z binding were introduced
into the system by thiophosphorylation, we carried
out titration experiments with tpS40-hTH1–50 and 14-3-3z
under the same conditions that were used for pS40-hTH1–
50. We determined a Kd value of 3.8 mM for tpS40-hTH1–
50 (Fig. 3 A), compared to 1.0 mM for the pS40-hTH1–50
(Fig. 1 D). This demonstrates that binding is slightly reduced
by the replacement of an oxygen atom with a sulfur atom.
Interestingly, binding of 14-3-3z to tpS40-hTH1–50 causes
a downfield 31
P chemical shift (Fig. 3 A), whereas binding
to pS40-hTH1–50 induces an upfield shift (see above, and
Fig. 1 D), illustrating that chemical shifts are extremely sensitive
to any changes and hard to predict.
Whereas 14-3-3z binding to the pS19pS40-hTH1–50
peptide yields three distinct pS40 phosphorous signals that
Biophysical Journal 107(9) 2185–2194
2190 Hritz et al.
exhibit slow exchange, pS40free
, pS40*, and pS40c
(Fig. 2 A),
binding to the pS19tpS40-hTH1–50 peptide results in only
two signals that are in fast-intermediate exchange (Fig. 3
B), caused in part by the smaller chemical shift difference between
free (pS40free
) and bound (pS40c
) resonances (0.60
p.p.m) compared to that for pS19pS40-hTH1–50 (0.94
p.p.m.). However, even with this difference, it seems reasonable
to assume that the binding scheme in Fig. 2 B can be
applied to pS19tpS40-hTH1–50, with Kd2 ¼ 3.8 mM instead
of 1 mM and Keq ¼ 0.26 (see below) rather than 1.5. Using
the well-separated pS19bound 31
P resonance in the thiophosphorylated
peptide (Fig. 3 B) proved beneficial for deconvoluting
the overlapping pS19 resonances in the 14-3-3z-bound
pS19pS40-hTH1–50 spectra (see Fig. S3).
Analysis of the binding equilibrium between
doubly phosphorylated hTH1 peptide to the
14-3-3z homodimer
Fig. 2 B illustrates a comprehensive equilibrium binding scenario
for the interaction between the doubly phosphorylated
hTH1 peptide and the 14-3-3z homodimer. Six possible complexes
are considered. Because insufficient experimental
data for the accurate determination of all individual dissociation
constants are available, it is assumed that each subunit
monomer in the 14-3-3z homodimer binds equivalent phosphate
groups on pS19pS40-hTH1–50 with identical affinity,
irrespective of the state of the other 14-3-3z monomer, i.e., no
cooperativity is present. Therefore, three independent equilibrium
constants, Kd1, Kd2, and Keq (Fig. 2 B), come into
play. Kd1 (0.15 5 0.03 mM) and Kd2 (1.0 5 0.1 mM) are
the experimentally determined dissociation constants for
the interaction between singly phosphorylated hTH1–50
peptides, pS19-hTH1–50 and pS40-hTH1–50, respectively,
and the 14-3-3z dimer (using the monomer at the concentration
unit); Keq is the equilibrium constant defined by the concentration
ratio of the ðP40
19
jÞ complex to the ðP/
19À40Þ
complex. This is treated as the only variable parameter in
this simplistic equilibrium binding model.
We used three approaches for determining Keq, as follows:
In the first approach, we derived a set of six equations for
the individual, independent equilibria, along with two equations
that are necessary to ensure mass conservation of the
peptide and 14-3-3z (see Fig. S2 A). These equations were
solved numerically in the software MATLAB. The resulting
solutions provide the molar concentration of peptide complexes
as a function of the 14-3-3z to doubly phosphorylated
peptide ratio, r (see Fig. S2 B). The equations in Fig. 2 C
link the theoretically predicted molar concentrations of
individual complexes (see Fig. S2 B) with the intensities
of five experimentally observable resonances. By fitting
the theoretical profiles (continuous lines in Fig. 2 D) with
the experimentally observed intensities (individual data
points depicted by x in Fig. 2 D), for several 14-3-3z to peptide
molar ratios, r, a Keq value of 1.5 was determined.
In the second approach, the value of Keq could, in principle,
also be determined from the observed difference in the
‘‘pS40c
’’ resonance position from that of ‘‘pS40bound
,’’ if the
resonance position of pS40 in the P/
19À40 complex is known.
Unfortunately, this is not known experimentally, but,
assuming that it is identical to that of (pS40*), then the
following equation holds:
dðpS40c
Þ ¼
Â
P/
19À40
Ã
dðpS40Ã
Þ þ
Â
P40
19
j
Ã
dðpS40bound
Þ
Â
P/
19À40
Ã
þ
Â
P40
19
j
à : (4)
Considering that
Keq ¼
Â
P40
19
j
Ã
Â
P/
19À40
Ã;
then
Keq ¼
dðpS40c
Þ À dðpS40Ã
Þ
dðpS40boundÞ À dðpS40cÞ
: (5)
If the values of d(pS40*), 3.93 p.p.m.; d(pS40c
), 3.43 p.p.m.;
and d(pS40bound
), 3.26 p.p.m. are applied to Eq. 5, a Keq
value of 3.0 is found.
A
B
FIGURE 3 Binding of thiophosphorylated hTH1–50 peptides to 14-3-3z.
(A) 31
P spectra of tpS40-hTH1–50 peptide (0.71 mM) for varying molar ratios
of 14-3-3z to the peptide (r); 0:1 (blue), 0.5:1 (red), 1:1 (dark green),
2:1 (purple), 4:1 (orange), and 8:1 (brown). (Inset) The chemical shift
changes upon 14-3-3z binding (Dd) are plotted as a function of molar ratio
(r). (Continuous line) Predicted Dd from fitting the experimental Dd to the
quadratic equation shown in Eq. 2. (B) 31
P spectra of pS19tpS40-hTH1–50
(0.83 mM) for varying molar ratios of 14-3-3z to the peptide (r): 0:1 (blue),
0.5:1 (red), 1:1 (dark green), 2:1 (purple), 3:1 (yellow), 4:1 (orange), and
5:1 (light green). To see this figure in color, go online.
Biophysical Journal 107(9) 2185–2194
Binding of Phospho-TH Peptides to 14-3-3 2191
In the third approach, the value of Keq can be estimated,
based on Kd2 and the local concentration of the pS40 phosphate
group in the P/
19À40 complex. In an extended chain
model of the doubly phosphorylated peptide, the distance
between the Ser19
and Ser40
phosphate groups is estimated
to be ~6 nm. The local concentration of the Ser40
phosphate
that resides within a sphere of 6 nm radius, if the Ser19
phosphate from the same peptide is fixed to the first binding
site on the 14-3-3z dimer, would become cloc ~ 1/
(6.022*1023
*4*103
*(6.10À9
)3
)M, i.e., 1.9 mM. Thus, an
estimated value of Keq is
Â
P40
19
j
ÃÂ
P/
19À40
Ã
¼ cloc=Kd2 ¼ 1:9:
The latter two estimated Keq values of 3.0 and 1.9 are very
similar to the value determined from fitting (Keq ¼1.5),
considering that Keq values of 3.0, 1.9, and 1.5 result in relative
populations of 75, 66, and 60%, respectively, of the P40
19
j
complex.
Using the simplified binding model (Fig. 2 B), with Kd1 ¼
0.15 mM, Kd2 ¼ 1.0 mM, and Keq ¼ 1.5, populations of
P/
40À19, P19À40
40À19, and P40À19
40À19 complexes were found to be
significantly lower than other complexes (<5.5%; see
Fig. S2 B). For example, at a 1:1 molar ratio of doubly phosphorylated
peptide to 14-3-3z (r ¼ 1), 37.0% of the peptide
is unbound; 17.7% of the peptide is in the P40
19
j complex;
11.8% is in P/
19À40 complex; and 24.1% (2 Â 12.05%) is
in the P19À40
19À40 complex. Together, all other species account
for 9.4% of the total population. At r ¼ 5, the theoretical
binding model predicts 52.5% of the peptide is in the P40
19
j
complex and 35% is in the P/
19À40 complex. Note that the
31
P resonances of the pS40 phosphate groups in these two
complexes are in fast exchange (Fig. 2 B), resulting in a
single pS40c
peak (Fig. 2 A).
DISCUSSION
In the prevailing model of 14-3-3 binding to doubly phosphorylated
partners, the partners contain a high affinity,
sometimes called a ‘‘gatekeeper’’ residue, which initially interacts
with 14-3-3, and a secondary, weaker epitope, which
only binds when the gatekeeper site is already bound to 14-
3-3 (23). hTH1 that is singly phosphorylated on Ser40
is
thought to be unable to bind 14-3-3z, whereas the pS19
phosphate is the high affinity epitope (18,19). Our NMR results
with the N-terminal 50 residue peptide, pS40-hTH1–
50, indicate, however, that binding of pS40 to 14-3-3z is
indeed possible, as evidenced by the change from the
free peptide phosphorous resonance to the bound one.
This interaction is clearly weaker (Kd ¼ 1.0 5 0.1 mM)
than that between 14-3-3z and pS19-hTH1–50, containing
the purported gatekeeper phosphate, which binds a factorof-10
tighter (Kd ¼ 0.15 5 0.03 mM). Therefore, even under
conditions where 14-3-3z is present in excess over the
pS19pS40-hTH1–50 peptide, two complexes, P40
19
j and
P/
19À40, form in a 60:40% (Keq ¼ 1.5) ratio, irrespective of
the 14-3-3z concentration. Our results show that it is
possible that multiple complexes for doubly phosphorylated
binding partners of 14-3-3z may exist, even if in the phosphorylated
full-length hTH1 additional mechanisms may
come into play and alter the relative binding affinities (18).
Indeed, for the full-length pS40-hTH1, the binding to 14-
3-3g was almost undetectable by native mass spectroscopy
and surface plasmon resonance, while for the doubly phosphorylated
enzyme Ser40
is involved in binding to 14-3-3g
(22). In addition, binding of one 14-3-3g dimer to one of
the monomers in tetrameric hTH1—which is organized as
a dimer of dimers—appears to interfere with the binding
of an additional 14-3-3g dimer to the adjacent monomer
in hTH1, but not to the dimer at the opposite end, as seen
in electron micrographs of the complex (22). Thus, in the
complex involving the full-length enzyme, steric factors
appear to be responsible for moving the equilibrium toward
a scenario where the pS19 site binds to a single monomer
in the 14-3-3 dimer. Whether this is the case for all 14-3-3
interactions with phosphorylated regulatory regions is unclear.
In general, the presence of additional interactions
and their contributions to the overall regulatory mechanisms
have to be ascertained and confirmed for each individual
system.
At this point, it may be of interest to analyze the underlying
thermodynamics that govern the observed interaction
between 14-3-3z and pS19pS40-hTH1–50 (or other similar
peptides, denoted as 1‒2, with two phosphorous sites), with
the P2
1
j bound form as the dominant complex when 14-3-3z
is in excess and P1À2
1À2 dominant when peptide is in excess.
For simplicity, we assume that the binding of site 1 (primary
site) is independent of that of site 2 (secondary site) and that
peptide binding to one binding site on the 14-3-3z dimer
leads to a global entropy decrease (DSG < 0), whereas
engagement of the second site on the same peptide only results
in a local entropy decrease (DSL < 0). The enthalpy
changes corresponding to the binding of the primary and
secondary sites to 14-3-3 protein are denoted as DH1 and
DH2. Given these assumptions, the condition where the
amount of P1À2
1À2 is larger than P2
1
j (when peptide is in excess
over 14-3-3z; r < 1), i.e.,
DG
À
P1À2
1À2
Á
À DG
À
P2
1
j
Á
<0;
results in
DH1 À DH2<TðDSG À DSLÞ: (6)
The opposite inequality
ðDH1 À DH2>TðDSG À DSLÞÞ
is expected for the case when the amount of P2
1
j is larger than
P1À2
1À2 (when r > 1). In the case that the two phosphorous
Biophysical Journal 107(9) 2185–2194
2192 Hritz et al.
groups exhibit comparable binding affinities (e.g., such is
the case for the PKCε-V3 peptide that contains tandem
repeat sequences (27)),
DG
À
P2
1
j
Á
À DG
À
P1À2
1À2
Á
¼ ðDH2 À DH1Þ þ TðDSG À DSLÞ:
Because DH1 À DH2 ~ 0 and T(DSG À DSL) < 0, the free
energy difference
À
DG
À
P2
1
j
Á
À DG
À
P1À2
1À2
ÁÁ
becomes negative, making the amount of P2
1
j complex larger
than the amount of P1À2
1À2 complex, irrespective of the r value.
Considering the large number of 14-3-3 binding partners
and the above-described different binding modes, both cases
are thermodynamically possible and may contribute to the
regulation imparted by 14-3-3 on its targets.
CONCLUSIONS
A detailed analysis of the interaction between singly or
doubly phosphorylated hTH1–50 peptides and 14-3-3z is
presented. Using phosphorous NMR, dissociation constants
for the binding of singly phosphorylated peptides, pS19hTH1–50
and pS40-hTH1–50, to 14-3-3z were determined.
They differ by a factor of 10, with the Ser19
phosphorylated
peptide exhibiting a higher affinity (Kd ¼ 0.15 5 0.03 vs.
1.0 5 0.1 mM). Furthermore, for the doubly phosphorylated
peptide, pS19pS40-hTH1–50, a mixture of distinct
complexes was observed, which permitted us to derive a
comprehensive equilibrium binding scheme. All six
possible different complexes are taken into account, and
the system is described by three equilibrium constants:
Kd1 ¼ 0.15 mM, Kd2 ¼ 1.0 mM, and Keq ¼ 1.5. For excess
pS19pS40-hTH1–50 peptide, the major complex is the 14-
3-3z dimer with two peptides bound via the phosphate group
on Ser19
ðP19À40
19À40Þ, whereas, for excess 14-3-3z, two major
complexes are seen, each one with a doubly phosphorylated
peptide bound to one 14-3-3z dimer, with the Ser19
phosphate
group bound to one 14-3-3z subunit, and the Ser40
phosphate group either bound to the second subunit ðP40
19
jÞ
or free in solution ðP/
19À40Þ in a 60:40 ratio.
For other systems, the underlying thermodynamics will
contribute to whether a doubly phosphorylated partner
will bind with a 1:1 stoichiometry to the 14-3-3z dimer or
if a mixture of complexes is formed. It is likely that most
14-3-3z binding partners will possess phosphorylated residues
that exhibit different affinities for the binding sites
on 14-3-3z. The presented methodology derived by monitoring
individual 31
P resonances has general applicability
for any complex system involving 14-3-3 and doubly phosphorylated
ligands, and it will be interesting to extend this
approach to full-length hTH1 or other target proteins.
SUPPORTING MATERIAL
Three figures are available at http://www.biophysj.org/biophysj/supplemental/
S0006-3495(14)01012-1.
AUTHOR CONTRIBUTIONS
J.H., A.M., and A.M.G. conceived this study; J.H. and I.-J.L.B. acquired
experimental data; J.H., T.K., I.-J.L.B., and A.M.G. performed data analysis
and interpretation; J.H., I.-J.L.B., and A.M.G. drafted the article; all authors
contributed to the final article.
We thank Jinwoo Ahn and Leonardus Koharudin for advice on protein purification,
Doug Bevan and Phil Greer for computer technical support, and
Michael J. Delk for NMR instrumental support. We also thank Teresa Brosenitsch
for critical reading of the manuscript. We are grateful to Dr. Damodaran
Krishnan, Department of Chemistry, University of Pittsburgh, for
access to the Bruker 600 MHz NMR spectrometer with a BBF probe.
J.H. acknowledges financial support by an International Outgoing Fellowship
of the European Community program ‘‘Support for Training and
Career Development of Researchers (Marie Curie)’’, under contract No.
PIOF-GA-2009-235902. This work was supported in part by startup funds
from the University of Pittsburgh School of Medicine.
REFERENCES
1. Morrison, D. K. 2009. The 14-3-3 proteins: integrators of diverse
signaling cues that impact cell fate and cancer development. Trends
Cell Biol. 19:16–23.
2. Gardino, A. K., and M. B. Yaffe. 2011. 14-3-3 proteins as signaling
integration points for cell cycle control and apoptosis. Semin. Cell
Dev. Biol. 22:688–695.
3. Freeman, A. K., and D. K. Morrison. 2011. 14-3-3 proteins: diverse
functions in cell proliferation and cancer progression. Semin. Cell
Dev. Biol. 22:681–687.
4. Johnson, C., S. Crowther, ., C. MacKintosh. 2010. Bioinformatic and
experimental survey of 14-3-3-binding sites. Biochem. J. 427:69–78.
5. Jin, J., F. D. Smith, ., T. Pawson. 2004. Proteomic, functional, and
domain-based analysis of in vivo 14-3-3 binding proteins involved in
cytoskeletal regulation and cellular organization. Curr. Biol.
14:1436–1450.
6. Obsil, T., and V. Obsilova. 2011. Structural basis of 14-3-3 protein
functions. Semin. Cell Dev. Biol. 22:663–672.
7. Yang, X., W. H. Lee, ., J. M. Elkins. 2006. Structural basis for protein-protein
interactions in the 14-3-3 protein family. Proc. Natl.
Acad. Sci. USA. 103:17237–17242.
8. Gardino, A. K. 2006. Structural determinants of 14-3-3 binding specificities
and regulation of subcellular localization of 14-3-3-ligand complexes:
a comparison of the x-ray crystal structures of all human 14-3-3
isoforms. Semin. Cancer Biol. 16:173–182.
9. Yaffe, M. B., K. Rittinger, ., L. C. Cantley. 1997. The structural basis
for 14-3-3: phosphopeptide binding specificity. Cell. 91:961–971.
10. Ganguly, S., J. L. Weller, ., D. C. Klein. 2005. Melatonin synthesis:
14-3-3-dependent activation and inhibition of arylalkylamine n-acetyltransferase
mediated by phosphoserine-205. Proc. Natl. Acad. Sci.
USA. 102:1222–1227.
11. Aitken, A. 2006. 14-3-3 proteins: a historical overview. Semin. Cancer
Biol. 16:162–172.
12. Ichimura, T., T. Isobe, ., H. Fujisawa. 1987. Brain 14-3-3 protein is an
activator protein that activates tryptophan 5-monooxygenase and tyrosine
3-monooxygenase in the presence of Ca2þ
calmodulin-dependent
protein kinase II. FEBS Lett. 219:79–82.
Biophysical Journal 107(9) 2185–2194
Binding of Phospho-TH Peptides to 14-3-3 2193
13. Nagatsu, T., M. Levitt, and S. Udenfriend. 1964. Tyrosine hydroxylase:
the initial step in norepinephrine biosynthesis. J. Biol. Chem.
239:2910–2917.
14. Grima, B., A. Lamouroux, ., J. Mallet. 1987. A single human gene
encoding multiple tyrosine hydroxylases with different predicted functional
characteristics. Nature. 326:707–711.
15. Kaneda, N., K. Kobayashi, ., T. Nagatsu. 1987. Isolation of a novel
cDNA clone for human tyrosine hydroxylase: alternative RNA splicing
produces four kinds of mRNA from a single gene. Biochem. Biophys.
Res. Commun. 146:971–975.
16. Fitzpatrick, P. F. 1999. Tetrahydropterin-dependent amino acid hydroxylases.
Annu. Rev. Biochem. 68:355–381.
17. Daubner, S. C., T. Le, and S. Wang. 2011. Tyrosine hydroxylase and
regulation of dopamine synthesis. Arch. Biochem. Biophys. 508:1–12.
18. Kleppe, R., K. Toska, and J. Haavik. 2001. Interaction of phosphorylated
tyrosine hydroxylase with 14-3-3 proteins: evidence for a phosphoserine
40-dependent association. J. Neurochem. 77:1097–1107.
19. Obsilova, V., E. Nedbalkova, ., T. Obsil. 2008. The 14-3-3 protein
affects the conformation of the regulatory domain of human tyrosine
hydroxylase. Biochemistry. 47:1768–1777.
20. Itagaki, C., T. Isobe, ., T. Ichimura. 1999. Stimulus-coupled interaction
of tyrosine hydroxylase with 14-3-3 proteins. Biochemistry.
38:15673–15680.
21. Skjevik, A. A., M. Mileni, ., A. Martinez. 2013. The N-terminal
sequence of tyrosine hydroxylase is a conformationally versatile motif
that binds 14-3-3 proteins and membranes. J. Mol. Biol. 426:150–168.
22. Kleppe, R., S. Rosati, ., A. Martinez. 2014. Phosphorylation dependence
and stoichiometry of the complex formed by tyrosine hydroxylase
and 14-3-3g. Mol. Cell. Proteomics. 13:2017–2030.
23. Yaffe, M. B. 2002. How do 14-3-3 proteins work? Gatekeeper phosphorylation
and the molecular anvil hypothesis. FEBS Lett. 513:53–57.
24. Kleppe, R., S. Ghorbani, ., J. Haavik. 2014. Modeling cellular signal
communication mediated by phosphorylation dependent interaction
with 14-3-3 proteins. FEBS Lett. 588:92–98.
25. Lian, L. Y., and G. C. K. Roberts. 1993. Effects of chemical exchange
on NMR spectra. In NMR of Macromolecules. A Practical Approach.
R. C. K. Robert, editor. IRL Press, Oxford, United Kingdom, pp.
153–182.
26. Bain, A. D. 2003. Chemical exchange in NMR. Prog. Nucl. Magn.
Reson. Spectrosc. 43:63–103.
27. Kostelecky, B., A. T. Saurin, ., N. Q. McDonald. 2009. Recognition
of an intra-chain tandem 14-3-3 binding site within PKCε. EMBO
Rep. 10:983–989.
28. Molzan, M., and C. Ottmann. 2012. Synergistic binding of the
phosphorylated S233-and S259-binding sites of C-RAF to one 14-3-3
z-dimer. J. Mol. Biol. 423:486–495.
Biophysical Journal 107(9) 2185–2194
2194 Hritz et al.
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
61
Paper 12
Louša, P.; Nedozrálová, H.; Župa, E.; Nováček, J.; Hritz, J.*: Phosphorylation of the regulatory
domain of human tyrosine hydroxylase 1 monitored using non-uniformly
sampled NMR. Biophysical Chemistry 2017, 223, 25-29
Phosphorylation of the regulatory domain of human tyrosine
hydroxylase 1 monitored using non-uniformly sampled NMR
Petr Louša, Hana Nedozrálová, Erik Župa, Jiří Nováček, Jozef Hritz ⁎
CEITEC MU, Masaryk University, Kamenice 753/5, 625 00 Brno, Czech Republic
H I G H L I G H T S
• Disordered part of regulatory domain of
human tyrosine hydroxylase 1 was
assigned.
• Transient alpha-helices are present next
to phosphorylation sites S40 and S19.
• The secondary structure does not
change after phosphorylation.
• The phosphorylation kinetic rates were
measured efficiently using time resolved
NMR.
G R A P H I C A L A B S T R A C T
a b s t r a c ta r t i c l e i n f o
Article history:
Received 30 December 2016
Accepted 24 January 2017
Available online 27 January 2017
Human tyrosine hydroxylase 1 (hTH1) activity is regulated by phosphorylation of its regulatory domain (RDhTH1)
and by an interaction with the 14-3-3 protein. The RD-hTH1 is composed of a structured region (66-
169) preceded by an intrinsically disordered protein region (IDP, hTH1_65) containing two phosphorylation
sites (S19 and S40) which are highly relevant for its increase in activity. The NMR signals of the IDP region in
the non-phosphorylated, singly phosphorylated (pS40) and doubly phosphorylated states (pS19_pS40) were
assigned by non-uniformly sampled spectra with increased dimensionality (5D). The structural changes induced
by phosphorylation were analyzed by means of secondary structure propensities. The phosphorylation kinetics of
the S40 and S19 by kinases PKA and PRAK respectively were monitored by non-uniformly sampled time-resolved
NMR spectroscopy followed by their quantitative analysis.
© 2017 Elsevier B.V. All rights reserved.
Keywords:
Human tyrosine hydroxylase
IDP
NMR
Phosphorylation
Kinetics
Time-resolved NMR
Non-uniform sampling
SSP
1. Introduction
Tyrosine hydroxylase (TH) is an enzyme which converts L-tyrosine
to L-DOPA. This reaction is the rate limiting step in the biosynthetic
pathway producing important catecholamine neurotransmitters: dopamine,
noradrenaline and adrenaline [1,2]. The human enzyme isoform 1
is a tetramer each consisting of three domains: an N-terminal regulatory
domain (RD-hTH1, 1-169 aa), a catalytic domain (170-450 aa), and a
short C-terminal tetramerization domain (451-497 aa) [3]. The first 65
residues of the regulatory domain form an intrinsically disordered protein
region (IDP, hTH1_65) important for regulation. The activity of
hTH1 is controlled by the phosphorylation of its IDP region (S19, S31,
S40) and by the interaction with 14-3-3 protein [4]. Phosphorylation
sites S19 and S40 are the most relevant phosphorylation sites regarding
14-3-3 binding [5,6].
Recently, the NMR structure of the ordered region (65-159) of the
dimeric regulatory domain of rat tyrosine hydroxylase (the sequence
Biophysical Chemistry 223 (2017) 25–29
⁎ Corresponding author.
E-mail address: jozef.hritz@ceitec.muni.cz (J. Hritz).
http://dx.doi.org/10.1016/j.bpc.2017.01.003
0301-4622/© 2017 Elsevier B.V. All rights reserved.
Contents lists available at ScienceDirect
Biophysical Chemistry
journal homepage: http://www.elsevier.com/locate/biophyschem
identity between rat and human RD-hTH1 is 81.8%) was determined by
Zhang et al. [7]. Authors claimed that the problems with unstable sample
and low signal dispersion of the IDP region prevented structural
characterization of the full RD. We overcame these problems by modifying
the preparation of full length RD-hTH1 (1-169) sample in non- and
phosphorylated states and by applying non-uniform sampling (NUS)
NMR approaches allowing much faster data collection in comparison
with the uniformly sampled experiments.
In the past, the rate of the phosphorylation of hTH1 was monitored
semi-quantitatively by Toska et al. [8] using radioactively labeled
ATP. Such methodology has several drawbacks, especially the
necessity of working with radioactive material, laborious sample
preparation and low temporal resolution. The alternative methodology
for a monitoring of protein phosphorylation is NMR spectroscopy
where individual NMR spectra are collected sequentially during
the course of a phosphorylation reaction [9]. The time resolution of
such an approach is on the order of the measurement time of one
particular NMR spectra. Another possibility for monitoring is to measure
one long 2D spectrum and afterwards analyze lineshapes of signals
modulated by the reaction kinetics [10]. In this way, the time
resolution can be reduced at cost of significant increase in difficulty
of analysis. The recent advances in NMR allow us to overcome
these limitations by applying non-uniform sampled time-resolved
NMR spectroscopy to monitor the reaction course with time resolution
down to seconds [11]. This approach was already successfully
utilized for the monitoring of the phosphorylation of cytoplasmic domain
of human B cell receptor protein CD79b [11].
In this study, we present the resonance assignment of the IDP region
(hTH1_65) of RD-hTH1 in the non-, singly- and doubly-phosphorylated
states using non-uniformly sampled NMR experiments with increased
dimensionality. Next, we analyze the structural changes induced by
the phosphorylation of S40 and S19 by PKA and PRAK kinase respectively
and the kinetics of these processes.
2. Experimental
2.1. Protein expression and purification
The regulatory domain (residues 1-169) of human tyrosine
hydroxylase 1 (RD-hTH1) in a pET15b plasmid containing a TEVcleavable
His-tag was expressed in E. coli BL21(DE3)RIL cells.
For preparation of 15
N-labeled and 13
C,15
N-labeled samples, the
cells were cultured in M9 medium with 15
NH4Cl, ampicillin
(100 mg/ml), chloramphenicol (35 mg/ml) and with addition of
13
C6-glucose for the double labeled sample. Cells grew at 37 °C
until OD600 ~ 0.8 then were induced with 0.5 mM IPTG and further
cultured at 18 °C for ~18 h. Cells were harvested and homogenized
in 50 mM Tris pH = 8, 150 mM NaCl, 3 mM NaN3. Cell lysate was centrifuged
for 1 h at 21,040g. Supernatant was then applied on a Ni2+
affinity column (HisTrap HP, GE Healthcare) equilibrated in 50 mM
Tris pH = 8, 500 mM NaCl, 3 mM NaN3. Sample was eluted by gradient
of elution buffer (equilibration buffer + 1 M imidazole) at its
~70% concentration. Sample was then gel filtrated on a Superdex
75 column (HiLoad 16/600 Superdex 75 pg, GE Healthcare) equilibrated
in 50 mM Tris pH = 8, 100 mM NaCl, 3 mM NaN3. The eluted
sample was treated with TEV protease (protein:protease ratio 20:1)
at 4 °C overnight and then dialysed into phosphate buffer (20 mM sodium
phosphate buffer pH = 6, 3 mM NaN3). The His-tag cleaved
protein was loaded on a cation exchange column (Resource S, GE
Healthcare) equilibrated in phosphate buffer (pH = 6.0), and sample
was eluted by a gradient of elution buffer (phosphate buffer, pH =
6.0 + 1 M NaCl) at conductivity 23–45 mS·cm−1
. Fractions containing
our protein sample were again dialysed into phosphate buffer
(pH = 6.0) and then concentrated for final gel filtration on a
Superdex 75 column equilibrated in phosphate buffer.
2.2. NMR samples and phosphorylation
The NMR backbone assignment of the IDP region (1–65 aa) was performed
on [15
N, 13
C] labeled samples of non- and doubly phosphorylated
RD-hTH1 at 1.0 mM concentration in 20 mM sodium phosphate
buffer pH = 6 with 8% D2O and 3 mM NaN3. The phosphorylation kinetic
studies were performed with [15
N] samples at 0.3 mM concentration
in phosphorylation buffer containing 50 mM sodium phosphate buffer
pH = 6, 10 mM ATP, 10 mM MgCl2 with 8% D2O and 3 mM NaN3. For
phosphorylation of S40, PKA (the catalytic subunit of cAMP-dependent
protein kinase, New England BioLabs Inc.) was used in 0.5 μg/ml
(13.2 nM) concentration. Afterwards, S19 phosphorylation using PRAK
(p38 regulated/activated protein kinase, obtained from University
Dundee, Scotland) was performed. The concentration of PRAK was
0.11 mg/ml (2.02 μM). Both phosphorylation reactions were monitored
over the course of 40 h.
2.3. NMR experiments
The assignment of non-phosphorylated RD-hTH1 was performed
using a Bruker 850 MHz US2
spectrometer equipped with cryogenic triple-resonance
probe head (5 mm CPTCI 1H/19F-13C/15N/D). The kinetic
measurements as well as assignment of phosphorylated RD-hTH1
were performed using a Bruker 600 MHz spectrometer equipped with
a cryogenic triple-resonance probe head (5 mm CPQCI 1H-31P/13C/
15N/D). Both probe heads are equipped with z-axis gradient coils. All
measurements were done at a temperature of 293.2 K.
For the assignment of non-phosphorylated and doubly S19_S40phosphorylated
RD-hTH1, 3D HNCO [12], 5D HN(CA)CONH and 5D
HabCabCONH [13,14] were measured. All experiments were carried
out with non-uniform sampling of the indirectly detected domains.
The time schedule was generated using Poisson disk sampling on a
grid, introducing distance constraints between points. The density of
points was set according to a Gaussian distribution (σ = 0.5).
The phosphorylation was monitored using 2D HSQC experiments. In
both cases, the maximal evolution times were set to 102 and 64 ms, respectively.
The time resolution was achieved using non-uniform sampling
in the indirect domain. The sampling schedule comprised 16,000
points in total for both phosphorylations. The size of the schedules
exceeded the size of regular Nyquist grid 125 times. The total measuring
time was 40 h.
2.4. Data processing
The non-uniform 3D HNCO spectra were processed using a Multidimensional
Fourier Transform [15], while 5D HN(CA)CONH and 5D
HabCabCONH spectra were processed using a Sparse Multidimensional
Fourier Transform (SMFT) algorithm [16]. The direct dimension was
square cosine weighted and zero-filled to 4096 complex points, followed
by a standard FFT. The 5D spectra were processed by the SMFT algorithm
using the program reduced, using fixed frequencies of 13
C′, 15
N
and 1
HN
identified in the 3D HNCO spectrum, providing sets of 2D slices.
The assignment and visualization of the NMR spectra was performed in
the software Sparky 3.115 [18].
The secondary structure propensities were calculated by the program
SSP [19] using chemical shifts of 1
Hα
, 13
Cα
and 13
Cβ
. The data
from RefDB [17] were used for random coil referencing. These values
were used for calculation of the phosphorylated state as well.
The time-resolved HSQC spectra were processed using a
coprocessed Multidimensional Decomposition (co-MDD) [11]. The
initial window size was set to 64 points. In the case of PRAK phosphorylation,
the window size was incremented by a factor of 1.05 to reduce
the fitting errors in subsequent analyses. The processing yielded 250
individual frames for PKA and 52 frames for PRAK phosphorylation.
26 P. Louša et al. / Biophysical Chemistry 223 (2017) 25–29
3. Results and discussion
The 15
N- and 13
C,15
N-labeled samples of RD-hTH1 (region 1-169)
protein were expressed and purified as described in the Methods section.
Final purity of prepared non-, singly and doubly phosphorylated
variants was verified by MALDI-TOF-MS spectroscopy (Fig. S1 in
Suppl. mat.). We want to emphasize that in order to study the IDP region
of RD-hTH1, very high purity of samples is needed because of proteolytic
degradation.
3.1. NMR assignment
The assignment of the disordered part (IDP, first 65 residues) within
the RD-hTH1 was performed by employing non-uniform high-dimensional
5D spectra HabCabCONH and HN(CA)CONH. NUS 5D NMR spectroscopy
allows measurement of well resolved spectra of the regions
with fast tumbling (i.e. slow relaxation) in a very efficient way. The
HN(CA)CONH spectrum provided sequential information leading to
linkage of several long fragments. This assignment was then verified
by amino acid classification based on chemical shifts of 1
Hα
, 1
Hβ
, 13
Cα
,
13
Cβ
derived from cross-sections of HabCabCONH spectra. Using this approach,
we were able to assign all amide resonances in region 1-65 and
most resonances of Hα
, Hβ
, Cα
, Cβ
, and carbonyl atoms (including prolines)
except for those directly preceding prolines, i.e. M1, T3, T8, S31,
and V60 and except for Hβ
, Cβ
atoms of L21 and I42 (Tables S1, S2, S3
in Suppl. mat.). In comparison to previously published NMR assignment
of RD of rat TH [7], hereby presented assignment for RD of human TH is
more complete, especially in the region around second phosphorylation
site (S40).
The described assignment procedure was applied to the non- and
double-phosphorylated variants of RD-hTH1. The single phosphorylated
variant (pS40) was assigned based on the very close similarities in
region 1-30 with respect to the non-phosphorylated variant, and region
30-65 was rather similar to the doubly phosphorylated variant. Fig. 1A
presents the superposed assigned HSQC spectra of the IDP region of
RD-hTH1 in the non-, singly- (pS40) and doubly- (pS19_pS40) phosphorylated
states.
Naturally, the largest changes in the chemical shifts (Fig. 1B) are
observed for the phosphorylated residue itself and its neighbors in the
primary sequence. The effect on longer distances seems to be more
Fig. 1. Superposition of HSQC spectra of differently phosphorylated RD-hTH1 (IDP region). The spectra of non-phosphorylated (black), single (pS40) phosphorylated (green) and doubly
(pS19_pS40) phosphorylated spectrum (red) are shown in Panel A. The assignment labels are black for peak positions in non-phosphorylated RD-hTH1, green for signals that changed
significantly during S40 phosphorylation (by PKA) and red for signals that changed during the subsequent S19 phosphorylation by PRAK. Changes of chemical shifts of RD-hTH1
observed in HSQC spectra during phosphorylation of S40 (green) and of S19 (red) are shown in Panel B. The asterisks denote the position of phosphorylation sites.
27P. Louša et al. / Biophysical Chemistry 223 (2017) 25–29
pronounced around the pS40 site with change observed even at M30.
This can be due to the transient secondary structure elements described
in the following section.
3.2. Secondary structural changes induced by phosphorylation
The secondary structure propensities (SSP) for non- and doubly
phosphorylated RD-hTH1 were determined by the approach described
in Methods. The phosphoserines were excluded from this analysis due
to large change in chemical shift of their beta carbon (Table 1). The
SSP values close to zero mean that the particular amino acids are in a
random-coil conformation; the larger positive and negative values indicate
the alpha-helical or beta-sheet conformations, respectively [19].
Fig. 2A indicates slight alpha-helical propensities in region 17-23 and
more pronounced in the region 35-55 for both non- and doubly phosphorylated
variants of RD-hTH1. The intensities of signals in 3D HNCO
spectra in Fig. 2B further support the existence of alpha-helical structure
in the region 35-55. The intensity is related to relaxation properties of
residues. The residues with slower tumbling, i.e. inside structured regions,
relax faster which manifests as lowered intensity of the resulting
signal. The highest intensity is found for the flexible N-terminal residues
as expected, while the lowest intensity corresponds to the alpha-helical
region suggested by SSP.
Both secondary structure propensities (Fig. 2A) as well as the peak
intensities (Fig. 2B) over residues within the IDP region are very similar
for non- and doubly phosphorylated RD-hTH1. There is slight increase
in alpha-helical propensity for region 48-58 and slight decrease of
alpha-helical propensity in region 18-22 around the second phosphorylation
site (pS19). This observation indicates negligible conformational
changes within the RD-hTH1 due to the phosphorylation of S19 and
S40. This is in agreement with the proposed molecular mechanism of
hTH1 activation suggesting that the phosphorylation at S40 and S19 induce
conformational changes of the whole RD with respect to the catalytic
domain rather than within the RD itself [20].
From a methodological point of view, we found it quite surprising
that in contrast to amide groups, the chemical shifts of nuclei that are
usually used for the SSP analysis (1
Hα
, 13
Cα
and 13
Cβ
) are relatively
insensitive to the phosphorylation of serines (Table 1). The standard
deviation of chemical shifts for SSP analysis is on the order of 1 ppm,
which is much more than the differences in Table 1. The only significant
difference is found only for the phosphorylated serine itself and
its beta carbon. This insensitivity of chemical shifts of regions around
phosphoserines in IDP regions allows us to perform SSP analysis
using non-phosphorylated reference values also for other phosphorylated
proteins without the need for intrinsic chemical shift
referencing [21].
3.3. Phosphorylation kinetics
In Fig. 3 we present the time progress of phosphorylation for involved
serine and surrounding well resolved residues. We plot together
decrease in intensity of the original peak and increase in intensity of
newly formed peak after phosphorylation. First, S40 was phosphorylated
using PKA. The progress of phosphorylation was monitored by
changes in the intensity of the HSQC spectrum, which provided the information
to derive a 0th
order rate constant of k = 0.128 ±
0.003 mM·h−1
(Fig. 3A). After 10 h when S40 was fully phosphorylated,
PRAK was added, and phosphorylation of S19 was observed. The intensity
changes can be described by 1st
order kinetics. Fitting provides a
rate constant of k = 0.556 ± 0.008 h−1
(Fig. 3B). After the normalization
of rate constants to micromolar kinase concentrations, we obtained
these values: 9.7 mM·h−1
/μM for PKA and 0.27 h−1
/μM for PRAK.
4. Conclusions
Advanced techniques of non-uniform sampled NMR experiments
were applied to gain information about the intrinsically disordered
region of the regulatory domain of human tyrosine hydroxylase 1
(RD-hTH1) and its phosphorylated variants. The assignment of
non-phosphorylated and doubly phosphorylated (pS19_pS40) samples
was performed using non-uniformly sampled 5D NMR spectroscopy.
These two sets of chemical shifts were also sufficient to
perform assignment of singly phosphorylated (pS40) RD-hTH1. Although
the phosphorylation has quite significant impact on amidic
chemical shifts of phosphoserine and its neighboring residues, it
has negligible effect on aliphatic chemical shifts with the exception
of Cβ
of the phosphoserine itself. Therefore, the aliphatic chemical
shifts could be employed to monitor alpha-helical propensity in the
region between residues 35 and 55 without the need of phosphoprotein
reference. Phosphorylation has no significant impact on the secondary
structure propensities of RD-hTH1.
Table 1
Differences of chemical shifts of nuclei used in SSP calculation for residues close to phosphorylation sites.
A17 V18 S19 E20 L21 R38 Q39 S40 L41 I42
ΔδCα
(ppm) 0.23 0.23 −0.08 −0.51 −0.17 −0.40 0.61 −0.04 −0.26 −0.04
ΔδCβ
(ppm) −0.07 −0.19 1.78 0.25 0.14 0.03 −0.19 1.73 0.09 –
ΔδHα
(ppm) −0.03 −0.05 −0.01 0.00 −0.01 0.03 −0.09 −0.03 0.00 −0.02
Fig. 2. Secondary structure propensities (A) and intensities (B) of signals in 3D HNCO
spectra of non-phosphorylated (black bars) and doubly-phosphorylated (pS19_pS40,
grey bars) RD-hTH1 (IDP region). The asterisks denote phosphorylation sites.
28 P. Louša et al. / Biophysical Chemistry 223 (2017) 25–29
The kinetics of phosphorylation of S40 by PKA and of S19 by PRAK
were determined using non-uniformly sampled time-resolved NMR
spectroscopy. The phosphorylation of S40 was determined as a 0th
order reaction with normalized rate constant value 9.7 mM·h−1
/μM,
while phosphorylation of S19 was observed as 1st
order reaction with
value 0.27 h−1
/μM. The described time-resolved NMR spectroscopy approach
has general applicability over large time scales of various posttranslational
processes and can be easily employed in other protein
systems.
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.bpc.2017.01.003.
Acknowledgment
We thank Dr. Tanvir Shaikh for critical reading of the manuscript.
The project is financed from the SoMoPro II programme. The research
leading to this invention has acquired a financial grant from the People
Programme (Marie Curie action) of the Seventh Framework Programme
of EU according to the REA Grant Agreement No. 291782. The research is
further co-financed by the South-Moravian Region. The article/paper reflects
only the author's views and the Union is not liable for any use that
may be made of the information contained therein. In addition, this
work was also supported by the Czech Science Foundation (15-
34684L). The CIISB research infrastructure project LM2015043 funded
by MEYS CR is gratefully acknowledged for the partial financial support
of the measurements at the Josef Dadok National NMR Centre and
Proteomics Core Facilities, CEITEC – Masaryk University.
References
[1] T. Nagatsu, M. Levitt, S. Udenfriend, Tyrosine hydroxylase: the initial step in
norephinephrine biosynthesis, J. Biol. Chem. 239 (1964) 2910–2917.
[2] P.B. Molinoff, J. Axelrod, Biochemistry of catecholamines, Annu. Rev. Biochem. 40
(1971) 465–500.
[3] S.C. Daubner, D.L. Lohse, P.F. Fitzpatrick, Expression and characterization of catalytic
and regulatory domains of rat tyrosine hydroxylase, Protein Sci. 2 (1993) 1452–1460.
[4] P.F. Fitzpatrick, Tetrahydropterin-dependent amino acid hydroxylases, Annu. Rev.
Biochem. 68 (1999) 355–381.
[5] J. Hritz, I.-J. Byeon, T. Krzysiak, A. Martinez, V. Sklenář, A.M. Gronenborn, Dissection
of binding between a phosphorylated tyrosine hydroxylase peptide and 14-3-3zeta:
a complex story elucidated by NMR, Biophys. J. 107 (2014) 2185–2194.
[6] R. Kleppe, K. Toska, J. Haavik, Interaction of phosphorylated tyrosine hydroxylase
with 14-3-3 proteins: evidence for a phosphoserine 40-dependent association, J.
Neurochem. 77 (2001) 1097–1107.
[7] S. Zhang, T. Huang, U. Ilangovan, A.P. Hinck, P.F. Fitzpatrick, The solution structure of
the regulatory domain of tyrosine hydroxylase, J. Mol. Biol. 426 (2014) 1483–1497.
[8] K. Toska, R. Kleppe, C.G. Armstrong, N.A. Morrice, P. Cohen, J. Haavik, Regulation of
tyrosine hydroxylase by stress-activated protein kinases, J. Neurochem. 83 (2002)
775–783.
[9] F.X. Theillet, H.M. Rose, S. Liokatis, A. Binolfi, R. Thongwichian, M. Stuiver, P. Selenko,
Site-specific NMR mapping and time-resolved monitoring of serine and threonine
phosphorylation in reconstituted kinase reactions and mammalian cell extracts,
Nat. Protoc. 8 (2013) 1416–1432.
[10] I. Landrieu, L. Lacosse, A. Leroy, J.-M. Wieruszeski, X. Trivelli, A. Sillen, N. Sibille, H.
Schwalbe, K. Saxena, T. Langer, G. Lippens, NMR analysis of a tau phosphorylation
pattern, J. Am. Chem. Soc. 128 (2006) 3575–3583.
[11] M. Mayzel, J. Rosenlöw, L. Isaksson, V.Y. Orekhov, Time-resolved multidimensional
NMR with non-uniform sampling, J. Biomol. NMR 58 (2014) 129–139.
[12] L.E. Kay, M. Ikura, R. Tschudin, A. Bax, Three-dimensional triple-resonance NMR
spectroscopy of isotopically enriched proteins, J. Magn. Reson. 89 (1990)
496–514.
[13] K. Kazimierczuk, A. Zawadzka-Kazimierczuk, W. Kozminski, Non-uniform frequency
domain for optimal exploitation of non-uniform sampling, J. Magn. Reson. 205
(2010) 286–292.
[14] V. Motackova, J. Novacek, A. Zawadzka-Kazimierczuk, K. Kazimierczuk, L. Zidek, H.
Sanderova, L. Krasny, W. Kozminski, V. Sklenar, Strategy for complete NMR assignment
of disordered proteins with highly repetitive sequences based on resolutionenhanced
5D experiments, J. Biomol. NMR 48 (2010) 169–177.
[15] J. Stanek, W. Kozminski, Iterative algorithm of discrete Fourier transform for processing
randomly sampled NMR data sets, J. Biomol. NMR 47 (2010) 65–77.
[16] K. Kazimierczuk, A. Zawadzka, W. Kozminski, Narrow peaks and high dimensionalities:
exploiting the advantages of random sampling, J. Magn. Reson. 197 (2009)
219–228.
[17] H. Zhang, S. Neal, D.S. Wishart, RefDB: a database of uniformly referenced protein
chemical shifts, J. Biomol. NMR 25 (2003) 173–195.
[18] T.D. Goddard and D. G. Kneller, SPARKY 3, University of California, San Francisco,
USA.
[19] J.A. Marsh, V.K. Singh, Z. Jia, J.D. Forman-Kay, Sensitivity of secondary structure propensities
to sequence differences between α- and γ-synuclein: implications for fibrillation,
Protein Sci. 15 (2006) 2795–2804.
[20] S.C. Daubner, T. Lee, S. Wang, Tyrosine hydroxylase and regulation of dopamine synthesis,
Arch. Biochem. Biophys. 508 (2011) 1–12.
[21] K. Modig, V.W. Jürgensen, K. Lindorff-Larsen, W. Fieber, H. Bohr, F.M. Poulsen, Detection
of initiation sites in protein folding of the four helix bundle ACBP by chemical
shift analysis, FEBS Lett. 581 (25) (2007) 4965–4971.
Fig. 3. Progress of phosphorylation by PKA kinase is shown in Panel A. The intensities of signals belonging to non-phosphorylated protein (squares) and S40-phosphorylated (triangles) are
plotted against reaction time. Other residues close to S40 were not used due to severe overlap, leading to errors in analysis. Panel B: Progress of phosphorylation by PRAK. Intensities
belonging to singly (pS40) (squares) and to doubly (pS19_pS40) phosphorylated sample (triangles) are plotted against reaction time. Other residues were strongly influenced by
signal overlap and therefore not used in subsequent analysis.
29P. Louša et al. / Biophysical Chemistry 223 (2017) 25–29
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
62
Paper 13
Zapletal, V.; Mládek, A.; Melková, K.; Louša, P.; Nomilner, E.; Jaseňáková, Z.; Kubáň, V.;
Makovická, M.; Laníková, A.; Žídek L.; Hritz, J.* Choice of force field for proteins containing
structured and intrinsically disordered regions. Biophys. J. 2020, 118, 1621–
1633
Article
Choice of Force Field for Proteins Containing
Structured and Intrinsically Disordered Regions
Vojtech Zapletal,1,2
Arnost Mla´dek,2
Katerina Melkova´,1,2
Petr Lousa,2
Erik Nomilner,1
Zuzana Jasena´kova´,1,2
Vojtech Kuba´n,1,2
Marketa Makovicka´,1
Alice Lanı´kova´,1
Luka´s Zı´dek,1,2
and Jozef Hritz2,*
1
National Centre for Biomolecular Research, Faculty of Science and 2
Central European Institute of Technology, Masaryk University, Brno,
Czech Republic
ABSTRACT Biomolecular force fields optimized for globular proteins fail to properly reproduce properties of intrinsically
disordered proteins. In particular, parameters of the water model need to be modified to improve applicability of the force fields
to both ordered and disordered proteins. Here, we compared performance of force fields recommended for intrinsically disordered
proteins in molecular dynamics simulations of three proteins differing in the content of ordered and disordered regions
(two proteins consisting of a well-structured domain and of a disordered region with and without a transient helical motif and
one disordered protein containing a region of increased helical propensity). The obtained molecular dynamics trajectories were
used to predict measurable parameters, including radii of gyration of the proteins and chemical shifts, residual dipolar couplings,
paramagnetic relaxation enhancement, and NMR relaxation data of their individual residues. The predicted quantities
were compared with experimental data obtained within this study or published previously. The results showed that the NMR
relaxation parameters, rarely used for benchmarking, are particularly sensitive to the choice of force-field parameters, especially
those defining the water model. Interestingly, the TIP3P water model, leading to an artificial structural collapse, also resulted
in unrealistic relaxation properties. The TIP4P-D water model, combined with three biomolecular force-field parameters
for the protein part, significantly improved reliability of the simulations. Additional analysis revealed only one particular force
field capable of retaining the transient helical motif observed in NMR experiments. The benchmarking protocol used in our
study, being more sensitive to imperfections than the commonly used tests, is well suited to evaluate the performance of newly
developed force fields.
INTRODUCTION
The fact that many proteins of biological relevance contain
considerably large intrinsically disordered regions (IDRs),
contradicting the classical structure-function paradigm,
has been accepted by the structural biology community
during the past two decades (1–3). Intrinsically disordered
proteins (IDPs) represent a relatively diverse class of molecules
differing in their biophysical properties. One polypeptide
chain often contains fully structured domains
together with IDRs. Moreover, IDRs are not random
polymer chains but exhibit various degree of partial
ordering. They typically contain a short segment with an
increased propensity to form secondary transient structures,
described in the literature as prestructured motifs (4), preformed
structural elements (5), or molecular recognition
features (6). Such hybrid systems present methodological
challenges because an IDR tethered to a well-ordered
domain is a molecule consisting of regions with highly
diverse dynamics. As a result, it is difficult for experimental
and computational methods to accurately capture
both types of behavior in these systems.
Submitted April 17, 2019, and accepted for publication February 5, 2020.
*Correspondence: jozef.hritz@ceitec.muni.cz
Editor: Rohit Pappu.
SIGNIFICANCE We compared the performance of several force fields in molecular dynamics simulations of three
proteins differing in the content of ordered and disordered regions. From the obtained trajectories, we predicted a set of
measurable quantities and compared them with their experimental values. Among the predicted parameters, NMR
relaxation data were particularly sensitive to the choice of force field parameters, especially those defining the water model.
The presented benchmarking protocol will help to select force fields that reliably simulate properties of physiologically
important intrinsically disordered proteins.
Biophysical Journal 118, 1621–1633, April 7, 2020 1621
https://doi.org/10.1016/j.bpj.2020.02.019
Ó 2020 Biophysical Society.
NMR represents a method of choice for studies of proteins
with IDRs at atomic resolution, as disordered systems
are difficult to investigate using single-crystal x-ray diffraction
or single-particle reconstruction of cryo-electron microscopic
images. Currently available NMR methods provide
sufficient resolution to overcome the narrow distribution
of chemical shifts of IDPs (7,8). However, it should be noted
that the resolution improvement of IDP-targeted NMR experiments
relies on the slow relaxation of IDRs. Therefore,
the sensitivity of such experiments is often too low for the
rapidly relaxing signals of amino acids in the well-ordered
regions of hybrid proteins.
Molecular dynamics (MD) simulations could, in principle,
serve as an ideal tool to study behavior of hybrid proteins at an
atomic level. Moreover, most of the NMR parameters can be
reliably predicted from structural models, which allows for
direct comparison of MD results with experimental data. In
practice, several problems complicate MD simulations of
IDPs. The energy landscapes of IDRs are expected to be
weaklyfunneledsuchthatthesearchforaspecificfunctionally
competent conformation could be extremely inefficient (9). In
thisstudy, weexamined the applicability ofMDsimulations to
hybrid proteins and assessed their reliability by predicting
several measurable parameters from the obtained trajectories
and comparing them with experimental data. Moreover, we
provide an example of prediction of measurable parameters
as a guide to select optimal setup for future experiments,
such as positions with paramagnetic relaxation enhancement
(PRE) labels. We tested the currently available force fields
Amber99SB-ILDN (A99), CHARMM22* (C22*), and
CHARMM36m (C36m) in combination with explicit solvent
models TIP3P, TIPS3P, and TIP4P-D. Chemical shift, residual
dipolar coupling (RDC), PRE, relaxation rate, and smallangle
x-ray scattering (SAXS) experimental data were used
for validation. It should be pointed out that our goal was not
to describe the properties of the studied proteins as faithfully
as possible but to look for features that would distinguish
the performance of various force fields in MD simulations of
a microsecond time range. The hybrid proteins investigated
in this study included 1) d subunit of RNA polymerase from
Bacillus subtilis (dRNAP), 2) regulatory domain of human
tyrosine hydroxylase (RD-hTH), and 3) a fragment consisting
of residues 159–254 of rat microtubule-associated protein 2c
(MAP2c159–254
). dRNAP makes RNA polymerase sensitive
to the concentration of initiating nucleoside triphosphates,
which is important for rapid changes in gene expression
(10). It consists of two domains of a similar size. The N-terminal
half of dRNAP folds into a well-ordered, mostly a-helical
domain, whereas the C-terminal half is disordered and highly
negatively charged, with the exception of a lysine-rich motif
96
KAKKKKAKK104
and C-terminal Lys173 (11). Experimental
data showed that the lysine stretch makes transient
electrostatic contacts with various residues in the acidic C-terminal
region (1). No sign of formation of transient a-helical
structures was observed in the C-terminal domain, which prefers
extended backbone conformations, presumably because
of the electrostatic repulsion of the acidic side chains.
RD-hTH catalyzes hydroxylation of L-tyrosine to L-3,4-dihydroxyphenylalanine
(L-DOPA) and is a key and rate-limiting
enzyme in biosynthesis of important catecholamine neurotransmitters
(12,13). Its N-terminal region ($40% of its
sequence) is disordered but contains a segment with $80%
propensity to form four turns of a-helix (14). MAP2c159–254
corresponds to the central region of MAP2c, where proteins
regulating the microtubule-stabilizing activity of MAP2c
bind in a phosphorylation-dependent manner (15,16).
MAP2c159–254
is mostly disordered but exhibits an $20% propensity
to form four turns of an a-helix presumably important
for intermolecular interactions (16,17).
MATERIALS AND METHODS
NMR spectroscopy
NMR assignments of dRNAP and of the disordered part of RD-hTH
were published previously (14,18). MAP2c159–254
was assigned as
described for full-length MAP2c (19). PRE data of dRNAP were published
previously (11). RDCs were calculated as a difference between
splitting observed in in-phase/anti-phase (IPAP) spectra (20) obtained
for proteins in stretched 5% polyacrylamide gel and in isotropic medium.
The RD-hTH data were acquired at 20
C on a 600 and 850 MHz Bruker
Avance III spectrometer (Bruker, Billerica, MA), and the MAP2c159–254
data were acquired at 27
C on a 600 MHz Bruker Avance III spectrometer.
The backbone amide 15
N relaxation data were published previously
for dRNAP (21) and measured using standard pulse sequences (22) for
1.0 mM [15
N]-RD-hTH and 0.34 mM [15
N]-MAP2c159–254
at 27
C on
850 and 950 MHz Bruker Avance III spectrometers, respectively. The
interscan delays were set to 1.5, 2, 6, and 25 s for R1, R2, and steady-state
heteronuclear Overhauser effect (ssNOE) measurements without and
with ssNOE transfer, respectively. The relaxation delays for the R1
experiment were 11.2, 16.8, 28.0, 44.8, 61.6, 95.2, 196, and 308 ms,
and the delays for the R2 experiment were 0, 14.4, 28.8, 43.2, 57.6,
86.4, and 129.6 ms. R1 and R2 data were fitted using a two-parameter
exponential. The errors were obtained using the bootstrap procedure
(23). The ssNOE values were calculated as a ratio between signal
intensities obtained from spectra with or without 1
H saturation. The errors
were derived from background noise levels in each individual
spectrum.
SAXS
The SAXS data sets were collected using a BioSAXS-1000 (Rigaku, Tokyo,
Japan) instrument with an x-ray beam wavelength of 1.54 A˚ at 27
C. The
distance between the sample and the detector (PILATUS 100K; Dectris,
Baden-Daettwil, Switzerland) was 0.48 m, covering a scattering vector
(q ¼ 4psin(q)/l) range from 0.009 to 0.65 A˚ À1
. For solvent and sample,
one two-dimensional image was collected with 1 h exposure time per image.
Radial averaging of two-dimensional scattering images and the solvent
subtractions were performed using SAXSLab3.0.0r1 (Rigaku). All data sets
were truncated to a maximal scattering vector of 0.3 A˚ À1
for further analysis.
Radii of gyration were determined using PRIMUS from ATSAS v2.7.2
(24) with Guinier analysis (implemented in the PRIMUS Guinier Wizard)
in the range 2–52 for dRNAP and in the range 9–30 for MAP2c159–254
;
the globular particle type was used for both proteins. The molecular form
factor analysis (25) was performed online at http://sosnick.uchicago.edu/
SAXSonIDPs.
Zapletal et al.
1622 Biophysical Journal 118, 1621–1633, April 7, 2020
Computational details
The MD simulations were performed using Amber99SB-ILDN (26),
CHARMM22* (27), and CHARMM36m (28) force-field parameters for
the protein atoms and the TIP3P, TIPS3P (29,30), or TIP4P-D (31) water
models. The proteins were solvated using a rhombic dodecahedral box of
waters with a minimal distance between the box walls and solute of 2 nm.
The charge of the system was neutralized by adding ClÀ
and Naþ
ions,
and the concentration of salt was adjusted to 100 mM. All simulations
were performed under periodic boundary conditions. Before the MD
runs, in vacuo and solvent energy minimizations with the steepest descent
algorithm were carried out. The lengths of bonds with hydrogen atoms
were constrained using the LINCS algorithm (32). An integration time
step of 2 fs was used. A cutoff of 1.0 nm was applied for the LennardJones
interactions and short-range electrostatic interactions. Long-range
electrostatic interactions were calculated by particle mesh Ewald summation
with a grid spacing of 0.12 nm and a fourth-order interpolation (33).
The four-step 8-ns-long equilibration protocol consisted of the following
parts: 1) 2-ns relaxation of water molecules at 300 K during the NVT
equilibration with restrained (1000 kJ molÀ1
nmÀ2
) solute coordinates,
2) 2-ns NVT (300 K) run with restrained (1000 kJ molÀ1
nmÀ2
) backbone
atom coordinates, 3) 2-ns NpT (300 K, 1 atm) run with restrained (1000 kJ
molÀ1
nmÀ2
) backbone atom coordinates, and 4) 2 ns of unrestrained NpT
simulation (300 K, 1 atm). The length of the follow-up production NpT
simulations (300 K, 1 atm) was 200 ns. The simulations using the
TIP4P-D water model were further prolonged to 1 ms, and additional independent
simulations (500 ns for dRNAP and MAP2c159–254
, 400 ns
for RD-hTH) starting from different initial conditions were run for all
listed force fields and protein systems. The temperature and pressure
were maintained using the Berendsen coupling scheme (34) during the
equilibration steps; the production NpT simulations were performed using
the velocity rescaling thermostat with a stochastic term (35) and the
Parrinello-Rahman barostat algorithms (36). Atomic coordinates were recorded
every 1 ps.
Predictions of NMR parameters
Chemical shifts were calculated using the prediction algorithm SPARTAþ
(37) for each structure, and averaged secondary chemical shifts (SCSs)
were calculated by subtracting the random-coil values (38). RDC calculations
using a local alignment window were performed. For calculations using
a local alignment window, the RDC, calculated using the program
PALES (39), for the central amino acid of the local 15-amino-acid segment
was calculated for each conformer (40). The resulting RDC profile along
the primary sequence was calculated by averaging each value over the
whole trajectory and multiplying by the corresponding scaled absolute
value of the generic baseline to account for long-range effects (41). The
scale was chosen so that the lowest root mean-square deviations (RMSDs)
from the experimental values were obtained in the disordered regions. PRE
was calculated for the spin label used in the experiments, i.e., for thiol-reactive
methanethiosulfonate (MTSL) attached to a side chain of cysteine
introduced by site-directed mutagenesis. Sterically allowed MTSL sidechain
conformations were sampled using previously published rotameric
distributions (42) and built explicitly for each spin-label site of each individual
structure backbone. 600 side-chain conformers were calculated,
and the sterically allowed conformers were retained. Relaxation effects
were averaged over these conformers as described by Salmon et al. (41).
The strategy described previously (43) was used to calculate relaxation
rates. The autocorrelation function Ci(t) was calculated from subtrajectories
of each simulation, probing two timescales (tmax % 5 or 50 ns).
Each trajectory was thus divided into the blocks of 10 and 100 ns, respectively,
and averaged. For each averaged block, Ci(t) was described as a sum
of N ¼ 512 exponentials eÀt=tc;i
whose amplitudes Ai were obtained using a
Tikhonov regularization procedure (44). For each averaged block, the spectral
densities are then defined as
JiðuÞ ¼
XN
i ¼ 1
Aitc;i
1 þ u2t2
c;i
(1)
and used to predict spin relaxation rates.
RESULTS AND DISCUSSION
Impact of selected water model
Our first goal was to examine the performance of various
models of water in simulations of three hybrid proteins
used as test molecules in this study. It is well described in
the literature (31) that the water models typically used in
MD simulations (e.g., TIP3P) significantly underestimate
London dispersion interactions. To prevent this problem,
TIPS3P and TIP4P-D water models were also tested, and
the reliability of the behavior of IDPs or the IDRs was
analyzed. Modified TIP3P containing nonzero van der
Waals parameters was introduced originally in 1998 (30)
and shown to provide more realistic results for IDPs when
combined with C36m (28). Simulations using this model
typically result in extended states when other models of water
tend to produce ensembles that are structurally too
compact relative to experiments (31).
Our preliminary 200-ns MD runs, using the TIP3P water
model in combination with A99 (26), confirmed the artificial
behavior of TIP3P.
The inferior performance of TIP3P was manifested most
clearly by the calculated radius of gyration (Rg) of
MAP2c159–254
. All simulations started with conformations
having Rg close to the experimental value of 2.5 nm obtained
from SAXS. During the initial 100 ns of simulations with
TIP3P, Rg dropped to $1.5 nm regardless of the protein
force field used (Fig. 1 b). This result confirms that attractive
interactions leading to formation of compact structures are
unnaturally enhanced when TIP3P is used, as was also reported
previously in (31,45). We also ran 200-ns simulations
with the TIPS3P model (30) in combination with the C22*
(26) and C36m (28) force fields. Remarkably, TIPS3P did
not prevent the artificial compaction of MAP2c159–254
(Fig. 1 b; a comparison of trajectories calculated using
C36m with TIP3P and TIPS3P is presented in Fig. S1). A
similar trend was observed also for RD-hTH (Fig. 1 c),
although a direct comparison with the experiment was not
possible because of the RD-hTH dimerization in real
samples.
Finally, we performed MD simulations with the TIP4P-D
water model combined with A99, C22*, and C36m. In agreement
with the literature (31) and in a sharp contrast with the
simulations run with TIP3P and TIPS3P, Rg of MAP2c159–254
did not drop below the value calculated from the SAXS
data (Fig. 1 e). In the case of RD-hTH, TIP4P-D also prevented
unexpected compaction but only in combination
with C36m (see the discussion of the impact of the protein
force field parameters in the following section).
Force Field for Disordered Regions
Biophysical Journal 118, 1621–1633, April 7, 2020 1623
Interestingly, the water model influenced simulated Rg of
dRNAP less than Rg of MAP2c159–254
and RD-hTH. We
observed the artificial collapsed structure only rarely in simulations
of dRNAP with TIP3P or TIPS3P (see A99 data in
Fig. 1 a). The most likely explanation is that the C-terminal
IDR of dRNAP is very strongly negatively charged, preventing
its collapsed structure regardless of applied water model
(Fig. S2). To examine the influence of water models in the
simulations of dRNAP more closely, we also calculated
NMR relaxation parameters from the MD trajectories. As
discussed below in more details, the prediction of relaxation
parameters is a challenging task. Therefore, we hoped that a
comparison of calculated and experimental relaxation parameters
might reveal more subtle effects. Indeed, we
observed much lower predicted values of ssNOE for residues
134–173 of dRNAP (Fig. 2 a). Such values indicate
an artificially high disorder of the highly acidic region
further than 30 residues from the positively charged lysine
stretch (residues 96–104). Importantly, this effect was
observed not only for the original TIP3P model but also
for the modified version TIPS3P (red and blue traces in
Fig. 2 a). Limitations of TIPS3P were already noticed by
Huang et al., who reported that this model with C36m provided
correct Rg for the arginine-serine but underestimated
the Rg of the Thermotoga maritima cold-shock protein and
of the N-terminal domain of HIV-1 integrase (28). A further
modification of TIPS3P improved the Rg prediction for the
cold-shock proteins but worsened the agreement for the
other two proteins (28).
In conclusion, significant differences between water
models were already observed during the first 200 ns of
the simulations. Considering the results for TIP3P, TIPS3P,
and TIP4P-D, we continued our study only with the TIP4PD
model, extending the MD simulations to the microsecond
range.
Impact of force fields on global shape of proteins
We first examined the global shape of studied proteins in
terms of Rg and SAXS curves. The Rg-values calculated
from 1-ms (Fig. 1) and 0.5-ms (Figs. S3 and S4) A99,
C22*, and C36m trajectories of dRNAP (simulated using
TIP4P-D) oscillated between 3 and 6 nm. The experimental
Rg-value, obtained by the Guinier analysis of the SAXS
curve, was (3.45 5 0.3) nm. The Rg alone did not reveal
any systematic difference among the force fields, except
for an extending event observed around 550 ns in the
C36m simulation.
The Rg-values calculated from MD trajectories of
MAP2c159–254
oscillated between 2 and 4.5 nm, close to
the experimental value of (2.5 5 0.3) nm (Fig. 1 e).
Figs. S5 and S6 show the comparisons between the predicted
and measured SAXS profiles for dRNAP and
MAP2c159–254
and for individual force fields. The lowest
Q-factors, indicating the best fit (Table 1), were obtained
FIGURE 1 Simulated Rg of dRNAP (a and d), MAP2c159–254
(b and e),
and RD-hTH (c and f) obtained using the TIP3P (TIPS3P in the case of
C22* and C36m) water model (a–c) and TIP4P-D water model (d–f).
Experimental data are shown in gold, and values calculated using A99,
C22*, and C36m are shown in green, red, and blue, respectively. To see
this figure in color, go online.
-2.0
-1.0
0.0
1.0
ssNOE
0.0
2.0
4.0
6.0
Γx/s-1
0.0
1.0
2.0
R1/s-1
a TIP3P
TIPS3P
TIPS3P
b TIP3P
TIPS3P
TIPS3P
c TIP3P
TIPS3P
TIPS3P
d TIP4P-D
e TIP4P-D
f TIP4P-D
100 120 140 160
Residue number
100 120 140 160
Residue number
FIGURE 2 ssNOE (a and d) and relaxation rates Gx (b and e) and R1 (c
and f) of the C-terminal IDR of dRNAP obtained with the 5-ns sliding window
using the TIP3P (TIPS3P in the case of C22* and C36m) water model
(a–c) and TIP4P-D water model (d–f). Experimental data are shown in gold,
and values calculated using A99, C22*, and C36m are shown in green, red,
and blue, respectively. To see this figure in color, go online.
Zapletal et al.
1624 Biophysical Journal 118, 1621–1633, April 7, 2020
for the C22* force field. Inspection of individual SAXS
profiles revealed that C36m and A99 overestimated
SAXS intensities for MAP2c159–254
and underestimated
SAXS intensities for dRNAP in the medium- and low-q
region (0.1 nmÀ1
< q < 1 nmÀ1
; see Table 1), whereas
C22* predicted the medium- and low-q SAXS intensities
well.
We also fitted the experimental and predicted SAXS
profiles to molecular form factors (MFFs) developed by
Riback et al. (25). Fitting the experimental MAP2c159–254
data to an MFF provided Rg ¼ (2.87 5 0.3) nm and a value
of the Flory exponent typical for a fully unfolded polypeptide
(n ¼ 0.59 5 0.02). Very similar values were obtained
by fitting the profile simulated by C22*, Rg ¼ (2.952 5
0.003) nm and n ¼ 0.603 5 0.001. As expected, the MFF
did not fit well the SAXS profile of dRNAP, containing a
large well-ordered domain.
The effect of a force field on the global conformations of
RD-hTH was examined as well (Fig. 1 f). Compact structures
with Rg z 2 nm were formed after 500 ns
in simulations with A99 and C22* force fields but not
with C36m. Therefore, it seems that CH36m in combination
with TIP4P-D most efficiently prevents the collapse of
structures of RD-hTH and MAP2c, representing proteins
not exhibiting extraordinary electrostatic repulsion.
Impact of force fields on long-range contacts
After evaluating the effect of different force-field parameters
on the global shapes of the studied proteins, we
compared the ability of the force fields to properly reproduce
contacts between residues further apart in the
sequence. We started by inspecting dRNAP as a test system
for which long-range contacts are observed experimentally,
yet no transient helicity is observed in the C-terminal IDR.
Residue pairwise distance maps represent an efficient
way to evaluate contacts formed during individual MD
runs. Rectangular boxes in the maps presented in Fig. 3
highlight distances 1) between the lysine-rich stretch
96
KAKKKKAKK104
and highly acidic residues in the C-terminal
region and 2) between the ordered N-terminal and
disordered C-terminal domains. Differences between distances
represented by colors document that the relative
average distances varied depending on the force field
used. For example, the lysine tract interacted with the residues
in the vicinity of Glu120 more strongly in the A99
simulation than in the runs with the C22* and C36m force
fields (blue rectangles in Fig. 3). The average distances between
the N-terminal domain and vicinity of Glu120 (red
rectangles in Fig. 3) also differed. We want to emphasize
that the reliability of distance maps is influenced not only
by the capabilities of individual force fields applied for
TABLE 1 Comparison of Calculated RMSD from Experimental Values and Normalized scores
Metric dRNAP RD-hTH MAP2c159–254
A99 C22* C36m A99 C22* C36m A99 C22* C36m
RMSD:
DdCa
/ppm 0.92 0.88 0.82 0.92 0.84 0.64 0.74 0.44 0.55
DdCb
/ppm 0.86 0.78 0.79 0.63 0.52 0.43 0.45 0.45 0.41
DdC(O)/ppm 0.74 0.67 0.62 0.96 0.90 0.56 0.84 0.55 0.65
DdN/ppm 1.71 1.92 1.67 3.55 3.40 2.88 1.54 1.90 1.07
Q for D(NHN
) 0.30 0.33 0.33 1.21 1.12 0.89 0.76 0.69 0.74
Local PREa
by MTSL at L110C 0.25 0.41 0.26 n.d.b
n.d. n.d. n.d. n.d. n.d.
Local PRE by MTSL at L132C 0.07 0.17 0.16 n.d. n.d. n.d. n.d. n.d. n.d.
Local PRE by MTSL at L151C 0.12 0.13 0.14 n.d. n.d. n.d. n.d. n.d. n.d.
Local PRE by MTSL at L168C 0.06 0.06 0.07 n.d. n.d. n.d. n.d. n.d. n.d.
PRE (all data) 0.08 0.10 0.09 n.d. n.d. n.d. n.d. n.d. n.d.
ssNOE 0.15 0.17 0.20 0.11 0.38 0.25 0.15 0.20 0.16
R2 or Gx/sÀ1
1.50 1.56 2.46 2.62 2.32 0.93 1.13 1.32 1.09
R1/sÀ1
0.12 0.15 0.14 0.19 0.24 0.13 0.21 0.19 0.23
Q for SAXS (0.1 nmÀ1
< q < 1 nmÀ1
) 0.057 0.040 0.084 n.d. n.d. n.d. 0.070 0.048 0.075
Q for SAXS (all data) 0.060 0.045 0.085 n.d. n.d. n.d. 0.084 0.066 0.083
Score:
sCS 1.11 1.08 1.00 1.46 1.32 1.00 1.42 1.22 1.11
sRDC 1.00 1.09 1.08 1.33 1.20 1.00 1.06 1.00 1.36
sPRE (all data) 1.00 1.25 1.11 n.d. n.d. n.d. n.d. n.d. n.d.
srelax 1.00 1.13 1.37 1.74 2.57 1.44 1.04 1.16 1.07
sNMR 1.00 1.13 1.44 1.54 1.89 1.22 1.06 1.12 1.07
sSAXS (all data) 1.32 1.00 1.87 n.d. n.d. n.d. 1.28 1.00 1.26
sall 1.08 1.08 1.21 1.55 1.78 1.17 1.24 1.15 1.11
Rg/nm 4.13 4.03 4.66 n.d. n.d. n.d. 2.60 2.94 2.78
Rg penalty 0.13 0.10 0.28 n.d. n.d. n.d. 0 0.06 0
scombined 1.21 1.18 1.49 1.55 1.78 1.17 1.24 1.21 1.11
a
Calculated for PRE of residues 84–104 because of the indicated spin label.
b
Input experimental data not determined.
Force Field for Disordered Regions
Biophysical Journal 118, 1621–1633, April 7, 2020 1625
the dRNAP in combination with the TIP4P-D water model
but also by limited sampling within the trajectories of the
cumulative length of 1.5 ms.
To directly compare the contacts observed during the
MD calculations with the experimental data, we simulated
PRE of individual residues of dRNAP for a series of spinlabel
positions examined in a previous experimental
study (11). The results showed that A99 realistically reproduced
experimentally observed contacts of the lysine
stretch 96
KAKKKKAKK104
with labels placed at L110C
and L132C in the highly acidic C-terminal sequence
(Fig. 4, a and b). The experimentally detected contact
with L151C was also predicted. In agreement with the
distance map, C22* overestimated contacts of the lysines
with the closest label at L110C and underestimated
contacts with more distant labels, including L132C.
C36m did not overestimate contacts with the label at
L110C but underestimated contacts with the label at
L132C similarly to C22* and A99 (see RMSD for PRE
in the lysine stretch for individual spin labels, listed in
Table 1).
In conclusion, correct prediction of electrostatic contacts
between residues apart in the sequence is a challenging task.
In the case of dRNAP, A99 with TIP4P-D performed best,
being able to predict reliably distances between amino acids
separated by up to 50 residues in the sequence.
Impact of protein force-field parameters on local
conformations
In the next step, we analyzed the accuracy of description of
local backbone conformations of dRNAP in the simulations.
For this purpose, we compared experimental values
of several NMR parameters reflecting the local backbone
conformation with the values calculated from snapshots
of the MD simulations. First, we checked values of RDC.
In principle, RDCs depend both on local conformation
and on the overall shape of the molecule, determining the
distribution of orientations of the molecule in a partially
aligned environment (20). However, it is very demanding
to achieve a good sampling of conformations and orientations
to faithfully reproduce experimental data without
any prior information. Therefore, we used the knowledge
of long-range electrostatic contacts in the dRNAP molecule,
obtained experimentally as PRE, and applied the local
averaging window (40) to predict RDC values. The predicted
RDC values are plotted in Fig. 5 a. The prediction
varied especially in the vicinity of the lysine stretch, in
FIGURE 3 Maps describing the distances between residue pairs of dRNAP simulated using A99 (a and b), C22* (c and d), and C36m (e and f) force fields
in combination with TIP4P-D. The scales shown at the right indicate color coding of mean inter-residue Ca
-Ca
distances (lower row) and of populations of
events when the distance between any pair of residue atoms was shorter than 0.4 nm (upper row). To see this figure in color, go online.
Zapletal et al.
1626 Biophysical Journal 118, 1621–1633, April 7, 2020
which A99 and C36m achieved better agreement with the
experiment than C22*.
The second examined NMR parameter reflecting local
conformation was the chemical shift. In comparison with
RDC, the chemical shifts are measured with higher precision,
and their prediction does not require a prior knowledge
of molecular orientation. SCSs (deviations of predicted
chemical shifts from their random-coil values (38)) are
compared with the experimental data in Fig. 5, b–e. Values
provided by all force fields agree well with the experimental
chemical shifts, except for some mismatch of data predicted
by C22* for the lysine stretch.
In summary, all force fields predicted local conformation
of the extended disordered region of dRNA reasonably well.
The slightly worse prediction in the lysine-rich motif by
C22* most likely reflects less accurate description of electrostatic
contacts by the force field.
Simulation of transient helical regions
In the next step, we tested how different force fields describe
transient a-helical elements in (partially) disordered proteins.
A propensity to adopt a-helical secondary structure
represents another level of complexity of IDP conformations
not present in the mostly extended C-terminal domain
of dRNAP. To explore its effect on the simulations, we inspected
MD trajectories of RD-hTH and MAP2c159–254
.
These proteins were chosen so that they differ in a-helical
propensity.
Experimental values of chemical shifts (17,19) indicate
that RD-hTH and MAP2c159–254
form a-helices in the regions
40–53 and 200–216 with $80 and 20% propensity,
respectively (cf. Fig. 5). We performed a set of simulations
with different force fields, starting from structures containing
ideal a-helices in the experimentally identified a-helical
regions, and observed the stability of the helices (Figs. S10–
S12). During MD simulations using A99 and C22* force
fields, the initially present a-helix unfolded in less than
80 ns (Fig. S12). Huang et al. (28) reported that C36m optimized
with TIPS3P 1) correctly simulated transient a-helices
and 2) provided correct Rg for the arginine-serine
peptide but not for other two tested IDPs. Therefore, we
were curious whether C36m would keep its ability to maintain
the transient a-helices with TIP4P-D. For RD-hTH,
C36m with TIP4P-D maintained the a-helical conformation
in the whole 1-ms trajectory and for 220 ns in an independent
400-ns run (Fig. S13). In the case of MAP2c159–254
, the transient
a-helix unfolded after $35, 85, and 830 ns in three independent
runs starting from conformations including the
helix at the beginning (Fig. S9). The lower stability of the
MAP2c159–254
helix in simulations is in agreement with its
low (20%) population observed experimentally. Formation
of the well-defined helix was not observed in the remaining
465, 415, and 170 ns of the trajectories or during two 0.5-ms
runs started from conformations without the helix. However,
temporary formation of the helix (for $50 ns) was
observed in simulations of MAP2c159–254
using C22* (Fig.
S12 a) and A99 (Fig. S12 d). It should be emphasized that
the calculated trajectories do not fully sample the equilibrium
canonical ensembles. Therefore, we do not expect to
observe quantitative agreement between the experimental
and simulated populations of transient a helices.
The ability of the force fields to reliably describe transient
a-helices was directly reflected by the predicted
SCS values (Fig. 5). Outside of the helical region, a good
agreement of the predicted and experimental chemical
shifts was obtained for all force fields tested. Deviations
from the experimental values were observed for the A99
and C22* simulations in the region where the originally
present a-helix unfolded. For the C36m simulations,
SCSs typical for a-helices were obtained in the regions
where the a-helix was modeled (Fig. 5, g–i and l–n). Quantitatively,
the values of predicted versus SCSs corresponded
to 87% populations of the helix in the simulations vs.
$80% population in the real sample of RD-hTH. In the
case of MAP2c159–254
, data from all trajectories were combined
in a ratio corresponding to the experimentally estimated
20% population of the helix.
In conclusion, the ability of C36m and TIP4P-D to keep
the transient a-helices during the simulation and to prevent
the artificial collapse of IDR structures suggests that this
combination is most robust for our studied proteins.
This is noteworthy considering that C36m was not optimized
in combination with TIP4P-D, but with TIPS3P and
0.0
0.2
0.4
0.6
0.8
1.0
0.0
0.2
0.4
0.6
0.8
1.0
0.0
0.2
0.4
0.6
0.8
1.0
PRE
0.0
0.2
0.4
0.6
0.8
1.0
a
b
c
d
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170
Residue number
FIGURE 4 Simulated PRE of dRNAP with the spin label at L110C (a),
L132C (b), L151C (c), and E168C (d). Experimental data are shown in
gold, and values calculated from MD simulations using the TIP4P-D water
model combined with the A99, C22*, and C36m force fields are shown in
green, red, and blue, respectively. To see this figure in color, go online.
Force Field for Disordered Regions
Biophysical Journal 118, 1621–1633, April 7, 2020 1627
its variants (28). We believe that the C36m and TIP4P-D
combination is a promising candidate for a benchmarking
on a wider range of proteins, as was done, e.g., by Robustelli
et al. (45).
It should be noted that the tests discussed so far utilize a
limited set of the most readily available experimental
values. We already discussed that NMR relaxation data
distinguished the performance of TIP3P and TIP4P-D water
models better than Rg in the A99 and C36m simulations of
dRNAP (details are presented in the section Impact of
Selected Water Model). In the next section, we examine
whether extending the benchmarking to NMR relaxation
helps to discriminate the ability of force fields to describe
dynamic properties of hybrid proteins.
Simulation of NMR relaxation data
To test the examined force fields within the MD framework
more thoroughly, we compared their abilities to reproduce
NMR relaxation rates. The NMR signal relaxes because of
the local magnetic fields that fluctuate as a result of stochastic
molecular motions. The stochastic reorientation of vectors
describing the interactions contributing to relaxation
is described by the correlation function. In real samples,
large numbers of molecules with an almost isotropic distribution
of orientations are measured. Consequently, the correlation
functions have a simple analytical form (series of
exponential functions). In simulations, the sufficient sampling
of the orientations is difficult to achieve (in principle,
data of large sets of independent trajectories should be averaged),
which deteriorates the calculated correlation function
regardless of the force field employed. In our analyses of
limited numbers of trajectories, the ensemble averaging
was approximated by calculating averages of correlation
functions of different regions of the trajectories. The simulations
should be also sufficiently long because the correlation
function reflects only the effects of motions on the
timescale covered by the simulation. The obtained predictions
must be interpreted carefully to not confuse the artifacts
of the force fields with the effects of the sampling
scheme used.
The dRNAP molecule is particularly well suited to test
the ability of force fields to predict relaxation rates because
its domains greatly differ in their stochastic motions. Dynamics
of the well-ordered N-terminal domain is dominated
by overall tumbling and probes the ability of the force fields
to reproduce hydrodynamic properties. In contrast, internal
motions contribute most significantly to the relaxation rates
-12
-10
-8
-6
-4
-2
0
2
4
a
D(NH
N
)/Hz
-3
-2
-1
0
1
2
3 b
ΔδC
α
/ppm
0
1
2
3
c
ΔδC
β
/ppm
-3
-2
-1
0
1 d
ΔδC(O)/ppm
-2
0
2
4
6
8
10
12 e
ΔδN/ppm
-4
-2
0
2
4
6 f
D(NH
N
)/Hz
-3
-2
-1
0
1
2
3 g
ΔδC
α
/ppm
0
1
2
3
h
ΔδC
β
/ppm
-3
-2
-1
0
1
2 i
ΔδC(O)/ppm
-2
0
2
4
6
8
10
12 j
ΔδN/ppm
-1
0
1
k
D(NH
N
)/Hz
-3
-2
-1
0
1
2
3 l
ΔδC
α
/ppm
0
1
2
3
m
ΔδC
β
/ppm
-3
-2
-1
0
1
2 n
ΔδC(O)/ppm
-2
0
2
4
6
8
10
12 o
ΔδN/ppm
100 120 140 160
Residue number
0 20 40 60
Residue number
160 180 200 220 240
Residue number
δRNAP:
551381
RD-hTH:
061561
:159-254MAP2c
452951
FIGURE 5 Values of RDC (a, f, and k) and SCS (b–e, g–j, and l–o) in C-terminal IDR (residues 85–173) of dRNAP (a–e), N-terminal IDR (residues 1–65)
of RD-hTH (f–j), and MAP2c159–254
(k–o) simulated with the TIP4P-D water model. Experimental data are shown in gold, and values calculated using A99,
C22*, and C36m are shown in green, red, and blue, respectively. The random-coil limits are shown in gray. The MD simulations started from structures with
a-helices modeled for residues 40–53 of RD-hTH and 200–216 of MAP2c159–254
. For the sake of clarity, the statistical errors are not displayed here but
separately in Figs. S7–S9. To see this figure in color, go online.
Zapletal et al.
1628 Biophysical Journal 118, 1621–1633, April 7, 2020
of residues in the disordered C-terminal domain. Analysis of
the simulated trajectories showed that correlation functions
calculated for 5-ns sliding time windows (solid lines in
Fig. 6, a–c) describe relaxation in the C-terminal IDR sufficiently
well but fail to match the experimental data in the
well-ordered region, where slower motions, most notably
the overall tumbling, dominate the dynamics. Smooth profiles
of the calculated NMR relaxation parameters, resembling
the experimental profiles in the IDR, document that
the use of a short sliding window allowed us to average a
sufficient number of correlation functions and to capture
most important modes of motion.
To obtain relaxation rates close to the experimental
values also in the well-ordered N-terminal region, the
time window was extended to 50 ns, and averages of 12
correlation functions were calculated (dashed lines in
Fig. 6, a–c). Prediction of the cross-correlated Gx rate,
which is most sensitive to slow motions, was most informative
(Gx was used to monitor the slow motions instead
of the more frequently measured R2 rates because Gx could
be obtained with a higher accuracy than R2 for dRNAP,
which has an extremely poor dispersion of chemical shifts
in its C-terminal IDR (21)). Predicted and experimental
values were comparable for tested force fields, albeit scattered
for individual residues because of the contribution of
slow motions that were not sufficiently averaged in the
small set of the 30 independent correlation functions.
The match of the experimental and (average) simulated
Gx-values is in the line with the fact that TIP4P-D reproduces
the water diffusion coefficient more reliably than
the TIP3P model (31).
The general agreement was also good in the disordered
region for all three force field, but significant differences
were observed when different regions of the dRNAP
sequence were compared. The trend of ssNOE values,
most sensitive to fast motions, was best reproduced by
A99 (green line in Fig. 6 a). Also, the predictions by
CHARMM force fields reflected their abilities to predict
long-range contacts. Somewhat higher ssNOE values (and
elevated Gx) in the vicinity of residues 100 and 125 indicated
that C36m slightly overestimated partial ordering in the
most rigid regions of the C-terminal domain, where longrange
contacts were predicted (see the section Impact of
the Force Fields on Long-Range Contacts). This is in agreement
with the difficulty of predicting contacts between residues
far in the sequence, as discussed above. C22*
overestimated ssNOE around residues 115 and 90, in agreement
with its tendency to prefer contacts of the lysine
stretch with residues closer in the sequence.
In conclusion, we calculated NMR relaxation rates from
MD trajectories using two time windows (50 and 5 ns) to
cover different timescales of motions in the ordered and
disordered regions, respectively. The results independently
confirmed the observed moderate differences in predicting
long-range contacts and showed that all force fields describe
well the hydrodynamic properties for the TIP4P-D water
model. For dRNAP with a well-ordered domain and highly
charged disordered domains not forming transient helical
-1
0
1 assNOE
0
5
10
15
20
b
Γx/s
-1
0
1
2
c
R1/s
-1
-1
0
1 d
ssNOE
0
5
10
15
20
e
R2/s-1
0
1
2
f
R1/s-1
-1
0
1 g
ssNOE
0
5
10
15
20
h
R2/s
-1
0
1
2
i
R1/s
-1
20 40 60 80 100 120 140 160
Residue number
0 20 40 60
Residue number
160 180 200 220 240
Residue number
FIGURE 6 ssNOE (a, d, and g) and relaxation rates Gx (b), R2 (e and h), and R1 (c, f, and i) of dRNAP (a–c), N-terminal IDR (residues 1–65) of RD-hTH
(d–f), and MAP2c159–254
(g–i). Experimental data are shown in gold, and values calculated using A99, C22*, and C36m are shown in green, red, and blue,
respectively. Solid and dashed lines indicate data obtained from correlation functions calculated using 5- and 50-ns windows, respectively. For the sake of
clarity, the statistical errors are not displayed here but separately in Figs. S13–S15. To see this figure in color, go online.
Force Field for Disordered Regions
Biophysical Journal 118, 1621–1633, April 7, 2020 1629
structures, A99 performed best. The C36m force field
described long-range electrostatic contacts slightly worse,
but its accuracy was acceptable. C22* somewhat overestimated
electrostatic contacts between residues close in
the sequence, which resulted in noticeable, but not dramatic,
deviations of simulated NMR parameters from the
experiment.
NMR relaxation data were also predicted for RD-hTH
and MAP2c159–254
. In the case of RD-hTH, the comparison
was possible only in the N-terminal IDR because
the protein dimerizes in real samples. As a consequence,
relaxation of the well-ordered portion of RD-hTH is
incomparable (faster) with that simulated for the monomer.
Moreover, broadening of the peaks of the well-ordered
region did not allow us to obtain sufficiently
sensitive NMR spectra under the conditions used. Comparison
of the simulated and experimental relaxation rates
in the N-terminal IDR of RD-hTH (Fig. 6, d–f) and
MAP2c159–254
(Fig. 6, g–i) led to the same general conclusions
as for dRNAP. However, the simulation of RD-hTH
allowed us to address a particular feature not manifested
by dRNAP, namely the effect of formation of the transient
a-helix on the calculated relaxation rates. In the experimental
data, the propensity to form an a-helix is reflected
by elevated R2 values. Comparison of the relaxation rates
calculated from the C36m trajectories (in which the helix
was present during most of the simulation time) with those
obtained from the A99 and C22* runs (in which the helix
quickly unfolded) revealed that the presence of the a-helix
in the simulated structures is needed to reproduce the
values of R2 (Fig. 6). The 5-ns sliding window was sufficient
to match the experimental R2 profile in the C36m
simulations, indicating that the dynamics of the IDR of
RD-hTH is dominated by relatively short correlation
times. Less frequent conformational changes occurring
in the a-helical region were not sampled sufficiently and
resulted in high standard deviations of R2. Outside of
the transient a-helix, the data obtained with the 50-ns window
matched the experimental profile well.
In the case of MAP2c159–254
, which has a much lower
a-helical propensity, the increase of R2 is hardly visible in
the experimental data. Similarly to RD-hTH, relaxation
data predicted using the 5-ns window matched the experimental
values reasonably well.
Quantitative comparison of the force-field
reliability
To express the discussed differences of the force field performance
quantitatively, we applied the metrics developed
by Robustelli et al. (45) and calculated normalized forcefield
scores based on the RMSDs of the experimentally obtained
parameters from the corresponding values predicted
from the simulations (Table 1). The calculated combined
force-field scores (scombined) show that none of the tested
force fields provided superior prediction of all parameters
for all proteins.
In the case of dRNAP with a highly charged IDR forming
no transient a-helices, relative accuracy of predicting experimental
data varied for different parameters. Predicted
chemical shifts, reflecting the local backbone conformation,
was similar for all force fields (best for C36m). A99 best reproduced
NMR data sensitive to long-range intramolecular
interactions (RDC, PRE, and relaxation data, described by
sNMR). The chemical shift RMSD values are within the reported
standard deviation of the SPARTAþ predictor
(2.45, 1.09, 0.94, and 1.14 ppm for backbone N, C(O), Ca
,
and Cb
nuclei (37)). The quality factor Q of RDC is comparable
with typical RMSDs of data predicted from x-ray
structures with 2-A˚ resolution (46). PRE and relaxation
data deviated from the experimental ranges by less than
10%, with the exception of Gx, which is particularly difficult
to predict, as discussed above. C22* provided the best prediction
of parameters describing the overall shape of
dRNAP (Rg and SAXS profiles), with the Q-factor of
SAXS data equal to 5%. If the individual scores are averaged
with the same weights, the overall scores sall are better
for C22* and A99 than for C36m.
In the case of RD-hTH, with the experimentally determined
80% propensity to form an a-helix in its IDR,
C36m predicted all experimental data much better than
A99 or C22*. The superior performance of C36m, reflected
by low sall, can be clearly attributed to its ability to maintain
the experimentally observed transient a-helix. The quantitative
parameters showed that C36m predicted the experimental
data well (compared with the reference values
discussed above), with the exception of RDC.
In the case of MAP2c159–254
, which lacks a well-ordered
domain and exhibits only $20% propensity to form an a-helix,
all force fields predicted the experimental data with
similar accuracy, reflected by small differences between
their scores. It documents that the ability of C36m to maintain
transient a-helices loses its significance if the populations
of the helical structures are low. The performance of
all force fields was good, based on comparison with the
reference values discussed in the details of the dRNAP
simulations.
In conclusion, the quantitative comparison confirmed the
superior performance of C36m in the case when a transient
a-helix was present. A99 predicted most accurately parameters
influenced by long-range electrostatic interactions
(sNMR ¼ 1 for dRNAP).
Prediction of suitable spin-label positions
Calculation of already measured experimental parameters is
important for benchmarking of the simulations as illustrated
above. However, prediction of so far unknown measurable
values is also very useful because it can facilitate experimental
design. Selection of residues for placement of
Zapletal et al.
1630 Biophysical Journal 118, 1621–1633, April 7, 2020
paramagnetic labels can serve as an example. Preparation of
paramagnetically labeled samples is a time-consuming procedure,
including site-directed mutagenesis, expression, purification,
and paramagnetic spin labeling of the protein
before the NMR PRE measurements. A choice of the label
position not providing information about long-range contacts
thus represents a considerable waste of time and sources.
Reliable MD simulations can be used for in silico
prediction how much structural information a label in a
certain position can provide and how much such labeling
perturbs the native structural ensemble. An example of
such prediction for all solvent accessible positions within
the RD-hTH is presented in Fig. 7.
Based on the simulations using C36m and TIP4P-D, we
calculated PRE profiles for all possible positions. To
localize solvent accessible positions reporting on longrange
contacts with the disordered region, we defined the
following score. Integrals of the areas between PRE
profiles and a threshold of 0.8 were summed in the disordered
region (residues 1–70), except for 510 residues in
the vicinity of the spin label. In addition, the score was
set to zero for residues with solvent accessibility lower
than 0.6 according to Fraczkiewicz and Braun (47) because
the label should be freely accessible on the surface of
protein and should not interfere with any protein conformation.
The analysis presented in Fig. 7 a indicates four to
five areas well suited for attachment of spin labels for
future PRE experiments. Predicted PRE profiles for labels
placed in the centers of the suggested areas are presented in
Fig. 7, b–e.
CONCLUSIONS
The goal of our study was to assess the applicability of
currently available MD approaches to hybrid proteins consisting
of ordered and disordered regions. We tested the performance
of the A99, C22*, and C36m force fields in
combination with the TIP3P and TIP4P-D water models.
The performance was examined for mostly disordered
MAP2c159–254
containing a low population of prestructured
a-helix, for dRNAP consisting of comparably large well-ordered
and disordered domains without transient a-helices,
and for RD-hTH consisting of ordered and disordered domains
with a highly populated a-helical prestructured motif.
Considering the functional importance of transient helical
elements, we paid particular attention to the ability to preserve
the a-helix during the simulation. The generated structural
ensembles were used for predicting a variety of NMR
and SAXS parameters and subsequently compared with the
experimental data.
The TIP4P-D water model performed substantially better
than TIP3P or TIPS3P. The differences between force fields
were less distinct. The optimal (most universal) combination
was CHARMM36m with the TIP4P-D water model,
which most efficiently prevented artificial collapse of disordered
regions and retained transient a-helical structure elements
within the disordered regions. The study also showed
that the performance of different force fields and models can
vary depending on the actual physical properties of investigated
IDRs.
Considering the fast pace of force-field development
(45,48), the major value of this study is not identification
of the best force field available at the time of testing but
presentation of a generally applicable benchmarking
approach, including NMR relaxation rates. An important
feature of the approach is the combination of checked parameters
that report on abilities of the force fields to reproduce
a broad range of physical properties of the studied
molecules.
SUPPORTING MATERIAL
Supporting Material can be found online at https://doi.org/10.1016/j.bpj.
2020.02.019.
AUTHOR CONTRIBUTIONS
J.H. and L.Z. designed the research. V.Z., A.M., P.L., A.L., E.N., and V.K.
carried out calculations, performed the experiment, and analyzed the data.
Z.J., M.M., and K.M. prepared the samples and performed the experiment.
V.Z., A.M., P.L., J.H., and L.Z. wrote the article, and all authors reviewed
the manuscript.
ACKNOWLEDGMENTS
This research was funded by the Ministry of Education, Youth, and
Sport of the Czech Republic (MEYS CR), grant numbers LTC17078
0.0
3.0
6.0
9.0
12.0
15.0
18.0
Score
0.0
0.5
1.0
0.0
0.5
1.0
PRE
0.0
0.5
1.0
0.0
0.5
1.0
a
b
c
d
e
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160
Residue number
FIGURE 7 Predicted preferential positions of spin labels in RD-hTH (a)
and simulated PRE profiles for four selected spin-label positions (b–e). To
see this figure in color, go online.
Force Field for Disordered Regions
Biophysical Journal 118, 1621–1633, April 7, 2020 1631
(Inter-Excellence Inter-Cost), LTAUSA18168 (Inter-Excellence Inter-Action),
and LQ1601 (National Sustainability Programme II Project CEITEC
2020). Computational resources were provided by CESNET (LM2015042)
and the CERIT Scientific Cloud (LM2015085) under the program ‘‘Projects
of Large Research, Development, and Innovations Infrastructures’’ funded
by MEYS CR and by IT4Innovations National Supercomputing Center
(LM2015070) under the program ‘‘Large Infrastructures for Research,
Experimental Development and Innovations’’ funded by MEYS CR. The
Czech Infrastructure for Integrative Structural Biology (CIISB) research
infrastructure project LM2018127 funded by MEYS CR is gratefully
acknowledged for the partial financial support of the measurements at the
Josef Dadok National NMR Centre and at X-Ray Diffraction and BioSAXS
Core Facilities, CEITEC-Masaryk University.
REFERENCES
1. Uversky, V. N. 2002. Natively unfolded proteins: a point where biology
waits for physics. Protein Sci. 11:739–756.
2. Dunker, A. K., C. J. Brown, ., Z. Obradovic. 2002. Intrinsic disorder
and protein function. Biochemistry. 41:6573–6582.
3. Tompa, P. 2011. Unstructural biology coming of age. Curr. Opin.
Struct. Biol. 21:419–425.
4. Chi, S.-W., D.-H. Kim, ., K. H. Han. 2007. Pre-structured motifs in
the natively unstructured preS1 surface antigen of hepatitis B virus.
Protein Sci. 16:2108–2117.
5. Fuxreiter, M., I. Simon, ., P. Tompa. 2004. Preformed structural elements
feature in partner recognition by intrinsically unstructured proteins.
J. Mol. Biol. 338:1015–1026.
6. Vacic, V., C. J. Oldfield, ., A. K. Dunker. 2007. Characterization of
molecular recognition features, MoRFs, and their binding partners.
J. Proteome Res. 6:2351–2366.
7. Nova´cek, J., L. Zidek, and V. Sklena´r. 2014. Toward optimal-resolution
NMR of intrinsically disordered proteins. J. Magn. Reson. 241:41–52.
8. Nowakowski, M., S. Saxena, ., W. Kozminski. 2015. Applications of
high dimensionality experiments to biomolecular NMR. Prog. Nucl.
Magn. Reson. Spectrosc. 90–91:49–73.
9. Papoian, G. A. 2008. Proteins with weakly funneled energy landscapes
challenge the classical structure-function paradigm. Proc. Natl. Acad.
Sci. USA. 105:14237–14238.
10. Rabatinova´, A., H. Sanderova´, ., L. Kra´sny´. 2013. The d subunit of
RNA polymerase is required for rapid changes in gene expression
and competitive fitness of the cell. J. Bacteriol. 195:2603–2611.
11. Papouskova´, V., P. Kadera´vek, ., L. Zı´dek. 2013. Structural study of
the partially disordered full-length d subunit of RNA polymerase
from Bacillus subtilis. ChemBioChem. 14:1772–1779.
12. Nagatsu, T., M. Levitt, and S. Udenfriend. 1964. Tyrosine hydroxylase.
The initial step in norepinephrine biosynthesis. J. Biol. Chem.
239:2910–2917.
13. Molinoff, P. B., and J. Axelrod. 1971. Biochemistry of catecholamines.
Annu. Rev. Biochem. 40:465–500.
14. Lousa, P., H. Nedozra´lova´, ., J. Hritz. 2017. Phosphorylation of the
regulatory domain of human tyrosine hydroxylase 1 monitored using
non-uniformly sampled NMR. Biophys. Chem. 223:25–29.
15. Jansen, S., K. Melkova´, ., L. Zı´dek. 2017. Quantitative mapping of
microtubule-associated protein 2c (MAP2c) phosphorylation and regulatory
protein 14-3-3z-binding sites reveals key differences between
MAP2c and its homolog Tau. J. Biol. Chem. 292:6715–6727.
16. Melkova´, K., V. Zapletal, ., L. Zı´dek. 2018. Functionally specific
binding regions of microtubule-associated protein 2c exhibit distinct
conformations and dynamics. J. Biol. Chem. 293:13297–13309.
17. Melkova´, K., V. Zapletal, ., L. Zı´dek. 2019. Structure and functions of
microtubule associated proteins Tau and MAP2c: similarities and differences.
Biomolecules. 9:E105.
18. Mota´ckova´, V., J. Nova´cek, ., V. Sklena´r. 2010. Strategy for complete
NMR assignment of disordered proteins with highly repetitive
sequences based on resolution-enhanced 5D experiments. J. Biomol.
NMR. 48:169–177.
19. Nova´cek, J., L. Janda, ., V. Sklena´r. 2013. Efficient protocol for backbone
and side-chain assignments of large, intrinsically disordered proteins:
transient secondary structure analysis of 49.2 kDa microtubule
associated protein 2c. J. Biomol. NMR. 56:291–301.
20. Ottiger, M., F. Delaglio, and A. Bax. 1998. Measurement of J and
dipolar couplings from simplified two-dimensional NMR spectra.
J. Magn. Reson. 131:373–378.
21. Srb, P., J. Nova´cek, ., L. Zı´dek. 2017. Triple resonance 15
N NMR
relaxation experiments for studies of intrinsically disordered proteins.
J. Biomol. NMR. 69:133–146.
22. Korzhnev, D. M., M. Billeter, ., V. Y. Orekhov. 2001. NMR studies of
Brownian tumbling and internal motions in proteins. Prog. Nucl. Magn.
Reson. Spectrosc. 38:197–266.
23. Efron, B. 1979. Bootstrap methods: another look at the jackknife. Ann.
Stat. 7:1–26.
24. Petoukhov, M. V., D. Franke, ., D. I. Svergun. 2012. New developments
in the ATSAS program package for small-angle scattering data
analysis. J. Appl. Cryst. 45:342–350.
25. Riback, J. A., M. A. Bowman, ., T. R. Sosnick. 2017. Innovative scattering
analysis shows that hydrophobic disordered proteins are
expanded in water. Science. 358:238–241.
26. Lindorff-Larsen, K., S. Piana, ., D. E. Shaw. 2010. Improved sidechain
torsion potentials for the Amber ff99SB protein force field. Proteins.
78:1950–1958.
27. Piana, S., K. Lindorff-Larsen, and D. E. Shaw. 2011. How robust are
protein folding simulations with respect to force field parameterization?
Biophys. J. 100:L47–L49.
28. Huang, J., S. Rauscher, ., A. D. MacKerell, Jr. 2017. CHARMM36m:
an improved force field for folded and intrinsically disordered proteins.
Nat. Methods. 14:71–73.
29. Jorgensen, W. L. 1981. Quantum and statistical mechanical studies of
liquids. 10. Transferable intermolecular potential functions for water,
alcohols, and ethers. Application to liquid water. J. Am. Chem. Soc.
103:335–340.
30. MacKerell, A. D., D. Bashford, ., M. Karplus. 1998. All-atom empirical
potential for molecular modeling and dynamics studies of proteins.
J. Phys. Chem. B. 102:3586–3616.
31. Piana, S., A. G. Donchev, ., D. E. Shaw. 2015. Water dispersion interactions
strongly influence simulated structural properties of disordered
protein states. J. Phys. Chem. B. 119:5113–5123.
32. Hess, B., H. Bekker, ., J. G. E. M. Fraaije. 1997. LINCS: a linear
constraint solver for molecular simulations. J. Comput. Chem.
18:1463–1472.
33. Essmann, U., L. Perera, ., L. G. Pedersen. 1995. A smooth particle
mesh Ewald method. J. Chem. Phys. 103:8577–8593.
34. Berendsen, H. J. C., J. P. M. Postma, ., J. R. Haak. 1984. Molecular dynamics
with coupling to an external bath. J. Chem. Phys. 81:3684–3690.
35. Bussi, G., D. Donadio, and M. Parrinello. 2007. Canonical sampling
through velocity rescaling. J. Chem. Phys. 126:014101.
36. Parrinello, M., and A. Rahman. 1981. Polymorphic transitions in single
crystals: a new molecular dynamics method. J. Appl. Phys. 52:7182–7190.
37. Shen, Y., and A. Bax. 2010. SPARTAþ: a modest improvement in
empirical NMR chemical shift prediction by means of an artificial neural
network. J. Biomol. NMR. 48:13–22.
38. Nielsen, J. T., and F. A. A. Mulder. 2018. POTENCI: prediction of temperature,
neighbor and pH-corrected chemical shifts for intrinsically
disordered proteins. J. Biomol. NMR. 70:141–165.
39. Zweckstetter, M. 2008. NMR: prediction of molecular alignment from
structure using the PALES software. Nat. Protoc. 3:679–690.
40. Nodet, G., L. Salmon, ., M. Blackledge. 2009. Quantitative description
of backbone conformational sampling of unfolded proteins at
Zapletal et al.
1632 Biophysical Journal 118, 1621–1633, April 7, 2020
amino acid resolution from NMR residual dipolar couplings. J. Am.
Chem. Soc. 131:17908–17918.
41. Salmon, L., G. Nodet, ., M. Blackledge. 2010. NMR characterization
of long-range order in intrinsically disordered proteins. J. Am. Chem.
Soc. 132:8407–8418.
42. Sezer, D., J. H. Freed, and B. Roux. 2008. Simulating electron spin
resonance spectra of nitroxide spin labels from molecular dynamics
and stochastic trajectories. J. Chem. Phys. 128:165106.
43. Salvi, N., A. Abyzov, and M. Blackledge. 2016. Multi-timescale dynamics
in intrinsically disordered proteins from NMR relaxation and
molecular simulation. J. Phys. Chem. Lett. 7:2483–2489.
44. Urbanczyk, M., D. Bernin, ., K. Kazimierczuk. 2013. Iterative thresholding
algorithm for multiexponential decay applied to PGSE NMR
data. Anal. Chem. 85:1828–1833.
45. Robustelli, P., S. Piana, and D. E. Shaw. 2018. Developing a molecular
dynamics force field for both folded and disordered protein states.
Proc. Natl. Acad. Sci. USA. 115:E4758–E4766.
46. Bax, A. 2003. Weak alignment offers new NMR opportunities to study
protein structure and dynamics. Protein Sci. 12:1–16.
47. Fraczkiewicz, R., and W. Braun. 1998. Exact and efficient analytical
calculation of the accessible surface areas and their gradients for macromolecules.
J. Comput. Chem. 19:319–333.
48. Song, D., R. Luo, and H.-F. Chen. 2017. The IDP-specific force field
ff14IDPSFF improves the conformer sampling of intrinsically disordered
proteins. J. Chem. Inf. Model. 57:1166–1178.
Force Field for Disordered Regions
Biophysical Journal 118, 1621–1633, April 7, 2020 1633
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
63
Paper 14
Pavlíková Přecechtělová, J.; Mládek, A.; Zapletal, V.; Hritz, J. Quantum Chemical Calculations
of NMR Chemical Shifts in Phosphorylated Intrinsically Disordered Proteins,
JCTC 2019, 15, 5642-5658
Quantum Chemical Calculations of NMR Chemical Shifts in
Phosphorylated Intrinsically Disordered Proteins
Jana Pavlíková Přecechtělová,*,†,‡
Arnošt Mládek,†
Vojtěch Zapletal,†,§
and Jozef Hritz†
†
CEITEC and §
NCBR, Faculty of Science, Masaryk University, Kamenice 753/5, 625 00 Brno, Czech Republic
‡
Faculty of Pharmacy in Hradec Králové, Charles University, Akademika Heyrovského 1203, 500 05 Hradec Králové, Czech
Republic
*S Supporting Information
ABSTRACT: Quantum mechanics (QM) calculations are
applied to examine 1
H, 13
C, 15
N, and 31
P chemical shifts of
two phosphorylation sites in an intrinsically disordered
protein region. The QM calculations employ a combination
of (1) structural ensembles generated by molecular dynamics,
(2) a fragmentation technique based on the adjustable density
matrix assembler, and (3) density functional methods. The
combined computational approach is used to obtain chemical
shifts (i) in the S19 and S40 residues of the nonphosphorylated
and (ii) in the pS19 and pS40 residues of
the doubly phosphorylated human tyrosine hydroxylase 1 as the system of interest. We study the effects of conformational
averaging and explicit solvent sampling as well as the effects of phosphorylation on the computed chemical shifts. Good to great
quantitative agreement with the experiment is achieved for all nuclei, provided that the systematic error cancellation is
optimized by the choice of a suitable NMR standard. The effect of the standard reference on the computed 15
N and 31
P
chemical shifts is demonstrated by employing three different referencing methods. Error bars associated with the statistical
averaging of the computed 31
P chemical shifts are larger than the difference between the 31
P chemical shift of pS19 and pS40.
The sequence trend of 31
P shifts therefore could not be reliably reproduced. On the contrary, the calculations correctly predict
the change of the 13
C chemical shift for CB induced by the phosphorylation of the serine residues. The present work
demonstrates that QM calculations coupled with molecular dynamics simulations and fragmentation techniques can be used as
an alternative to empirical prediction tools in the structure characterization of intrinsically disordered proteins.
■ INTRODUCTION
Motivation. Intrinsically disordered proteins (IDPs) or
regions1
within the so-called hybrid proteins2
regulate a vast of
molecular processes in neurobiochemistry and their misfunction
leads to neurodenerative diseases, for example,
Parkinson’s and Alzheimer disease.3
IDPs are characterized
by polypeptide chains that do not fold into a stable and welldefined
tertiary structure. More than 30% of the proteins in the
eukaryotic genomes contain long (>30 residue) intrinsically
disordered regions4
and they are predicted to be common in
functional proteins.5,6
A typical example of a hybrid protein comprising the
structured as well as disordered regions is the human tyrosine
hydroxylase 1 (hTH1).7
It controls the rate-limiting step in the
synthesis of dopamine and adrenaline. The hTH1 contains the
regulatory domain (RD) composed from ∼70 amino acid IDP
region and the structured part (∼110 amino acids). Its
function is regulated by the phosphorylation of RD-hTH1
within its IDP region and the subsequent binding to the 14-3-3
proteins.8−10
The structural characterization of IDPs is extremely
challenging due to their flexibility that renders the applicability
of standard techniques such as X-ray crystallography and cryoelectron
microscopy limited.11
An efficient technique suitable
for the structural characterization of this challenging class of
proteins is nuclear magnetic resonance (NMR) spectroscopy.12
The structural interpretation of measured NMR data
combined with computational tools provides an ensemble of
diverse structures that characterize the flexible nature of
IDPs.13
The obtained structural ensembles are verified through
a comparison of experimental NMR properties including
chemical shifts (CSs) with software-aided predictions.14
Unfortunately, the currently available software tools,15−20
applicable to thousands of large biomolecular structures, are
not tailored to predict CSs for phosphorylated IDPs. The
absence of a suitable prediction tool is caused by the lack of
relevant experimental CS data that could be used as training
sets in the design of prediction algorithms. Instead, CSs
obtained from quantum mechanical (QM) calculations21−24
can potentially be used to aid the structural characterization of
phosphorylated IDPs.
In recent years, automatic fragmentation techniques have
emerged25−30
that allow for feasible QM calculations of NMR
Received: March 9, 2018
Published: September 5, 2019
Article
pubs.acs.org/JCTCCite This: J. Chem. Theory Comput. 2019, 15, 5642−5658
© 2019 American Chemical Society 5642 DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
DownloadedviaMASARYKUNIVonOctober13,2019at08:09:36(UTC).
Seehttps://pubs.acs.org/sharingguidelinesforoptionsonhowtolegitimatelysharepublishedarticles.
CSs31−36
in large biomolecular systems such as proteins,
protein−protein complexes, and nucleic acids. Fragment-based
approaches perform very well, yet their accuracy still does not
surpass that of empirical methods, especially for sensitive
nuclei such as HN
and 15
N.32
Fragmentation techniques thus
become a valuable prediction alternative in cases when
empirical parameterization is prevented by the lack of relevant
experimental NMR data. One such case is the prediction of 31
P
NMR CSs for phosphorylated proteins. Previous validation
studies25,31,32,34−36
that employed fragment-based methods, for
example, the adaptive density matrix assembler (ADMA),25
focused exclusively on the computation of conventional
protein nuclei, that is, 1
H, 13
C, 15
N in nonphosphorylated
well-structured proteins. It thus remains unknown how the
ADMA approach performs for CSs in proteins that are
intrinsically disordered andin additionmodified by a
phosphorylation. Of particular interest is what results can be
achieved if the ADMA methodology is combined with density
functional theory (DFT) and explicit solvent sampling by
molecular dynamics (MD). In this respect, three principal
questions arise: (1) do the MD/ADMA/DFT calculations of
1
H, 13
C, and 15
N CSs for the phosphorylated IDPs perform
equally well as in a previous study of N-methyl acetamide
(NMA) and an example of a structured protein?32
(2) What is
the accuracy of the MD/ADMA/DFT scheme for 31
P CSs? (3)
How does the phosphorylation affect the computed CSs of
other nuclei? The span of phosphorus CSs observed in
proteins is only a few ppm.37−39
This raises a great challenge
for QM calculations that have to be very accurate to be able to
reproduce CS variations due to structural changes. Previous
studies for proteins32,40−43
and nucleic acids44−47
indicate that
QM CS calculations have to be combined with MD
simulations to reflect the ensemble nature of the protein
conformation and solvent distribution. This becomes especially
crucial for flexible IDPs. Yet, a study combining MD and a
fragment-based approach has so far been performed only for
NMA, a textbook minimal-size model32,43
mimicking the
protein peptide bond, and a small well-structured protein.32
The example applications demonstrate that the accuracy of the
combined MD/QM approach is affected by the quality of MD
geometries. Because the simulation of IDPs requires the use of
specifically designed water models48
and/or even force fields
designed for that purpose,49
the MD-rooted errors in IDP
geometries can differ from those of well-structured proteins.
Computational methodologies are typically validated against
experimental data. The quantitative agreement is measured by
the root-mean-square-deviation (rmsd) while the qualitative
agreement is reflected by the correlation coefficient (R). A
previous study has shown that the dynamical averaging yields a
reasonable correlation of the computed and experimental 15
N
CSs while the rmsd’s remain high. It was proposed that the
problem could be alleviated by the use of a suitable secondary
standard.50−52
This referencing choice typically helps to
suppress systematic errors stemming from the computational
approach as well as from the fact that the computed and
experimental chemical shieldings are often obtained in
different phases (gas phase vs solution).53
Despite this
experience, computational studies of 15
N CSs in proteins
have so far employed secondary standards rather rarely.51
31
P CS calculations also suffer from referencing issues. The
quest for a suitable 31
P NMR reference compound is a longstanding
problem.53
The experimental 31
P NMR spectroscopy
experiments typically employs the 85% water solution of
H3PO4 as the standard reference. In calculations, however, the
referencing to the gas-phase computed value of the 31
P
chemical shielding in H3PO4 has been long perceived as
problematic for two reasons.54
First, the proper species in the
concentrated phosphoric acid solution are not known
experimentally. Second, water as a polar solvent can be
involved in associations with the H3PO4 molecules. Therefore,
31
P CS studies typically avoid the use of the calculated H3PO4
as the NMR standard. Instead, a secondary referencing scheme
employing PH3 has been proposed52
and extensively applied to
reference the computed 31
P CSs in inorganic55,56
and
organophosphorus53,57
compounds as well as in biomole-
cules.44,45,58,59
Nevertheless, a recent study carried out by
Fukal et al.46
shows that referencing to the computed H3PO4
in fact leads to a better agreement of the calculated 31
P CSs
with experiment. The study also proposes an alternative
relative referencing scheme, which obviates the use of an
external reference.
Objectives. In the present work, we extend on the previous
experience with QM calculations employing ADMA25,31,32
to
obtain NMR CSs in proteins. We aim to address three areas of
ADMA calculations that have not been studied in previous
works. First, the performance of the MD/ADMA/DFT
computational scheme is examined for 1
H, 13
C, and 15
N CSs
in an example IDP region. We analyze the performance in
terms of absolute agreement with experiment, CS variations
within the structural ensemble, as well as in terms of statistical
errors. Conclusions based on the results obtained here are
compared to those previously drawn for NMA and an example
of a structured protein.32
Second, we examine the effect of the
conformational averaging and solvent on protein 31
P CSs for
the first time. Finally, the effect of phosphorylation on the
computed 1
H, 13
C, and 15
N CSs is inspected. In order to
understand the quantitative differences between the computed
15
N and 31
P CSs and experiment, we additionally aim to
analyze three referencing schemes for each nuclei and to
identify the extent of systematic error cancellations affecting
the computed CSs.
In order to implement the defined objectives, we perform
MD simulations for the nonphosphorylated and doubly
phosphorylated hTH1 and apply ADMA to construct fragments
of two phosphorylation sites of hTH1, the serine
residues at positions 19 and 40. The two residues will be in the
following referred to as S19/S40 in the nonphosphorylated
hTH1 and pS19/pS40 in the doubly phosphorylated hTH1.
Density functional calculations are performed for the two
residues of interest.
■ METHODS
MD Simulations. The structure of the doubly phosphorylated
hTH1 used as a starting structure for the MD simulations
is provided in the Supporting Information. The starting
structure of the nonphosphorylated hTH1 was produced by
replacing (two) phosphate groups by hydrogens. MD
simulations of the nonphosphorylated and doubly phosphorylated
hTH1 were carried out using the protein Amber99SBILDN
force field,60,61
and parameters for the phosphorylated
serine residues (charge −2) were taken from the work of
Homeyer et al.62
The systems were solvated using a rhombic
dodecahedral box of TIP4P-D waters48
with a minimum
distance between the box walls and solute of 4 nm. The
TIP4P-D water model was designed to produce more reliable
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5643
conformational ensembles of IDPs.48
The charge of the system
was neutralized by adding Na+
and Cl−
ions; the concentration
of the salt was adjusted to the physiological concentration of
150 mM. All simulations were performed under periodic
boundary conditions. Prior to the MD runs, energy
minimizations with a steepest descent algorithm were
performed. The lengths of hydrogen atom covalent bonds
were constrained using the LINCS algorithm.63
An integration
time step of 2 fs was used. A cutoff of 1.0 nm was applied for
the Lennard-Jones interactions and short-range electrostatic
interactions. Long-range electrostatic interactions were calculated
by particle-mesh Ewald summation with a grid spacing of
0.12 nm and a fourth order interpolation.64
The 8 ns long
equilibration protocol consisted of 4 steps during which
backbone restraints were progressively released, starting at
1000 kJ mol−1
nm−2
. Two follow-up production NpT
simulations (300 K, 1 atm) were then carried out. The
resulting MD trajectories were 10 ns (simulation A) and 100
ns (simulation B) long. While both the temperature and
pressure were maintained using a Berendsen coupling
scheme,65
the production simulations were performed using
Nose−Hoover thermostat66
and Parrinello−Rahman barostat
algorithms.67
Atomic coordinates were recorded every 1 ps.
DFT Calculations of NMR CSs. MD snapshots have been
extracted in 100 ps steps from both MD trajectories. A total of
100 snapshots from the simulation A (A/100 ensemble) have
been obtained and further used. Calculations building on
simulation B employed the first 100 (B/100 ensemble) or 500
frames (B/500 ensemble). For each snapshot, the ADMA
fragmentation25
procedure has been performed using a Python
suite of codes. ADMA divides the protein into small fragments
in such a way that one fragment is generated for alanine,
glycine, and proline residues, while two fragments are
generated for the remaining amino acids (one for the backbone
part and one for the side chain part of the residue in question).
Fragments are then embedded into the surroundings of
neighboring macromolecule moieties, water molecules, and
ions to build the so-called parent molecules. The detailed
description of the procedure for the construction of parent
molecules can be found elsewhere.25,32
The surroundings
within the radius r from atoms of the original fragment is taken
into account, where we use r = 5 Å (for A/100 ensemble) and
r = 3.5 Å (for B/100 and B/500 ensembles). The size of the
parent molecules is 120−216 atoms for the former or 58−132
atoms for the latter, respectively. The geometries of parent
molecules constructed from the MD snapshots are directly
used as input coordinates for CS calculations, that is, no
geometry optimization is performed. The CS calculations
employed the Gaussian1668
implementation of the GIAO
formalism within the coupled perturbed DFT method.69−73
The bp8674,75
and b3lyp76−78
density functionals as
implemented in Gaussian have been used. The former has
previously provided favorable agreement between computed
and experimental 31
P CSs in nucleic acids44,45
while the latter
has been recommended by preceding validation studies of CSs
in structured proteins.31,32
For calculations with the A/100
ensemble, the DFT functionals were combined with the 6-
311G(d)79,80
and 6-31G(d)81,82
basis sets to keep the QM
methodology consistent with works31,32
demonstrating the
results of the MD/ADMA/DFT framework in ordered protein
structures. Using the same methodology facilitates a comparison
between the results achieved here and those obtained
previously.31,32
Additionally, b3lyp/6-311++G(d,p) and
b3lyp/pcs4/6-311++G(d,p) levels of theory were applied to
B/100 and B/500 ensembles to account for the effects of larger
basis sets and larger statistical data sets. The pcs4/6-311+
+G(d,p) notation refers to the pcs4 basis set placed only on
atoms of the original fragment in question (N, HN
, CA, HA,
C′, O for the backbone fragment and CB, HB1, HB2, O, P,
O1P, O2P, O3P for the side chain fragment, see Figure 1) and
6-311++G(d,p) basis set placed on atoms that were included
in the parent molecule as surroundings. In order to account for
the solvent effects, two sets of calculations have been carried
out for A/100. In the first one, water molecules and ions were
removed from the parent molecules and replaced by an implicit
solvent. The conductor-like implicit solvent model developed
within the framework of the polarizable continuum model83,84
based on the self-consistent reaction field has been applied.
Results obtained with this setup are referred to as “implicit”
throughout the text. The second set of calculations has been
performed with both implicit and explicit solvent as well as
with ions and will be from now on referred to as “explicit”.
From the NMR calculation of each parent molecule, we only
extract the chemical shielding of atoms belonging to the
original fragment (i.e., atoms constituting the surroundings are
disregarded). The chemical shielding σ is converted to the CS
δ using various referencing schemes. 1
H and 13
C CSs are
referenced to tetramethylsilane (TMS) computed at the same
level of theory as the atom of interest (X)
δ σ σ= −X
calc
TMS
calc
X (1)
Unless otherwise stated, 15
N CS calculations employ a
secondary standard referencing scheme based on methyl-
amine51
δ σ σ σ σ= − + −( )X
calc
CH NH (gas)
calc
X
calc
NH (liq)
exp
CH NH (gas)
exp
3 2 3 3 2
(2)
where σCH NH (gas)
calc
3 2
is the chemical shielding of methylamine
computed in the gas phase, σ = 244.6NH (liq)
exp
3
ppm is the
absolute 15
N chemical shielding of liquid ammonia at 25 °C,85
and σ = 249.5CH NH (gas)
exp
3 2
ppm is the experimental 15
N chemical
shielding of methylamine.86
The referencing for 31
P employs
the absolute experimental chemical shielding of the 85%
H3PO4 (328.4 ppm)87
as the standard reference
δ σ σ= −X
calc
85% H PO (liq)
exp
X
calc
3 4 (3)
Alternative referencing schemes have also been used (see
Results and Discussion) for 15
N as well as 31
P CSs to inspect
how the choice of the NMR standard influences the agreement
of the computed and experimental data.
Because we here focus on the two phosphorylation sites of
hTH1, the CS calculations are performed for four fragments of
Figure 1. Labeling of atoms in the phosphorylated serine.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5644
the nonphosphorylated and doubly phosphorylated hTH1.
The four fragments correspond to the backbone and side-chain
part of S19 and S40 or pS19 and pS40, respectively. The total
number of geometries subject to CS calculations is (100
snapshots) × (4 fragments) or (500 snapshots) × (4
fragments) per each of the two systems, respectively.
For every atom of a given fragment, the CS is calculated as
the statistical average, x̅ = ∑i=1
N
xi/N, where xi is the value of the
CS for the atom in question in the i-th snapshot and N = 100 is
the number of CS values that correspond to the individual
snapshots of the MD trajectory. The statistical distribution of
the calculated CSs within the chosen set of MD snapshots is
evaluated by the standard deviation of the sample mean sx̅
defined as
∑=
−
− ̅̅
=
s
N N
x x
1
( 1)
( )x
i
N
i
1
2
(4)
In order to estimate the error caused by using a sample,
which is much smaller than the true population of the IDP in
question, we report the 95% confidence interval (CI)
expressed as
= ̅ ± ̅
x zsCI x (5)
where z = 1.96 is the z-value from the standard normal z-table.
The term zsx̅
is called the maximum error of estimate (MEE).
The averaged CSs are compared with the experimental data
taken from the Supporting Information of Louša et al.9
(1
H,
13
C, and 15
N CSs) and from Hritz et al.8
(31
P CSs). CSs are
obtained for HA, HB1, HB2, HN
, CA, CB, C′, N, and P atoms
(Figure 1).
■ RESULTS AND DISCUSSION
Effects of Conformational Changes and Solvent on
the CSs. In order to get an idea how changes in the protein
conformation and the constant rearrangement of solvent
molecules during the simulation time affect the computed
CSs, we will first inspect the ranges of CSs obtained for the
MD ensembles. The conformational changes are easy to track
through the time dependence of the backbone torsion angles.
Figure 2a shows the time evolution of the torsion angles ϕ and
ψ for the residues pS19 and pS40 in the doubly
phosphorylated hTH1. It reveals instant, fast fluctuations of
the torsion angles in the equilibrated state while no major
conformational switches are observed. The CSs of atoms in the
backbone fluctuate similarly (for an example of 15
N CSs, see
Figure 2b), though not necessarily in direct correlation with
the torsion angle changes. This is obvious as many other effects
(besides the backbone conformation) influence the CSs, for
example, side-chain conformation, interactions of the protein
with the explicit solvent, interactions with other protein
moieties, and so forth. Particularly sensitive to the influence of
the explicit solvation are the polar protons, HN
, as they are
directly involved in the hydrogen bonding with the solvent
molecules. Figure 2c shows the time-dependence of the HN
CS
and does not reveal any patterns that would indicate unusual
solvent arrangements. The observed variations in CSs reflect
most importantly the combined effect of conformational
changes and changes in the positions and orientation of
water molecules. The impact of this combined effect is large,
not only for the polar protons but for other atoms as well (see
the discussion below) and leads to CSs spanning several to
tens of ppm depending on the type of the nucleus in question.
In general, the CSs of the explicitly solvated pS19 and pS40
fragments as extracted from the MD-generated ensemble of
hTH1 structures are in line with previous findings obtained for
the NMA,32
a minimal-size model of the protein peptide bond.
The nonpolar protons of pS19 and pS40 (i.e., HA, HB1, and
HB2) span a range of ∼3−4 ppm (Figure 3) when calculated
with bp86/6-311G(d) in the explicit solvent and >6 ppm when
Figure 2. Time evolution of (a) the backbone torsion angles ϕ(C′−
N−CA−C′), ψ(N−CA−C′−N), (b) HN
CSs, and (c) 15
N CSs in the
(1) pS19 and (2) pS40 residues of the doubly-phosphorylated hTH1
(trajectory A).
Figure 3. 1
H CSs of (a) HA and (b) HB1, HB2 atoms as obtained
from calculations in the implicit and explicit solvent. The histograms
include CSs of both pS19 and pS40. The dashed and solid vertical
lines mark the experimental values of the 1
H CSs in pS19 and pS40,
respectively. The experimental HB1 and HB2 CSs are identical and
differ negligibly between pS19 and pS40 (see Table S1 of the
Supporting Information).
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5645
calculated with b3lyp/pcs4/6-311++G(d,p). We recall that the
two sets of calculations build on different ensembles of parent
molecules constructed with r = 5 Å and r = 3.5 Å cutoff for the
surroundings, respectively. It is thus both, the basis set size and
the solvent shell size, that are responsible for the distinctions in
the variations of proton CSs. HA CSs are larger by about ∼0.5
ppm than the CSs of HB1, HB2. The distributions of nonpolar
proton CSs obtained with b3lyp/6-311G(d) in implicit and
explicit solvents overlap to a large extent. Yet, the addition of
explicit water molecules to the implicit solvent model causes a
notable shift of the CS histogram by up to 0.5 ppm to larger
CSs and thus also to an improved agreement with experimental
values (see Figure 3 and Table S1).
For the polar protons, HN
, the histograms of computed CSs
are shown in Figure 4. It was observed already previously32
that calculations of HN
CSs in the implicit solvent provide
rather narrow distributions while the combination of the
implicit and explicit solvents leads to the distribution
broadening. We can see this behavior for IDPs here as well
(Figure 4). The HN
CSs computed with bp86/6-311G(d) span
a range of up to 3 ppm in the implicit solvent with the average
value being around 5 ppm (see Table 1). Upon adding the
explicit solvent on top of the implicit solvent model, the span
of HN
CSs increases to ∼4.5 ppm. Consequently, the statistical
errors increase as well. Note that MEEs are larger for the polar
protons (up to 0.19 ppm, Table 1) than for the nonpolar ones
(up to 0.11 ppm; Table S1). The inclusion of the explicit water
molecules shifts the CS histogram toward a better agreement
with the experiment. Further improvement is achieved by
extending the basis set. At the b3lyp/pcs4/6-311++G(d,p)
level, the CS histogram spans more than 6 ppm and leads to
overall best statistical averages in terms of agreement with
experiment. The computed average CSs amount to 7.90
(pS19) and 7.33 ppm (pS40). This is underestimated by 0.7
ppm and ∼1.5 ppm compared to the experimental values of 8.6
ppm (pS19) and 8.8 ppm (pS40), respectively (Table 1). It
was proposed before32
that the remaining error in the
computed HN
CSs results from too long hydrogen bonds
caused by the AMBER force field parameterization, especially
by the approximations used in the parameterization of the
TIP3P water model. This hypothesis was confirmed by
combined ab initio MD/DFT calculations of NMR CSs in
NMA.43
Here, we employ the TIP4P-D model for the solvent,
which is more suitable for IDPs than TIP3P; however, it suffers
from similar problems with respect to NMR CS calculations
using MD geometries. CS results obtained with B/100 and B/
500 ensembles demonstrate that the augmentation of the
statistical data set significantly reduces the MEE. For HN
CSs,
the MEEs approximately halved after increasing the number of
MD snapshots from 100 to 500.
The 13
C CSs for both CA and CB obtained here with bp86/
6-311G(d) and the A/100 ensemble span a range of ∼15−20
ppm (Figure 5). The typical MEEs then amount to 0.6 ppm
(Table 2). On the contrary, the CSs computed using the B/
100 ensemble display larger variations of ∼25 ppm. The
increased fluctuations are likely caused by the fact that only the
first solvation shell was explicitly included in the QM
calculations. Because the two carbon atoms do not directly
interact with the solvent molecules, the addition of the explicit
solvent does not significantly change the shape and location of
the 13
C CS distributions on the CS scale. Yet, the histograms
obtained with explicit solvation seem to be closer to the
approximate Gaussian shape, and there is an obvious difference
in the percentage occurrences of the individual CS values when
comparing the two solvent model setups, respectively. The
average CS of CA is little affected by the explicit solvation
(Table 2(a)). For CB (Table 2(b)), explicit solvent decreases
the CS by 2.0 (pS19) and 1.2 ppm (pS40) compared to
calculations in the implicit solvent. Consequently, the
difference between the calculated and experimental CSs
decreases from 3.9 ppm (pS19) and 2.7 ppm (pS40) to 1.9
and 1.5 ppm, respectively, upon explicit solvation. These
findings are in line with the downfield shift of methyl carbons
observed in NMR calculations of N-methyl-acetamide.32
Interestingly, calculations employing B/100 and B/500
ensembles provide much worse estimates of CA CSs than
the calculations building on molecular structures constructed
from A/100 snapshots. This is likely caused by the differences
in the sampling of torsion angles within the A and B trajectory,
respectively. The histogram of b3lyp/pcs4/6-311++G(d,p)Figure
4. 1
H CSs of HN
atoms as obtained from calculations in the
implicit and explicit solvent. The histograms include CSs of both pS19
and pS40, respectively. The dashed and solid vertical lines mark the
experimental values of the HN
CSs in pS19 and pS40, respectively.
Table 1. 1
H CSs (in ppm) of HN
Atoms Calculated by Different QM Setupsa
method ensemble solvent δ(pS19)/ppm δ(pS40)/ppm
b3lyp/pcs4/6-311++G(d,p) B/100 explicit 7.90 ± 0.25 7.33 ± 0.23
b3lyp/6-311++G(d,p) B/100 explicit 7.43 ± 0.28 7.12 ± 0.32
b3lyp/6-311++G(d,p) B/500 explicit 7.42 ± 0.13 7.23 ± 0.16
bp86/6-311G(d) A/100 implicit 5.06 ± 0.11 5.02 ± 0.12
bp86/6-311G(d) A/100 explicit 6.26 ± 0.16 6.28 ± 0.19
b3lyp/6-311G(d) A/100 explicit 6.14 ± 0.15 6.19 ± 0.18
b3lyp/6-31G(d) A/100 explicit 6.08 ± 0.14 6.13 ± 0.17
experimentb
aq solution 8.599 8.841
a
The CSs are reported as 95% CIs (eq 5). b
See the Supporting Information of ref 9.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5646
computed CSs is located significantly further away from the
experimental values. Therefore, the results obtained at the
lower level of theory (bp86/6-311G(d)) using the large (r = 5
Å) radius for the explicit water surroundings seem to be more
reliable. The effect of replacing the bp86 functional by b3lyp
on CA and CB CSs is about the same as the effect of replacing
the 6-311++G(d,p) basis set on fragment atoms by pcs4 (1−2
ppm). We will show below that the influence of the functional
and basis set on carbonyl carbons is significantly larger.
The span of carbonyl carbon CSs is comparable to that of
CA and CB and amounts to ∼30 ppm in both implicit and
explicit solvents (Figure 6) and the MEEs are up to 0.6 ppm
(Table 3). The distributions of CSs in the explicit solvent is
shifted downfield compared to the results in the implicit
solvent. Consequently, the average CS values increase by ∼1
and ∼2 ppm for pS19 and pS40 (Table 3), respectively. Frank
et al.31
recommended the b3lyp functional for the calculation
of carbonyl carbon CSs. We therefore also provide the results
obtained at the b3lyp/6-311G(d) level of theory for
comparison. Note that the change of the functional from
bp86 to b3lyp causes a larger CS change than the addition of
the explicit solvent (6.6 ppm vs 0.9 ppm for pS19). The results
in Table 3 also confirm previous suggestions32
that a basis set
larger than 6-31G(d) is needed to obtain an accurate
prediction for carbonyl carbon CSs. For an example of the
pS19 residue, the b3lyp/6-31G(d) level of theory underestimates
the carbonyl carbon CSs by 13 ppm. On the
contrary, the b3lyp/6-311++G(d,p) level of theory overestimates
the computed CSs by ∼3 ppm only and gives the
Figure 5. 13
C CSs of (a) CA atoms and (b) CB atoms as obtained
from calculations in the implicit and explicit solvent. The histograms
include CSs of both pS19 and pS40, respectively. The dashed and
solid vertical lines mark the experimental values of the CA/CB CSs in
pS19 and pS40, respectively.
Table 2. 13
C CSs (in ppm) of CA and CB Atoms Calculated by Different QM Setupsa
method ensemble solvent δ(pS19)/ppm δ(pS40)/ppm
(a) CA CSs
b3lyp/pcs4/6-311++G(d,p) B/100 explicit 67.3 ± 1.0 64.5 ± 1.1
b3lyp/6-311++G(d,p) B/100 explicit 64.8 ± 1.6 62.4 ± 1.6
b3lyp/6-311++G(d,p) B/500 explicit 64.7 ± 0.6 64.3 ± 0.7
bp86/6-311G(d) A/100 implicit 59.1 ± 0.6 57.4 ± 0.6
bp86/6-311G(d) A/100 explicit 58.7 ± 0.6 57.2 ± 0.6
b3lyp/6-311G(d) A/100 explicit 57.6 ± 0.5 56.1 ± 0.5
b3lyp/6-31G(d) A/100 explicit 54.3 ± 0.5 52.8 ± 0.5
experimentb
aq solution 59.1 59.8
(b) CB CSs
b3lyp/pcs4/6-311++G(d,p) B/100 explicit 67.9 ± 0.9 66.7 ± 1.5
b3lyp/6-311++G(d,p) B/100 explicit 65.1 ± 1.2 63.5 ± 2.3
b3lyp/6-311++G(d,p) B/500 explicit 66.2 ± 0.6 65.7 ± 0.7
bp86/6-311G(d) A/100 implicit 70.7 ± 0.6 69.0 ± 0.6
bp86/6-311G(d) A/100 explicit 68.7 ± 0.6 67.8 ± 0.6
b3lyp/6-311G(d) A/100 explicit 66.6 ± 0.6 65.8 ± 0.6
b3lyp/6-31G(d) A/100 explicit 62.4 ± 0.6 61.5 ± 0.5
experimentb
aq solution 66.8 66.3
a
The CSs are reported as 95% CIs (eq 5). b
See the Supporting Information of ref 9.
Figure 6. 13
C CSs of carbonyl carbons as obtained from calculations
in the implicit and explicit solvent. The histograms include CSs of
both pS19 and pS40, respectively. The dashed and solid vertical lines
mark the experimental values of the carbonyl 13
C CSs in pS19 and
pS40, respectively.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5647
overall best agreement with the experiment. Introducing pcs4
on backbone or side chain atoms of the phosphorylated serine
increases the computed CS by as much as 7−8 ppm. Carbonyl
carbons are thus sensitive to the choice of the DFT functional
(hybrid vs non-hybrid) as well as the basis set.
The variations of the CSs within the MD ensemble of pS19
and pS40 fragments are particularly large for the nitrogen CSs.
The range of 15
N CSs spans over 40−50 ppm (Figure 7),
regardless of the level of theory and solvent model used.
Consequently, the average 15
N CSs have the largest MEEs out
of all nuclei (up to 2.4 ppm), see Table 4. The inclusion of
explicit water molecules causes a notable shift of the 15
N CS
distributions toward larger CS values. Consequently, the
average 15
N CSs of 111.8 ppm (pS19) and 112.0 ppm
(pS40) obtained with bp86/6-311G(d) in the implicit solvent
increase to 118.4 and 118.2 ppm, respectively, upon explicit
solvation. The agreement with the experimental values 119.7
(pS19) and 117.2 ppm (pS40) thus significantly improves.
Note that results obtained with the b3lyp/6-311G(d) level of
theory provides 15
N CSs smaller by ∼3 ppm than the bp86computed
counterparts, and the replacement of bp86 by b3lyp
increases the difference between the computed and experimental
data. The augmentation of the basis set for fragment
atoms from 6-311++G(d,p) to pcs4 leads to an overestimation
of the computed CSs by up to 5 ppm, which again worsens the
agreement with experiment. Comparison of the basis set,
solvent, and DFT functional effect reveals that the effects are of
about the same order, which is in line with previous findings.51
Cai et al.51
pointed out that the difficulty of the theoretical 15
N
CS prediction lies in its complexity. The 15
N CS is affected by
a multitude of effects (conformation, H-bonding, neighboring
residue types, solvent, electrostatic interactions), none of
which appears dominant.
It is striking that the b3lyp/6-31G(d) (Table 4) functional/
basis set combination is the most unfavorable model chemistry
tested here, judged by the agreement with the experiment.
Previous study31
of CSs in proteins recommended b3lyp/6-
31G(d) as a method of choice that benefits from error
cancellations and gives rather an accurate prediction of the 15
N
CSs in NMA and leads to a reasonable correlation of the
computed and experimental 15
N CSs in the HA2 domain. A
closer analysis reveals that the seeming discrepancy between
the conclusions made here and the conclusions of Exner et
al.32
can be explained by the different choice of the referencing
method as will be discussed in further details in a separate
section below.
The computed 31
P CSs in pS19 and pS40 change within
20−25 ppm range in the explicit solvent. The extent of the
variations is sizable considering the fact that the span of
experimental 31
P CSs in O-phosphorylated amino acids,37
peptides,38
and proteins39
is typically very small (a few ppm).
A large part of the CS changes is therefore likely to stem from
fluctuations in the H-bonding network rather than from the
Table 3. 13
C CSs (in ppm) of C′ Atoms Calculated by Different QM Setupsa
method ensemble solvent δ(pS19)/ppm δ(pS40)/ppm
b3lyp/pcs4/6-311++G(d,p) B/100 explicit 182.9 ± 1.3 183.8 ± 1.3
b3lyp/6-311++G(d,p) B/100 explicit 176.0 ± 1.9 177.6 ± 1.8
b3lyp/6-311++G(d,p) B/500 explicit 177.6 ± 0.9 177.5 ± 0.8
bp86/6-311G(d) A/100 implicit 169.6 ± 0.5 168.9 ± 0.4
bp86/6-311G(d) A/100 explicit 170.5 ± 0.6 170.9 ± 0.5
b3lyp/6-311G(d) A/100 explicit 177.1 ± 0.6 177.2 ± 0.5
b3lyp/6-31G(d) A/100 explicit 161.3 ± 0.5 160.9 ± 0.5
experimentb
aq solution 174.3 174.4
a
The CSs are reported as 95% CIs (eq 5). b
See the Supporting Information of ref 9.
Figure 7. 15
N CSs (in ppm) of backbone N atoms as obtained from
calculations in the implicit and explicit solvent. The histograms
include CSs of both pS19 and pS40, respectively. The dashed and
solid vertical lines mark the experimental values of the 15
N CSs in
pS19 and pS40, respectively.
Table 4. 15
N CSs (in ppm) of Backbone N Atoms Calculated by Different QM Setupsa
method ensemble solvent δ(pS19)/ppm δ(pS40)/ppm
b3lyp/pcs4/6-311++G(d,p) B/100 explicit 123.9 ± 2.1 124.7 ± 1.9
b3lyp/6-311++G(d,p) B/100 explicit 119.0 ± 2.1 120.2 ± 2.4
b3lyp/6-311++G(d,p) B/500 explicit 118.5 ± 1.2 120.0 ± 1.0
bp86/6-311G(d) A/100 implicit 111.8 ± 1.5 112.0 ± 1.6
bp86/6-311G(d) A/100 explicit 118.4 ± 1.8 118.2 ± 1.7
b3lyp/6-311G(d) A/100 explicit 115.1 ± 1.7 115.2 ± 1.6
b3lyp/6-31G(d) A/100 explicit 95.4 ± 1.3 95.6 ± 1.2
experimentb
aq solution 119.7 117.2
a
The CSs are reported as 95% CIs (eq 5) and referenced to NH3/CH3NH2 (eq 2). b
See the Supporting Information of ref 9.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5648
changes in the conformation of the phosphoserine residues.
The plot of the 31
P CS histograms in the implicit and explicit
solvents (Figure 8) corroborates the hypothesis as the 20 ppm
CS span in the explicit solvent reduces to 10 ppm after
removing the explicit solvent molecules from the calculations.
A similar CS behavior has already been observed for the polar
protons (see above). As a result of forming hydrogen bonds
between phosphate oxygens and water hydrogens, the sharp
and high histogram of CSs transforms into a broad one with
overall lower percentage occurrences. Also, note that the use of
the explicit solvent shifts the CS distributions more upfield and
as a result, the computed CS ensemble better fits the
experimental 31
P CSs of pS19 and pS40 (Figure 8). The
average CSs in the implicit solvent are 16.2 ± 0.4 ppm (pS19)
and 16.1 ± 0.4 ppm (pS40). H-bonding interactions between
the solute and explicit solvent decrease the average CSs by ∼10
ppm, which leads to δ(pS19) = 5.8 ± 0.7 ppm and δ(pS40) =
6.0 ± 0.8 ppm (Table 5), respectively. The quantitative
agreement with the experimental values of 3.76 ppm (pS19)
and 4.18 ppm (pS40), respectively, is therefore significantly
improved. The best results, obtained with b3lyp/6-311G(d),
differ from experiment by only 0.3 and 0.8 ppm, respectively.
However, note that explicit solvation increases the MEEs from
0.4 ppm in the implicit solvent to 0.8 ppm in the explicit
solvent. The difference between the experimental CSs of pS19
and pS40 amounts to 0.42 ppm and is thus about the same as
the error of the explicit solvent calculations. This makes
reproducing the sequence trends hardly possible, as will be
further demonstrated in a separate section below.
31
P CS calculations that employ pcs4 differ by as much as 18
ppm in the ensemble average from the calculations that apply
6-311++G(d,p) to all atoms. The basis set effect is thus almost
twice as large as the effect of explicit solvation. The agreement
with experimental data significantly deteriorates upon the
replacement of 6-311++G(d,p) on atoms of the original
fragments, which indicates an accumulation of systematic
errors. The compensation of systematic errors is analyzed in
more detail below.
Previous MD/DFT studies25,31,32,88
of CSs in proteins opted
to apply only double-zeta quality basis sets to make the
computations tractable even for an increasing number of MD
snapshots and their fragments. Because this option might be
unavoidable also in future calculations of 31
P CSs in proteins,
we test the performance of the 6-31G(d) basis set (along with
the b3lyp functional) here as well. For the 31
P CSs referenced
using 85% H3PO4, the results computed with b3lyp/6-31G(d)
differ from experiment by >70 ppm (Table 5) for both pS19
and pS40. However, we will demonstrate below that the
agreement with the experimental data can be dramatically
influenced by the choice of the reference compound and/or
the referencing approach. The section Choice of the
Referencing Method for 31
P CSs discusses the 31
P CS
referencing in detail and explains the reasons for the observed
deviations of the computed and experimental 31
P NMR CS
data.
Choice of the Referencing Method for 15
N CSs. For
15
N, three methods have become common in the literature to
reference the computed CSs: (method 1) referencing to the
absolute 15
N chemical shielding of liquid ammonia employing
the calculated 15
N chemical shielding of CH3NH2
51
(see eq 2)
used as a secondary standard (the multistandard approach),50
(method 2) referencing to the absolute 15
N chemical shielding
of liquid ammonia at 25 °C43,85,88
δ σ σ= −X
calc
NH (liq)
exp
X
calc
3 (6)
and (method 3) referencing to the 15
N chemical shielding of
NH3 calculated at the same level of theory as the molecule of
interest25,31,32,34
δ σ σ= −X
calc
NH (gas)
calc
X
calc
3 (7)
The three methods/references will be in the following
referred to as calc-CH3NH2, liq-NH3, and calc-NH3,
respectively. In order to inspect the effect of referencing, we
calculated the 15
N CSs using the three referencing methods
listed above and compare the outcomes for three combinations
of the DFT functional and basis set. Figure 9 demonstrates that
CS distribution obtained with bp86/6-311G(d)/calc-CH3NH2
Figure 8. 31
P CSs as obtained from calculations in the implicit and
explicit solvents. The CSs were referenced to 85% H3PO4 (eq 3). The
histograms include CSs of both pS19 and pS40, respectively. The
dashed and solid vertical lines mark the experimental values of the 31
P
CSs in pS19 and pS40, respectively.
Table 5. 31
P CSs (in ppm) of P Atoms in Phosphate Groupsa
method ensemble solvent δ(pS19)/ppm δ(pS40)/ppm
b3lyp/pcs4/6-311++G(d,p) B/100 explicit 14.5 ± 1.0 15.0 ± 1.1
b3lyp/6-311++G(d,p) B/100 explicit −1.9 ± 1.5 −2.9 ± 1.7
b3lyp/6-311++G(d,p) B/500 explicit −3.0 ± 0.9 −2.2 ± 0.7
bp86/6-311G(d) A/100 implicit 16.2 ± 0.4 16.1 ± 0.4
bp86/6-311G(d) A/100 explicit 5.8 ± 0.7 6.0 ± 0.8
b3lyp/6-311G(d) A/100 explicit 3.5 ± 0.6 3.4 ± 0.7
b3lyp/6-31G(d) A/100 explicit −67.2 ± 1.0 −68.7 ± 1.0
experimentb
aq solution 3.76 4.18
a
The CSs are reported as 95% CIs (eq 5) and referenced to 85% H3PO4 (eq 3). b
ref 8.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5649
covers about the same range as the distribution obtained with
b3lyp/6-31G(d)/calc-NH3. For the calculations carried out
with the larger basis set, the best agreement between the
statistically averaged CSs and the experimental values of 119.7
ppm (pS19) and 117.2 ppm (pS40) is achieved when bp86/6-
311G(d)/calc-CH3NH2 is used. This yields δiso(pS19) = 118.4
ppm and δiso(pS40) = 118.2 ppm (Table 6), respectively.
Similarly, a favorable agreement with the experiment is
achieved with b3lyp/6-311++G(d,p).
In the particular case of using methylamine as a secondary
standard in 15
N CS calculations, the referencing follows eq 2.
Assuming that the 15
N chemical shielding only depends on the
local electron density, deviations of the computed 15
N
chemical shielding should be the same for all nitrogens with
similar chemical surroundings.51
If this is indeed the case,
subtracting σCH NH (gas)
calc
3 2
and σcalc
is supposed to mitigate
systematic errors due to the finite basis set and approximations
made in the DFT exchange−correlation functional. The
subsequent addition of the experimental CS of methylamine
with respect to liquid ammonia, typically used89
in
experimental protein NMR to reference 15
N CSs, makes the
computed results well suited for a comparison with
experimental data. This holds provided that the solution
NMR conditions are modeled by the explicit solvent in
calculations of σcalc
. Note that although eq 2 is designed to
suppress systematic errors of the calculation, only a part of the
error is eliminated. The value of σCH NH (gas)
calc
3 2
is obtained here
from a simple static DFT calculation, whereas the calculation
of σX
calc
employs DFT as well as MD. As a result, only the
systematic errors associated with the DFT calculations can
cancel out. Errors introduced by the approximations inherent
to the MD cannot be mitigated unless the calculation of the
standard includes the MD averaging as well. Because this is not
the case here, the errors caused by the MD force field
deficiencies still affect the data computed for hTH1.
Referencing the computed 15
N CSs to the calculated value of
the 15
N chemical shielding in NH3 (eq 7) can be interpreted as
calculating the CS using
δ σ σ σ σ= − + −( )X
calc
NH (gas)
calc
X
calc
NH (liq)
exp
NH (gas)
exp
3 3 3 (8)
where the term in the round brackets is neglected (set to zero).
In fact, abandoning the gas-to-liquid shift of NH3 is a very
rough approximation because σNH3(liq)
exp
= 244.6 ppm85
while
σNH3(gas)
exp 86
and therefore σNH3(liq)
exp
− σNH3(gas)
exp
= −19.9 ppm.
Figure 9. 15
N CSs obtained for the MD ensemble of the doubly
phosphorylated hTH1. The CSs are referenced using three
referencing schemes: (a) referencing to liquid ammonia using the
secondary standard CH3NH2 computed at the same level of theory,
(b) referencing to liquid ammonia, and (c) referencing to NH3
computed at the same level of theory. The histograms include CSs of
both pS19 and pS40, respectively. The dashed and solid vertical lines
mark the experimental values of the 15
N CSs in pS19 and pS40,
respectively.
Table 6. 15
N CSs (in ppm) of Backbone N Atoms as Obtained Using Different References for the Conversion from the
Chemical Shielding to the CSa
method ensemble calc-CH3NH2
b
liq-NH3
c
calc-NH3
d
15
N CSs/ppm
pS19
b3lyp/pcs4/6-311++G(d,p) B/100 123.9 ± 2.1 139.8 ± 2.1 151.7 ± 2.1
b3lyp/6-311++G(d,p) B/100 119.0 ± 2.1 132.8 ± 2.1 146.3 ± 2.1
bp86/6-311G(d) A/100 118.4 ± 1.8 126.9 ± 1.8 151.1 ± 1.8
b3lyp/6-311G(d) A/100 115.1 ± 1.7 124.1 ± 1.7 148.0 ± 1.7
b3lyp/6-31G(d) A/100 95.4 ± 1.3 106.2 ± 1.3 116.6 ± 1.3
experimente
119.7
pS40
b3lyp/pcs4/6-311++G(d,p) B/100 124.7 ± 1.9 140.6 ± 1.9 152.5 ± 1.9
b3lyp/6-311++G(d,p) B/100 120.2 ± 2.4 134.0 ± 2.4 147.5 ± 2.4
bp86/6-311G(d) A/100 118.2 ± 1.7 126.7 ± 1.7 150.9 ± 1.7
b3lyp/6-311G(d) A/100 115.2 ± 1.6 124.1 ± 1.6 148.0 ± 1.6
b3lyp/6-31G(d) A/100 95.6 ± 1.2 106.4 ± 1.2 116.8 ± 1.2
experimente
117.2
a
The CSs are reported as 95% CIs (eq 5). b
CSs are referenced to liquid ammonia using the secondary standard CH3NH2 computed at the same
level of theory (see eq 2). c
Liquid ammonia is used as a reference (see eq 6). d
NH3 computed at the same level of theory is used as a reference (see
eq 7). e
See the Supporting Information of ref 9.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5650
Referencing according to eq 7 thus introduces a systematic
overestimation by ∼20 ppm. Overall, the agreement between
the calculated 15
N CSs referenced using eq 7 and experimental
data is affected in three ways. First, the computed and
experimental data are referenced to different standards (gasphase
vs liquid NH3). Second, the chemical environment of
15
N in NH3 resembles the chemical environment of 15
N in the
protein amide group less than, for example, CH3NH2 or other
potential secondary standards. As a result, the compensation of
systematic errors of the DFT calculation is likely less effective.
Third, as in the case of referencing according to eq 2, the
computed results suffer from the MD force field inaccuracies.
The approach of eq 7 therefore implies extra sources of errors
when compared to the approach of eq 2 and is thus expected
to provide inferior results in terms of agreement with
experimental data. While this is indeed the case for 15
N CSs
obtained here with the larger 6-311G(d) basis set, the b3lyp/6-
31G(d) calculations perform better with the referencing to the
computed gas-phase value of NH3 (Table 6). It can be
assumed that with this type of referencing, the errors
introduced by the use of the small 6-31G(d) basis set is
accidentally compensated by the errors stemming from
neglecting the gas-to-liquid shift of the 15
N chemical shielding
in NH3 and therefore from the use of a different standard in
calculations and experiment, respectively. An analysis of the
computed 15
N chemical shielding in NH3, CH3NH2, pS19, and
pS40 (Table S2) reveals that this is indeed the case.
Truncation of the basis set leads to an increase of the 15
N
chemical shielding in pS19 and pS40 by ∼18 ppm, which is
accompanied by the decrease of the 15
N chemical shielding by
∼2 ppm in CH3NH2 and by as much as ∼13.5 ppm in NH3,
respectively. Consequently, the computed 15
N CSs in pS19
and pS40 decrease by 20 ppm when referenced to calcCH3NH2
and by more than 30 ppm when referenced to calcNH3.
This leads to b3lyp/6-311G(d)/calc-CH3NH2 results
being too underestimated while b3lyp/6-31G(d)/calc-3NH3
results agree well with experiment (Table 6).
The systematic errors associated with the calc-NH3
referencing were already noticed by Frank et al.31
and Exner
et al.32
It was suggested that the problem could potentially be
solved in future studies by using a secondary standard. The
results presented here strongly support the proposed solution.
The accuracy of the computed 15
N CSs referenced to the
calculated chemical shielding of NH3 can be significantly
improved by adding the gas-to-liquid shift of NH3. Yet, the
compensation of errors originating from the QM methodology
is likely to work better between the protein NH groups and
CH3NH2, especially if large basis sets are employed. The
present work thus highly supports the previous suggestion32
that very accurate 15
N CSs can be achieved if the dynamical
averaging of conformational changes and solvent rearrangements
is applied along with large basis sets and standards
similar to the groups typical for proteins.
Choice of the Referencing Method for 31
P CSs. For the
sake of computing 31
P CSs, we test three referencing methods
here in analogy with the referencing methods discussed above
for 15
N CSs. (Method 1) references the 31
P CSs to 85% H3PO4
using the secondary standard PH3 as proposed by van
Wüllen52
δ σ σ σ σ= − + −( )X
calc
PH (gas)
calc
X
calc
H PO (85% solution)
exp
PH (gas)
exp
3 3 4 3
(9)
where σPH (gas)
calc
3
is the chemical shielding of 31
P in PH3
calculated at the same level of theory as the parent molecules
constructed from the protein structure, σH PO (85% solution)
exp
3 4
is the
absolute experimental chemical shielding of the 85% H3PO4
(328.4 ppm),87
and σPH (gas)
exp
3
is the absolute experimental
chemical shielding of PH3 (594.5 ppm).87
(Method 2)
references to the experimental absolute 31
P chemical shielding
of 85% H3PO4 (eq 3). To the best of our knowledge, the
approach of eq 3 has not been commonly applied. Yet, we
propose to employ it here as an equivalent of the widely
accepted use of liquid NH3 in 15
N CS referencing. Finally,
(method 3) references the 31
P CSs to the chemical shielding of
31
P in H3PO4 calculated at the same level of theory
δ σ σ= −X
calc
H PO (gas)
calc
X
calc
3 4 (10)
The three referencing methods will be in the following
referred to as calc-PH3, liq-H3PO4, and calc-PH3, respectively.
In Figure 10, we compare the distributions of the 31
P CSs
calculated with the three referencing schemes. By far, the best
agreement with the experiment is achieved using the liq-H3PO4
referencing and the b3lyp/6-311G(d) level of theory, for which
we obtain δiso(pS19) = 3.5 ± 0.6 ppm and δiso(pS40) = 3.4 ±
0.7 ppm (Table 7) while the experimental values are δiso(pS19)
= 3.76 ppm and δiso(pS40) = 4.18 ppm. Note that with the liqH3PO4
referencing, the cancellation of errors stemming from
the QM methodology cannot occur as eq 3 subtracts a
calculated chemical shielding from an experimental one. This
inconsistency obviously leads to an error cancellation as well,
which is reflected in the very favorable agreement with
Figure 10. 31
P CSs obtained using different references for the
conversion from the chemical shielding to the CS. CSs are given in
ppm. (a) CSs are referenced to 85% H3PO4 using the secondary
standard PH3 computed at the same level of theory, (b) liquid 85%
H3PO4 is used as a reference, and (c) H3PO4 computed at the same
level of theory is used as a reference. The histograms include CSs of
both pS19 and pS40, respectively. The dashed and solid vertical lines
mark the experimental values of the 31
P CSs in pS19 and pS40,
respectively.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5651
experimental data. Yet, the errors have different origins, one
being, for example, the neglect of the ∼37 ppm difference
between the calculated (557.2 ppm, see the b3lyp/6-311G(d)
result in Table S3) and experimental (594.5 ppm) chemical
shielding of PH3 (eq 9).
For the referencing with calc-PH3 (Figure 10a) and calcH3PO4
(Figure 10c), respectively, a very different dependence
of the calculated CSs on the QM methodology is observed.
This is a consequence of the fact that the chemical shielding of
PH3 and H3PO4 changes differently when one of the three
methods is replaced by the other (Table S2). The cancellation
of the QM-related errors does not work the same. For instance,
when going from bp86/6-311G(d) to b3lyp/6-311G(d), the
chemical shielding of PH3 decreases by 4.8 ppm while the
chemical shielding of H3PO4 increases by 2.4 ppm.
Simultaneously, the chemical shielding of pS19/pS40 increases
by 2.4/2.6 ppm. The chemical shieldings of PH3 and the
protein phosphates, respectively, thus change in opposite
directions. This leads to the 7 ppm difference between bp86/6-
311G(d) and b3lyp/6-311G(d) results already mentioned
above. In contrast, the chemical shieldings of H3PO4 and the
protein phosphates change in the same direction by an equal
amount. The results obtained with bp86/6-311G(d) and
b3lyp/6-311G(d) are equal as a result when the referencing to
H3PO4 is employed. The described changes of the chemical
shieldings and CSs demonstrate the effect of the DFT
approximation.
The influence of the basis set on the 31
P NMR parameters
can be traced by comparing the b3lyp results in the last two
columns of Table S2. The replacement of 6-311G(d) by 6-
31G(d) increases the chemical shielding of PH3 by 29.3 ppm
and the chemical shielding of H3PO4 by as much as 84.2 ppm.
This is to be compared with the increase of the pS19 and pS40
chemical shielding by 70.7 and 72.1 ppm, respectively. PH3
thus eliminates only less than a half of the error caused by the
truncation of the basis set, thereby leaving >40 ppm difference.
On the contrary, H3PO4 overcompensates the error merely by
10 ppm. The described effects overall lead to the trend
unveiled by Figure 10. Under the H3PO4 referencing, the 31
P
CSs obtained with the smaller 6-31G(d) basis set display a
superior agreement with the experimental data than the 31
P
CSs obtained with 6-311G(d). We have witnessed a similar
behavior for 15
N CSs referenced to NH3 computed at the same
level of theory as the atoms of interest (see the paragraph
Choice of the Referencing Method for 15
N CSs).
The 31
P CSs results just described can be easily understood.
The chemical environment of phosphorus in the phosphoserine
phosphate closely resembles that in H3PO4. Consequently,
the systematic error in the chemical shielding is also similar,
which facilitates the cancellation of systematic errors due to the
finite basis set and approximations to the exchange−
correlation energy of DFT. The compensation of errors
between PH3 and the protein phosphates is less effective. In
general, the results presented here further support the findings
of Fukal et al.,46
who performed a benchmarking study of the
31
P chemical shielding in PH3 and H3PO4. The dependence of
δiso(31
P) on the basis set was found not very systematic, which
led to notable variations of CSs in the phosphates of interest.
Effects of Systematic Errors on the Computed CSs.
The agreement between the CSs computed here and the
experimental data is affected by several systematic errors. The
main source of errors are (1) the basis set dependence of the
calculated chemical shielding, (2) the sample size, that is, the
number of MD frames used for the statistical averaging, and
(3) the quality of molecular geometries as provided by the MD
force field. In the case of calculations employing the B/100
ensemble of MD frames, the error associated with the
truncation of the explicit solvation to the first solvation shell
adds to the list of systematics errors.
The basis set related systematic errors can be compensated
by the choice of a suitable NMR reference, as we have
thoroughly demonstrated above. The error compensation
works only partially and especially in the case of 31
P its extent
is hard to estimate in advance. In agreement with our findings,
Fukal et al.46
previously pointed out that the chemical
shielding of 31
P in the reference compounds (PH3, H3PO4)
show a different dependence on the basis set than the molecule
of interest. This is the consequence of the differences in the
underlying electronic structures. As a result, 31
P CSs referenced
with the calculated chemical shielding of PH3 and H3PO4,
Table 7. 31
P CSs (in ppm) of P Atoms in Phosphate Groups as Obtained Using Different References for the Conversion from
the Chemical Shielding to the CSa
31
P CSs/ppm
method ensemble calc-PH3
b
liq-H3PO4
c
calc-H3PO4
d
pS19
b3lyp/pcs4/6-311++G(d,p) B/100 −25.2 ± 1.0 14.5 ± 1.0 −36.5 ± 1.0
b3lyp/6-311++G(d,p) B/100 −33.0 ± 1.5 −1.9 ± 1.5 −43.9 ± 1.5
bp86/6-311G(d) A/100 −26.7 ± 0.7 5.8 ± 0.7 −39.0 ± 0.7
b3lyp/6-311G(d) A/100 −33.8 ± 0.6 3.5 ± 0.6 −39.0 ± 0.6
b3lyp/6-31G(d) A/100 −75.2 ± 1.0 −67.2 ± 1.0 −25.5 ± 1.0
experimente
3.76
pS40
b3lyp/pcs4/6-311++G(d,p) B/100 −24.7 ± 1.1 15.0 ± 1.1 −36.0 ± 1.1
b3lyp/6-311++G(d,p) B/100 −34.0 ± 1.7 −2.9 ± 1.7 −44.9 ± 1.7
bp86/6-311G(d) A/100 −26.5 ± 0.8 6.0 ± 0.8 −38.8 ± 0.8
b3lyp/6-311G(d) A/100 −33.9 ± 0.7 3.4 ± 0.7 −39.1 ± 0.7
b3lyp/6-31G(d) A/100 −76.7 ± 1.0 −68.7 ± 1.0 −27.0 ± 1.0
experimente
4.18
a
The CSs are reported as 95% CIs (eq 5). b
CSs are referenced to 85% H3PO4 using the secondary standard PH3 computed at the same level of
theory (see eq 9). c
85% solution of H3PO4 is used as a reference (see eq 3). d
H3PO4 computed at the same level of theory is used as a reference
(see eq 10). e
ref 8.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5652
respectively, display numerical instabilities. That is why we
propose to use the liq-H3PO4 referencing as a more
transparent alternative to the established referencing schemes.
In order to inspect the influence of the statistical data set
size, the ensemble of MD frames subject to MD/ADMA/DFT
calculations was enlarged from 100 to 500 MD frames. Using
the B/500 ensemble, CSs were calculated with b3lyp/6-311+
+G(d,p). To make the calculations computationally tractable,
only water molecules of the first solvation shell were included
in the molecular clusters. This corresponds to applying the r =
3.5 Å cutoff in the construction of parent molecules. Although
samples >500 MD frames would lead to even beer converged
ensemble averages (Figure 11a,b), the MEEs are converged
sufficiently attfter 500 frames. The MEE for the 15
N and 31
P
CS does not exceed 1.2 and 0.8 ppm, respectively. The
additional 400 MD frames reduce the MEE approximately by a
factor of 2 (Figure 11c,d). The width of the CI for 15
N and 31
P
CSs reduces by ∼2−3 and ∼1−2 ppm, respectively.
Consequently, the effect of the sample size on 15
N CSs is
comparable to the effect of the basis set or the DFT functional
(Table 4). On the contrary, the influence of the sample size on
31
P CSs is much smaller than the basis set effect (Table 5).
Because molecular geometries are overtaken directly from
the MD simulation without any further geometry optimization,
they suffer from incorrect bond lengths and angles caused by
the approximations inherent to the force field of the classical
MD simulation. While we are fully aware of the problem, there
are several reasons for which we opt to dismiss the geometry
optimization step here. First, the computed CSs are to be
compared with data from NMR spectroscopy experiments. We
therefore strive to calculate the properties of a thermodynamic
ensemble of structures that is substantially different from an
ensemble of energy-minimized structures resulting from a
geometry optimization. Second, while a partial geometry
optimization could serve here as a compromise solution, its
successful execution is often not very straightforward. The
parent molecules built from protein fragments tend to be
sizable and have numerous degrees of freedom both in the
protein of interest and in the H-bonding network. In such a
case, the choice of geometrical parameters to be frozen/
optimized during the partial optimization is not obvious. The
goal of the partial optimization is to preserve the overall
geometry of the MD snapshot and improve the bond lengths
and valence angles at the same time. Several different
combinations of frozen and relaxed parameters are possible,
each introducing a substantial bias on the resulting geometries.
Consequently, the NMR parameters computed for the partially
optimized molecular geometries can significantly differ
depending on the choice of frozen parameters. In addition,
the number of fixed torsion angles often becomes large, which
causes the system to be overconstrained, and the optimization
fails to converge as a result. Further potential complications
include the disruption of a realistic orientation of water
molecules with respect to the solute. This could have severe
consequences, given the demonstrated importance of the
explicit solvent for the quantitative agreement with experiment.
Equally importantly, the geometry optimization raises the
computational time/memory/storage requirements of the
MD/DFT calculations. The computational costs thus quickly
become too high and make the MD/DFT calculations
intractable, especially if a large number of fragments need to
be calculated. For these reasons, the state-of-the-art MD/DFT
studies of NMR CSs in biomolecules opt to dismiss the
geometry optimization step, especially if realistic protein
structures rather than model systems are studied. In the recent
work by Fukal et al.,46
geometry optimization was carried out
for snapshots obtained from an MD simulation of
diethylphosphate. Depending on the referencing scheme used
for 31
P CSs, the statistically averaged CSs computed with
b3lyp/IGLO-III increased by 10−20 ppm. However, we have
to point out that the optimization employed only implicit
solvation.
While a geometry reoptimization is beyond the scope of the
present work, two alternative optimization approaches can
potentially be applied in future to avoid the difficulties
described above. First, the use of a normal mode geometry
Figure 11. Dependence of the average 15
N (a) and 31
P (b) CS and the associated MEE (c,d) on the number of MD frames used in the statistical
averaging. The dashed and solid lines in pictures (a,b) indicate the experimental values of the corresponding CSs in pS19 and pS40, respectively.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5653
optimization90
can be considered to preserve the overall
geometry of MD snapshots. Another option would be to apply
a penalty function to restrain the key internal coordinates.91
The two approaches can potentially be applied provided that
the number of MD frames stays within reasonable limits.
CS Differences between pS19 and pS40. It was
proposed46
that errors related to the 31
P CS referencing and
the quality of MD geometries can potentially be eliminated or
reduced by calculating the relative Δδ(31
P) CS, defined as
Δδ(31
P) = δ(31
P,phosphate 1) − δ(31
P,phosphate 2), which is
equivalent to the difference between the chemical shieldings
Δδ(31
P) ≡ δ(31
P,phosphate 1) − δ(31
P,phosphate 2). Because
the pS19 and pS40 satisfy the condition of chemical
equivalence, they are well suited for the relative Δδ calculation
that avoids the use of an external reference. Table 8 shows the
MD-averaged 31
P CSs of pS19 and pS40. The CS difference
δ(pS19) − δ(pS40) of −0.2 ppm does closely approach the
experimental CS difference of −0.42 ppm. However, MEEs
obtained for the 31
P CSs of both residues are as large as 0.8
ppm. The average CS of pS19 lies within the error bar of pS40
and vice versa, and the two 31
P CSs are therefore identical
numbers in practice. The size of the error is too large also for
the chemical shielding. The difference between the chemical
shielding of pS19 (322.6 ppm) and pS40 (322.4 ppm) is 0.2
ppm while the MEE amounts to 0.8 ppm. The average
chemical shielding for one residue thus falls within the error
bar of the other residue. Calculating the relative Δδ is therefore
not relevant here. To be able to reproduce the very small size
of the 31
P CS difference (Δδexp(pS19 − pS40) = −0.42 ppm),
the MEEs would have to reduce significantly. This could be
potentially achieved46
by the geometry optimization of MD
snapshots as discussed above.
Effect of Phosphorylation. It is known that the
phosphorylation of amino acids influences the conformation
of peptides.92−94
The recent NMR study of the RD of hTH1
demonstrated that phosphorylation slightly decreases the
alpha-helical propensity around the S19 phosphorylation site.
Figure 12 shows the backbone torsion angles as obtained from
the MD simulation of the nonphosphorylated and doubly
phosphorylated hTH1. The ϕ torsion angle in S19 spans a
broad interval of −20° to 180°, whereas the same angle in S40
covers a smaller interval of (50°, 180°). In contrast, the range
of ψ is narrower in pS19 (80−180°) than in pS40 (−10° to
180° degrees). Differences between the CSs of the nonphosphorylated
and doubly phosphorylated hTH1 thus reflect
not only the direct electronic effect of the phosphate group but
also the effect of the induced conformational change. In the
experiment, the most significant CS change (2.7 and 2.8 ppm
for pS19 and pS40, respectively) upon phosphorylation is
observed for the CB carbon (Table 9). The MD/DFT
calculations correctly predict the downfield shift induced by
phosphorylation as well and estimate it to 3.6 ppm (pS19) and
2.3 ppm (pS40), respectively. The CB CS can thus safely be
used as a probe of effects coupled with the phosphorylation.
The same observation was recently reported in Louša et al.9
based on the experimental NMR data. The CB CS change
should therefore be incorporated into the databases underlying
the CS prediction programs.
■ CONCLUSIONS
We have demonstrated that the accuracy of the MD/ADMA/
DFT approach for the calculation of CSs in phosphorylation
sites of IDPs is affected by many factors, including the basis set,
molecular geometries, statistical sample size, CS referencing,
Table 8. Comparison of the 31
P CSs and Chemical
Shieldings of pS19 and pS40, Obtained at the bp86/6-
311G(d) Level
δliq‑H3PO4
/ppma
σ/ppm δexp/ppmb
pS19 5.8 ± 0.8 322.6 ± 0.8 3.76
pS40 6.0 ± 0.8 322.4 ± 0.8 4.18
ΔpS19−pS40
c
−0.2 0.2 −0.42
a
CSs referenced to the 85% H3PO4 (eq 3). b
Experimental CSs taken
from ref 8. c
ΔpS19−pS40 is the difference between the 31
P CS or
chemical shielding of pS19 and pS40, respectively, that is, ΔpS19−pS40 =
δ(pS19) − δ(pS40) = σ(pS19) − σ(pS40).
Figure 12. Ramachandran plot of torsion angles in (a) residues S19
and S40 of the nonphosphorylated hTH1 and in (b) residues pS19
and pS40 of the doubly phosphorylated hTH1.
Table 9. CSs in the S19, S40, and pS19, pS40 Residues of
the Nonphosphorylated and Doubly Phosphorylated hTH1,
Respectivelya
non-phosphorylated phosphorylated
atom δcalc/ppmb
δexp/ppmc
δcalc/ppmb
δexp/ppmc
S19 pS19
P n.a. n.a. 5.8 ± 0.7 3.76
HN
6.77 ± 0.18 8.429 6.26 ± 0.16 8.599
N 120.5 ± 1.3 119.0 118.4 ± 1.8 119.7
C′ 168.3 ± 0.6 174.9 170.5 ± 0.6 174.3
Ca 58.5 ± 0.7 58.27 58.7 ± 0.6 59.13
Ha 4.63 ± 0.08 4.454 4.52 ± 0.10 4.316
Cb 65.1 ± 0.6 64.02 68.7 ± 0.6 66.75
Hb1 4.38 ± 0.07 3.848 3.81 ± 0.10 3.905
Hb2 4.05 ± 0.09 3.917 3.56 ± 0.09 3.905
S40 pS40
P n.a. n.a. 6.0 ± 0.8 4.18
HN
6.75 ± 0.17 8.392 6.28 ± 0.19 8.841
N 120.6 ± 1.6 117.1 118.2 ± 1.7 117.2
C′ 169.9 ± 0.5 174.7 170.9 ± 0.6 174.4
Ca 58.6 ± 0.7 58.88 57.2 ± 0.6 59.78
Ha 4.61 ± 0.08 4.394 4.5 ± 0.09 4.238
Cb 65.6 ± 0.6 63.63 67.9 ± 0.6 66.30
Hb1 4.31 ± 0.06 3.873 3.85 ± 0.10 3.904
Hb2 4.11 ± 0.08 3.873 3.85 ± 0.11 3.904
a
The CSs are reported as 95% CIs (eq 5). 15
N CSs were referenced to
NH3/CH3NH2 (eq 2) and 31
P CSs were referenced to 85% H3PO4
(eq 3). b
Computed CSs obtained with bp86/6-311G(d). c
Experimental
1
H, 13
C, and 15
N CSs are taken from the Supporting
Information of ref 9, 31
P CSs are taken from ref 8.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5654
and the size of explicitly treated surroundings. Calculations of
CSs for atoms affected by hydrogen-bonding with other
protein moieties and solvent are particularly difficult to
compute. This finding is in line with conclusions previously
made for the NMA model32,43
of the protein peptide bond.
The MD ensemble of pS19 and pS40 geometries within the
hTH1 peptide (region 1−50) augmented by the explicit
solvent and protein surroundings displays large CS variations.
The nonpolar (HA, HB1, HB2) and polar (HN
) proton CSs
span a range of ∼3−4 ppm and up to 6 ppm, respectively,
depending on the conformation of the fragment in question
and the arrangement of the surrounding explicit water
molecules. The 13
C CSs vary within up to 20 ppm for the
aliphatic (CA, CB) and up to 30 ppm for the carbonyl (C′)
carbons. Particularly large ensemble variations have been
observed for 15
N CSs that range within as much as 50 ppm.
Finally, the conformational and solvent shell changes lead to
31
P CSs fluctuating within 25 ppm. The range of 31
P CSs
obtained computationally thus substantially exceeds the
several-ppm-wide range found in the experiment.
The CS variations in the MD structural ensemble are
reflected by MEEs, which were systematically higher for
average CSs computed with the B/100 ensemble than for the
A/100 ensemble. Nevertheless, MEEs improved with averaging
over the extended B/500 set of MD snapshots. The resulting
MEE is about the same for the nonpolar and polar proton CSs
(up to 0.20 ppm). MEE for 13
C CSs amounts to 0.7−0.9 ppm,
whereas it is as large as 1.2 ppm for 15
N CSs. This indicates
that reproducing the sequence trends will be more difficult for
15
N CSs than for 13
C CSs, which agrees with previous findings
of other authors.32,35
The average 31
P CSs are calculated with
an MEE of 0.9 ppm, which is close to the 0.42 ppm difference
between the experimental 31
P CSs of pS19 and pS40. This
makes the discrimination between the two residues of hTH1 a
difficult task within the computational setup employed here.
Provided a suitable reference is chosen, the use of an explicit
solvent in ensemble calculations uniformly improves the
quantitative agreement with experiment for all nuclei
considered in this work. The H-bonding interactions
significantly broaden the CS distributions of polar protons
and phosphoruses. Consequently, the CS ranges increase by
1.5 and 10 ppm for HN
and P atoms, respectively, compared to
implicit solvent-only calculations.
Regardless of the level of theory used, the results for
nonpolar proton CSs deviate from the experimental values by
∼0.5 ppm or less. If a large basis set is applied, rather good
ensemble averages are obtained also for the polar protons. The
b3lyp/pcs4/6-311++G(d,p) level of theory underestimates the
CSs for polar protons by 1−2 ppm. The best predictions of
CSs for the aliphatic carbons (CA, CB) were achieved with the
modest b3lyp/6-311G(d) theoretical description. b3lyp/6-
311++G(d,p) overestimates the carbonyl carbon CSs by 3−4
ppm and the replacement of the 6-311++G(d,p) basis set on
atoms of the original fragment deteriorates the results. The
absolute agreement with the experiment for 15
N and 31
P CSs
depends on the efficiency of the error cancellation, which is
strongly affected by the choice of the referencing method. The
best agreement of the computed 15
N CSs with the experiment
has been found using a secondary reference scheme employing
liquid ammonia and CH3NH2 as the primary and secondary
standards, respectively. Within this referencing scheme, the
15
N CSs computed with b3lyp/6-311++G(d,p) differ from the
experiment by no more than ∼3 ppm. For 31
P, referencing to
the 85% solution of H3PO4 yields computed CSs that differ
from experiment by 6−7 ppm (b3lyp/6-311++G(d,p)) or less
than 1 ppm (b3lyp/6-311G(d)).
15
N CSs computed with bp86/6-311G(d) are overestimated
by ∼30 ppm when referenced to NH3 calculated at the same
level theory. 20 ppm out of the 30 ppm overestimation comes
from the neglect of the gas-to-liquid shift of NH3. 31
P CSs are
underestimated by 30 ppm or more when referenced to either
PH3 secondary reference or H3PO4 computed at the same level
of theory. This systematic error, partly caused by inaccurate
MD molecular geometries, is eliminated here by changing the
reference to the absolute 31
P chemical shielding of 85%
H3PO4. In this way, a favorable quantitative agreement with
experiment is achieved by neglecting the ∼37 ppm difference
between the calculated and experimental 31
P chemical
shielding of PH3.
Calculations employing b3lyp/6-311G(d) and 85% H3PO4
reference provide the 31
P CSs of 3.5 ± 0.6 and 3.4 ± 0.7 ppm
in pS19 and pS40, respectively. Because of the small difference
between the two MD-averaged values and large standard
deviations of the mean, the MD/ADMA/DFT calculations
were not able to discriminate the 31
P CSs of the two residues
in question.
Upon the phosphorylation of hTH1, the computed CB CS
increases by 3.6 ppm in the pS19 and by 2.3 ppm in the pS40
residue, respectively. This is in good agreement with the
experimental 2.7 and 2.8 ppm downfield shift of pS19 and
pS40, respectively. It can be thus concluded that unlike the
standard CS prediction protocols designed for well-structured
proteins, the MD/ADMA/DFT calculations are capable of
correctly predicting the phosphorylation-induced CS changes.
■ ASSOCIATED CONTENT
*S Supporting Information
The Supporting Information is available free of charge on the
ACS Publications website at DOI: 10.1021/acs.jctc.8b00257.
1
H CSs of HA, HB1, and HB2; calculated 15
N chemical
shielding in the reference compounds as well in the pS19
and pS40 residues; calculated 31
P chemical shielding in
the reference compounds as well in the pS19 and pS40
residues; time dependence of the 31
P chemical shielding
in pS19 and pS40; time dependence of the difference
between the 31
P chemical shielding of pS19 and pS40;
and structures of the doubly phosphorylated and
nonphosphorylated hTH1 used as starting coordinates
for the MD simulation (ZIP)
■ AUTHOR INFORMATION
Corresponding Author
*E-mail: precechj@faf.cuni.cz. Phone: +420 549 49 3847.
ORCID
Jana Pavlíková Přecechtělová: 0000-0003-0844-2666
Jozef Hritz: 0000-0002-4512-9241
Notes
The authors declare no competing financial interest.
■ ACKNOWLEDGMENTS
The authors thank Thomas E. Exner (University of Tübingen,
Germany) for providing the Python suite of codes for the
ADMA workflow procedure and the related technical
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5655
assistance. This research was financed by Czech Science
Foundation (no. GF15-34684L) and by the Ministry of
Education, Youth and Sports of the Czech Republic (MEYS
CR) under the National Sustainability Programme II, project
CEITEC 2020 (LQ1601). Computational resources were
provided by the projects CESNET (LM2015042) and
CERIT Scientific Cloud (LM2015085) under the program
“Projects of Large Research, Development, and Innovations
Infrastructures” funded by MEYS CR, and by the IT4Innovations
National Supercomputing Center (LM2015070) under
the program “Large Infrastructures for Research, Experimental
Development and Innovations” funded by MEYS CR.
■ REFERENCES
(1) Oldfield, C. J.; Dunker, A. K. Intrinsically Disordered Proteins
and Intrinsically Disordered Protein Regions. Annu. Rev. Biochem.
2014, 83, 553−584.
(2) Dunker, A. K.; Babu, M. M.; Barbar, E.; Blackledge, M.; Bondos,
S. E.; Dosztányi, Z.; Dyson, H. J.; Forman-Kay, J.; Fuxreiter, M.;
Gsponer, J.; Han, K.-H.; Jones, D. T.; Longhi, S.; Metallo, S. J.;
Nishikawa, K.; Nussinov, R.; Obradovic, Z.; Pappu, R. V.; Rost, B.;
Selenko, P.; Subramaniam, V.; Sussman, J. L.; Tompa, P.; Uversky, V.
N. What’s in a Name? Why These Proteins are Intrinsically
Disordered. Intrinsically Disord. Proteins 2013, 1, No. e24157.
(3) Dyson, H. J.; Wright, P. E. Intrinsically Unstructured Proteins
and Their Functions. Nat. Rev. Mol. Cell Biol. 2005, 6, 197−208.
(4) Ward, J. J.; Sodhi, J. S.; McGuffin, L. J.; Buxton, B. F.; Jones, D.
T. Prediction and Functional Analysis of Native Disorder in Proteins
from the Three Kingdoms of Life. J. Mol. Biol. 2004, 337, 635−645.
(5) Uversky, V. N. Natively Unfolded Proteins: A Point Where
Biology Waits for Physics. Protein Sci. 2002, 11, 739−756.
(6) Dunker, A. K.; Brown, C. J.; Lawson, J. D.; Iakoucheva, L. M.;
Obradović, Z. Intrinsic Disorder and Protein Function. Biochemistry
2002, 41, 6573−6582.
(7) Steinacker, P.; Aitken, A.; Otto, M. 14-3-3 Proteins in
Neurodegeneration. Semin. Cell Dev. Biol. 2011, 22, 696−704.
(8) Hritz, J.; Byeon, I.-J. L.; Krzysiak, T.; Martinez, A.; Sklenar, V.;
Gronenborn, A. M. Dissection of Binding between a Phosphorylated
Tyrosine Hydroxylase Peptide and 14-3-3ζ: A Complex Story
Elucidated by NMR. Biophys. J. 2014, 107, 2185−2194.
(9) Louša, P.; Nedozrálová, H.; Župa, E.; Nováček, J.; Hritz, J.
Phosphorylation of the Regulatory Domain of Human Tyrosine
Hydroxylase 1 Monitored Using Non-Uniformly Sampled NMR.
Biophys. Chem. 2017, 223, 25−29.
(10) Shimada, T.; Fournier, A. E.; Yamagata, K. Neuroprotective
Function of 14-3-3 Proteins in Neurodegeneration. BioMed Res. Int.
2013, 2013, 564534.
(11) Oroguchi, T.; Ikeguchi, M.; Sato, M. Towards the Structural
Characterization of Intrinsically Disordered Proteins by SAXS and
MD Simulation. J. Phys.: Conf. Ser. 2011, 272, 012005.
(12) Kosol, S.; Contreras-Martos, S.; Cedeño, C.; Tompa, P.
Structural Characterization of Intrinsically Disordered Proteins by
NMR Spectroscopy. Molecules 2013, 18, 10802−10828.
(13) Ytreberg, F. M.; Borcherds, W.; Wu, H.; Daughdrill, G. W.
Using chemical shifts to generate structural ensembles for intrinsically
disordered proteins with converged distributions of secondary
structure. Intrinsically Disord. Proteins 2015, 3, No. e984565.
(14) Nielsen, J. T.; Eghbalnia, H. R.; Nielsen, N. C. Chemical Shift
Prediction for Protein Structure Calculation and Quality Assessment
Using an Optimally Parameterized Force Field. Prog. Nucl. Magn.
Reson. Spectrosc. 2012, 60, 1−28.
(15) Shen, Y.; Bax, A. SPARTA+: A Modest Improvement in
Empirical NMR Chemical Shift Prediction by Means of an Artificial
Neural Network. J. Biomol. NMR 2010, 48, 13−22.
(16) Han, B.; Liu, Y.; Ginzinger, S. W.; Wishart, D. S. SHIFTX2:
Significantly Improved Protein Chemical Shift Prediction. J. Biomol.
NMR 2011, 50, 43−57.
(17) Meiler, J. PROSHIFT: Protein Chemical Shift Prediction Using
Artificial Neural Networks. J. Biomol. NMR 2003, 26, 25−37.
(18) Vila, J. A.; Arnautova, Y. A.; Martin, O. A.; Scheraga, H. A.
Quantum-Mechanics-Derived 13
Cα Chemical Shift Server (CheShift)
for Protein Structure Validation. Proc. Natl. Acad. Sci. U.S.A. 2009,
106, 16972−16977.
(19) Xu, X.-P.; Case, D. A. Automated Prediction of 15
N, 13
Cα
, 13
Cβ
and 13
C′ Chemical Shifts in Proteins Using a Density Functional
Database. J. Biomol. NMR 2001, 21, 321−333.
(20) Larsen, A. S.; Bratholm, L. A.; Christensen, A. S.; Channir, M.;
Jensen, J. H. ProCS15: a DFT-Based Chemical Shift Predictor for
Backbone and Cβ atoms in Proteins. PeerJ 2015, 3, No. e1344.
(21) Sumowski, C. V.; Hanni, M.; Schweizer, S.; Ochsenfeld, C.
Sensitivity of ab Initio vs Empirical Methods in Computing Structural
Effects on NMR Chemical Shifts for the Example of Peptides. J. Chem.
Theory Comput. 2014, 10, 122−133.
(22) Mulder, F. A. A.; Filatov, M. NMR Chemical Shift Data and Ab
Initio Shielding Calculations: Emerging Tools for Protein Structure
Determination. Chem. Soc. Rev. 2010, 39, 578−590.
(23) Sitkoff, D.; Case, D. A. Density Functional Calculations of
Proton Chemical Shifts in Model Peptides. J. Am. Chem. Soc. 1997,
119, 12262−12273.
(24) Case, D. A. Chemical Shifts in Biomolecules. Curr. Opin. Struct.
Biol. 2013, 23, 172−176.
(25) Frank, A.; Onila, I.; Möller, H. M.; Exner, T. E. Toward the
Quantum Chemical Calculation of Nuclear Magnetic Resonance
Chemical Shifts of Proteins. Proteins: Struct., Funct., Bioinf. 2011, 79,
2189−2202.
(26) He, X.; Wang, B.; Merz, K. M. Protein NMR Chemical Shift
Calculations Based on the Automated Fragmentation QM/MM
Approach. J. Phys. Chem. B 2009, 113, 10380−10388.
(27) He, X.; Zhu, T.; Wang, X.; Liu, J.; Zhang, J. Z. H. Fragment
Quantum Mechanical Calculation of Proteins and Its Applications.
Acc. Chem. Res. 2014, 47, 2748−2757.
(28) Zhu, T.; Zhang, J. Z. H.; He, X. In Advance in Structural
Bioinformatics; Wei, D., Xu, Q., Zhao, T., Dai, H., Eds.; Springer
Netherlands: Dordrecht, 2015; pp 49−70.
(29) Jose, K. V. J.; Raghavachari, K. Fragment-Based Approach for
the Evaluation of NMR Chemical Shifts for Large Biomolecules
Incorporating the Effects of the Solvent Environment. J. Chem. Theory
Comput. 2017, 13, 1147−1158.
(30) Gordon, M. S.; Fedorov, D. G.; Pruitt, S. R.; Slipchenko, L. V.
Fragmentation Methods: A Route to Accurate Calculations on Large
Systems. Chem. Rev. 2012, 112, 632−672.
(31) Frank, A.; Möller, H. M.; Exner, T. E. Toward the Quantum
Chemical Calculation of NMR Chemical Shifts of Proteins. 2. Level of
Theory, Basis Set, and Solvents Model Dependence. J. Chem. Theory
Comput. 2012, 8, 1480−1492.
(32) Exner, T. E.; Frank, A.; Onila, I.; Möller, H. M. Toward the
Quantum Chemical Calculation of NMR Chemical Shifts of Proteins.
3. Conformational Sampling and Explicit Solvents Model. J. Chem.
Theory Comput. 2012, 8, 4818−4827.
(33) Victora, A.; Möller, H. M.; Exner, T. E. Accurate Ab Initio
Prediction of NMR Chemical Shifts of Nucleic Acids and Nucleic
Acids/Protein Complexes. Nucleic Acids Res. 2014, 42, No. e173.
(34) Zhu, T.; He, X.; Zhang, J. Z. H. Fragment Density Functional
Theory Calculation of NMR Chemical Shifts for Proteins with
Implicit Solvation. Phys. Chem. Chem. Phys. 2012, 14, 7837−7845.
(35) Zhu, T.; Zhang, J. Z. H.; He, X. Automated Fragmentation
QM/MM Calculation of Amide Proton Chemical Shifts in Proteins
with Explicit Solvent Model. J. Chem. Theory Comput. 2013, 9, 2104−
2114.
(36) Swails, J.; Zhu, T.; He, X.; Case, D. A. AFNMR: Automated
Fragmentation Quantum Mechanical Calculation of NMR Chemical
Shifts for Biomolecules. J. Biomol. NMR 2015, 63, 125−139.
(37) Potrzebowski, M. J.; Assfeld, X.; Ganicz, K.; Olejniczak, S.;
Cartier, A.; Gardiennet, C.; Tekely, P. An Experimental and
Theoretical Study of the 13
C and 31
P Chemical Shielding Tensors
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5656
in O-Phosphorylated Amino Acids. J. Am. Chem. Soc. 2003, 125,
4223−4232.
(38) Hoffmann, R.; Reichert, I.; Wachs, W. O.; Zeppezauer, M.;
Kalbitzer, H. R. 1
H and 31
P NMR Spectroscopy of Phosphorylated
Model Peptides. Int. J. Pept. Protein Res. 1994, 44, 193−198.
(39) Matheis, G.; Whitaker, J. R. 3
1P NMR Chemical Shifts of
Phosphate Covalently Bound to Proteins. Int. J. Biochem. 1984, 16,
867−873.
(40) Scheurer, C.; Skrynnikov, N. R.; Lienin, S. F.; Straus, S. K.;
Brüschweiler, R.; Ernst, R. R. Effects of Dynamics and Environment
on N-15 Chemical Shielding Anisotropy in Proteins. A Combination
of Density Functional Theory, Molecular Dynamics Simulation, and
NMR Relaxation. J. Am. Chem. Soc. 1999, 121, 4242−4251.
(41) Vícha, J.; Babinský, M.; Demo, G.; Otrusinová, O.; Jansen, S.;
Pekárová, B.; Žídek, L.; Munzarová, M. L. The Influence of Mg2+
Coordination on 13
C and 15
N Chemical Shifts in CKI1RD Protein
Domain from Experiment and Molecular Dynamics/Density Functional
Theory Calculations. Proteins: Struct., Funct., Bioinf. 2016, 84,
686−699.
(42) Dračínský, M.; Bouř, P. Computational Analysis of Solvent
Effects in NMR Spectroscopy. J. Chem. Theory Comput. 2010, 6, 288−
299.
(43) Dračínský, M.; Möller, H. M.; Exner, T. E. Conformational
Sampling by Ab Initio Molecular Dynamics Simulations Improves
NMR Chemical Shift Predictions. J. Chem. Theory Comput. 2013, 9,
3806−3815.
(44) Přecechtělová, J.; Novák, P.; Munzarová, M. L.; Kaupp, M.;
Sklenář, V. Phosphorus Chemical Shifts in a Nucleic Acid Backbone
from Combined Molecular Dynamics and Density Functional
Calculations. J. Am. Chem. Soc. 2010, 132, 17139−17148.
(45) Přecechtělová, J.; Munzarová, M. L.; Vaara, J.; Novotný, J.;
Dračínský, M.; Sklenář, V. Toward Reproducing Sequence Trends in
Phosphorus Chemical Shifts for Nucleic Acids by MD/DFT
Calculations. J. Chem. Theory Comput. 2013, 9, 1641−1656.
(46) Fukal, J.; Páv, O.; Buděšínský, M.; Šebera, J.; Sychrovský, V.
The Benchmark of 31
P NMR Parameters in Phosphate: a Case Study
on Structurally Constrained and Flexible Phosphate. Phys. Chem.
Chem. Phys. 2017, 19, 31830−31841.
(47) Benda, L.; Schneider, B.; Sychrovský, V. Calculating the
Response of NMR Shielding Tensor σ(31P) and 2J(31P,13C)
Coupling Constants in Nucleic Acid Phosphate to Coordination of
the Mg2+ Cation. J. Phys. Chem. A 2011, 115, 2385−2395.
(48) Piana, S.; Donchev, A. G.; Robustelli, P.; Shaw, D. E. Water
Dispersion Interactions Strongly Influence Simulated Structural
Properties of Disordered Protein States. J. Phys. Chem. B 2015, 119,
5113−5123.
(49) Huang, J.; Rauscher, S.; Nawrocki, G.; Ran, T.; Feig, M.; de
Groot, B. L.; Grubmüller, H.; MacKerell, A. D., Jr CHARMM36m: an
Improved Force Field for Folded and Intrinsically Disordered
Proteins. Nat. Methods 2017, 14, 71−73.
(50) Sarotti, A. M.; Pellegrinet, S. C. A Multi-Standard Approach for
GIAO 13
C NMR Calculations. J. Org. Chem. 2009, 74, 7254−7260.
(51) Cai, L.; Fushman, D.; Kosov, D. S. Density Functional
Calculations of 15
N Chemical Shifts in Solvated Dipeptides. J. Biomol.
NMR 2008, 41, 77−88.
(52) van Wüllen, C. A Comparison of Density Functional Methods
for the Calculation of Phosphorus-31 NMR Chemical Shifts. Phys.
Chem. Chem. Phys. 2000, 2, 2137−2144.
(53) Latypov, S. K.; Polyancev, F. M.; Yakhvarov, D. G.; Sinyashin,
O. G. Quantum Chemical Calculations of 31
P NMR Chemical Shifts:
Scopes and Limitations. Phys. Chem. Chem. Phys. 2015, 17, 6976−
6987.
(54) Chesnut, D. B. Theoretical Study of 31
P NMR Chemical
Shielding Models for Concentrated Phosphoric Acid Solution. J. Phys.
Chem. A 2005, 109, 11962−11966.
(55) Patchkovskii, S.; Ziegler, T. Phosphorus NMR Chemical Shifts
with Self-Interaction Free, Gradient-Corrected DFT. J. Phys. Chem. A
2002, 106, 1088−1099.
(56) Pascual-Borràs, M.; López, X.; Poblet, J. M. Accurate
Calculation of 31
P NMR Chemical Shifts in Polyoxometalates. Phys.
Chem. Chem. Phys. 2015, 17, 8723−8731.
(57) Chernyshev, K. A.; Larina, L. I.; Chirkina, E. A.; Krivdin, L. B.
The Effects of Intramolecular and Intermolecular Coordination on
31
P Nuclear Shielding: Phosphorylated Azoles. Magn. Reson. Chem.
2012, 50, 120−127.
(58) Přecechtělová, J.; Munzarová, M. L.; Novák, P.; Sklenář, V.
Relationships between P-31 Chemical Shift Tensors and Conformation
of Nucleic Acid Backbone: A DFT Study. J. Phys. Chem. B 2007,
111, 2658−2667.
(59) Přecechtělová, J.; Padrta, P.; Munzarová, M. L.; Sklenář, V. P-
31 Chemical Shift Tensors for Canonical and Non-Canonical
Conformations of Nucleic Acids: A DFT Study and NMR
Implications. J. Phys. Chem. B 2008, 112, 3470−3478.
(60) Lindorff-Larsen, K.; Piana, S.; Palmo, K.; Maragakis, P.; Klepeis,
J. L.; Dror, R. O.; Shaw, D. E. Improved Side-Chain Torsion
Potentials for the Amber ff99SB Protein Force Field. Proteins: Struct.,
Funct., Bioinf. 2010, 78, 1950−1958.
(61) Hornak, V.; Abel, R.; Okur, A.; Strockbine, B.; Roitberg, A.;
Simmerling, C. Comparison of Multiple Amber Force Fields and
Development of Improved Protein Backbone Parameters. Proteins
2006, 65, 712−725.
(62) Homeyer, N.; Horn, A. H. C.; Lanig, H.; Sticht, H. AMBER
Force-Field Parameters for Phosphorylated Amino Acids in Different
Protonation States: Phosphoserine, Phosphothreonine, Phosphotyrosine,
and Phosphohistidine. J. Mol. Model. 2006, 12, 281−289.
(63) Hess, B.; Kutzner, C.; van der Spoel, D.; Lindahl, E.
GROMACS 4: Algorithms for Highly Efficient, Load-Balanced, and
Scalable Molecular Simulation. J. Chem. Theory Comput. 2008, 4,
435−447.
(64) Essmann, U.; Perera, L.; Berkowitz, M. L.; Darden, T.; Lee, H.;
Pedersen, L. G. A Smooth Particle Mesh Ewald Method. J. Chem.
Phys. 1995, 103, 8577−8593.
(65) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.;
DiNola, A.; Haak, J. R. Molecular Dynamics with Coupling to an
External Bath. J. Chem. Phys. 1984, 81, 3684−3690.
(66) Evans, D. J.; Holian, B. L. The Nose-Hoover Thermostat. J.
Chem. Phys. 1985, 83, 4069−4074.
(67) Parrinello, M.; Rahman, A. Polymorphic transitions in single
crystals: A new molecular dynamics method. J. Appl. Phys. 1981, 52,
7182−7190.
(68) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.;
Robb, M. A.; Cheeseman, J. R.; Scalmani, G.; Barone, V.; Petersson,
G. A.; Nakatsuji, H.; Li, X.; Caricato, M.; Marenich, A. V.; Bloino, J.;
Janesko, B. G.; Gomperts, R.; Mennucci, B.; Hratchian, H. P.; Ortiz, J.
V.; Izmaylov, A. F.; Sonnenberg, J. L.; Williams-Young, D.; Ding, F.;
Lipparini, F.; Egidi, F.; Goings, J.; Peng, B.; Petrone, A.; Henderson,
T.; Ranasinghe, D.; Zakrzewski, V. G.; Gao, J.; Rega, N.; Zheng, G.;
Liang, W.; Hada, M.; Ehara, M.; Toyota, K.; Fukuda, R.; Hasegawa, J.;
Ishida, M.; Nakajima, T.; Honda, Y.; Kitao, O.; Nakai, H.; Vreven, T.;
Throssell, K.; Montgomery, J. A., Jr.; Peralta, J. E.; Ogliaro, F.;
Bearpark, M. J.; Heyd, J. J.; Brothers, E. N.; Kudin, K. N.; Staroverov,
V. N.; Keith, T. A.; Kobayashi, R.; Normand, J.; Raghavachari, K.;
Rendell, A. P.; Burant, J. C.; Iyengar, S. S.; Tomasi, J.; Cossi, M.;
Millam, J. M.; Klene, M.; Adamo, C.; Cammi, R.; Ochterski, J. W.;
Martin, R. L.; Morokuma, K.; Farkas, O.; Foresman, J. B.; Fox, D. J.
Gaussian 16, Revision A.03; Gaussian, Inc.: Wallingford, CT, 2016.
(69) London, F. Théorie quantique des courants interatomiques
dans les combinaisons aromatiques. J. Phys. Radium 1937, 8, 397−
409.
(70) McWeeny, R. Perturbation Theory for the Fock-Dirac Density
Matrix. Phys. Rev. 1962, 126, 1028−1034.
(71) Ditchfield, R. Self-Consistent Perturbation Theory of
Diamagnetism. Mol. Phys. 1974, 27, 789−807.
(72) Wolinski, K.; Hinton, J. F.; Pulay, P. Efficient Implementation
of the Gauge-Independent Atomic Orbital method for NMR
Chemical Shift Calculations. J. Am. Chem. Soc. 1990, 112, 8251−8260.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5657
(73) Cheeseman, J. R.; Trucks, G. W.; Keith, T. A.; Frisch, M. J. A
Comparison of Models for Calculating Nuclear Magnetic Resonance
Shielding Tensors. J. Chem. Phys. 1996, 104, 5497−5509.
(74) Becke, A. D. Density-Functional Exchange-Energy Approximation
with Correct Asymptotic Behavior. Phys. Rev. A: At., Mol.,
Opt. Phys. 1988, 38, 3098−3100.
(75) Perdew, J. P. Density-Functional Approximation for the
Correlation Energy of the Inhomogeneous Electron Gas. Phys. Rev.
B: Condens. Matter Mater. Phys. 1986, 33, 8822−8824.
(76) Becke, A. D. Density-Functional Thermochemistry. III. The
Role of Exact Exchange. J. Chem. Phys. 1993, 98, 5648−5652.
(77) Lee, C.; Yang, W.; Parr, R. G. Development of the ColleSalvetti
Correlation-Energy Formula into a Functional of the Electron
Density. Phys. Rev. B: Condens. Matter Mater. Phys. 1988, 37, 785−
789.
(78) Miehlich, B.; Savin, A.; Stoll, H.; Preuss, H. Results Obtained
with the Correlation Energy Density Functionals of Becke and Lee,
Yang and Parr. Chem. Phys. Lett. 1989, 157, 200−206.
(79) McLean, A. D.; Chandler, G. S. Contracted Gaussian Basis Sets
for Molecular Calculations. I. Second Row Atoms, Z=11−18. J. Chem.
Phys. 1980, 72, 5639−5648.
(80) Krishnan, R.; Binkley, J. S.; Seeger, R.; Pople, J. A. SelfConsistent
Molecular Orbital Methods. XX. A Basis Set for
Correlated Wave Functions. J. Chem. Phys. 1980, 72, 650−654.
(81) Hariharan, P. C.; Pople, J. A. The Influence of Polarization
Functions on Molecular Orbital Hydrogenation Energies. Theor.
Chim. Acta 1973, 28, 213−222.
(82) Francl, M. M.; Pietro, W. J.; Hehre, W. J.; Binkley, J. S.;
Gordon, M. S.; DeFrees, D. J.; Pople, J. A. Self-Consistent Molecular
Orbital Methods. XXIII. A Polarization-Type Basis Set for SecondRow
Elements. J. Chem. Phys. 1982, 77, 3654−3665.
(83) Barone, V.; Cossi, M. Quantum Calculation of Molecular
Energies and Energy Gradients in Solution by a Conductor Solvent
Model. J. Phys. Chem. A 1998, 102, 1995−2001.
(84) Cossi, M.; Rega, N.; Scalmani, G.; Barone, V. Energies,
Structures, and Electronic Properties of Molecules in Solution with
the C-PCM Solvation Model. J. Comput. Chem. 2003, 24, 669−681.
(85) Jameson, C. J.; Jameson, A. K.; Oppusunggu, D.; Wille, S.;
Burrell, P. M.; Mason, J. 15
N Nuclear Magnetic Shielding Scale from
Gas Phase Studies. J. Chem. Phys. 1981, 74, 81−88.
(86) Cramer, C. J. Essentials of Computational Chemistry: Theories
and Models, 2nd ed.; Wiley: New York, 1961; p 347.
(87) Jameson, C. J.; De Dios, A.; Keith Jameson, A. Absolute
Shielding Scale for 31
P from Gas-Phase NMR studies. Chem. Phys.
Lett. 1990, 167, 575−582.
(88) Vícha, J.; Babinský, M.; Demo, G.; Otrusinová, O.; Jansen, S.;
Pekárová, B.; Žídek, L.; Munzarová, M. L. The Influence of Mg2+
Coordination on 13
C and 15
N Chemical Shifts in CKI1RD Protein
Domain from Experiment and Molecular Dynamics/Density Functional
Theory Calculations. Proteins: Struct., Funct., Bioinf. 2016, 84,
686−699.
(89) Wishart, D. S.; Bigam, C. G.; Yao, J. 1
H, 13
C and 15
N Chemical
Shift Referencing in Biomolecular NMR. J. Biomol. NMR 1995, 6,
135−140.
(90) Bouř, P.; Keiderling, T. A. Partial Optimization of Molecular
Geometry in Normal Coordinates and Use as a Tool for Simulation of
Vibrational Spectra. J. Chem. Phys. 2002, 117, 4126−4132.
(91) Kruse, H.; Šponer, J. Towards Biochemically Relevant QM
Computations on Nucleic Acids: Controlled Electronic Structure
Geometry Optimization of Nucleic Acid Structural Motifs Using
Penalty Restraint Functions. Phys. Chem. Chem. Phys. 2015, 17,
1399−1410.
(92) Broncel, M.; Wagner, S. C.; Paul, K.; Hackenberger, C. P. R.;
Koksch, B. Towards Understanding Secondary Structure Transitions:
Phosphorylation and Metal Coordination in Model Peptides. Org.
Biomol. Chem. 2010, 8, 2575−2579.
(93) Tholey, A.; Lindemann, A.; Kinzel, V.; Reed, J. Direct Effects of
Phosphorylation on the Preferred Backbone Conformation of
Peptides: A Nuclear Magnetic Resonance Study. Biophys. J. 1999,
76, 76−87.
(94) Fujitani, N.; Kanagawa, M.; Aizawa, T.; Ohkubo, T.; Kaya, S.;
Demura, M.; Kawano, K.; Nishimura, S.-i.; Taniguchi, K.; Nitta, K.
Structure Determination and Conformational Change Induced by
Tyrosine Phosphorylation of the N-Terminal Domain of the α-Chain
of Pig Gastric H+/K+-ATPase. Biochem. Biophys. Res. Commun. 2003,
300, 223−229.
Journal of Chemical Theory and Computation Article
DOI: 10.1021/acs.jctc.8b00257
J. Chem. Theory Comput. 2019, 15, 5642−5658
5658
STRUCTURAL AND INTERACTION PROPERTIES OF INTRINSICALLY DISORDERED PROTEINS
(IDPS) DETERMINED BY EXPERIMENTAL NMR AND COMPUTATIONAL STUDIES
64
Paper 15
Jansen, S.;Melková, K.; Trošanová, Z.; Hanáková, K.; Zachrdla, M.; Nováček, J.; Župa, E.;
Zdráhal, Z.; Hritz, J.*; Žídek, L.*: Quantitative Mapping of MAP2c Phosphorylation and
14-3-3ζ Binding Sites Reveals Key Differences Between MAP2c and Tau. J. Biol. Chem.
2017, 292, 6715-6727
Quantitative mapping of microtubule-associated protein 2c
(MAP2c) phosphorylation and regulatory protein 14-3-3␨binding
sites reveals key differences between MAP2c and its
homolog Tau
Received for publication,December 7, 2016, and in revised form, March 1, 2017 Published, Papers in Press,March 3, 2017, DOI 10.1074/jbc.M116.771097
Se´verine Jansen‡§
, Katerˇina Melkova´‡§
, Zuzana Trosˇanova´‡§
, Katerˇina Hana´kova´§
, Milan Zachrdla‡§
, Jirˇí Nova´cˇek§
,
Erik Zˇupa‡§
, Zbyneˇk Zdra´hal§
, Jozef Hritz‡§1
, and Luka´sˇ Zˇídek‡§2
From the ‡
National Centre for Biomolecular Research, Faculty of Science, and the §
Central European Institute of Technology,
Masaryk University, Kamenice 5, CZ-62500 Brno, Czech Republic
Edited by Norma Allewell
Microtubule-associated protein 2c (MAP2c) is involved in
neuronal development and is less characterized than its homolog
Tau, which has various roles in neurodegeneration. Using
NMR methods providing single-residue resolution and quantitative
comparison, we investigated molecular interactions
important for the regulatory roles of MAP2c in microtubule
dynamics. We found that MAP2c and Tau significantly differ in
the position and kinetics of sites that are phosphorylated by
cAMP-dependent protein kinase (PKA), even in highly homo-
logousregions.Wedeterminedthebindingsitesofunphosphorylated
and phosphorylated MAP2c responsible for interactions
with the regulatory protein 14-3-3␨. Differences in phosphorylation
and in charge distribution between MAP2c and Tau suggested
that both MAP2c and Tau respond to the same signal
(phosphorylation by PKA) but have different downstream
effects, indicating a signaling branch point for controlling
microtubule stability. Although the interactions of phosphorylated
Tau with 14-3-3␨ are supposed to be a major factor in
microtubule destabilization, the binding of 14-3-3␨ to MAP2c
enhanced by PKA-mediated phosphorylation is likely to influence
microtubule-MAP2c binding much less, in agreement with
the results of our tubulin co-sedimentation measurements. The
specific location of the major MAP2c phosphorylation site in a
region homologous to the muscarinic receptor-binding site of
Tau suggests that MAP2c also may regulate processes other
than microtubule dynamics.
Cytoskeletal microtubule-associated proteins (MAPs)3
are
proteins of critical importance for regulating the stability and
dynamics of microtubules (1). MAP2 and Tau represent MAP
subfamilies expressed in neurons; MAP2 is localized in dendrites,
whereas Tau is found mainly in axons (2). Tau and
MAP2 belong to the class of intrinsically disordered proteins
(IDPs), which lack a unique structure and which exist in multiple,
quickly interconverting conformations (3–7). NMR is the
method of choice for structural investigation of this type of
protein. Tau and MAP2 differ in their N-terminal projection
domains, which contain acidic and proline-rich subdomains,
whereas the C-terminal parts, containing the microtubulebinding
domain (MTBD) and the C-terminal region, are homologous
(8). Tau is expressed in several splice variants. The
human brain isoforms differ in the number of microtubulebinding
regions (MTBRs) in the C-terminal portion and in the
presence of two inserts near the N terminus. For the sake of
simplicity, only the 441-residue variant lacking exons 6, 8, and
10 (9) is discussed in this paper. This variant has been studied in
detail and was shown to form paired helical filaments and
neurofibillary tangles in brains of patients suffering from
Alzheimer’s disease (10). The MAP2 family is composed of two
high-molecular-weight proteins, MAP2a and MAP2b, each
consisting of 1830 amino acids, and two low-molecular-weight
proteins, MAP2c and MAP2d, consisting of 467 and 498 amino
acids, respectively. The MAP2 isoforms differ mainly in their
projection domains (8), with MAP2c being the shortest functional
isoform. MAP2c mainly is expressed perinatally (2). Postnatally,
its expression is restricted to regions exhibiting postnatal
plasticity, such as the olfactory bulb (11), suggesting a role in
neuronal development.
The phosphorylation of MAPs, which regulates their binding
to microtubules (12–17) and consequently the microtubule
dynamics (18), is implicated in neuronal development and
plasticity (19). In vivo MAPs are substrates of various protein
kinases, such as cAMP-dependent protein kinase A (PKA), protein
kinase C (PKC), cdc2 kinase, and the like (16, 17, 20, 21).
Phosphorylation of Tau weakens its interaction with microtubules
and increases its affinity to regulatory 14-3-3 proteins
(vide infra).
This work was supported by Grants GA15-14974S (to S. J., K. M., K. H., M. Z.,
and L. Z.) and GF15-34684L (Z. T., E. Z., and J. H.) from the Czech Science
Foundation. The authors declare that they have no conflicts of interest
with the contents of this article.
The chemical shifts were deposited in the Biological Magnetic Resonance Bank
under BMRB accession no. 26960.
1
To whom correspondence may be addressed. Tel.: 420-54949-3847; E-mail:
jozef.hritz@ceitec.muni.cz.
2
To whom correspondence may be addressed: National Centre for Biomolecular
Research, Masaryk University, Kamenice 5, CZ-62500 Brno, Czech
Republic. Tel.: 420-54949-8393; Fax: 420-54949-2556; E-mail: lzidek@
chemi.muni.cz.
3
The abbreviations used are: MAP, microtubule-associated protein; IDP,
intrinsically disordered protein; MTBD, microtubule-binding domain;
MTBR, microtubule-binding region; SSP, secondary structure propensity;
MST, microscale thermophoresis; TCEP, tris(2-carboxyethyl)phosphine.
cros
ARTICLE
J. Biol. Chem. (2017) 292(16) 6715–6727 6715
© 2017 by The American Society for Biochemistry and Molecular Biology, Inc. Published in the U.S.A.
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/DownloadedfromatMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/DownloadedfromatMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
The 14-3-3 family includes highly conserved ubiquitous proteins
(22) mostly expressed in the brain, and in particular in
regions exhibiting neuroplasticity, and having important roles
in neuronal development (23). The seven 14-3-3 isoforms present
in mammals are ␤, ␥, ⑀, ␨, ␶, ␴, and ␩. The 14-3-3␨ isoform
was reported to be associated with microtubules in the brain
(24). In vitro, 14-3-3␨ interacts with unphosphorylated Tau;
however the phosphorylation of Tau increases its affinity for
14-3-3␨ (24–27).
Our goal was to compare the phosphorylation of Tau and
MAP2c by PKA and the interactions of Tau and MAP2c with
14-3-3␨. To the best of our knowledge, no data have been published
on the interaction of 14-3-3␨ with MAP2c until now. A
prerequisite for such studies is the knowledge of the phosphorylated
residues of MAP2c. MAP2c phosphorylation by PKA has
been studied previously, but the reported data are partially contradictory
(28, 29). Therefore, we re-addressed this issue and
determined the PKA phosphorylation sites of MAP2c by NMR
and mass spectrometry (MS). We then characterized the interaction
of MAP2c with 14-3-3␨ at a molecular level, and we
studied the effect of 14-3-3␨ on tubulin polymerization induced
by MAP2c. We found that the same regions of unphosphorylated
MAP2c and Tau interact with 14-3-3␨ and that the binding
affinity of MAP2c is greatly increased by phosphorylation,
as described previously for Tau (25, 39). However, NMR
analysis of the phosphorylation kinetics revealed that,
despite their highly homologous sequences, PKA phosphorylates
MAP2c and Tau in a different manner. Consequently,
the high-affinity binding sites of phosphorylated MAP2c and
Tau differ substantially.
Results
NMR assignment of the phosphorylated MAP2c
Chemical shifts of unphosphorylated MAP2c were assigned
in our previous study (30). The same assignment strategy for
PKA-phosphorylated MAP2c was used in this study. The
resonance frequencies of the phosphorylated MAP2c were
assigned using 5D CACONCACO, 3D (H)CANCO, and 5D
HC(CC-TOCSY)CACON NMR experiments. The spectra
weremeasuredona1.1mM [13
C,15
N]MAP2csamplephosphorylated
by PKA for 24 h. The assigned backbone chemical shifts
were similar to those obtained for unphosphorylated MAP2c
other than the phosphorylated residues and their neighbors.
The major changes in the CACONCACO spectrum were
observed near Ser-435, where six neighboring residues showed
a significant change in backbone chemical shifts, up to 1 ppm
for 15
N and 0.3 ppm for protons.
Identification of phosphorylation sites
The phosphorylated MAP2c residues were first detected by
NMR spectroscopy. A 1
H,15
N HSQC spectrum of [15
N,13
C]
MAP2c was measured after 24 h of phosphorylation. Four new
intense peaks and several weaker signals appeared at proton
frequencies downfield from 8.6 ppm, corresponding to a chemical
shift of the NH groups influenced by phosphorylation (Fig.
1). The 5D CACONCACO spectra of phosphorylated MAP2c
provided an unambiguous assignment of the major new peaks
in the 1
H,15
N HSQC spectrum to pSer-435, pSer-184, pThr-
220, and to Arg-221 following pThr-220. The 3D HNCACB
spectrum allowed us to classify the types of amino acids preceding
the four additional phosphoserines/phosphothreonines
detected in the 1
H,15
N HSQC spectrum.
The phosphorylation sites were also identified by MS (Table
1). After incubation with PKA for 24 h, MAP2c was digested
with trypsin. The phosphorylated peptides were separated from
the unphosphorylated ones by TiO2 fractionation, and the
phosphopeptides were subjected to LC-MS/MS analysis. In
agreement with the NMR data, the most intense peaks corresponded
to peptides containing pSer-184 and pSer-435. Phosphopeptides
containing pThr-220 were not detected in our
tryptic digests (possibly because of the high frequency of tryptic
cleavage sites in the vicinity of Thr-220 resulting in peptides too
short for analysis under the conditions used in our study) but
9.0
9.0
8.5
8.5
8.0
8.0
δ2 (1H) / ppm
125 125
120 120
115 115
δ1(15N)/ppm
pS189
pS435
pS184
pT220
pS319
pS212 or pS215
S222
R221
pX
9.0
9.0
8.8
8.8
8.6
8.6
δ2 (1H) / ppm
122 122
120 120
118 118
116 116
114 114
δ1(15N)/ppm
pS189
pS435
pS184
pT220
pS319
pS212 or pS215
pX
Figure 1. Overlaid 1
H,15
N HSQC spectra of unphosphorylated [15
N,13
C]
MAP2c (blue), and [15
N,13
C]MAP2c phosphorylated by PKA for 24 h (red).
The well-resolved peaks of phosphorylated residues (pSer-184, pSer-189,
pThr-220, and pSer-435) at proton frequencies downfield from 8.6 ppm and
of their neighbors (Arg-221 and Ser-222) are labeled. Bottom, close-up of the
region of phosphorylated residues plotted with a lowered signal threshold
level to show minor peaks. pX, this peak could not be assigned.
Phosphorylation of MAP2c and interaction with 14-3-3
6716 J. Biol. Chem. (2017) 292(16) 6715–6727
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
were unambiguously identified previously (29). The peak areas
of the obtained phosphopeptides relative to the sum of the areas
of peaks containing pSer-435 are presented in Fig. 2. The phosphopeptides
identified by MS are presented in Table 1.
Verification of NMR assignment of phosphorylated residues by
site-directed mutagenesis
Because the sensitivity of the 5D CACONCACO spectra
was not sufficient to assign minor phosphorylation sites, we
designed several mutants (S184D, S189D/S435D, S199D/
S435D, T220E, S319D/S435D, T320E/S435D, S342D/S435D,
S350D/S435D, S367D/S435D, S382D/S435D, and S435D)
basedonourMSdataandonphosphorylationsitesreportedinthe
literature(28).Themutantswereuniformly15
N-labeledandphosphorylated,
and the 1
H,15
N HSQC spectra were recorded. In the
spectra of the S184D, T220E, S435D, S189D/S435D, and S319D/
S435Dmutants,thepeaksofthemutatedphosphorylatedresidues
disappeared, confirming our previous assignment and allowing us
to assign two minor peaks to pSer-189 and pSer-319. No significant
changes were observed in the region of the phosphorylated
serines and threonines in the HSQC spectra of the other mutants,
suggesting that their degree of phosphorylation is low.
Sequence-based prediction of phosphorylation sites
In principle, the target sites for specific kinases are encoded
in the amino acid sequence and by their accessibility to the
kinase. However, currently available Web server predictors of
phosphorylation sites have been optimized for structured proteins,
whereas their reliability for IDPs is not well documented.
Fig. 3 compares the PKA phosphorylation sites of MAP2c
determined by NMR in this study with three popular Web
server predictors. One can see that the PKA phosphorylation of
Ser-435 was predicted by all three predictors. Sites Ser-184,
Ser-189, and Thr-220 were predicted by Scansite3 (31) and
GPS3 (32) but not by Kinasephos2 (33). This limited comparison
indicates that Scansite3 and GPS3 can identify the most
important PKA phosphorylation sites within the intrinsically
disordered protein MAP2c. On the other hand, all three predictors
generated significant number of false-positive predictions
with respect to the NMR data (Fig. 3). Interestingly, the same
type of prediction for Tau protein shows better agreement
between GPS3 and Kinasephos2 in contrast to Scansite3 (Fig.
3). New experimental results are therefore needed to improve
the reliability of prediction of phosphorylated sites within IDPs.
0.0
0.2
0.4
0.6
0.8
1.0
1.2
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
P P P P P P PP PPP P
RelativeIntensity
0.0
0.2
0.4
0.6
0.8
1.0
1.2
160 170 180 190 200 210 220 230 240 250 260 270 280 290 300 310
P P P P PP P P P P PP P P P P P P P PP P P P P
RelativeIntensity
0.0
0.2
0.4
0.6
0.8
1.0
1.2
320 330 340 350 360 370 380 390 400 410 420 430 440 450 460
P P P P P P
RelativeIntensity
Residue number
Figure 2. Phosphorylation of MAP2c. The dots represent the ratio Uu/Up, where Uu and Up are peak intensities of unphosphorylated residues in the unphosphorylated
and phosphorylated MAP2c samples, respectively. Triangles represent the ratio Pp/(Up ϩ Pp), where Pp are peak intensities of phosphorylated
residues in the phosphorylated MAP2c sample. The bars represent the sum of areas of peptide peaks containing the given phosphorylated residue in mass
spectra (relative to the sum of areas of peptides peaks containing Ser-435). Letter P (above the plot), indicates proline residues.
Table 1
Phosphopeptides identified by MS in percentage compared with the
most intense peak, Ser-435
Only the phosphopeptides that have an intensity higher than 1% of the intensity of
the most intense peak are presented here. In the absence of PKA, none of the
phosphopeptides identified has an intensity higher than 1% of the most intense peak.
The phosphorylated residues are indicated in bold.
Residue Peptide % Intensity
Ser-184 183
SSLPRPSSILPPR195
65
Ser-199 196
RGVSGDREENSFSLNSSISSAR217
5
Ser-212 197
GVSGDREENSFSLNSSISSAR217
1
Ser-214 197
GVSGDREENSFSLNSSISSAR217
2
Ser-231 228
AGKSGTSTPTTPGSTAITPGTPPSYSSR255
2
Ser-319 315
SKIGSTDNIK324
7
Ser-367 363
VKIESVKLDFK373
2
Ser-435 433
RLSNVSSSGSINLLESPQLATLAEDVT
AALAK464
100
Ser-435, Ser-438 433
RLSNVSSSGSINLLESPQL451
3
Ser-435, Ser-442 433
RLSNVSSSGSIN444
1
Phosphorylation of MAP2c and interaction with 14-3-3
J. Biol. Chem. (2017) 292(16) 6715–6727 6717
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
Quantitative determination of phosphorylation
Strong signals indicating a high degree of phosphorylation
at Ser-184 and Ser-435 were observed using both NMR and
MS methods. NMR was used to determine the degree of
phosphorylation quantitatively. A comparison of the peak
intensities in the 3D HNCO spectra of phosphorylated and
unphosphorylated MAP2c revealed three regions showing a
significant decrease in the intensity of the peaks, corresponding
to unphosphorylated residues in the phosphorylated sample
(Fig. 2). The intensity levels decreased to ϳ40% in the vicinity
of Thr-220, to less than 10% in the vicinity of Ser-184 and
below the detection limit in the vicinity of Ser-435. Such a comparison
provided an overall quantitative picture of the phosphorylation
and allowed us to estimate the degree of phosphorylation
of individual amino acids (with the exception of residues
close in sequence to Ser-184, Thr-220, and Ser-435, where the
effect of possible minor phosphorylation was obscured by the
decrease in the signal intensity due to a major phosphorylation
site). In summary, both NMR and MS identified Ser-184 and
Ser-435 as major PKA phosphorylation sites of MAP2c, and
significant phosphorylation was observed also at Thr-220
(Fig. 2).
Kinetics of phosphorylation
The kinetics of phosphorylation was followed by real-time
NMR spectroscopy from 2 min to 30 h after the addition of
PKA. The signal of the first phosphorylated residue, pSer-435,
already appeared in the first spectrum recorded after the addition
of PKA and reached saturation after 1.5 h of incubation
(Fig. 4). The kinetic curve was exponential even for short reaction
times, indicating that PKA was not saturated by MAP2c.
The other residues were phosphorylated much more slowly,
and their buildup curves exhibited lag phases documenting
competition of the phosphorylation sites for PKA. After the lag
phase, the ratio of phosphorylation rates of Ser-184, Ser-189,
and Thr-220 was almost constant (ϳ4:1:2).
To compare the phosphorylation of individual sites quantitatively,
apparent rate constants (kobs) were estimated as described
under“Experimentalprocedures.”Theevaluationofkobs wasfacilitated
by the kinetic separation of Ser-435, allowing us to assume
that the other sites were unphosphorylated during phosphorylation
of Ser-435 by the aforementioned kinetic partitioning of Ser-
184, Ser-189, and Thr-220. The estimated kobs values were 9700 Ϯ
1500 M
Ϫ1
sϪ1
for Ser-435, 200 Ϯ 14 M
Ϫ1
sϪ1
for Ser-184, 50 Ϯ 6
M
Ϫ1
sϪ1
forSer-189,and95Ϯ21M
Ϫ1
sϪ1
forThr-220.Insummary,
P1 P2
MTBR1 MTBR2
MTBR3 MTBR4
Muscarinic receptor binding site
Tau:
Tau:
Tau:
Tau:
Tau:
Tau:
Tau:
MAP2c:
MAP2c:
MAP2c:
MAP2c:
MAP2c:
MAP2c:
MAP2c:
aggregation seed
aggregation seed
61
106
141
217
297
377
441
78
156
227
275
331
403
467
MAEPRQEFEVMEDHAGTYGLG- - - - - - - - DRKDQGG- - - - - - - - - - - YTMHQDQEGDTDAGLKESPLQTPTEDGSEEPGS
ETSDAKSTPTAEDVTAPLVD - - - - - - - - - - - - - - - - EGAPGKQAAAQP - - - - HT - E I P - - - - - - - - - EGTTA - EEAG IGD
TP - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - SLEDEAAGHVTQARMVSKSKDGTGSDDKK
AKGADGKTK I ATPRGAAPPGQKGQANATR I PAKTPPAPKTPPSSGEPPKSGDRSGYSS - - PGSPGTPGSRSRTPSL - - PT
PPTREPKKVAVVRTPPKSPSSAKSRLQTAPVPMPDLKNVKSK IGSTENLKHQPGGGKVQ I I NKKLDLSNVQSKCGSKDN I
KHVPGGGSVQ I VYKPVDLSKVTSKCGSLGN I HHKPGGGQVEVKSEKLDFKDRVQSK I GSLDN I THVPGGGNKK I ETHKLT
FRENAKAKTDHGAE I VYKSPVVSGDTSPRHLSNVSSTGS I DMVDSPQLATLADEVSASLAKQGL
MADERKDEGKAPHWTSASLTEAAAHPHSPEMKDQGGSGEGLSRSANGFPYREEEEGAFGEHGSQGTYSDTKENG I N- -GE
LTSADRE - - TAEEVSAR I VQVVTAEAVAVLKGEQEKEAQHKDQPAALPLAAEETVNLPPSPPPSPASEQTAALEEATSGE
SAQAPSAFKQAKDKVTDGI TKSPEKRSSLPRPSS I LPPRRGVSGDREENSFSLNSS I - - - - SSARRTTRS - - - - - EP I RR
A - - - - - - - - - - - - - - - - - - -GKSGTSTPTTPGSTA I TPGTPPS - - - - - - - - - - - - YSSRTPGTPGTP - SYPRTPGTPKSG
I LVPSEKKVA I I RTPPKSPATPKQ- LRL I NQPLPDLKNVKSK IGSTDN I KYQPKGGQ- - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - VQ I VTKK I DLSHVTSKCGSLKN I RHRPGGGRVK I ESVKLDFKEKAQAKVGSLDNAHHVPGGGNVK I DSQKLN
FREHAKARVDHGAE I I TQSPSRSSVASPRRLSNVSSSGS I NLLESPQLATLAEDVTAALAKQGL
Fyn-binding sitejaw jaw
jaw
Figure3.Comparisonofsequences,phosphorylationsites,andelectrostaticpropertiesofTauandMAP2c.Experimentallydeterminedphosphorylation
sites are indicated by background colors: red, class I; orange, class II; green, class III (according to the classification used by Laudrieu et al. (43)). The predicted
phosphorylation sites are indicated by dots below the sequences: red, prediction by Scansite3; green, prediction by GPS3; blue, prediction by Kinasephos2.
Residues predicted by 1433pred (48) to interact with 14-3-3␨ are underlined. Regions with a propensity to form helical, ␤-strand, and polyproline II structures
according to Mukrasch et al. (38) and Nova´cˇek et al. (30) are indicated by magenta, cyan, and gray bars above the sequences, respectively. The relative
electrostatic potential is shown as color-coded boxes below the sequences for unphosphorylated (upper row) and phosphorylated (lower row) form of each
protein. The potential is approximated by ⌺jCQi/(d0 ϩ d1͉ni Ϫ nj͉), where Qi and nj are the charge and sequential number of the i-th residue, C is a constant
including the electric permittivity, and dk are distance constants. The ratio d1/d0 was set to 2.0, and the colors were chosen so that red and blue correspond to
the highest negative and positive potential, respectively, which makes the color code independent of C/d0. Gaps in the sequences were inserted manually to
optimize the alignment of regions with similar trends to form transient secondary structures. The annotation of functionally important regions of Tau, defined
by the double-headed arrows above the sequences, was taken from the literature (50, 38, 9).
Phosphorylation of MAP2c and interaction with 14-3-3
6718 J. Biol. Chem. (2017) 292(16) 6715–6727
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
the real-time NMR measurements revealed a dramatic difference
between phosphorylation of Ser-435 and other residues.
Effect of phosphorylation on the secondary structure
propensities of MAP2c
To determine the effect of phosphorylation on the secondary
structure, secondary structure propensities (SSP) of unphosphorylated
and phosphorylated MAP2c were calculated. To
identify SSP changes close to the phosphorylation sites, the SSP
was calculated from the chemical shifts of 13
C␣
, 13
C␤
, and 1
H␣
,
recently found to be insensitive to the presence of the phosphate
group attached to neighboring residues (34). The most
significant changes were observed in two regions close to the
major phosphorylation sites. In the vicinity of pSer-184 and
especially pThr-220, the SSP value indicates that phosphorylation
stabilizes the already existing secondary structures (Fig. 5).
The opposite effect was observed in a region preceding pSer-
435, where phosphorylation alters the SSP of residues Arg-
425–Pro-431 from the propensity to form an extended secondary
structure in the unphosphorylated state to a helical
propensity in the phosphorylated state (Fig. 5). The less pronounced
changes observed for residues distant from the phosphorylation
sites (including the N-terminal region) indicate
that phosphorylation by PKA also influences the intramolecular
interactions of MAP2c.
Apparent dissociation constants of MAP2c-14-3-3␨ complexes
The affinities of 14-3-3␨ to MAP2c in phosphorylated and
non-phosphorylated forms were compared by microscale thermophoresis
(MST). To achieve a sufficient sensitivity, MAP2c
was labeled fluorescently. MAP2c contains a single cysteine
residue (Cys-348), but it is located in a potential binding region
(MTBR3). Therefore, an E52C/C348S MAP2c construct was
prepared and labeled specifically by the fluorescent dye Alexa
Fluor 647 at the introduced Cys-52.
Titration of E52C/C348S MAP2c with 14-3-3␨ revealed that
both unphosphorylated and phosphorylated MAP2c bind 14-3-
3␨, but phosphorylation greatly enhances the affinity. The titration
curve of phosphorylated E52C/C348S MAP2c exhibited a
well-defined inflection point in the low micromolar range of
14-3-3␨ concentrations and another less-defined inflection
point at a concentration 2 orders of magnitude higher (Fig. 6).
Fitting the first sigmoidal region separately (for 14-3-3␨ monomer
concentrations lower than 100 ␮M) resulted in an apparent
dissociation constant (expressed for the 14-3-3␨ monomer concentration
as described under “Experimental procedures”) of
0.57 Ϯ 0.37 ␮M and stoichiometry of n ϭ 1.53 Ϯ 0.14 (14-3-3␨
monomer/MAP2c molecule). The fractional stoichiometry
may indicate the contribution of several binding modes, but the
size of the experimental errors made a precise determination of
the stoichiometry difficult. Fitting the second region gives KЈD Ϸ
280 Ϯ 120 ␮M, but full saturation could not be achieved.
The binding curve of unphosphorylated E52C/C348S
MAP2c showed an interaction at least 1 order of magnitude
weaker than observed for the phosphorylated form. Fitting the
data provided KЈD ϭ 92 Ϯ 12 ␮M, but the shape of the titration
curve indicates that this number is probably affected by additional
binding events that occur at a higher 14-3-3␨ concentration
and cannot be fully separated.
The obtained apparent dissociation constants are useful for
a rough comparison of the binding affinities. However, they
should not be overinterpreted because the MAP2c-14-3-3␨
interaction is more complex than the simple binding model
used for the data fitting (35).
Identification of 14-3-3␨-binding sites in MAP2c
NMR spectroscopy was used to identify individual residues
of MAP2c that interact with 14-3-3␨. Unphosphorylated
[13
C,15
N]MAP2c and [13
C,15
N]MAP2c phosphorylated for 24 h
by PKA were mixed with 14-3-3␨ in ratios of 1:0.125, 1:0.25,
1:0.5, 1:1, and 1:2 (MAP2c:14-3-3␨ monomer). A 3D HNCO
spectrum was recorded before titration and after each addition
of 14-3-3␨. When titrating unphosphorylated MAP2c with
14-3-3␨ (Fig. 7), we observed a gradual diminution of the intensity
of the peaks of the residues in the MTBD (Thr-296–Ser-
380) and in the C-terminal region with a strong helical propensity
(Ile-443–Leu-467), indicating that these regions bind
14-3-3␨ and become less flexible. Note that the broad region of
residues with reduced peak intensity due to the interaction with
14-3-3␨ is interrupted by several short stretches of amino acids
already exhibiting significant line broadening in free MAP2c,
and therefore it is little affected by 14-3-3␨ binding. This effect
is particularly notable for the PXGG motifs immediately preceding
the regions with high ␤-strand propensity in the MTBD.
These regions initiate aggregation in Tau (36, 37). On the other
hand, a decrease in the peak intensity was also observed for
several residues in the N-terminal portion of MAP2c, indicating
that interdomain interactions exist in MAP2c as in Tau (38).
The addition of 14-3-3␨ to phosphorylated MAP2c resulted,
in addition to the changes described above, in a strong decrease
of peak heights of amino acids in the vicinity of pSer-184 and
pThr-220 in the proline-rich domain and of pSer-435 in the
region corresponding to the muscarinic receptor-binding site
of Tau (Fig. 3). The fact that the peak intensities were substantially
reduced already in the presence of substoichiometric
amounts of 14-3-3␨ suggests that the spectra of the 14-3-3␨bound
MAP2c are also influenced by an intermediate chemical
exchange. Therefore, we did not use our NMR data for quantitative
determination of the binding affinity.
0.0
1.0
2.0
3.0
0 5 10 15 20 25 30
Peakvolume/arbitraryunits
Time / hrs
pS435
pS184
pT220
pS189
Figure 4. Volumes of pSer-435 (black), pSer-184 (dark gray), pThr-220
(medium gray), and pSer-189 (light gray) peaks in the SOFAST-HMQC
spectra recorded during MAP2c phosphorylation by PKA.
Phosphorylation of MAP2c and interaction with 14-3-3
J. Biol. Chem. (2017) 292(16) 6715–6727 6719
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
Effect of 14-3-3␨ on MAP2c-induced tubulin polymerization
The effect of 14-3-3␨ on the polymerization of tubulin
induced by unphosphorylated and phosphorylated MAP2c was
measured by co-sedimentation assays (Fig. 8). To measure the
direct effect of MAP2c and 14-3-3␨ on tubulin polymerization,
tubulin was not stabilized by Taxol in the assay. 14-3-3␨ alone
was not able to induce the polymerization of microtubules (data
not shown). About 20% of the tubulin polymerized spontaneously.
In the presence of 2 ␮M unphosphorylated and phosphorylated
MAP2c, tubulin polymerization increased by 13 Ϯ 2%
and 12 Ϯ 1%, respectively. The addition of up to 2 ␮M 14-3-3␨
(monomer concentration) had a negligible effect, but 20 ␮M
14-3-3␨ reduced the amount of tubulin in the pellet to a level
comparable with the background of its spontaneous polymerization.
The effect of MAP2c phosphorylation was negligible.
Thus the data show that phosphorylation of MAP2c by PKA
does not influence its affinity to tubulin under the conditions
used and that 14-3-3␨ competes with tubulin for binding unphosphorylated
and PKA-phosphorylated MAP2c with a comparable
efficiency.
Discussion
The key role of phosphorylation in regulating the interactions
of MAPs with microtubules has been well-known for
decades (12–17). 14-3-3 proteins are proposed to be involved
in the phosphorylation-dependent control of microtubule
dynamics by competing for phosphorylated Tau with tubulin
(24, 39, 40). PKA seems to be the key kinase in this mechanism.
Phosphorylation of Tau by PKA or PKB, but not by GSK-3␤,
CDK2, or CK1, enhances the interaction with 14-3-3␨ (25),
whereas Tau phosphorylated by PKA binds to tubulin with
7-fold lower affinity (41, 20).
In our study, we searched for possible differences between
Tau and its homolog, MAP2c, that would implicate distinct
regulatory roles for these proteins. As phosphorylation by PKA
is sufficient to form the sites critical for 14-3-3␨ binding in Tau,
we also used PKA in our studies of MAP2c.
Phosphorylation of MAP2c by PKA has been studied already
in the past. However, early studies of MAP2c phosphorylation
resulted in partially contradictory conclusions. Using two-dimensional
electrophoresis combined with mass spectrometry
and site-directed mutagenesis, Ozer and Halpain (28) identified
Ser-319, Ser-350, and Ser-382 as early phosphorylation sites.
Later, Ser-184, Thr-220, and Ser-435 were reported by Alexa et
al. (29) to be the major phosphorylation sites in MAP2c, in
agreement with the prediction based on the sequence (42). A
single completely phosphorylated residue, Thr-220, was found
in a 20-kDa peptide (Asn-205–Glu-366) of MAP2c phosphorylated
by PKA (29). Although the Asn-205–Glu-366 peptide
also contains Ser-319, Ser-350, and Ser-382, phosphorylation of
these serines was not observed by Alexa et al. (29). Therefore,
we revisited the topic of phosphorylation of MAP2c by PKA
and, taking advantage of the recent methodological progress in
MS and NMR spectroscopy, clarified which amino acids are
predominantly phosphorylated under the given conditions.
Our results confirm that Ser-184, Thr-220, and Ser-435 are
-0.2
0.0
0.2
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
SSP
-0.2
0.0
0.2
160 170 180 190 200 210 220 230 240 250 260 270 280 290 300 310
* *
SSP
-0.2
0.0
0.2
320 330 340 350 360 370 380 390 400 410 420 430 440 450 460
*
SSP
Residue number
Figure5.EffectsofphosphorylationonthesecondarystructurepropensityofunphosphorylatedMAP2c(blue)andphosphorylatedMAP2c(red).The
symbols above the graphs indicate continuous regions of unphosphorylated (blue) and phosphorylated (red) with the propensity to form helical (empty boxes)
or extended (arrows) structures (see “Experimental procedures” for details). Asterisks indicate the major PKA phosphorylation sites.
0.870
0.875
0.880
0.885A
0.825
0.830
0.835
0.840
0.845
0.850
0.01 0.1 1 10 100 1000
B
Normalizedfluorescence
14-3-3ζ monomer concentration / μ M
Figure 6. Microscale thermophoretic analysis of interaction between
5␮M E52C/C348S MAP2c and 14-3-3␨ in 50 mM Tris buffer. A, unphosphorylated
MAP2c; B, phosphorylated MAP2c. The plots show interactions in the
14-3-3␨ monomer concentration range from 36.6 nM to 1.2 mM. The mean
values Ϯ S.D. for each concentration point were calculated from triplicate
measurements.
Phosphorylation of MAP2c and interaction with 14-3-3
6720 J. Biol. Chem. (2017) 292(16) 6715–6727
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
phosphorylated most efficiently. The Web server predictors
also identified Ser-184, Thr-220, and Ser-435 as phosphorylation
sites, but the number of false positive predictions documents
that experimental determination of phosphorylation
sites is still necessary (Fig. 3).
The locations of the phosphorylation sites in MAP2c, and
especially the kinetics of phosphorylation, differ significantly
from Tau. Our results can be compared quantitatively with the
data published by Landrieu et al. (43), which were obtained
under similar conditions.
The major phosphorylation site of Tau (class I according to
the classification introduced by Landrieu et al. (43), corresponding
to kobs Ϸ 104
M
Ϫ1
sϪ1
and highlighted by the red background
in Fig. 3) is 211
RTPpSLP216
in the P2 region of the
proline-rich domain (Fig. 3), representing a typical 14-3-3␨binding
motif, RS/TXpSXP (25). The P2 region is involved in
interactions with tubulin (38, 44), and phosphorylation of Ser-
214 reduces tubulin binding (44) and the ability of Tau to promote
microtubule assembly (45). The corresponding sequence
in MAP2c, 255
RTPGTP260
, does not contain a phosphorylatable
serine or threonine in the position of pSer-214 as in Tau. This
correlates with the fact that the phosphorylation by PKA inhibits
the microtubule-stabilizing (growth promoting) activity of
Tau (46) but not of MAP2 (47). However, the proline-rich
region of MAP2c also contains a similar phosphorylated 14-3-
3␨-binding site (48) but in a different position (181
KRSpSLP186
),
in a region exhibiting a certain sequence homology with the P1
region (see Fig. 3) of Tau. Ser-184 of MAP2c is phosphorylated
more slowly than Ser-214 of Tau, exhibiting the second highest
phosphorylation rate among the MAP2c phosphorylation sites
(class II, with kobs Ϸ 102
M
Ϫ1
sϪ1
, yellow background in Fig. 3).
Both Ser-214 of Tau and Ser-184 of MAP2c are accompanied by
weaker phosphorylation sites (Ser-208 and Ser-189, respectively)
in close proximity.
0.0
1.0
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
A P P P P P P PP PPP P
I/I0
0.0
1.0
160 170 180 190 200 210 220 230 240 250 260 270 280 290 300 310
P P P P PP P P P P PP P P P P P P P PP P P P P
I/I0
0.0
1.0
320 330 340 350 360 370 380 390 400 410 420 430 440 450 460
P P P P P P
I/I0
0.0
1.0
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
B P P P P P P PP PPP P
I/I0
0.0
1.0
160 170 180 190 200 210 220 230 240 250 260 270 280 290 300 310
P P * P P PP * P P P P PP P P P P P P P PP P P P P
I/I0
0.0
1.0
320 330 340 350 360 370 380 390 400 410 420 430 440 450 460
P P P P P * P
I/I0
Residue number
Figure 7. Intensities of the peaks in MAP2c HNCO spectra upon addition of 14-3-3␨ compared with intensities of the peaks in free MAP2c with the
MAP2c:14-3-3␨ monomer ratios of 1:0.125 (blue), 1:0.25 (cyan), 1:0.5 (green), 1:1 (orange), and 1:2 (red). A, unphosphorylated MAP2c; B, phosphorylated
MAP2c. The empty boxes and arrows above the graphs indicate propensity to form helical and extended structures, respectively, according to the data
presented in Fig. 5. Letter P and asterisks (above the plot), indicate proline residues and major phosphorylation sites, respectively.
0
10
20
0 0.02 0.2 2 20
Tubulininpellet/%
14-3-3ζ monomer concentration / μ M
Figure 8. Results of the tubulin-MAP2c co-sedimentation assay. The relative
amount of tubulin in the pellet (after substraction of the background of
spontaneous tubulin polymerization) with increasing concentrations of
14-3-3␨ in the presence of unphosphorylated MAP2c and phosphorylated
MAP2c is shown as white and black boxes, respectively. The error bars correspond
to the standard deviations calculated from triplicate measurements.
Only the effect of 20 ␮M 14-3-3␨ is statistically significant at ␣ ϭ 0.05.
Phosphorylation of MAP2c and interaction with 14-3-3
J. Biol. Chem. (2017) 292(16) 6715–6727 6721
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
The proline-rich domain of MAP2c contains another class II
phosphorylation site, Thr-220, in a region that has little homology
with Tau (Fig. 3). The SSP calculated from the chemical
shifts indicates that phosphorylation by PKA stabilizes the secondary
structure in the vicinity of Thr-220 (Fig. 5). This site has
been studied extensively by Alexa et al. (29), who found that
pThr-220 is extraordinarily sensitive to phosphatases and that
MAP2c phosphorylated at Ser-184 and Ser-435, but not at Thr-
220, stabilizes microtubules significantly more strongly than
the other combinations of phosphorylation.
The remaining phosphorylation sites are located in a C-terminal
region with a high sequence homology between MAP2c
and Tau. The site with the second highest phosphorylation rate
(class II) in Tau is 321
KCGpS324
in MTBR3 (Fig. 3), representing
the typical KXGS motif responsible for microtubule binding
(16, 17). Ozer and Halpain (28) detected a rapid PKA phosphorylation
of the corresponding residue Ser-350 (and Ser-319 and
Ser-382 in other KXGS motifs) in MAP2c and found that the
mutation of serines in the KXGS motifs of MAP2c has a great
impact on interactions with microtubules. However, our quantitative
data clearly show that the degree of phosphorylation at
Ser-350 is low under our conditions. On the contrary, PKA in
our study preferentially phosphorylated Ser-435 in the C-terminal
region of MAP2c, corresponding to the muscarinic
receptor-binding site of Tau (Fig. 3). The rate of Ser-435 phosphorylation
was comparable with that of Ser-214 of Tau (class
I). Despite the high sequence homology, the corresponding serine
(Ser-409) is only a minor phosphorylation site (class III,
indicated by the green background in Fig. 3) of Tau.
The listed striking differences between phosphorylation patterns
and kinetics of MAP2c and Tau, determined by the same
experimental approach (real-time NMR spectroscopy), are surprising
considering the high sequence homology (especially in
MTBDs) and have an important implication for 14-3-3␨ binding.
Our interpretation is that these highly homologous regions
have different local structures within MAP2c and Tau with a
direct consequence: differing accessibility for PKA. This explanation
is in agreement with the current view of IDPs as proteins
that exhibit structural features significantly different from a
random coil organization.
A recent study by Joo et al. (39) investigating the interaction
of phosphorylated Tau with 14-3-3␴ allowed us to make a
direct comparison of the interactions of 14-3-3 with MAP2c
and Tau. Unphosphorylated forms of MAP2c and Tau bind the
14-3-3 proteins, but phosphorylation by PKA significantly
increases the binding affinities. The phosphorylation-independent
14-3-3␨-binding sites of MAP2c are located in the MTBD
and in the C-terminal domain. The same regions have been
shown to bind 14-3-3␨ in unphosphorylated Tau (24, 25). Residue-specific
data for unphosphorylated Tau are not reported
by Joo et al. (39), but their analysis of the set of all residues of
phosphorylated Tau influenced by 14-3-3␴ binding leads to the
same conclusion.
Phosphorylated forms of both MAP2c and Tau seem to
interact with 14-3-3 proteins predominantly via two phosphoserines,
one located in the proline-rich domain (25, 27, 39) and
the other in the C-terminal portion (27, 39). As discussed above,
the phosphorylation sites of MAP2c and Tau differ significantly
even in regions of high sequential homology, which has a direct
impact on phosphorylation-dependent 14-3-3 binding.
Charge distribution along the MAP2c and Tau sequences
provides a physicochemical explanation of the observed differences
in 14-3-3␨ binding. Positively charged regions of Tau are
known to bind to acidic residues of tubulin and polyanions (49,
50), and MAP2c is likely to behave in a similar manner. We
propose that the stretches of positively charged residues in
MAP2c and Tau are also involved in the interaction with electronegative
regions within the highly acidic 14-3-3␨ protein
(pI Ϸ 4.7). To explain the observed differences in the interactions
of MAP2c and Tau, one should search for regions of the
proteins that differ in charge distributions.
Fig. 3 shows that the electrostatic potential in the compared
MAPs is similar in MTBR1 and MTBR4 but very different in
MTBR3. A long stretch of positively charged residues is present
in this region of MAP2c. As PKA did not phosphorylate Ser-350
under the conditions of our study, phosphorylation did not
influence the charge distribution in that region of MAP2c. In
contrast, the corresponding MTBR3 region in Tau is efficiently
phosphorylated at Ser-324 (class II kinetics), which interrupts a
stretch of positively charged residues with a negative patch.
Therefore, the possible electrostatic interaction with acidic
regions of 14-3-3 proteins is likely to be greatly suppressed by
phosphorylation in the case of Tau but is little affected by phosphorylation
in the case of MAP2c. This working hypothesis
is consistent with our finding that 14-3-3␨ competes with
MAP2c-induced tubulin polymerization regardless of MAP2c
phosphorylation (Fig. 8), in contrast to the model of the Tauregulated
microtubule polymerization (51).
Another region with differing electrostatic potential between
MAP2c and Tau is located 30 to 40 residues downstream of
MTBR4. Phosphorylation of Ser-435, representing the major
PKA target in MAP2c (but not in Tau), significantly reduces the
positive charge in the mentioned region of MAP2c. This phosphorylation
also induced the propensity to form a helical structure
in the region Arg-425–Pro-431 (Fig. 5). The specific
phosphorylation of MAP2c Ser-435 and its physicochemical
consequences are particularly interesting because this region is
responsible for the ability of Tau to act as a muscarinic agonist
(52).
The observed phosphorylation patterns and interactions
with 14-3-3 are related to the biological functions of Tau and
MAP2c. Phosphorylation of Tau and MAP2c is controlled by
various neurotransmitter receptors (53–55). PKA plays a key
role in the signaling cascades triggered by receptors activating
adenyl cyclase directly but is also involved, via Ca2ϩ
-stimulated
adenyl cyclases, in pathways employing Ca2ϩ
as a second messenger
(56). PKA phosphorylates Tau and MAP2c directly but
also activates downstream kinases able to phosphorylate residues
not targeted by PKA (57–60). The effects of direct phosphorylation
by PKA addressed in this study differ not only
between Tau and MAP2 isoforms but also between different
activities of the MAPs. PKA moderately decreases the binding
of Tau and MAP2 isoforms to microtubules (29, 41, 47), suppresses
the microtubule-nucleating activity of both Tau and
high-molecular-weight MAP2 (46, 47), and inhibits the microPhosphorylation
of MAP2c and interaction with 14-3-3
6722 J. Biol. Chem. (2017) 292(16) 6715–6727
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
tubule-stabilizing activity of Tau (46), but it does not affect
the microtubule-stabilizing activity of high-molecular-weight
MAP2 (47). Differences in the phosphorylation of Tau and
MAP2c also influence the interactions with 14-3-3 proteins,
which are proposed to regulate interactions with microtubules
and modulate phosphorylation of Tau by various kinases (for
review, see Ref. 51). Moreover, binding to microtubules is in
equilibrium with the interactions of MAPs with other components
of the cytoskeleton. Phosphorylation in MTBRs promotes
MAP2c localization to actin-rich regions of neurons
(28), with a direct impact on neuronal development and plasticity
(19, 61, 62). It is evident that the kinetics of phosphorylation
is an important factor in determining the balance of the
aforementioned effects and contributes to the specificity of the
control of microtubule stability by Tau and MAP2c under different
physiological conditions and in different regions of neurons
(8). Some of the functional differences may be directly
related to the observed distinct kinetics of phosphorylation:
PKA phosphorylation of Ser-214 in Tau inhibits the microtubule-stabilizing
activity of Tau but does not affect the microtubule-stabilizing
activity of MAP2 (46, 47); and the kinetics of
phosphorylation and dephosphorylation of Thr-220 in MAP2c
should be comparable if Thr-220 dephosphorylation plays a
regulatory role (29).
In conclusion, PKA phosphorylates Tau and MAP2c differently
even in highly homologous regions, which is reflected by
the interactions of different sites of Tau and MAP2c with
14-3-3 proteins. The involvement of two proteins responding
differently to the same signal (phosphorylation by PKA) may
represent a branching point in the signaling pathways controlling
microtubule stability and other important events such as
activation of cholinergic receptors.
Experimental procedures
Preparation of recombinant proteins
MAP2c expression and purification was performed as
described previously (1, 30). The protein was then phosphorylated,
or dialyzed, against NMR buffer consisting of 50 mM
MOPS, pH 6.9, 150 mM NaCl, and 0.7 mM TCEP. MAP2c was
expressed in M9 medium containing 15
N NH4Cl and/or 13
C
glucose for NMR measurement.
MAP2c was phosphorylated at 30 °C with 650 units of the
catalytic subunit of PKA (New England Biolabs)/mg of MAP2c
in a buffer containing 50 mM Tris-HCl, 10 mM MgCl2, 0.1 mM
EDTA, 2 mM DTT, pH 7.5, and 20 mM ATP. After 24 h of
incubation, PKA was deactivated by heating to 95 °
C for 20 min.
For NMR measurement, the protein was dialyzed against the
NMR buffer.
14-3-3␨ was purified as described previously (35). Two surface-exposed
cysteines were mutated for alanine (C25A and
C189A). This allowed us to work with highly concentrated samples
over longer times without the risk of disulfide bond formation.
We showed earlier that the mutation of these two exposed
cysteines does not change the fold or stability of the protein
(35). The purity of the final protein samples, including phosphorylated
and labeled samples, was checked by MALDI-MS.
Mutagenesis
Single-point mutants of MAP2c (S435D, S184D, T220E, and
C348S) were produced using the QuikChange Lightning sitedirected
mutagenesis kit (Agilent Technologies, Santa Clara,
CA) following the manufacturer’s protocol and using MAP2c in
the pET3a vector as a template. To prepare the double mutants
(S189D/S435D, S199D/S435D, S319D/S435D, T320E/S435D,
S342D/S435D, S350D/S435D, S367D/S435D, and S382D/
S435D), S435D MAP2c in pET3a was used as a template, and
C348S MAP2c was used as template for the double mutant
E52C/C348S. The results of the mutations were confirmed by
sequencing. For NMR measurement, the MAP2c mutants were
expressed in 1 liter of 15
N M9 medium. Phosphorylation was
performed as for the wild-type MAP2c.
Mass spectrometry
MAP2c at 0.2 mM concentration was phosphorylated for 24 h
with 650 units of PKA/mg of MAP2c. Phosphorylated MAP2c
and 0.16 mM unphosphorylated MAP2c were processed by the
filter-aided sample preparation (FASP) method (63, 64). Proteins
were alkylated and digested by trypsin on the filter unit
membrane, and the resulting peptides were eluted by ammonium
bicarbonate. One-tenth of the peptide mixture was analyzed
directly, and the rest of the sample was used for phosphopeptide
enrichment. Both peptide mixtures were separately
analyzed on a LC-MS/MS system (RSLCnano connected to
Orbitrap Elite, Thermo Fisher Scientific).
The MS data were acquired using a data-dependent strategy
by selecting up to the top six precursors based on precursor
abundance in the survey scan (350–2000 m/z). High-resolution
HCD or ETD MS/MS spectra were acquired in the Orbitrap
analyzer. The analysis of the mass spectrometric RAW data files
was carried out using Proteome Discoverer software (version
1.4, Thermo Fisher Scientific) with an in-house Mascot (version
2.4.1, Matrix Science) search engine utilization. Peptides
with a false discovery rate lower than 1%, rank 1, and search
engine rank 1 and with at least 6 amino acids were considered.
Quantitative information assessment was performed using
Skyline (Skyline Software Systems).
NMR spectroscopy
NMR experiments were performed using a 950-MHz Bruker
Avance III spectrometer equipped with a 1
H-13
C/15
N/D TCI
cryogenic probe head with z axis gradients and a 700-MHz
Bruker Avance III spectrometer equipped with a 1
H/13
C/15
N
TXO cryogenic probe head with z axis gradients. All experiments
were performed at 27 °
C with the temperature calibrated
according to the chemical shift differences of pure methanol
peaks. The indirect dimensions in 3D and 5D experiments were
acquired in a non-uniformly sampled manner. On-grid Poisson
disk sampling with a Gaussian probability distribution (65) was
applied.
The 1
H-15
N HSQC (66, 67) spectrum of wild-type MAP2c
was recorded with spectral widths set to 11,904 Hz in the direct
dimension and to 2500 Hz in the indirect dimension. 2048 and
128 complex points were acquired in the direct and the indirect
dimensions, respectively. The 1
H-15
N HSQC spectra of the
MAP2c mutants were recorded with spectral widths set to
Phosphorylation of MAP2c and interaction with 14-3-3
J. Biol. Chem. (2017) 292(16) 6715–6727 6723
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
17,045 Hz in the direct dimension and to 1000 Hz in the indirect
dimension. 2048 and 256 complex points were acquired in the
direct and the indirect dimensions, respectively. 16 scans with
recycle delay set to 1 s were recorded.
The 1
H-15
N SOFAST-HMQC (68) spectra were recorded
with spectral widths set to 13,297 Hz in the direct dimension
and to 2500 Hz in the indirect dimension. 2048 and 64 complex
points were acquired in the direct and the indirect dimensions,
respectively. 16 scans with the recycle delay set to 228 ms were
recorded in each experiment.
The 3D (CACO)NCACO spectrum (30) was acquired with
spectral widths set to 7042 (acquired dimension (aq)) ϫ 2000
(15
N) ϫ 4000 (13
C␣
) Hz and with maximal evolution times of 46
ms (15
N) and 26 ms (13
C␣
) in the indirectly detected dimensions.
The overall number of 1024 complex points was acquired in the
acquisition dimension, and 3000 hypercomplex points were
randomly distributed over the indirectly detected dimensions.
The 5D CACONCACO spectrum (30) was acquired with spectral
widths set to 7042 (aq) ϫ 4000 (13
C␣
) ϫ 2000 (15
N) ϫ 2000
(13
CЈ) ϫ 4000 (13
C␣
) Hz. The maximal evolution times in the indirectly
detected dimensions were set to 26 ms for the 13
C␣
dimensions,
46 ms for the 15
N dimension, and 28 ms for the 13
CЈ dimension.
The overall number of 1024 complex points was acquired in
the acquisition dimension, and 3000 hypercomplex points were
distributed over the indirectly detected dimensions.
The 3D (H)CANCO spectrum (69) was acquired with spectral
widths set to 7042 (aq) ϫ 2000 (15
N) ϫ 4000 (13
C␣
) Hz and
with maximal evolution times of 32 ms (15
N) and 26 ms (13
C␣
)
in the indirectly detected dimensions. The overall number of
1024 complex points was acquired in the acquisition dimension,
and 1500 hypercomplex points were distributed over the
indirectly detected dimensions.
The 5D HC(CC-TOCSY)CACON (30) experiment was measured
with the spectral widths set to 7042 (aq) ϫ 2500 (15
N) ϫ
4000 (13
C␣
) ϫ 12500 (13
Cali
) ϫ 5000 (1
Hali
) Hz where “ali” indicates
nuclei in the aliphatic side chain. The maximal evolution
times in the indirectly detected dimensions were set to 46 ms
for the 15
N dimension, 26 ms for the 13
C␣
dimension, 8 ms for
the 13
Cali
dimension, and 10 ms for the 1
Hali
dimension. The
overall number of 1024 complex points was acquired in the
acquisition dimension, and 2000 hypercomplex points were
distributed over the indirectly detected dimensions. The 3D
(HC(CC-TOCSY))CACON (30) experiment was acquired
with spectral widths set to 7042 (aq) ϫ 1956 (15
N) ϫ 4000
(13
C␣
) Hz and maximal evolution times of 46 ms for 15
N and 26
ms for 13
C␣
indirectly detected dimensions. The overall number
of 1024 complex points was acquired in the acquisition
dimension, and 2000 hypercomplex points were distributed
over the indirectly detected dimensions.
The 3D HNCO (70) spectra were acquired with spectral
widths set to 18939 (aq) ϫ 2000 (15
N) ϫ 2000 (13
CЈ) Hz and
maximal evolution times of 120 ms for 15
N and 80 ms for 13
CЈ
indirectly detected dimensions. The overall number of 2048
complex points was acquired in the acquisition dimension, and
2000 hypercomplex points were distributed over the indirectly
detected dimensions.
The 3D HNCACB experimental data (71) were acquired with
spectral widths set to 18939 (aq) ϫ 2500 (15
N) ϫ 17921 (13
CЈ)
Hz. The number of recorded complex points was 2048, 40, and
128 for the 1
H, 15
N, and 13
C dimensions, respectively.
Uniformly sampled data processing and direct dimension pro-
cessingofnon-uniformlysampleddataweredoneusingNMRPipe
software (72). Multidimensional Fourier transform with iterative
algorithm for artifact suppression (73) was employed to process
indirect dimensions in three-dimensional experiments. Indirect
dimensionsinfive-dimensionalexperimentswereprocessedusing
thesparsemultidimensionalFouriertransform(74).Spectralanalysis
was done using the software Sparky 3.115 (T. D. Goddard and
D. G. Kneller, University of California, San Francisco).
The transient secondary structure propensities of phosphorylated
and unphosphorylated MAP2c were calculated using the
program SSP (75) as described previously (30). To obtain values
not biased by the direct chemical effect of the presence of the
phosphate group, SSP values were calculated only from the
chemical shifts of 13
C␣
, 13
C␤
, and 1
H␣
, which are insensitive to
the presence of the phosphate group attached to neighboring
residues, with the exception of 13
C␤
of the phosphorylated side
chain (34). Residues pSer-184, pThr-220, and pSer-435 were
excluded from the analysis. Regions of propensity to form helical
structures were defined as stretches of at least three residues
with SSP higher than 0.07 (interrupted by no more than
two residues with SSP lower than 0.07). Regions of propensity
to form extended structures were defined as stretches of at least
three residues with SSP lower than Ϫ0.07 (interrupted by no
more than two residues with SSP higher than Ϫ0.07).
Phosphorylation kinetics
Real-time phosphorylation was followed by recording 1
H,
15
N-SOFAST-HMQC spectra at 27 °
C in 5-min intervals for
30 h after the addition of PKA. [15
N]MAP2c at a 0.65 mM concentration
in 50 mM MOPS, pH 6.9, 150 mM NaCl, 0.7 mM
TCEP, 10 mM MgCl2, 20 mM ATP, and 0.1 mM EDTA was
mixed with 12,500 units of PKA directly in the NMR tube.
According to the manufacturer, the PKA concentration was
ϳ0.1 ␮M. The first SOFAST-HMQC experiment was started 2
min after the addition of PKA. The relative concentrations of
individual phosphorylation sites were assumed to be proportional
to the corresponding peak volumes, and PKA was
assumed to be saturated by ATP and magnesium. Phosphorylation
rates were modeled as
Ϫ
d͓S͔i
dt
ϭ
d͓P͔i
dt
ϭ
kcat,icPKA͓S͔i/Km,i
1 ϩ ⌺i͓S͔i/Km,i
(Eq. 1)
where t is time, [S]i and [P]i are concentrations of the i-th
unphosphorylated and phosphorylated phosphorylation sites,
respectively, kcat,i and Km,i are catalytic and Michaelis constants,
respectively, and cPKA is the total concentration of PKA.
The apparent rate constants, kobs,i, were defined as
kobs,i ϭ
kcat,i/Km,i
1 ϩ ⌺icMAP2c/Km,i
(Eq. 2)
where cMAP2c is the total concentration of MAP2c.
The kobs value for Ser-435 was estimated by fitting the peak
volumes to the integrated form of Equation 1, assuming that
concentrations of other phosphorylation sites were equal to
Phosphorylation of MAP2c and interaction with 14-3-3
6724 J. Biol. Chem. (2017) 292(16) 6715–6727
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
cMAP2c. The value of kcat could not be reliably separated from
kobs for the obtained data. To estimate kobs for Ser-184, Ser-189,
and Thr-220, the ratios of peak volumes of pSer-184, pSer-189,
and pThr-220 were assumed to be constant, and peak volumes,
V, were fitted to the equation
V ϭ V͑0͒ͩ1 Ϫ
a exp͑Ϫbt͒ Ϫ b exp͑Ϫat͒
a Ϫ b ͪ (Eq. 3)
where a Ͻ b, a ϭ kobscPKA at [Ser-435] ϭ 0, and b approximates
the rate of the preceding phosphorylation of Ser-435.
Interaction with 14-3-3␨ by NMR
A sample containing 550 ␮l of 0.69 mM unphosphorylated
[15
N,13
C]MAP2c was titrated with 17, 34, 69, 139, and 280 ␮l
of 2.5 mM 14-3-3␨ (monomer concentration). After each
addition, a 3D HNCO spectrum was recorded. 500 ␮l of 0.61
mM phosphorylated [15
N,13
C]MAP2c was titrated with 17,
34, 68, 136, and 272 ␮l of 2.2 mM 14-3-3␨. For each titration
point, 3D HNCO and 2D 1
H,15
N HSQC spectra were
measured.
Microscale thermophoresis
Binding experiments between 14-3-3␨ and MAP2c in the
unphosphorylated and phosphorylated forms were measured
by MST at 20 °C. E52C/C348S MAP2c was specifically labeled
at position Cys-52 with the fluorophore Alexa Fluor 647 C2
maleimide (AF647, Thermo Fisher Scientific). Titration experiments
were performed in buffer containing 50 mM Tris, pH 7.5.
In addition, 0.5 mg/ml BSA and 0.05% Tween 20 were added to
prevent aggregation of the studied proteins on standard capillary
walls. Binding studies were performed at 30% laser power
with a Monolith NT.115 device (NanoTemper Technologies,
Munich, Germany) in combination with three different MST
power setups (at 40, 60, and 80%). The data were fitted to the
following model,
⌽ ϭ ⌽0 ϩ ⌬
x ϩ nc ϩ KDЈ Ϫ ͱ͑x ϩ nc ϩ KDЈ͒2
Ϫ 4ncx
2n
(Eq. 4)
where ⌽ ϭ F/Fmax is the normalized fluorescence, c is the total
concentration of MAP2c, x is the total monomer concentration
of 14-3-3␨, and KЈD is the apparent dissociation constant
expressed for the 14-3-3␨ monomer concentration.
Microtubular co-sedimentation assay
Co-sedimentation assays were adapted from Valencia et al.
(76). Samples of 5 ␮M porcine brain tubulin in 30 ␮l of tubulin
buffer (80 mM PIPES, pH 6.9, 2 mM MgCl2, and 0.5 mM EGTA)
were polymerized by the addition of 2 ␮M MAP2c and incubated
for 15 min at 37 °
C. 20 nM, 200 nM, 2 ␮M, and 20 ␮M
concentrations of 14-3-3␨ (monomer concentrations) in tubulin
buffer were added to the MAP2c-tubulin samples before the
incubation. Microtubule bundles were pelleted by centrifugation
(50,000 ϫ g, 30 min, 37 °
C). The pellet was washed by tubulin
buffer and adjusted to the same volume as the supernatant
with the buffer. Both supernatant and pellet were mixed with
the SDS gel loading buffer. Equal volumes of supernatant and
pellet were separated by SDS-PAGE. The Coomassie Brilliant
Blue-stained protein bands were analyzed using the densitometer
software QuantiScan 3.0.
Author contributions—S. J., L. Z., J. H., J. N., and K. M. conceived
and designed the research. S. J., E. Z., K. M., and Z. T. prepared the
protein samples. Z. T. and K. M., respectively, performed the MST
and co-sedimentation assays. L. Z., M. Z., S. J., and K. M. performed
the NMR experiments, and K. H. and Z. Z. performed the MS experiments.
S. J., L. Z., and J. H. wrote the paper. All authors reviewed the
results and approved the final version of the manuscript.
Acknowledgments—CIISB research infrastructure project LM2015043,
funded by Ministry of Education, Youth and Sports of the Czech
Republic (MEYS CR), is gratefully acknowledged for partial financial
support of the measurements at the Josef Dadok National NMR Centre,
Biomolecular Interactions and Crystallization and Proteomics
Core Facilities, Central European Institute of Technology (CEITEC),
Masaryk University. We thank Tanvir Shaikh for language corrections
and Arnosˇt Mla´dek and Irmgard Fischer for technical help.
References
1. Gamblin, T. C., Nachmanoff, K., Halpain, S., and Williams, R. C. (1996)
Recombinant microtubule-associated protein 2c reduces the dynamic instability
of individual microtubules. Biochemistry 35, 12576–12586
2. Jalava, N. S., Lopez-Picon, F. R., Kukko-Lukjanov, T. K., and Holopainen,
I. E. (2007) Changes in microtubule-associated protein-2 (MAP2) expression
during development and after status epilepticus in the immature rat
hippocampus. Int. J. Dev. Neurosci. 25, 121–131
3. Dunker, A. K., Obradovic, Z., Romero, P., Garner, E. C., and Brown, C. J.
(2000) Intrinsic protein disorder in complete genomes. Genome Inform.
11, 161–171
4. Dunker, A. K., Oldfield, C. J., Meng, J., Romero, P., Yang, J. Y., Chen, J. W.,
Vacic, V., Obradovic, Z., and Uversky, V. N. (2008) The unfoldomics decade:
an update on intrinsically disordered proteins. BMC Genomics 9, S1
5. Dyson, H. J., and Wright, P. E. (2005) Intrinsically unstructured proteins
and their functions. Nat. Rev. Mol. Cell Biol. 6, 197–208
6. Tompa, P. (2005) The interplay between structure and function in intrinsically
unstructured proteins. FEBS Lett. 579, 3346–3354
7. Fink, A. L. (2005) Natively unfolded proteins. Curr. Opin. Struct. Biol. 15,
35–41
8. Dehmelt, L., and Halpain, S. (2005) The MAP2/Tau family of microtubule-associated
proteins. Genome Biol. 6, 204
9. Su¨ndermann, F., Fernandez, M. P., and Morgan, R. O. (2016) An evolutionary
roadmap to the microtubule-associated protein MAP Tau. BMC
Genomics 17, 264
10. Grundke-Iqbal, I., Iqbal, K., Quinlan, M., Tung, Y. C., Zaidi, M. S., and
Wisniewski, H. M. (1986) Microtubule-associated protein Tau: a component
of Alzheimer paired helical filaments. J. Biol. Chem. 261, 6084–6089
11. Viereck, C., Tucker, R. P., and Matus, A. (1989) The adult rat olfactory
system expresses microtubule-associated proteins found in the developing
brain. J. Neurosci. 9, 3547–3557
12. Vallee, R. (1980) Structure and phosphorylation of microtubule-associated
protein 2 (MAP2). Proc. Natl. Acad. Sci. U.S.A. 77, 3206–3210
13. Yamauchi, T., and Fujisawa, H. (1983) Disassembly of microtubules by the
action of calmodulin-dependent protein kinase (kinase II) which occurs
only in the brain tissues. Biochem. Biophys. Res. Commun. 110, 287–291
14. Burns, R. G., Islam, K., and Chapman, R. (1984) The multiple phosphorylation
of the microtubule-associated protein MAP2 controls the MAP2:
tubulin interaction. Eur. J. Biochem. 141, 609–615
15. Hoshi, M., Akiyama, T., Shinohara, Y., Miyata, Y., Ogawara, H., Nishida,
E., and Sakai, H. (1988) Protein-kinase-C-catalyzed phosphorylation of
the microtubule-binding domain of microtubule-associated protein 2 inPhosphorylation
of MAP2c and interaction with 14-3-3
J. Biol. Chem. (2017) 292(16) 6715–6727 6725
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
hibits its ability to induce tubulin polymerization. Eur. J. Biochem. 174,
225–230
16. Ainsztein, A. M., and Purich, D. L. (1994) Stimulation of tubulin polymerization
by MAP-2: control by protein kinase C-mediated phosphorylation
at specific sites in the microtubule-binding region. J. Biol. Chem. 269,
28465–28471
17. Illenberger, S., Drewes, G., Trinczek, B., Biernat, J., Meyer, H. E., Olmsted,
J. B., Mandelkow, E. M., and Mandelkow, E. (1996) Phosphorylation of
microtubule-associated proteins MAP2 and MAP4 by the protein kinase
p110mark: phosphorylation sites and regulation of microtubule dynamics.
J. Biol. Chem. 271, 10834–10843
18. Drewes, G., Ebneth, A., and Mandelkow, E. M. (1998) MAPs, MARKs and
microtubule dynamics. Trends Biochem. Sci. 23, 307–311
19. Sa´nchez, C., Díaz-Nido, J., and Avila, J. (2000) Phosphorylation of microtubule-associated
protein 2 (MAP2) and its relevance for the regulation of
the neuronal cytoskeleton function. Prog. Neurobiol. 61, 133–168
20. Illenberger, S., Zheng-Fischho¨fer, Q., Preuss, U., Stamer, K., Baumann, K.,
Trinczek, B., Biernat, J., Godemann, R., Mandelkow, E. M., and Mandelkow,
E. (1998) The endogenous and cell cycle-dependent phosphorylation
of Tau protein in living cells: implications for Alzheimer’s disease.
Mol. Biol. Cell 9, 1495–1512
21. Avila, J., Domínguez, J., and Díaz-Nido, J. (1994) Regulation of microtubule
dynamics by microtubule-associated protein expression and phosphorylation
during neuronal development. Int. J. Dev. Biol. 38, 13–25
22. Aitken, A., Collinge, D. B., van Heusden, B. P., Isobe, T., Roseboom, P. H.,
Rosenfeld, G., and Soll, J. (1992) 14-3-3 proteins: a highly conserved, widespread
family of eukaryotic proteins. Trends Biochem. Sci. 17, 498–501
23. Skoulakis, E. M., and Davis, R. L. (1998) 14-3-3 proteins in neuronal development
and function. Mol. Neurobiol. 16, 269–284
24. Hashiguchi, M., Sobue, K., and Paudel, H. K. (2000) 14-3-3␨ is an effector
of Tau protein phosphorylation. J. Biol. Chem. 275, 25247–25254
25. Sadik, G., Tanaka, T., Kato, K., Yamamori, H., Nessa, B. N., Morihara, T.,
and Takeda, M. (2009) Phosphorylation of Tau at Ser214 mediates its
interaction with 14-3-3 protein: implications for the mechanism of tau
aggregation. J. Neurochem. 108, 33–43
26. Sluchanko, N. N., Seit-Nebi, A. S., and Gusev, N. B. (2009) Effect of phosphorylation
on interaction of human Tau protein with 14-3-3␨. Biochem.
Biophys. Res. Commun. 379, 990–994
27. Sluchanko, N. N., Seit-Nebi, A. S., and Gusev, N. B. (2009) Phosphorylation
of more than one site is required for tight interaction of human Tau
protein with 14-3-3␨. FEBS Lett. 583, 2739–2742
28. Ozer, R. S., and Halpain, S. (2000) Phosphorylation-dependent localization
of microtubule-associated protein MAP2c to the actin cytoskeleton.
Mol. Biol. Cell 11, 3573–3587
29. Alexa, A., Schmidt, G., Tompa, P., Ogueta, S., Va´zquez, J., Kulcsa´r, P.,
Kova´cs, J., Dombra´di, V., and Friedrich, P. (2002) The phosphorylation
state of threonine-220, a uniquely phosphatase-sensitive protein kinase A
site in microtubule-associated protein MAP2c, regulates microtubule
binding and stability. Biochemistry 41, 12427–12435
30. Novácˇek, J., Janda, L., Dopitova´, R., Žídek, L., Sklenárˇ, V. (2013) Efficient
protocol for backbone and side-chain assignments of large, intrinsically
disordered proteins: transient secondary structure analysis of 49.2-kDa
microtubule-associated protein 2c. J. Biomol. NMR 56, 291–301
31. Obenauer, J. C., Cantley, L. C., and Yaffe, M. (2003) Scansite 2.0: proteome-wide
prediction of cell signaling interactions using short sequence
motifs, Nucleic Acids Res. 31, 3635–3641
32. Xue, Y., Liu, Z., Cao, J., Ma, Q., Gao, X., Wang, Q., Jin, C., Zhou, Y., Wen,
L., and Ren, J. (2011) GPS 2.1: enhanced prediction of kinase-specific
phosphorylation sites with an algorithm of motif length selection. Protein
Eng. Des. Sel. 24, 255–260
33. Wong, Y. H., Lee, T. Y., Liang, H. K., Huang, C., Yang, Y. H., Chu, C. H.,
Huang, H. D., Ko, M. T., and Hwang, J. K. (2007) KinasePhos 2.0: a Web
server for identifying protein kinase-specific phosphorylation sites based
on sequences and coupling patterns. Nucleic Acids Res. 35, 588–594
34. Lousˇa, P., Nedozra´lova´, H., Župa, E., Novácˇek, J., and Hritz, J. (2017) Phosphorylation
of the regulatory domain of human tyrosine hydroxylase 1
monitored using non-uniformly sampled NMR. Biophys. Chem. 223,
25–29
35. Hritz, J., Byeon, I. J., Krzysiak, T., Martinez, A., Sklenar, V., and Gronenborn,
A. M. (2014) Dissection of binding between a phosphorylated tyrosine
hydroxylase peptide and 14-3-3␨: a complex story elucidated by
NMR. Biophys. J. 107, 2185–2194
36. von Bergen, M., Friedhoff, P., Biernat, J., Heberle, J., Mandelkow, E. M.,
and Mandelkow, E. (2000) Assembly of Tau protein into Alzheimer paired
helical filaments depends on a local sequence motif ((306)VQIVYK(311))
forming beta structure. Proc. Natl. Acad. Sci. U.S.A. 97, 5129–5134
37. Xie, C., Miyasaka, T., Yoshimura, S., Hatsuta, H., Yoshina, S., Kage-Nakadai,
E., Mitani, S., Murayama, S., and Ihara, Y. (2014) The homologous
carboxyl-terminal domains of microtubule-associated protein 2 and Tau
induce neuronal dysfunction and have differential fates in the evolution of
neurofibrillary tangles. PLoS One 9, e89796
38. Mukrasch, M. D., Bibow, S., Korukottu, J., Jeganathan, S., Biernat, J.,
Griesinger, C., Mandelkow, E., and Zweckstetter, M. (2009) Structural
polymorphism of 441-residue Tau at single residue resolution. PLOS Biol.
7, e34
39. Joo, Y., Schumacher, B., Landrieu, I., Bartel, M., Smet-Nocca, C., Jang, A.,
Choi, H. S., Jeon, N. L., Chang, K. A., Kim, H. S., Ottmann, C., and Suh,
Y. H. (2015) Involvement of 14-3-3 in tubulin instability and impaired
axon development is mediated by Tau. FASEB J. 29, 4133–4144
40. Sluchanko, N. N., and Gusev, N. B. (2010) 14–3-3 proteins and regulation
of cytoskeleton. Biochemistry (Mosc.) 75, 1528–1546
41. Ackmann, M., Wiech, H., and Mandelkow, E. (2000) Nonsaturable binding
indicates clustering of Tau on the microtubule surface in a paired
helical filament-like conformation. J. Biol. Chem. 275, 30335–30343
42. Kennelly, P. J., and Krebs, E. G. (1991) Consensus sequences as substrate
specificity determinants for protein kinases and protein phosphatases.
J. Biol. Chem. 266, 15555–15558
43. Landrieu, I., Lacosse, L., Leroy, A., Wieruszeski, J. M., Trivelli, X., Sillen,
A., Sibille, N., Schwalbe, H., Saxena, K., Langer, T., and Lippens, G. (2006)
NMR analysis of a Tau phosphorylation pattern. J. Am. Chem. Soc. 128,
3575–3583
44. Sillen, A., Barbier, P., Landrieu, I., Lefebvre, S., Wieruszeski, J. M., Leroy,
A., Peyrot, V., and Lippens, G. (2007) NMR investigation of the interaction
between the neuronal protein Tau and the microtubules. Biochemistry 46,
3055–3064
45. Yoshida, H., and Goedert, M. (2006) Sequential phosphorylation of tau
protein by cAMP-dependent protein kinase and SAPK4/p38delta or JNK2
in the presence of heparin generates the AT100 epitope. J. Neurochem. 99,
154–164
46. Brandt, R., Lee, G., Teplow, D. B., Shalloway, D., and Abdel-Ghany, M.
(1994) Differential effect of phosphorylation and substrate modulation on
Tau’s ability to promote microtubule growth and nucleation. J. Biol. Chem.
269, 11776–11782
47. Itoh, T. J., Hisanaga, S., Hosoi, T., Kishimoto, T., and Hotani, H. (1997)
Phosphorylation states of microtubule-associated protein 2 (MAP2) determine
the regulatory role of MAP2 in microtubule dynamics. Biochemistry
36, 12574–12582
48. Madeira, F., Tinti, M., Murugesan, G., Berrett, E., Stafford, M., Toth, R.,
Cole, C., MacKintosh, C., and Barton, G. J. (2015) 14-3-3-Pred: improved
methods to predict 14–3-3-binding phosphopeptides. Bioinforma 31,
2276–2283
49. Mukrasch, M. D., Biernat, J., von Bergen, M., Griesinger, C., Mandelkow,
E., and Zweckstetter, M. (2005) Sites of Tau important for aggregation
populate ␤-structure and bind to microtubules and polyanions. J. Biol.
Chem. 280, 24978–24986
50. Mukrasch, M. D., von Bergen, M., Biernat, J., Fischer, D., Griesinger, C.,
Mandelkow, E., and Zweckstetter, M. (2007) The “jaws” of the Tau-microtubule
interaction. J. Biol. Chem. 282, 12230–12239
51. Sluchanko, N. N., and Gusev, N. B. (2011) Probable participation of
14–3-3 in Tau protein oligomerization and aggregation. J. Alzheimers Dis.
27, 467–476
52. Go´mez-Ramos, A., Díaz-Herna´ndez, M., Rubio, A., Miras-Portugal, M. T.,
and Avila, J. (2008) Extracellular Tau promotes intracellular calcium increase
through M1 and M3 muscarinic receptors in neuronal cells. Mol.
Cell. Neurosci. 37, 673–681
Phosphorylation of MAP2c and interaction with 14-3-3
6726 J. Biol. Chem. (2017) 292(16) 6715–6727
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
53. Gardiner, J., Overall, R., and Marc, J. (2011) The microtubule cytoskeleton
acts as a key downstream effector of neurotransmitter signaling. Synapse
65, 249–256
54. Ovsepian, S. V., O’Leary, V. B., and Zaborszky, L. (2016) Cholinergic
mechanisms in the cerebral cortex: beyond synaptic transmission. Neuroscientist
22, 238–251
55. Busceti, C. L., Di Pietro, P., Riozzi, B., Traficante, A., Biagioni, F., Nistico`,
R., Fornai, F., Battaglia, G., Nicoletti, F., and Bruno, V. (2015) 5-HT(2C)
serotonin receptor blockade prevents tau protein hyperphosphorylation
and corrects the defect in hippocampal synaptic plasticity caused by a
combination of environmental stressors in mice. Pharmacol. Res. 99,
258–268
56. Wang, H., and Zhang, M. (2012) The role of Ca2ϩ
-stimulated adenylyl
cyclases in bidirectional synaptic plasticity and brain function. Rev. Neurosci.
23, 67–78
57. Vossler, M. R., Yao, H., York, R. D., Pan, M. G., Rim, C. S., and Stork, P. J.
(1997) cAMP activates MAP kinase and Elk-1 through a B-Raf- and Rap1dependent
pathway. Cell 89, 73–82
58. Kim, H. A., DeClue, J. E., and Ratner, N. (1997) cAMP-dependent protein
kinaseAisrequiredforSchwanncellgrowth:interactionsbetweenthecAMP
and neuregulin/tyrosine kinase pathways. J. Neurosci. Res. 49, 236–247
59. Blanco-Aparicio, C., Torres, J., and Pulido, R. (1999) A novel regulatory
mechanism of MAP kinases activation and nuclear translocation mediated
by PKA and the PTP-SL tyrosine phosphatase. J. Cell Biol. 147,
1129–1136
60. Ambrosini, A., Tininini, S., Barassi, A., Racagni, G., Sturani, E., and Zippel,
R. (2000) cAMP cascade leads to Ras activation in cortical neurons. Brain
Res. Mol. Brain Res. 75, 54–60
61. Mohan, R., and John, A. (2015) Microtubule-associated proteins as direct
crosslinkers of actin filaments and microtubules. IUBMB Life 67, 395–403
62. Elie, A., Prezel, E., Gue´rin, C., Denarier, E., Ramirez-Rios, S., Serre, L.,
Andrieux, A., Fourest-Lieuvin, A., Blanchoin, L., and Arnal, I. (2015) Tau
co-organizes dynamic microtubule and actin networks. Sci. Rep. 5, 9964
63. Wis´niewski, J. R., Ostasiewicz, P., and Mann, M. (2011) High recovery
FASP applied to the proteomic analysis of microdissected formalin fixed
paraffin embedded cancer tissues retrieves known colon cancer markers. J.
Proteome Res. 10, 3040–3049
64. Wis´niewski, J. R., Zougman, A., Nagaraj, N., and Mann, M. (2009) Universal
sample preparation method for proteome analysis. Nat. Methods 6,
359–362
65. Kazimierczuk, K., Zawadzka, A., and Koz´min´ski, W. (2008) Optimization
of random time domain sampling in multidimensional NMR. J. Magn.
Reson. 192, 123–130
66. Bodenhausen, G., and Ruben, D. J. (1980) Natural abundance nitrogen-15
NMR by enhanced heteronuclear spectroscopy. Chem. Phys. Lett. 69,
185–189
67. Sklenar, V., Piotto, M., Leppik, R., and Saudek, V. (1993) Gradient-tailored
water suppression for 1
H-15
N HSQC experiments optimized to retain full
sensitivity. J. Magn. Reson. 102, 241–245
68. Schanda, P., and Brutscher, B. (2005) Very fast two-dimensional NMR
spectroscopy for real-time investigation of dynamic events in proteins on
the time scale of second. J. Am. Chem. Soc. 127, 8014–8015
69. Bermel, W., Bertini, I., Felli, I. C., and Pierattelli, R. (2009) Speeding up
(13)C direct detection biomolecular NMR spectroscopy. J. Am. Chem. Soc.
131, 15339–15345
70. Kay, L. E., Ikura, M., Tschudin, R., and Bax, A. (1990) Three-dimensional
triple-resonance NMR spectroscopy of isotopically enriched proteins. J.
Magn. Reson. 89, 496–514
71. Sattler, M., Schleucher, J., and Griesinger, C. (1999) Heteronuclear multidimensional
NMR experiments for the structure determination of proteins
in solution employing pulsed field gradients. Prog. Nucleic Magn.
Reson. Spectrosc. 34, 93–158
72. Delaglio, F., Grzesiek, S., Vuister, G. W., Zhu, G., Pfeifer, J., and Bax, A.
(1995) NMRPipe: a multidimensional spectral processing system based on
UNIX pipes. J. Biomol. NMR 6, 277–293
73. Stanek, J., and Koz´min´ski, W. (2010) Iterative algorithm of discrete Fourier
transform for processing randomly sampled NMR data sets. J. Biomol.
NMR 47, 65–77
74. Kazimierczuk, K., Zawadzka, A., and Koz´min´ski, W. (2009) Narrow peaks
and high dimensionalities: exploiting the advantages of random sampling.
J. Magn. Reson. 197, 219–228
75. Marsh, J. A., Singh, V. K., Jia, Z., and Forman-Kay, J. D. (2006) Sensitivity
of secondary structure propensities to sequence differences between
␣- and ␥-synuclein: implications for fibrillation. Protein Sci. 15,
2795–2804
76. Valencia, R. G., Walko, G., Janda, L., Novacek, J., Mihailovska, E., Reipert,
S., Andra¨-Marobela, K., and Wiche, G. (2013) Intermediate filament-associated
cytolinker plectin 1c destabilizes microtubules in keratinocytes.
Mol. Biol. Cell 24, 768–784
Phosphorylation of MAP2c and interaction with 14-3-3
J. Biol. Chem. (2017) 292(16) 6715–6727 6727
atMASARYKOVAUNIVERZITAonJanuary10,2018http://www.jbc.org/Downloadedfrom
VOLUME 292 (2017) PAGES 6715–6727
DOI 10.1074/jbc.A116.771097
Quantitative mapping of microtubule-associated protein
2c (MAP2c) phosphorylation and regulatory protein
14-3-3␨-binding sites reveals key differences between
MAP2c and its homolog Tau.
Se´verine Jansen, Katerˇina Melkova´, Zuzana Trosˇanova´, Katerˇina Hana´kova´,
Milan Zachrdla, Jirˇi Nova´cˇek, Erik Zˇupa, Zbyneˇk Zdra´hal, Jozef Hritz,
and Luka´sˇ Zˇídek
PAGE 6720:
There was an error in Fig. 6. Panels A and B were inadvertently
exchanged. The correct legend should read as follows. Microscale thermophoretic
analysis of interaction between 5 ␮M E52C/C348S
MAP2c and 14-3-3␨ in 50 mM Tris buffer. A, phosphorylated MAP2c;
B, unphosphorylated MAP2c. The plots show interactions in the
14-3-3␨ monomer concentration range from 36.6 nM to 1.2 mM. The
mean values Ϯ S.D. for each concentration point were calculated from
triplicate measurements.
The results are described correctly in the text, and the error does not
affect the results or conclusions of the work.
ADDITIONS AND CORRECTIONS
10316 J. Biol. Chem. (2017) 292(24) 10316–10316
© 2017 by The American Society for Biochemistry and Molecular Biology, Inc. Published in the U.S.A.
CONCLUSIONS
65
7 Conclusions
The 15 selected articles of the applicant are included in this habilitation thesis within
four scientific chapters that are preceded by a general theoretical chapter. The theoretical
chapter focuses on the various aspects of MD that are relevant to the selected
scientific topics. In the general theoretical chapter, the applicant also incorporated
some of his observations from the ongoing teaching process and thus this chapter, as
well as the whole thesis, can be useful for students interested in the given methodologies.
The main applicant’s aim went beyond the simple proclamation that biomolecules
and their complexes are dynamic. It is shown here how this fact, when properly implemented
in a variety of computational approaches leads to much more reliable results
in terms of the agreement with experimental data.
The most valuable outcome from the first scientific chapter is the observed large sensitivity
of docking reliability, not only due to conformational changes but also due to
thermal motion within the protein target. This fact was presented and applied in the
docking results for cytochrome P450 2D6 combining MD and molecular docking. The
available crystal structure of cytochrome P450 2D6 was only in the apo form where
the active site was too small for the majority of known substrates, leading to only 20%
reliability of structural binding poses. The described approach allowed efficient incorporation
of induced fit effects when combined with the designed decision tree which
provided a high degree of robustness. The designed decision tree, divides potential
drug-like molecules, based on their size and number of hydrophobic regions, into three
classes. When each class is then docked into the provided three different structures of
cytochrome P450 2D6, the prediction reliability increased to 80% without any additional
computation costs. The speed and simplicity of approaches are particularly important
for the pharmaceutical application and indeed this approach became very popular
in the medicinal chemistry field and was applied to a large variety of protein targets.
The applicant wants to emphasize that while described modifications provide
CONCLUSIONS
66
improved reliability of molecular docking in terms of structural information, they were
still absolutely unsatisfactory in terms of the prediction of binding affinities from docking
simulations dependent on the scoring function.
Binding affinities along with other properties such as conformer populations, protein
stability, solubility, etc. are governed by the enthalpic and entropic components of the
Gibbs free energy differences. This is the reason why free energy is central to physical
chemistry and related thermodynamically oriented computational chemistry. It is
known from statistical physics that free energy is directly related to the states within
statistical ensembles and therefore the capability to sample sufficiently by particular
computational tools is so important. Because the sampling capability of standard MD
is quite small, dozens of different alternative approaches have been developed in the
community over the last decades. The second scientific chapter presents the applicant’s
contributions within the scope of Hamiltonian replica exchange molecular dynamics.
Here two particular modifications of the hamiltonian within the H-REMD
scheme are stressed. The first modification used so-called increased softness of nonbonded
interactions between particular sets of atoms. Its applicability was shown on
GTP and 8-Br-GTP molecules where non-bonded interactions between base and sugar
parts were softened to allow the faster transition between anti and syn conformations.
The resulting sampling enhancement was over three orders of magnitude faster than
regular MD while providing the production of a statistical ensemble of the same quality.
The second modification of Hamiltonian within H-REMD was inspired by umbrella
sampling methods, in which the ligand was pulled from and into the binding site by
using un-directional distance-field distance-restraints. It was applied to the binding/unbinding
phenomena of phosphopeptides to the 14-3-3 protein. In addition to the
preferred binding pathways, the binding affinities were predicted with a reasonable
agreement to the experimental data. On the other hand, it should be also pointed out
that this approach is computationally very demanding because at every distance both
the ligand and protein degrees of freedom need to be sampled sufficiently.
CONCLUSIONS
67
In many applications (e.g. when addressing a set of similar molecules differing by a
functional group) it is sufficient to calculate relative rather than absolute binding affinities.
In such cases, alchemical free energy perturbation approaches can be used as is
described in the third scientific chapter. However, these calculations may suffer a failure
of convergence if molecules contain high intramolecular energy barriers. Here, the
applicant designed an enhanced sampling-one step perturbation (ES-OS) method to
tackle this problem in a highly efficient way. The core idea is to design a very specific
and completely artificial reference state (using two different sets of soft-core nonbonded
interactions). If the reference state is well designed then its single MD run is
sufficient to calculate relative free energies between similar compounds in the studied
set. The practical applicability of the ES-OS method was shown for the set of C8-analogs
of GTP for which conformational probabilities and relative lipophilicity, solubility and
binding affinities with respect to the FtsZ protein target were calculated and their comparison
with the available experimental data shows good agreement despite the very
limited computational costs concerning other computational methods. The wider use
of this method in computational medicinal chemistry is prohibited by the need to design
a specific reference state that is generally far from trivial.
In the fourth scientific chapter focused on structural and interaction properties of IDPs,
the applicant applied primarily solution NMR rather than computational methods. The
reasons for this are manifold. In the case of monitoring kinetic profiles of IDPs phosphorylation,
comparable data obtained by simulations would require an advanced
combination of MD with time-dependent QM methods with still quite questionable outcomes.
On the other hand, NMR spectroscopy allows measuring these data in a few
days when using 15N labeled IDP of interest in sufficient concentration and the kinase
of interest. In the case of binding affinities – computational methods are applicable in
a similar manner to that described in the second and third scientific chapters but the
situation starts to differ when having longer IDP phosphorylated on multiple sites
CONCLUSIONS
68
where each of them is capable of interacting with globular proteins such as 14-3-3 proteins.
In such cases, avidity effects start to be important and very difficult to be incorporated
within the computational predictions. Such situations are also quite challenging
for the proper interpretation of NMR data as the applicant showed for the doubly
phosphorylated fragment (each phosphorylated site is different) of TH interacting with
14-3-3zeta homodimer having two binding sites. Similar studies are very rare in literature
and therefore more data should be produced to have good experimental references
for future computational studies. The last significant issue is the fact that biomolecular
force-field parameters were optimized and tested for globular proteins over
a half-century ago. Recently, it turned out that most of them simply overstabilize protein
structures and when applied to IDPs – artificial collapse tends to be observed. In
our recent publicationP13, we tested several popular biomolecular force-fields and paying
attention to particular parametrizations of water models and when comparing with
a whole range of experimental, mostly NMR data, it turned out that most of them are
inadequate for selected IDP proteins. For only very few combinations did we observe
decent but far from perfect agreement with the experimental data. In addition to more
reliable force-field parameters, there is the obvious challenge for the enhanced sampling
of larger IDP proteins when attempting to simulate them in an explicit water environment.
The applicant with his students and colleagues continues to work on addressing
both of those aspects. In the longer-term perspective, we plan to apply such
improved approaches to studies of conformation changes of the Tau protein from its
native to the pathological fibrils observed in Alzheimer's disease patients.
BIBLIOGRAPHY
69
Bibliography
1. Ponder JW, Case DA. Protein Simulations. Adv Protein Chem. 2003;66:27-85.
http://www.sciencedirect.com/science/article/pii/S006532330366002X
2. Berendsen HJC. Simulating the Physical World: Hierarchical Modeling from
Quantum Mechanics to Fluid Dynamics. Vol 9780521835.; 2007.
doi:10.1017/CBO9780511815348
3. LEACH AR. Molecular Modelling : Principles and Applications. Essex: Longman;
1996.
4. Frenkel, D.; Smit B. Understanding Molecular Simulation: From Algorithms to
Applications. Academic Press, Boston. Academic Press; 2002.
5. Schlick T. Molecular Modeling and Simulation: An Interdisciplinary Guide.
Springer-Verlag, New York; 2002.
6. Zuckerman DM. Statistical Physics of Biomolecules; CRC Press.; 2010.
7. Allison JR. Computational methods for exploring protein conformations.
Biochem Soc Trans. 2020;48(4):1707-1724. doi:10.1042/BST20200193
8. Swendsen, R.H.; Wang JS. Replica Monte Carlo simulation of spin glasses. Phys
Rev Lett. 1986;57:2607–2609.
9. Sugita, Y.; Okamoto Y. Replica-exchange molecular dynamics method for
protein folding. Chem Phys Lett. 1999;314:141–151.
10. Hansmann UHE. Parallel tempering algorithm for conformational studies of
biological molecules. Chem Phys Lett. 1997;281:140-150.
11. Fukunishi, H.; Watanabe, O.; Takada S. On the Hamiltonian replica exchange
method for efficient sampling of biomolecu-lar systems: Application to protein
structure prediction. J Chem Phys. 2002;116:9058.
12. Sugita, Y.; Kitao, A.; Okamoto Y. Multidimensional replica-exchange method for
free-energy calculations. J Chem Phys. 2000;113:6042.
13. Abrams, C.; Bussi G. Enhanced Sampling in Molecular Dynamics Using
Metadynamics, Replica-Exchange, and Temperature-Acceleration. Entropy.
2014;16:163-199.
14. Hritz J, Oostenbrink C. Hamiltonian replica exchange molecular dynamics using
BIBLIOGRAPHY
70
soft-core interactions. J Chem Phys. 2008;128(14):1-10.
doi:10.1063/1.2888998
15. Prakash, M.K., Barducci, A.; Parrinello M. Replica temperatures for uniform
exchange and efficient roundtrip times in explicit solvent parallel tempering
simulations. J Chem Theory Comput. 2011;7:2025–2027.
doi:10.1021/ct200208h
16. Spiwok, V.; Sucur, Z.; Hosek P. Enhanced sampling techniques in biomolecular
simulations. Biotech Adv. 2015;33:1130-1140.
17. Affentranger, R.; Tavernelli, I.; Di Iorio EE. A novel Hamiltonian replica
exchange MD protocol to enhance protein conformational space sampling. J
Chem Theory Comput. 2006;2:217–228. doi:10.1021/ct050250b
18. Meli, M.; Colombo GA. Hamiltonian Replica Exchange Molecular Dynamics (MD)
Method for the Study of Folding, Based on the Analysis of the Stabilization
Determinants of Proteins. Int J Mol Sci. 2013;14:12157-12169.
19. de Ruiter, A.; Oostenbrink C. Protein–Ligand Binding from Distancefield
Distances and Hamiltonian Replica Ex-change Simulations. J Chem Theory
Comput. 2013;9:883-892.
20. Hritz, J.; Oostenbrink C. Optimization of Replica Exchange Molecular Dynamics
by Fast Mimicking. J Chem Phys. 2007;127:204104.
21. Rowland, P.; Blaney, F. E.; Smyth, M. G.; Jones, J. J.; Leydon VR., Oxbrow, A. K.;
Lewis, C. J.; Tennant, M. G.; Modi S. E, D.S.; Chenery, R.J.; Bridges AM. Crystal
structure of human cytochrome P450 2D6. J Biol Chem. 2006;281:7614–7622.
22. Lui VWY, Peyser ND., Ng PKS, et al. Frequent mutation of receptor protein
tyrosine phosphatases provides a mechanism for STAT3 hyperactivation in
head and neck cancer. Proc Natl Acad Sci U S A. 2014;111(3):1114-1119.
doi:10.1073/pnas.1319551111
23. Oostenbrink C.; de Ruiter A.; Hritz J.; Vermeulen NPE. Malleability and
versatility of Cytochrome P450 active sites studied by molecular simulations.
Curr Drug Metab. 2012;13:190-196.
24. Beutler, T. C.; Mark, A. E.; van Schaik, R. C.; Gerber, P. R.; van Gunsteren WF.
Avoiding singularities and numerical instabilities in free energy calculations
based on molecular simulations. Chem Phys Lett. 1994;222:529-539.
doi:10.1016/0009-2614(94)00397-1
25. Kumar, S.; Rosenberg, J. M.; Bouzida, D.; Swendsen, R. H.; Kollman PA. The
BIBLIOGRAPHY
71
weighted histogram analysis method for free-energy calculations on biomolecules.
I. The method. J Comput Chem. 1992;13:1011–1021.
26. Torrie, G.M.; Valleau JP. Nonphysical sampling distributions in monte carlo freeenergy
estimation: umbrella sampling. J Comput Phys. 1977;23:187–199.
doi:10.1016/0021-9991(77)90121-8
27. Kaestner J. Umbrella sampling. Comput Mol Sci. 2011;1:932-942.
doi:10.1002/wcms.66
28. Tembe, B.; McCammon JA. Ligand-receptor interactions. Comput Chem.
1984;8:281-283.
29. Zwanzig RW. High-temperature equation of state by a perturbation method. I.
Nonpolar gases. J Chem Phys. 1954;22:1420-1426.
30. Beveridge, D. L.; DiCapua FM. Free energy via molecular simulation:
applications to chemical and biomolecualr systems. Ann Rev Biophys Biophys
Chem. 1989;18:431-492.
31. Gilson, M. K.; Given, J. A.; Bush, B. L.; McCammon JA. The statisticalthermodynamic
basis for computation of binding affinities: a critical review.
Biophys J. 1997;72:1047-1069.
32. Kollman PA. Free energy calculations: applications to chemical and biochemical
phenomena. Chem Rev. 1993;93:2395-2417.
33. Tompa P. Intrinsically disordered proteins: a 10-year recap. Trends Biochem
Sci. 2012;37:509-516. doi:10.1016/j.tibs.2012.08.004
34. Uversky VN. A decade and a half of protein intrinsic disorder: biology still waits
for physics. Protein Sci. 2013;22:693-724.
35. Dunker AK, Lawson JD, Brown CJ, et al. Intrinsically disordered protein. J Mol
Graph Model. 2001;19(1):26-59. doi:10.1016/S1093-3263(00)00138-8
36. Ward JJ, Sodhi JS, McGuffin LJ, Buxton BF, Jones DT. Prediction and Functional
Analysis of Native Disorder in Proteins from the Three Kingdoms of Life. J Mol
Biol. 2004;337(3):635-645. doi:10.1016/j.jmb.2004.02.002
37. Robustelli P, Piana S, Shaw DE. Developing a molecular dynamics force field for
both folded and disordered protein states. Proc Natl Acad Sci U S A.
2018;115(21):E4758-E4766. doi:10.1073/pnas.1800690115
38. Piana, S.; Lindorff-Larsen, K.; Shaw DE. How robust are protein folding
BIBLIOGRAPHY
72
simulations with respect to force field parameterization? Biophys J.
2011;100:L47–L49.
39. Motáčková V, Nováček J, Zawadzka-Kazimierczuk A, et al. Strategy for complete
NMR assignment of disordered proteins with highly repetitive sequences based
on resolution-enhanced 5D experiments. J Biomol NMR. 2010;48(3):169-177.
doi:10.1007/s10858-010-9447-3
40. Novacek, J.; Janda, L.; Dopitova, R.; Zidek, L.; Sklenar V. Efficient protocol for
backbone and side-chain assignments of large, intrinsically disordered
proteins: transient secondary structure analysis of 49.2 kDa microtubule
associated protein 2c. J Biomol NMR. 2013;56:291–301.
41. Demo, G; Papouskova, V; Komarek, J; Kaderavek, P; Otrusinova, O; Srb, P;
Rabatinova, A; Krasny, L; Zidek, L; Sklenar V; Wimmerova M. X-ray vs. NMR
structure of N-terminal domain of delta-subunit of RNA polymerase. J Struct
Biol. 2014;187:174–186.