Articles
https://doi.org/10.1038/s41594-019-0300-4
1
MRC Laboratory of Molecular Biology, Cambridge, UK. 2
Institute of Clinical Neurobiology, University of Wuerzburg, Wuerzburg, Germany. 3
The Francis
Crick Institute, London, UK. 4
Department of Neuromuscular Disease, UCL Institute of Neurology, London, UK. 5
Division of Brain Sciences, Department
of Medicine, Imperial College London, London, UK. 6
Institute of Quantitative Biology, Biochemistry and Biotechnology, Edinburgh University, Edinburgh,
UK. 7
Department of Genetics, Environment and Evolution, UCL Genetics Institute, London, UK. 8
Institute of Molecular Biology GmbH, Mainz, Germany.
9
MRC Cancer Unit at the University of Cambridge, Cambridge, UK. 10
RNA Biology and Cancer Laboratory, Peter MacCallum Cancer Centre, Melbourne,
Australia. 11
Okinawa Institute of Science & Technology Graduate University, Okinawa, Japan. 12
Center for Motor Neuron Biology and Disease, Department
of Pathology and Cell Biology, Columbia University, New York, NY, USA. 13
Institute of Neuroscience, Newcastle University, Newcastle upon Tyne, UK.
14
Department of Biochemistry, University of Cambridge, Cambridge, UK. 15
Faculty of Computer and Information Science, University of Ljubljana, Ljubljana,
Slovenia. 16
These authors contributed equally: Michael Briese, Nejc Haberman, Christopher R. Sibley. *e-mail: jernej.ule@crick.ac.uk
S
plicing is a multi-step process in which small nuclear ribonucleoprotein
particles (snRNPs) and associated splicing
factors bind at specific positions around intron boundaries
to assemble an active spliceosome through a series of remodeling
steps. The splicing reactions are coordinated by dynamic pairings
between different snRNAs, between snRNAs and pre-mRNA and
by protein–RNA contacts1
. Spliceosome assembly begins with ATPindependent
binding of U1 snRNP at the 5′ splice site (SS) and of
U2 small nuclear RNA auxiliary factors 1 and 2 (U2AF1 and U2AF2,
also known as U2AF35 and U2AF65) to the 3′SS. ATP-dependent
remodeling then leads to the formation of complex A in which U2
snRNP contacts the BP, stabilized through interactions with the
U2AF and U2 snRNP splicing factor 3 (SF3a and SF3b) complex.
Next, U4/U6 and U5 snRNPs are recruited to form complex B. The
actions of many RNA helicases and pre-mRNA processing factor 8
(PRPF8) then facilitate rearrangements of snRNP interactions and
establishment of the catalytically competent Bact
and C complexes.
These catalyze the two trans-esterification reactions leading to lariat
formation, intron removal and exon ligation2
.
Transcriptome-wide studies of splicing reactions are valuable
to unravel the multi-component and dynamic assembly of the
spliceosome on the pre-mRNA substrate3–5
. Accordingly, ‘spliceosome
profiling’ has been developed through affinity purification
of the tagged U2·U5·U6·NTC complex from Schizosaccharomyces
pombe to monitor its interactions using an RNA footprinting-based
strategy3,4
. However, it is unclear if this method can be applied to
mammalian cells that might be more sensitive to the introduction
of affinity tags into splicing factors. Furthermore, no method has
simultaneously monitored the full complexity of the interactions of
diverse RNA binding proteins (RBPs) on pre-mRNAs from the earliest
to the latest stages of spliceosomal assembly.
Here, we have adapted the individual nucleotide resolution UV
crosslinking and immunoprecipitation (iCLIP) method6
to develop
spliceosome iCLIP. This approach identifies crosslinks of endogenous,
untagged spliceosomal factors on pre-mRNAs at nucleotide
resolution. In a previous study, we demonstrated the validity of this
approach by showing how PRPF8 remodels spliceosomal contacts
at 5′SS5
. Here, we comprehensively characterize spliceosome iCLIP
and show that it simultaneously maps the crosslink profiles of core
and accessory spliceosomal factors that are known to participate
across the diverse stages of the splicing cycle. Due to iCLIP’s nucleotide
precision, we distinguished seven binding peaks corresponding
to distinct RBPs that differ in their requirement for ATP or the factor
PRPF8. Spliceosome iCLIP also purifies intron lariats and identified
132,287 candidate BP positions. Compared to BPs identified
in previous RNA sequencing (RNA-seq) studies7–9
, those identified
by spliceosome iCLIP contain more canonical sequence and structural
features. We further examined the binding profiles of spliceosomal
RBPs around the BPs. This demonstrates that assembly of
SF3 and associated spliceosomal complexes tends to be determined
by a primary BP in most introns, even though alternative BPs are
detected by lariat-derived reads in RNA-seq. Moreover, we identify
complementary roles of U2AF and SF3 complexes in BP definition.
Taken together, these findings demonstrate the value of spliceosome
A systems view of spliceosomal assembly and
branchpoints with iCLIP
Michael Briese1,2,16
, Nejc Haberman3,4,16
, Christopher R. Sibley   1,4,5,6,16
, Rupert Faraway3,4
,
Andrea S. Elser3,4
, Anob M. Chakrabarti3,7
, Zhen Wang1
, Julian König1,8
, David Perera9
,
Vihandha O. Wickramasinghe9,10
, Ashok R. Venkitaraman9
, Nicholas M. Luscombe   3,7,11
,
Luciano Saieva12,13
, Livio Pellizzoni   12
, Christopher W. J. Smith   14
, Tomaž Curk   15
and Jernej Ule   1,3,4
*
Studies of spliceosomal interactions are challenging due to their dynamic nature. Here we used spliceosome iCLIP, which
immunoprecipitates SmB along with small nuclear ribonucleoprotein particles and auxiliary RNA binding proteins, to map spliceosome
engagement with pre-messenger RNAs in human cell lines. This revealed seven peaks of spliceosomal crosslinking
around branchpoints (BPs) and splice sites. We identified RNA binding proteins that crosslink to each peak, including known
and candidate splicing factors. Moreover, we detected the use of over 40,000 BPs with strong sequence consensus and structural
accessibility, which align well to nearby crosslinking peaks. We show how the position and strength of BPs affect the
crosslinking patterns of spliceosomal factors, which bind more efficiently upstream of strong or proximally located BPs and
downstream of weak or distally located BPs. These insights exemplify spliceosome iCLIP as a broadly applicable method for
transcriptomic studies of splicing mechanisms.
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb930
ArticlesNATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
iCLIP for transcriptome-wide studies of BP definition and spliceosomal
interactions with pre-mRNAs.
Results
Spliceosome iCLIP identifies interactions between splicing factors,
snRNAs and pre-mRNAs. SmB/B′ proteins are part of the
highly stable Sm core common to all spliceosomal snRNPs except
U6 (ref. 1
). To adapt iCLIP for the study of a multi-component
machine like the spliceosome, we immunopurified endogenous
SmB/B′ proteins10
using a range of conditions with differing stringency
of detergents and salt concentrations in the lysis and washing
steps (Supplementary Table 1, Fig. 1a and Supplementary Fig. 1a,b).
First, to enable denaturing purification, we generated HEK293 cells
stably expressing Flag-tagged SmB and used 6 M urea during cell
lysis to minimize co-purification of additional proteins11
(‘stringent’
purification, Supplementary Table 1), followed by dilution of the
lysis buffer (see Methods) to facilitate immunopurification of SmB
via the Flag tag. We observed a 25 kDa band corresponding to the
molecular weight of SmB–RNA complexes, which was absent when
UV light or anti-Flag antibody were omitted, or when cells not
expressing Flag-SmB were used (Supplementary Fig. 1c). Next, we
used standard nondenaturing iCLIP conditions, which uses a high
concentration of detergents in the lysis buffer, and wash buffer with
1 M NaCl (‘medium’ purification, Supplementary Table 1). This
disrupts most protein-protein interactions but can preserve stable
complexes such as snRNPs, as evident by the multiple radioactive
bands in addition to the 25 kDa SmB–RNA complex on treatment
with low RNase (Fig. 1b). Of note, similar profiles of protein–RNA
complexes were obtained when using different monoclonal SmB/B′
antibodies (Supplementary Fig. 1d). Last, we further decreased the
concentration of detergents in the lysis buffer, used 0.1 M NaCl in
the washing buffer (‘mild’ purification, Supplementary Table 1), and
used a low RNase treatment that leaves snRNAs generally intact so
that they could serve as a scaffold for purifying the multi-protein
spliceosomal complexes (Fig. 1a).
To produce complementary DNA (cDNA) libraries with spliceosome
iCLIP, we immunoprecipitated SmB/B′ under three different
stringency conditions from lysates of UV-crosslinked cells, and isolated
a broad size distribution of protein–RNA complexes to recover
the greatest possible diversity of spliceosomal protein–RNA interactions
(Fig. 1b and Supplementary Fig. 1c,d). An antibody against
endogenous SmB/B′ was used for medium and mild purification
from HEK293, K562 and HepG2 cells, and an anti-Flag antibody
for stringent purification from HEK293 cells expressing Flag-SmB
(Supplementary Tables 2 and 3). As in previous iCLIP studies6
, the
nucleotide preceding each cDNA was used for all analyses. When
stringent conditions were used, >75% of iCLIP cDNAs mapped to
snRNAs, probably corresponding to the direct binding of Flag-SmB
(Fig. 1c). However, the proportion of snRNA crosslinking reduced to
~40–60% under mild and medium conditions, with a corresponding
increase of crosslinking to introns and exons that probably reflects
binding of snRNP-associated proteins to pre-mRNAs (Fig. 1a,c).
Spliceosome iCLIP identifies seven crosslinking peaks on premRNAs.
Assembly of the spliceosome on pre-mRNA is guided by
three main landmarks: the 5′SS, 3′SS and BP. Therefore, we evaluated
if spliceosomal crosslinks are located at specific positions relative
to splice sites and computationally predicted BPs12
. For this
purpose, we performed spliceosome iCLIP from human Cal51 cells
that have been used previously as a model system to study the roles
of spliceosomal factors in the cell cycle5
. RNA maps of summarized
spliceosomal crosslinking revealed seven peaks around these
landmarks (Fig. 2a). Importantly, similar positional patterns were
also seen in HEK293, K562 and HepG2 cell lines (Supplementary
Fig. 2a). The centers of the peaks were 15 nucleotides upstream
of the 5′SS (peak 1), 10 nucleotides downstream of the 5′SS
(peak 2), 31 nucleotides downstream of the 5′SS (peak 3), 26 nucleotides
upstream of the BP (peak 4), 20 nucleotides upstream of the BP
(peak 5), 11 nucleotides upstream of the 3′SS (peak 6) and 3 nucleotides
upstream of the 3′SS (peak 7). We also observed an alignment
of cDNAs at the start of introns and at the BPs, which we refer
to as positions A and B, respectively (Fig. 2a and Supplementary
Fig. 2a). The crosslinking enrichment at most peaks was generally
stronger under mild conditions, especially at the 3′SS
(Supplementary Fig. 2a). This indicates that spliceosome iCLIP performed
under mild conditions is the most suitable for investigating
spliceosomal assembly on pre-mRNAs.
Spliceosome iCLIP monitors multiple stages of spliceosomal
remodeling. Next, we investigated whether spliceosome iCLIP is
able to monitor spliceosome assembly at different stages during the
splicing cycle. For this purpose, we knocked down (KD) PRPF8 in
Cal51 cells (Supplementary Fig. 2b) and performed spliceosome
iCLIP under mild conditions. As an integral component of the U4/
U6.U5 tri-snRNP, PRPF8 is essential for both catalytic reactions1
.
We previously showed that PRPF8 is required for efficient spliceosomal
assembly at 5′SS5
. Here, we additionally find that PRPF8
is essential for efficient spliceosomal assembly at peaks 4 and 5
(Fig. 2a). Moreover, we also observed a major decrease of reads
truncating at positions A and B, whereas crosslinking at peaks 2 and
6 is increased with PRPF8 KD.
To further investigate whether spliceosome iCLIP can monitor
distinctstagesofthesplicingreaction,weperformedanin vitrosplicing
assay in which an exogenous pre-mRNA splicing substrate was
incubated with HeLa nuclear extract in the presence or absence of
ATP. ATP is required for the progression of early, ATP-independent,
spliceosomal complexes to later assembly stages mediating the catalytic
splicing reactions. The RNA substrate was produced by in vitro
transcription of a minigene construct containing a short intron and
flanking exons from the human C6orf10 gene. Gel electrophoresis
analysis confirmed that the minigene RNA was efficiently spliced
in vitro in an ATP-dependent manner (Supplementary Fig. 2c). We
performed spliceosome iCLIP from the splicing reactions using
mild purification conditions (Supplementary Fig. 2d). Following
sequencing, the reads mapping to the exogenous splicing substrate
or spliced product represented ~1%, whereas the remaining reads
were derived from endogenous RNAs present in the nuclear extract
(Supplementary Table 4). The spliced product was detected with
exon-exon junction reads primarily in the presence of ATP (364
reads in +ATP versus 5 reads in −ATP condition) (Supplementary
Fig. 2e and Supplementary Table 4). As expected, given that the spliceosome
rapidly disassembles on completion of the splicing reaction,
very few reads mapped to the spliced (364 reads) compared
to unspliced substrate (48,584 reads) (Supplementary Table 4) in
the +ATP condition. It should be considered, however, that some
reads from the exogenous minigene could represent RNA that did
not enter the splicing pathway.
We visualized crosslinking on the substrate RNA, and marked
positions that correspond to peaks on the transcriptome-wide RNA
maps (Fig. 2b). While crosslinking peaks on a metagene plot might
not necessarily be representative of individual splicing substrates,
we nevertheless observed crosslinking in corresponding regions
of the C6orf10 substrate (comparing Fig. 2a,b). When comparing
crosslinking in the presence or absence of ATP, an unchanged
crosslinking profile was seen in regions of peaks 1, 2, 6 and 7, indicating
these are ATP-independent contacts of early spliceosomal
factors. In contrast, the presence of ATP led to a ~11-fold increase
of crosslinking in the region upstream of the BP where the PRPF8dependent
peaks 4 and 5 are located on endogenous transcripts
(Fig. 2b). This indicates that spliceosome iCLIP detects pre-mRNA
binding of factors contributing to early, ATP-independent and late,
ATP-dependent stages of spliceosomal assembly.
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb 931
Articles NATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
Following crosslinking, the peptide that remains bound to the
RNA after RBP digestion will normally terminate reverse transcription
to produce so-called ‘truncated cDNAs’13–15
. Accordingly,
analysis of data from iCLIP and derived methods, such as eCLIP16
,
generally refer to the nucleotide preceding the iCLIP read on the reference
genome as the ‘crosslink site’. However, in spliceosome iCLIP
we additionally expect cDNAs that truncate at the three-way junction
formed by intron lariats, where the 5′ end of the intron is linked
via a 2′–5′ phosphodiester bond to the BP (Fig. 2c). Following RNase
digestion, such lariat three-way-junction RNAs present two available
3′ ends for ligation of adapters, such that cDNAs can truncate at the
BP (position B) or at the start of the intron (position A). Interestingly,
the medium purification condition was optimal to produce cDNAs
truncating at positions A and B (Supplementary Fig. 2a), possibly
because spliceosomal C complexes containing lariat intermediates
are known to be stable under high-salt conditions17
. Note that peaks
A and B are higher in HEK293 compared to HepG2 and K562 cells
under medium purification conditions, and probably reflect differences
in lariat co-purification. Meanwhile, the number of cDNAs
truncating at positions A and B is dramatically decreased under conditions
that inhibit splicing progression and lariat formation; PRPF8
KD in vivo (twofold, Fig. 2a) or absence of ATP in vitro (≥18-fold,
Fig. 2b). This further confirms that spliceosome iCLIP can monitor
spliceosome assembly at distinct stages of the splicing cycle.
Specific RBPs are enriched at each peak of spliceosomal crosslinking.
Next, to identify RBPs that crosslink at peaks identified
by spliceosome iCLIP, we examined the eCLIP data for 110 RBPs
(from 157 eCLIP samples of 68 RBPs in the HepG2 and 89 RBPs
in the K562 cell line) provided by the ENCODE consortium16
. Of
note, comparisons between iCLIP and eCLIP are justified due to
their use of identical lysis and wash buffers (analogous to medium
stringency in the present study), use of truncated cDNAs to identify
crosslink sites and similar RNase digestion conditions, and comparable
crosslinking profiles for RBPs such as PTBP1 and U2AF2
(ref. 15
). Accordingly, we analyzed the eCLIP data to identify RBPs
with enriched normalized crosslinking at each spliceosomal iCLIP
peak. This identified a specific set of RBPs at each peak, with good
overlap between RBPs identified in K562 and HepG2 cells (Fig.
3 and Supplementary Dataset 1). As expected, SF3 components
SF3B4, SF3A3 and SF3B1 bind to peaks 4 or 5 (ref. 18
). U2AF2 binds
the polypyrimidine (polyY) tract (peak 6), and U2AF1 close to the
intron–exon junction (peak 7) (ref. 19
).
Spliceosome iCLIP identifies BPs with canonical sequence and
structural features. To determine whether spliceosome iCLIP
could experimentally identify human BPs, we used spliceosome
iCLIP data produced under medium purification from Cal51 cells.
Most cDNA starts in spliceosome iCLIP overlap with a uridine-rich
b
a
c
185
kDa
115
80
65
50
30
25
15
RNase I++ + ++ +
no Ab a-SmB/B′
SmB/B′
B
D1
D2
F
E
G
D3
snRNPs
Partial RNase digestion
immunoprecipitation
ligation of the 3′ adapter
pre-mRNA
UVUV
UV
UV
snRNA
3′
3′
5′ 3′
3′
5′
5′
3′
B
D1
D2
F
E
G
D3
B
D1
D2
F
E
G
D3
B
Medium
Stringent
Continue
to iCLIP
Mild
In vivo UV crosslinking
0
25
50
75
100
Percentageof
crosslinks
UTR CDSIntronsnRNA lncRNA
Stringent Medium Mild
snRNA
snRNA
pre-mRNA
snRNA
pre-mRNA
Fig. 1 | Spliceosome iCLIP identifies protein interactions with snRNAs and splicing substrates. a, Schematic representation of the spliceosome iCLIP
method performed under conditions of varying purification stringency. b, Autoradiogram of crosslinked RNPs immunopurified from HeLa cells under
medium conditions by a SmB/B′ antibody following digestion with high (++) or low (+) amounts of RNase I. The dotted line depicts the region typically
excised from the nitrocellulose membrane for spliceosome iCLIP. As a control, the antibody (Ab) was omitted during immunopurification. c, Genomic
distribution of spliceosome iCLIP cDNAs produced under stringent, medium and mild conditions from HEK293 cells. Data were mapped first to snRNAs,
allowing multiple mapping reads, and then to the genome, allowing only uniquely mapped reads. Proportions of cDNAs mapping to snRNAs, introns,
coding sequence of mRNAs (CDS), untranslated regions of mRNAs (UTR) and long noncoding RNAs (lncRNAs) are shown (but not the intergenic reads
and other types of RNAs). Data are shown as mean ± s.e.m. from three independent experiments for the medium and mild purification condition and two
independent experiments for the stringent purification condition. Source data for panel c are available online.
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb932
ArticlesNATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
motif (Fig. 4a), in agreement with an increased propensity of protein–RNA
crosslinking at uridine-rich sites14
. In contrast, cDNAs
ending at the last nucleotide of introns, which are thus probably
derived from intron lariats, have starts overlapping the YUNAY
motif matching the consensus BP sequence (Fig. 4b). Further, these
cDNAs have higher enrichment of mismatches of adenosines at
their first nucleotide (Supplementary Fig. 3a), which is consistent
with mismatch, insertion and deletion errors during reverse transcription
across the three-way junction of the BP9
. For comparison,
reads that start in regions where BPs are typically located, but which
do not align with intron ends, have less enrichment of the BP consensus
motif at their starts (Supplementary Fig. 3b,c). To identify a
confident set of putative BPs in a transcriptome-wide manner, we
therefore used the spliceosome iCLIP cDNAs that aligned with the
end of introns (Fig. 4b). These cDNAs started at adenines in 132,287
intronic positions, which we considered as BP candidates. The 41
read-length limited our analysis to the region where most BPs are
located, but more distal BPs cannot be identified by this approach.
For further study, we selected BPs with the highest number of truncated
cDNAs per intron. This identified candidate BPs in 43,637
introns of 9,565 genes.
To examine the BPs identified by spliceosome iCLIP, ‘iCLIP BPs’,
we compared them with the ‘computational BPs’ recently identified
withasequence-baseddeep-learningpredictor,LaBranchoR,which
predicted BPs for over 90% of 3′SS12
. We also compared with ‘RNAseq
BPs’, including the 138,314 BPs from 43,637 introns that were
identified by analysis of lariat-spanning reads from 17,164 RNAseq
datasets8
. Initially, 65% of iCLIP BPs overlapped with the topscoring
computational BPs (Supplementary Fig. 3d). Interestingly,
in cases where iCLIP and computational BPs were located less
than five nucleotides apart, they frequently occurred within
A-rich sequences (Supplementary Fig. 3e). This mismatch could
be of a technical nature, as truncation of iCLIP cDNAs may not
always be precisely aligned to BPs in the case of A-rich sequences.
Alternatively, more than one A might be capable of serving as the
BP. When allowing a one nucleotide shift for comparison between
methods, as has been done previously12
, 70% of iCLIP BPs overlapped
with the top-scoring computational BPs, while 26% overlapped
with the RNA-seq BPs (Fig. 4c and Supplementary Dataset
2). If the computational BPs overlapped either with an iCLIP BP
and/or RNA-seq BP, it generally had a strong BP consensus motif
(o-BP, Fig. 4d).
c
+ATP–ATP
1 2 3
4 5 B 6 7
A
A1 2 3 4/5 6B 7
a
b
RNase
digestion
Adapter
ligation
Reverse
transcription
Lariat
5′
3′ 3′
3′
5′
5′
cDNA truncating at B
cDNA truncating at A
Branchpoint
Branchpoint
4
8
0
0
4
8
4 5 6/7B
−50 −25 0 25 50 75
Position relative to 5′SS
−50 −25 0 25
Position relative to branchpoints
0
1
2
3
−80 −40 0
Position relative to 3′SS
Normalizedcoverageof
cDNAstarts
Normalizednumberof
cDNAstarts(log2)
Control
PRPF8 KD
Control
PRPF8 KD
Control
PRPF8 KD
Fig. 2 | Analysis of spliceosomal interactions with pre-mRNAs in vitro and in vivo. a, Metagene plots of spliceosome iCLIP from Cal51 cells. Plots are
depicted as RNA maps of summarized crosslinking at all exon-intron and intron–exon boundaries, and around BPs to identify major binding peaks, and
to monitor changes between control and PRPF8 KD cells. Crosslinking is regionally normalized to its average crosslinking across the −100..50 nucleotide
(nt) region relative to splice sites or BPs depending on the RNA map in order to focus the comparison on the relative positions of peaks. b, Normalized
spliceosome iCLIP cDNA counts on the C6orf10 in vitro splicing substrate. Exons are marked by gray boxes, introns by a line and the BPs by a green dot.
The positions of crosslinking peaks are marked by numbers and letters corresponding to the peaks in Fig. 2a. c, Schematic description of the three-way
junctions of intron lariats. The three-way junction is produced after limited RNase I digestion of intron lariats. This can lead to cDNAs that do not truncate
at sites of protein–RNA crosslinking, but rather at the three-way junction of intron lariats. These cDNAs initiate from the end of the intron and truncate at
the BP (position B), or initiate downstream of the 5′SS and truncate at the first nucleotide of the intron (position A).
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb 933
Articles NATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
To gain insight into the differences between the methods,
we focused on BPs that were identified by a single method and
located more than five nucleotides away from BPs identified by
other methods. Notably, the computational- or iCLIP-specific
BPs have a strong enrichment of the consensus YUNAY motif
(C-BP, i-BP, Fig. 4e,f,h,i). In contrast, RNA-seq-specific BPs contain
a larger proportion of noncanonical BP motifs, which agrees
with previous observations7,9,12
(Fig. 4g,j). To evaluate further, we
compared iCLIP BPs with two studies that identified 59,359 BPs
by exoribonuclease digestion and targeted RNA-seq9
, and 36,078
BPs by lariat-spanning reads refined by U2 snRNP/pre-mRNA
base-pairing models7
. Considering the introns that contained BPs
defined both by RNA-seq and iCLIP, we found 57% and 47% overlapping
BPs (Supplementary Fig. 3f–i). Again, the iCLIP-specific
BPs were more strongly enriched in the consensus YUNAY motif
compared to BPs specifically identified by either RNA-seq method
(Supplementary Fig. 3j–o). We also examined the local RNA structure
around each category of BPs. Overlapping, iCLIP-specific and
computational-specific BPs had a decreased pairing probability
at the position of the BP, which was not seen for the RNA-seqspecific
BPs (Fig. 4k,l). The difference in RNA-seq BPs derives
from the presence of noncanonical non-A branched BPs, which
have a generally increased pairing probability (Supplementary Fig.
3p,q). This indicates that the non-A BPs might be structurally less
accessible for pairing with U2 snRNP.
Alignment of RBP binding profiles signifies the functionality of
BPs. Peaks 4, 5 and position B align to BP position, and therefore
we could evaluate how the crosslinking profiles of RBPs binding at
these peaks align to the different classes of BPs. First, we examined
the crosslinking of SF3B4, which binds in the region of peak 4 as
part of the U2 snRNP complex that recognizes the BP1
. Analysis
of the o-BPs defines the peak of SF3B4 crosslinking at the 25th
nucleotide upstream of BPs (Fig. 5 and Supplementary Fig. 4a, b).
However, the peak of SF3B4 crosslinking is shifted from this 25th
position for the nonoverlapping method-specific BPs; it is generally
closer than 25 nucleotides to the BPs located upstream of another
BP (up BP), and further than 25 nucleotides away from BPs located
downstream of another BP (down BP) (Fig. 5). The shift from the
expected position is greatest for RNA-seq-specific BPs (R-BP), and
smallest for computationally predicted BPs, as evident by eCLIP
data from two cell lines (Fig. 5a,b). Moreover, the same result is seen
BUD13−HepG2
SF3B1−K562
SMNDC1−K562
GPKOW−K562
SMNDC1−HepG2
SF3B4−HepG2
SF3B4−K562
0 5 10
Normalized crosslink enrichment
Peak 4
EFTUD2−K562
XRN2−HepG2
SF3B1−K562
SMNDC1−K562
BUD13−HepG2
SMNDC1−HepG2
GPKOW−K562
SF3B4−HepG2
SF3A3−HepG2
SF3B4−K562
Normalized crosslink enrichment
Peak 5
U2AF1−K562
U2AF2−HepG2
U2AF2−K562
U2AF2−Hela*
0 5 10 15
Normalized crosslink enrichment
Peak 6
U2AF1−HepG2
U2AF2−HepG2
U2AF2−K562
U2AF1−K562
0 1 2 3 4
Normalized crosslink enrichment
Peak 7
PRPF8−K562
RBM22−K562
SUPV3L1−HepG2
PRPF8−HepG2
0 2.5 5 7.5
Normalized crosslink enrichment
Position B
SF3B4−K562
PRPF8−K562
SUPV3L1−HepG2
PRPF8−HepG2
Normalized crosslink enrichment
Position A
0 5 10 15 20
0 2.5 5 7.5
–29..–22nt –21..–16nt
–11..–9nt –3..–1nt
–1..1nt –1..1nt
Fig. 3 | Identification of RBPs overlapping with spliceosomal peaks at BPs and 3′SS. Enrichment of eCLIP crosslinking within each of the spliceosome
iCLIP peaks, which are defined by the positions marked in the figure. We first regionally normalized the crosslinking of each RBP to its average crosslinking
over −100..50 nt region relative to 3′SS, which generates the RNA maps as shown in Supplementary Figs. 5 and 6. We then ranked the RBPs according to
the average normalized crosslinking across the nucleotides within each peak. We analyzed peaks 4–7 and positions A and B, as marked on the top of each
plot. The top-ranking RBPs in each peak are shown on the left plot and the full distribution of RBP enrichments is shown on the right plot.
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb934
ArticlesNATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
a b d
e
c
f g
h i j
k l
Starts of iCLIP reads
that align with ends of introns
132,287 positions
–5 0 5
Starts of all iCLIP reads
19,743,890 positions
9,733
(o)
iCLIP BPs
iCLIP BP
RNA-seq BPs
RNA-seq BP
>5 nt
An example nomenclature of misaligned BPs (by >5 nt):
i-BPup R-BPdown
Computational: C-BPup
4,038 positions
Computational: C-BPdown
4,830 positions
RNA-seq: R-BPup
3,899 positions
RNA-seq: R-BPdown
3,657 positions
iCLIP: i-BPdown
1,613 positions
iCLIP: i-BPup
2,546 positions
Computational BPs
1,674
21,344 (o)20,818 (o)
11,412 (i) 18,065 (R)
23,079 (C)
–5 0 5 –5 0 5
–5 0 5
–5 0 5
–5 0 5 –5 0 5
−20 −10 0 10 20
Position relative to branchpoint
Averagepairingprobability
Overlapping o-BP
51,888 positions
−20 −10 0 10 20
Position relative to branchpoint
C-BPup R-BPup
i-BPup o-BP
0
1
–5 0 5
Probability
0
1
Probability
0
1
Probability
0
1
Probability
0
1
Probability
0
1
Probability
0
1
Probability
0
1
Probability
–5 0 5
0
1
Probability
0.45
0.50
0.55
0.60
C-BPdown R-BPdown
i-BPdown o-BP
Fig. 4 | Comparison of BPs identified by spliceosome iCLIP, RNA-seq lariat reads or computational prediction. a, Weblogo around the nucleotide
preceding all spliceosome iCLIP reads. b, Weblogo around the nucleotide preceding only those spliceosome iCLIP reads that align with ends of introns.
c, Introns that contain at least one BP identified either by published RNA-seq8
or by spliceosome iCLIP are used to examine the overlap between the top
BPs identified by RNA-seq (that is, the BP with most lariat-spanning reads in each intron), iCLIP (BP with most cDNA starts) or computational predictions
(highest scoring BP)12
. BPs that are 0 or 1 nt apart are considered as overlapping. d, Weblogo of o-BP category of BPs. e, Weblogo of C-BPup category
of BPs. f, Weblogo of i-BPup category of BPs. g, Weblogo of R-BPup category of BPs. h, Weblogo of C-BPdown category of BPs. i, Weblogo of i-BPdown
category of BPs. j, Weblogo of R-BPdown category of BPs. k,l, The 100 nt RNA region centered on the BP was used to calculate pairing probability with
the RNAfold program using default parameters25
, and the average pairing probability of each nucleotide around BPs is shown for the 40 nt region around
method-specific BPs located upstream (k) or downstream (l). C-BP, C-BPs that are >1 nt away from BPs defined by other methods in the same intron; i-BP,
i-BPs that are >1 nt away from BPs defined by other methods in the same intron; R-BP: R-BPs that are >1 nt away from BPs defined by other methods in the
same intron; o-BP: o-BPs with up to 1 nt shift. If a BP defined by one method is >5 nt upstream of a BP defined by another method, then ‘up’ is added to its
acronym, and if it is >5 nt downstream, ‘down’ is added.
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb 935
Articles NATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
with U2AF2, where the strongest shift away from expected positions
is seen for RNA-seq BPs and weakest for computational BPs
(Supplementary Fig. 4c,d). The cDNA starts from PRPF8 eCLIP are
highly enriched at position B, corresponding to the lariat-derived
cDNAs that truncate at BPs (Fig. 3). Interestingly, the PRPF8 cDNA
starts had the strongest peak at the overlapping BPs but also peaked
at all the remaining classes of BPs (Supplementary Fig. 4e,f). This
indicates that all classes of BPs contribute to lariat formation and
that the nonoverlapping BPs most probably act as alternative BPs
within the introns.
Effects of BP position on spliceosomal assembly. To assess how
BP positioning determines spliceosome assembly, we evaluated
binding profiles of the RBPs that are enriched at peaks 4–7 and at
positions A and B (Fig. 3). We divided BPs based on their distance
from 3′SS, and normalized RBP binding profiles within each subclass
of BP. This showed that crosslinking of U2AF1 and U2AF2
aligns to the region between the BPs and 3′SS, which is covered by
the polyY tract (Supplementary Figs. 5 and 6). While SF3B4 is the
primary RBP crosslinking at peak 4, and SF3A3 at peak 5, binding
of SMNDC1, SF3B1, EFTUD2, BUD13, GPKOW and XRN2 to
peaks 4 and 5 was also evident (Supplementary Figs. 5 and 6 and
Fig. 3). PRPF8, RBM22 and SUPV3L1 have their cDNA starts truncating
at positions A and B (Supplementary Figs. 5 and 6), corresponding
to the three-way junction formed by intron lariats (Fig.
2c). This is in agreement with the association of PRPF8 and RBM22
with intron lariats as part of the human catalytic step I spliceosome1
.
The positions of SF3B4 and SF3A3 crosslinking peaks also agree
with CryoEM studies of the human spliceosome that show closer
pre-mRNA binding of SF3A3 (also referred to as SF3a60) to the BP
compared to SF3B4 (also referred to as SF3b49)20
.
To quantify how BP positioning affects the intensity of RBP
binding, we divided BPs into ten equally sized groups based on
the distance from 3′SS. We then normalized the relative binding
intensity of each RBP at each position on the RNA maps across the
ten groups and revealed strong relationships between BP position
and binding intensity of certain RBPs (Fig. 6a and Supplementary
Fig. 7a). For example, if a BP is located distally from the 3′SS,
then U2AF components bind stronger to peaks 6 and 7. In contrast,
if a BP is located proximally to the 3′SS, then EFTUD2,
SF3 components and several other RBPs bind stronger to peaks
4 or 5 (Fig. 6b). Notably, increased BP distance causes increased
binding of BUD13 and GPKOW at peaks 6 or 7 and decreased
binding at peaks 4 and 5. The more efficient recruitment of U2AF
and associated factors to peaks 6 and 7 could be explained by the
long polyY-tracts at distal BPs (Supplementary Fig. 5), while their
decreased binding at proximal BPs appears to be compensated by
increased binding of SF3 and other U2 snRNP-associated factors
at peaks 4 and 5.
In contrast to effects on individual splicing factors, we did not
observe any effect of BP distance on the relative intensity of spliceosome
iCLIP crosslinking in peaks 4 and 5 compared to 6 and 7
(Fig. 6c). This indicates that the effects may be masked during later
stages of spliceosome assembly. To ask if this is the case, we turned
to PRPF8, a protein that is essential for later stages of spliceosomal
assembly, a role it plays together with EFTUD2 and BRR2 as part
of U5 snRNP1
. PRPF8 KD leads to decreased spliceosomal binding
at peaks 4 and 5, and this effect is stronger at distal compared to
proximal BPs (Fig. 6c). In conclusion, our results reveal differences
in the binding profiles of splicing factors in relation to BP distance,
but these differences are neutralized at full spliceosome assembly in
a manner that requires the presence of PRPF8.
Misaligned BP is upstream of
a BP identified by another method
UpUpUp Down DownDown
C-BP
SF3B4 eCLIP
from K562
SF3B4 eCLIP, K562
BP count
SF3B4 eCLIP, HepG2
SF3B4 eCLIP
from HepG2
a
b
R-BP
i-BP
o-BP
−35
−15
−25
eCLIPcDNAstartsrelativetoBP
−35
−15
−25
eCLIPcDNAstartsrelativetoBP
C-BP
R-BP
i-BP
o-BP
Misaligned BP is downstream of
a BP identified by another method
4,038 2,546 3,899 51,895 4,830 1,613 3,657
16,078 4,947 9,791 239,162 10,276 38,204 34,116
13,545 5,396 7,326 192,818 8,248 7,556 4,903
Fig. 5 | Spliceosome assembly at BPs identified by spliceosome iCLIP, RNA-seq lariat reads or computational prediction. a,b, Violin plots depicting the
positioning of SF3B4 cDNA starts relative to the indicated BP categories. SF3B4 eCLIP data were from, K562 (a) and HepG2 (b) cells. Box-plot elements
are defined by center line, median; box limits, upper and lower quartiles; and whiskers, 1.5× interquartile range. Each data point corresponds to an eCLIP
crosslink event, and the total number of eCLIP crosslinks that map in the area analyzed around each set of BPs (sample size) is shown under the plot.
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb936
ArticlesNATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
Effects of BP strength on spliceosomal assembly. To examine how
BP strength affects spliceosomal assembly we focused on BPs that
have been identified both by spliceosome iCLIP and computational
modeling and are located at 23–28 nucleotides upstream of the 3′SS.
Of note, this is the most common position of BPs (Supplementary
Dataset 3). As an estimate of BP strength, we used the BP score,
a
b
Peak 4, HepG2
BP distance:
Peak 5, HepG2 Peak 6, HepG2 Peak 7, HepG2
BP distance: BP distance: BP distance:
4/5 6B 74 5
B
Examined regions
SF3B4
XRN2
EFTUD2
SF3A3
RBM22
U2AF2
BUD13
SMNDC1
CDC40
TBRG4
U2AF1
SUPV3L1
SF3B4
SF3A3
XRN2
BUD13
U2AF2
SUPV3L1
EFTUD2
RBM22
SMNDC1
CDC40
U2AF1
TBRG4
TBRG4
SF3A3
BUD13
CDC40
EFTUD2
RBM22
SUPV3L1
SF3B4
U2AF1
SMNDC1
U2AF2
XRN2
XRN2
BUD13
RBM22
SMNDC1
SUPV3L1
SF3A3
TBRG4
SF3B4
CDC40
EFTUD2
U2AF1
U2AF2
0.5
0.7
0.9
1.1
1.3
0.8
1.0
1.2
1.4
1.6
0.8
1.0
1.2
0.6
1.0
1.5
2.0
c
0
1
2
3
NormalizedcoverageofcDNAstarts
EFTUD2−HepG2
U2AF1−HepG2
0
1
2
−40 −30 −20 −10
Position relative to BP
NormalizedcoverageofcDNAstarts
−30 −20 −10 0
Position relative to 3′SS
0
2
4
6
NormalizedcoverageofcDNAstarts
0
1
2
3
NormalizedcoverageofcDNAstarts
Spliceosome iCLIP
−Cal51, control
0
1
2
−40 −30 −20 −10
Position relative to BP
NormalizedcoverageofcDNAstarts
−30 −20 −10 0
Position relative to 3′SS
Spliceosome iCLIP
−Cal51, PRPF8 KD
SF3A3−HepG2
U2AF2−HepG2
SMNDC1−HepG2
XRN2−HepG2
0
1
2
3
NormalizedcoverageofcDNAstarts
BUD13−K562
GPKOW−K562
5
6
6 6
6
7
7
B
6
7
7
5
5
5
5
4
4
4
4
4
BP distance:
−40 −30 −20 −10
Position relative to BP
−30 −20 −10 0
Position relative to 3′SS
−40 −30 −20 −10
Position relative to BP
−30 −20 −10 0
Position relative to 3′SS
Fig. 6 | BP position defines the binding patterns of splicing factors at 3′SS. a, Heatmaps depicting the normalized crosslinking of RBPs in peak regions
around ten groups of BPs that were categorized according to the distance of the BP from 3′SS. Crosslinks were derived as cDNA starts from eCLIP of
HepG2 cells. b, RNA maps showing normalized crosslinking profiles of selected RBPs relative to BPs and 3′SS for the two deciles of BPs that are located
most proximal (interrupted light lines) or most distal (solid dark lines) from 3′SS. c, RNA maps showing crosslinking profile of spliceosome iCLIP from
control and PRPF8 KD Cal51 cells in the same format as panel b.
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb 937
Articles NATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
which was determined with a deep-learning model12
. This showed
strong correlation between BP strength and RBP binding intensities,
such that most RBPs have increased crosslinking at peaks 4
and 5 at BPs with very high scores, and, conversely, increased crosslinking
at peaks 6 and 7 at BPs with very low scores (Fig. 7a,b and
Supplementary Fig. 7b). Since SF3 components primarily bind at
U2AF2 1
SF3 U2AF2 1
SF3
SF3 U2AF2 1
Spliceosome iCLIP
−Cal51, control
Spliceosome iCLIP
−Cal51, PRPF8 KD
SF3B4−HepG2
U2AF1-K562
SF3B1−K562
XRN2−K562
Strong branchpoint
Weak branchpoint
SF3
U2AF2 1
Branchpoint
Peak 4, HepG2
BP score: BP score: BP score: BP score:
Peak 5, HepG2 Peak 6, HepG2 Peak 7, HepG2
b
a
c
d
0
1
2
8
6
4
2
0
NormalizedcoverageofcDNAstarts
0
1
2
3
2
1
0
NormalizedcoverageofcDNAstarts
−50 −25 0
Position relative to 3′SS
−50 −25 0
Position relative to 3′SS
NormalizedcoverageofcDNAstarts
NormalizedcoverageofcDNAstarts
BP score:
4/5 6B 7
6
7
6
7
6
B B
6
7
4
4
4
4
Examined region:
SF3A3
SUPV3L1
BUD13
CDC40
SF3B4
U2AF2
U2AF1
XRN2
SMNDC1
EFTUD2
TBRG4
RBM22
TBRG4
U2AF2
BUD13
CDC40
SF3B4
SF3A3
SUPV3L1
SMNDC1
EFTUD2
RBM22
U2AF1
XRN2
SF3B4
SF3A3
CDC40
SUPV3L1
SMNDC1
TBRG4
U2AF1
RBM22
EFTUD2
BUD13
U2AF2
XRN2
SF3B4
TBRG4
SF3A3
U2AF1
U2AF2
SMNDC1
XRN2
SUPV3L1
RBM22
EFTUD2
BUD13
CDC40
0.7
0.9
1.1
1.3
0.8
1.0
1.2
1.4
0.8
1.0
1.2
1.4
0.8
1.0
1.2
1.4
−50 −25 0
Position relative to 3′SS
−50 −25 0
Position relative to 3′SS
Fig. 7 | BP strength correlates with the binding of splicing factors. a, Heatmaps depicting the normalized crosslinking of RBPs in peak regions around ten groups
of BPs that were categorized according to the computational scores that define BP strength. Crosslinks were derived as cDNA starts from eCLIP of HepG2 cells.
b, RNA maps showing normalized crosslinking profiles of selected RBPs relative to 3′SS for the two deciles of BPs that are lowest scoring (interrupted light lines)
or highest scoring (solid dark lines). c, RNA maps showing crosslinking profile of spliceosome iCLIP from control and PRPF8 KD Cal51 cells in the same format as
panel b. d, Schematic representation of the effects that BP position and score have on the assembly of SF3 and U2AF complexes around BPs.
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb938
ArticlesNATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
peaks 4 and 5, and U2AF components at peaks 6 and 7, an over
fourfold change is seen in the ratio of crosslinking when comparing
the extreme deciles of BP strength (Supplementary Fig. 7c). We did
not observe any correlation between the polyY tract coverage and
BP score (Supplementary Fig. 7d), which indicates that BP strength
directly affects the RBP binding profiles.
Similar to the effects on individual splicing factors, the relative
intensity of spliceosome iCLIP crosslinking in peaks 4 and 5 correlated
with BP strength (Fig. 7c). PRPF8 KD decreased spliceosomal
binding at peaks 4 and 5 of both classes of BPs, and this led
to stronger crosslinking at peaks 6 and 7 relative to peaks 4 and 5
at weak BPs, even though the peaks 4 and 5 are usually stronger.
The signal at position B of weak BPs is almost completely lost with
PRPF8 KD, which probably reflects the absence of intron lariats due
to perturbed splicing of introns with weak BPs (Fig. 7c). In conclusion,
our results suggest that pre-mRNA binding of spliceosomal
factors at peaks 4 and 5 closely correlates with BP strength, which
indicates that recognition of weak BPs might be more sensitive to
perturbed spliceosome function.
Discussion
Here we established spliceosome iCLIP to study the interactions
of endogenous snRNPs and accessory splicing factors on premRNAs.
We identified peaks of spliceosomal protein-pre-mRNA
interactions, which precisely overlap with crosslinking profiles of
15 splicing factors. Interestingly, the contacts of RBPs in peaks 4
and 5 do not overlap with any sequence motif, and thus the constrained
conformation of the larger spliceosomal complex appears
to act as a molecular ruler that positions each associated RBP on
pre-mRNAs at a specific distance from BPs. Moreover, the presence
of lariat-derived reads in spliceosome iCLIP identified >40,000
BPs that have canonical sequence and structural features. Due to
the precise alignment of splicing factors relative to the positions of
BPs, we could use their binding profiles to show that the assembly
of U2 snRNP is primarily coordinated by the computationally predicted
BPs, while alternative BPs, identified only by iCLIP or RNAseq,
are more rarely used. Finally, we reveal the major effect of the
position and strength of BPs on spliceosomal assembly, which can
explain why distally located or weak BPs are particularly sensitive
to perturbed spliceosome function with PRPF8 KD. These findings
demonstrate the broad utility of spliceosome iCLIP for transcriptome-wide
analysis of spliceosomal assembly on nascent RNAs, as
well as for monitoring the use of BPs.
The value of spliceosome iCLIP for identifying BPs. Both RNAseq
and iCLIP identify BPs by analyzing cDNAs derived from
intron lariats. Thus, the efficiency of these methods depends on the
abundance of intron lariats, which depends on the kinetics of lariat
debranching. Several studies demonstrated that lariats formed at
noncanonical BPs are less efficiently debranched21–23
, and therefore
these noncanonical BPs are expected to be more efficiently detected.
This is especially true for RNA-seq-based methods because they
monitor steady-state RNA levels. In contrast, iCLIP only captures
lariats in complex with spliceosomes, thus minimizing bias for
lariats that are stable after their release from the spliceosome. This
could explain why the BPs identified by iCLIP contain a stronger
consensus sequence than BPs identified from lariat-spanning reads
in RNA-seq. The further value of spliceosome iCLIP is that, in addition
to experiments under the medium condition that permit BP
identification through lariat-derived cDNAs, experiments under
the mild condition identify the SF3 complex and other U2 snRNPassociated
RBPs that crosslink at peaks 4 and 5. These can crucially
be used to independently validate the functional role of BPs in the
assembly of U2 snRNP. Thus, the use of spliceosome iCLIP under
both conditions, combined with computational modeling of BPs12
,
is well suited to studying the functionality of BPs.
The role of BP position and strength in spliceosomal assembly.
We show that BP position and the computationally defined strength
of BPs correlate with the relative binding of splicing factors around
BPs. This is exemplified by strong binding of SF3 components at
strong BPs, or BPs located close to 3′SS, while U2AF components
bind stronger to weak BPs, or BPs located further from 3′SS (Fig.
7d). In the cases of SF3B1, BUD13 and GPKOW, we observed
enriched binding at peaks 4 and 5 as well as 6 and 7, with reciprocal
changes between the two peak regions dependent on BP features
(Figs. 6 and 7). These RBPs are not known to bind at peaks 6 or 7,
and it is plausible that the signal at some peaks represents binding
of U2AF or other spliceosomal factors that are co-purified during
eCLIP. It is presently not possible to fully distinguish between direct
and indirect binding from eCLIP data, because purified protein–
RNA complexes have not been visualized after their separation on
SDS-PAGE gels in eCLIP13
. Nevertheless, it is clear that BP characteristics
determine the balance between binding of SF3 and associated
factors at peaks 4 and 5 and of U2AF and associated factors at
peaks 6 and 7. This suggests further study of RBP binding profiles
around BPs could unravel a BP ‘code’ that facilitates specific stages
of BP recognition and function.
In conclusion, spliceosome iCLIP monitors concerted premRNA
binding of many types of spliceosomal complexes with
nucleotide resolution, allowing their simultaneous study due to the
distinct position-dependent binding pattern of components acting
at multiple stages of the splicing cycle. The method can now be used
to study the endogenous spliceosome and BPs across tissues, species
and stages of development without the need for the protein tagging
used in yeast3,4
. Further, several spliceosomal components, including
U2AF1, SF3B1 and PRPF8, are targets for mutations in myeloid
neoplasms, retinitis pigmentosa and other diseases24
. Spliceosome
iCLIP could now be used to monitor global impacts of these mutations
on spliceosome assembly in human cells. More generally, our
study demonstrates the value of iCLIP for monitoring the positiondependent
assembly and dynamics of multi-protein complexes on
endogenous transcripts.
Online content
Any methods, additional references, Nature Research reporting
summaries, source data, statements of code and data availability and
associated accession codes are available at https://doi.org/10.1038/
s41594-019-0300-4.
Received: 6 December 2018; Accepted: 14 August 2019;
Published online: 30 September 2019
References
	1.	 Fica, S. M. & Nagai, K. Cryo-electron microscopy snapshots of the
spliceosome: structural insights into a dynamic ribonucleoprotein machine.
Nat. Struct. Mol. Biol. 24, 791–799 (2017).
	2.	 Wahl, M. C., Will, C. L. & Lührmann, R. The spliceosome: design principles
of a dynamic RNP machine. Cell 136, 701–718 (2009).
	3.	 Chen, W. et al. Transcriptome-wide interrogation of the functional intronome
by spliceosome profiling. Cell 173, 1031–1044 e13 (2018).
	4.	 Burke, J. E. et al. Spliceosome profiling visualizes operations of a dynamic
RNP at nucleotide resolution. Cell 173, 1014–1030 e17 (2018).
	5.	 Wickramasinghe, V. O. et al. Regulation of constitutive and alternative mRNA
splicing across the human transcriptome by PRPF8 is determined by 5’ splice
site strength. Genome Biol. 16, 201 (2015).
	6.	 König, J. et al. iCLIP reveals the function of hnRNP particles in splicing at
individual nucleotide resolution. Nat. Struct. Mol. Biol. 17, 909–915 (2010).
	7.	 Taggart, A. J. et al. Large-scale analysis of branchpoint usage across species
and cell lines. Genome Res. 27, 639–649 (2017).
	8.	 Pineda, J. M. B. & Bradley, R. K. Most human introns are recognized via
multiple and tissue-specific branchpoints. Genes Dev. 32, 577–591 (2018).
	9.	 Mercer, T. R. et al. Genome-wide discovery of human splicing branchpoints.
Genome Res. 25, 290–303 (2015).
	10.	Carissimi, C., Saieva, L., Gabanella, F. & Pellizzoni, L. Gemin8 is required for
the architecture and function of the survival motor neuron complex. J. Biol.
Chem. 281, 37009–37016 (2006).
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb 939
Articles NATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
	11.	Huppertz, I. et al. iCLIP: protein-RNA interactions at nucleotide resolution.
Methods 65, 274–287 (2014).
	12.	Paggi, J. M. & Bejerano, G. A sequence-based, deep learning model accurately
predicts RNA splicing branchpoints. RNA 24, 1647–1658 (2018).
	13.	Lee, F. C. Y. & Ule, J. Advances in CLIP technologies for studies of
protein-RNA interactions. Mol. Cell 69, 354–369 (2018).
	14.	Sugimoto, Y. et al. Analysis of CLIP and iCLIP methods for nucleotideresolution
studies of protein-RNA interactions. Genome Biol. 13, R67 (2012).
	15.	Haberman, N. et al. Insights into the design and interpretation of iCLIP
experiments. Genome Biol. 18, 7 (2017).
	16.	Van Nostrand, E. L. et al. A large-scale binding and functional map of human
RNA binding proteins. Preprint at bioRxiv https://doi.org/10.1101/179648
(2017).
	17.	Bessonov, S., Anokhina, M., Will, C. L., Urlaub, H. & Luhrmann, R. Isolation
of an active step I spliceosome and composition of its RNP core. Nature 452,
846–850 (2008).
	18.	Gozani, O., Feld, R. & Reed, R. Evidence that sequence-independent binding
of highly conserved U2 snRNP proteins upstream of the branch site is required
for assembly of spliceosomal complex A. Genes Dev. 10, 233–243 (1996).
	19.	Zarnack, K. et al. Direct Competition between hnRNP C and U2AF65
Protects the Transcriptome from the Exonization of Alu Elements. Cell 152,
453–466 (2013).
	20.	Zhang, X. et al. Structure of the human activated spliceosome in three
conformational states. Cell Res. 28, 307–322 (2018).
	21.	Jacquier, A. & Rosbash, M. RNA splicing and intron turnover are greatly
diminished by a mutant yeast branch point. Proc. Natl Acad. Sci. USA 83,
5835–5839 (1986).
	22.	Hesselberth, J. R. Lives that introns lead after splicing. Wiley Inter. Rev. RNA
4, 677–691 (2013).
	23.	Talhouarne, G. J. S. & Gall, J. G. Lariat intronic RNAs in the cytoplasm of
vertebrate cells. Proc. Natl Acad. Sci. USA 115, E7970–E7977 (2018).
	24.	Scotti, M. M. & Swanson, M. S. RNA mis-splicing in disease. Nat. Rev. Genet.
17, 19–32 (2016).
	25.	Lorenz, R. et al. ViennaRNA Package 2.0. Algorithms Mol. Biol. 6, 26 (2011).
Acknowledgements
We thank M. Llorian for help with the in vitro splicing reactions, K. Zarnack and G. Rot
for help with the data analyses and L. Strittmatter and members of the Ule lab for helpful
discussions and comments on the manuscript. This work was supported primarily by
the European Research Council (grant nos. 206726-CLIP and 617837-Translate) and
the Slovenian Research Agency (grant nos. P2-0209, Z7-3665 and J7-5460). C.R.S. was
supported by an Edmond Lily Safra Fellowship and a Sir Henry Dale Fellowship jointly
funded by the Wellcome Trust and the Royal Society (grant no. 215454/Z/19/Z). A.S.E.
is supported by the Biotechnology and Biological Sciences Research Council (grant no.
BB/M009513/1). A.M.C. is supported by a Wellcome Trust PhD Training Fellowship for
Clinicians (no. 110292/Z/15/Z). D.P. and V.O.W. were supported by Medical Research
Council grants (nos. MC_UU_12022/1 and MC_UU_12022/8 to A.R.V). L.P. was
supported by the National Institute of Neurological Disorders and Stroke of the National
Institutes of Health (NIH-NINDS) (grant no. R01 NS102451). The Francis Crick
Institute receives its core funding from Cancer Research UK (grant no. FC001002),
the UK Medical Research Council (grant no. FC001002) and the Wellcome Trust
(grant no. FC001002).
Author contributions
M.B., C.R.S. and J.U. conceived the project, designed the experiments and wrote the
manuscript with the assistance of all co-authors. M.B., C.R.S., Z.W., R.F. and A.S.E.
performed experiments with assistance from J.U., J.K. and C.W.S. N.H. performed most
of the computational analyses with assistance from C.R.S., T.C., R.F., A.M.C. and N.M.L.
V.O.W., D.P. and A.R.V. provided crosslinked pellets from wild-type and PRPF8-depleted
Cal51 cells. L.S. and L.P. developed and characterized the monoclonal antibody 18F6.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
s41594-019-0300-4.
Correspondence and requests for materials should be addressed to J.U.
Peer review information Anke Sparmann was the primary editor on this article and
managed its editorial process and peer review in collaboration with the rest of the
editorial team.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
© The Author(s), under exclusive licence to Springer Nature America, Inc. 2019
Nature Structural & Molecular Biology | VOL 26 | OCTOBER 2019 | 930–940 | www.nature.com/nsmb940
ArticlesNATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
Methods
Cell culture. Flp-In HEK293 T-REx cells were from ThermoFisher (R78007),
K562, HepG2 and standard HEK293 cells were obtained from the Francis Crick
Cell Services Science Technology Platform, and Cal51 breast adenocarcinoma
cells were obtained from the line originators26
. All cell lines tested negative for
Mycoplasma contamination. HEK293 and HepG2 were cultured in DMEM
with 10% FBS (ThermoFisher) and 1× penicillin-streptomycin (ThermoFisher).
K562 cells were cultured in RPMI 1640 (IMDM, ATCC) with 10% FBS and 1×
penicillin-streptomycin. Cal51 cells were cultured in DMEM (ThermoFisher) with
10% FBS and 1× penicillin-streptomycin.
To generate a plasmid encoding 3× Flag epitope-tagged SmB, the SmB cDNA
was amplified using Phusion High-Fidelity DNA polymerase (NEB) with primers
carrying the KpnI and NotI restriction enzymes sites and cloned using a Rapid
DNA Ligation Kit (ThermoFisher Scientific) into a pcDNA5/FRT/TO vector
modified to encode 3× Flag peptide upstream of the multiple cloning site. To
produce stable cell lines expressing this construct, the pcDNA5/FRT/TO plasmid
with 3× Flag epitope-tagged SmB was co-transfected with pOG44 plasmid into
Flp-In HEK293 T-REx cells (ThermoFisher, R78007). Cells stably expressing
these proteins were selected by culturing in DMEM containing 10% FBS, 3 μg ml−1
Blasticidin S HCl, 200 μg ml−1
Hygromycine (InvivoGen). Flp-In 293 T-REx cells
(Life Technologies) were cultured in DMEM with 10% FBS, 3 μg ml−1
Blasticidin S
HCl (Life Technologies), 50 μg ml−1
Zeocin (Life Technologies). Doxycycline was
added to media 24 h before sample preparation to induce construct expression.
Cal51 breast adenocarcinoma cells were prepared as described previously5
.
For siRNA-mediated depletion of PRPF8, Cal51 cells were transfected using
DharmaFECT1 (Dharmafect) with 25 nM siRNA targeting human PRPF8.
Transfected cells were harvested 54 h later, exposed to UV-C light and used for
iCLIP as described below. For the collection of samples from different stages of the
cell cycle, Cal51 cells were synchronized in G1/S by standard double thymidine
block. Briefly, cells were treated with 1.5 mM thymidine for 8 h, washed and
released for 8 h, then treated again with thymidine for a further 8 h. Cells were also
collected 3 h (S-phase) and 7 h (G2) after release from the thymidine block.
Antibody production. For production of the anti-SmB/B′ monoclonal antibody
18F6, Balb/c females were primed with Immuneasy adjuvant (Qiagen) and 25 mg
of 6× His-SmB purified recombinant proteins. Following two boosts at 2-week
intervals, SP2 myeloma cells were fused with mouse splenocytes and hybridoma
supernatants were analyzed onto antigen-coated aminosilane modified slides
using an LS400 Scanner (Tecan) and the GenePix Pro v.4.1 software as described
previously10
. Hybridoma cells were subcloned by limiting dilution and further
screened by ELISA, Western blot and immunofluorescence analysis of HeLa cells.
In vitro splicing. For in vitro splicing reactions, a C6orf10 minigene construct
containing exon 8 and 9 and 150 nucleotides of the intron around both splice sites
was produced (Fig. 2b). The minigene plasmid was linearized and transcribed
in vitro using T7 polymerase with 32
P-UTP. The transcribed RNA was then
subjected to in vitro splicing reactions using HeLa nuclear extract. HeLa nuclear
extract was depleted of endogenous ATP by pre-incubation and, for each reaction,
10 ng of RNA was incubated with 60% HeLa nuclear extract at 30 °C with or
without additional 0.5 mM ATP for 1 h in a 20 µl reaction. Afterward, the reaction
mixture was UV-crosslinked at 100 mJ cm–2
and stored at −80 °C until further use.
To visualize the splicing reaction products, proteinase K was added to the reaction
mixture for 30 min at 37 °C. The resulting RNA was phenol-extracted, precipitated
and subjected to gel electrophoresis on a 5% polyacrylamide-urea gel.
Spliceosome iCLIP protocol. For each experiment, three biological replicate
samples of cDNA libraries were prepared (Supplementary Tables 2 and 3).
The iCLIP method was done as previously described11
, with the following
modifications. Crosslinked cells or tissue were dissociated in the lysis buffer
according to the stringency conditions (stringent, medium, mild; Supplementary
Table 1) followed by sonication, low RNase I (AM2295, 100 U µl−1
, ThermoFisher)
digestion and centrifugation. RNase at low concentration ensured that cDNAs
are of optimal size for comprehensive crosslink determination15
. For denaturing,
high-stringency experiment11
, M2 anti-Flag antibody (Sigma) was used against
the 3× Flag-SmB protein that had been stably integrated into HEK293 Flp-In
cells (Supplementary Fig. 1c). Urea buffer (6 M) was first used to lyse cell pellets,
before being diluted down 1:9 with a Tween-20-containing IP buffer to allow for
immunopurification without denaturing of the M2 anti-Flag antibody, and then
proceeded as described previously15
.
Standard iCLIP protocol11
was used for Cal51 cells under mild and medium
stringency conditions, and for the in vitro splicing reactions under mild
conditions, while an updated protocol was used for HEK293, HepG2 and K562
cells27
. For SmB/B′ immunopurification anti-SmB/B′ antibodies 12F5 (sc-
130670, Santa Cruz Biotechnology for Cal51 cells, and S0698, Sigma-Aldrich
for HEK293, HepG2 and K562 cells) or 18F6 (as hybridoma supernatant,
generated as described previously10
) were used, which are different clones
from the same immunization. These antibodies behave identically under
immunopurification conditions (Supplementary Fig. 1d). For spliceosome
iCLIP from in vitro splicing reactions (Supplementary Fig. 2c,d), lysates
were incubated with 50 μl monoclonal anti-SmB/B′ antibody 18F6, and for
immunoprecipitations from cell lysates, 12F5 anti-SmB/B′ antibody was used.
The antibody was bound to 100 μl protein G Dynabeads (ThermoFisher)
under rotation at 4 °C followed by washing. As described previously, following
immunopurification, RNA 3′ end dephosphorylation, ligation of the adapter
5′-rAppAGATCGGAAGAGCGGTTCAG/ddC/-3′ to the 3′ end and 5′ end
radiolabeling, protein–RNA complexes were size-separated by SDS-PAGE
and transferred onto nitrocellulose membrane. The regions corresponding
to 28–180 kDa were excised from the membrane to isolate the bound RNA
by proteinase K treatment. RNAs were reverse-transcribed in all experiments
using SuperScript III or IV reverse transcriptase (ThermoFisher) and custom
indexed primers (Supplementary Table 2). Resulting cDNAs were subjected
to electrophoresis on a 6% TBE-urea gel (ThermoFisher) for size selection.
Purified cDNAs were circularized, linearized and amplified for high-throughput
sequencing.
Identification of protein crosslink sites around splice sites, in particular at
the peaks 4 and 5, was most efficient under the mild purification condition
(Supplementary Fig. 2a). This condition was therefore used for the analysis of
spliceosomal assembly on PRPF8 KD in Cal51 cells (Fig. 2a), and in the in vitro
splicing reactions in HeLa nuclear extract (Fig. 2b). For the identification of BPs,
we additionally used the medium condition, since it increases the frequency
of cDNAs truncating at peak B (Supplementary Fig. 2a). For this purpose, we
performed spliceosome iCLIP under medium purification conditions from Cal51
cells synchronized in G1, S and G2 phase. To maximize cDNA coverage, data from
all synchronized cells were merged with the control Cal51 cells under the mild
condition for BP identification.
Mapping of Sm iCLIP reads. We mapped iCLIP data to the GRCh38 primary
assembly and GENCODE v.27 gene annotations using STAR (v.2.2.1).
Experimental and random barcode sequences of iCLIP sequenced reads were
removed before mapping (Supplementary Table 2). Following mapping, we used
random barcodes to quantify the number of unique cDNAs at each genomic
position by collapsing cDNAs with the same random barcode that mapped
to the same starting position to a single cDNA. For analysis of crosslinking to
snRNAs, we first mapped to a transcriptome of all annotated snRNA sequences
in GENCODE v.27 using Bowtie2 (v.2.3.4.3) and kept the primary alignment.
Unmapped reads were then mapped with STAR as previously described and
intersected with GENCODE v.27 for subtype analysis, with reads from Bowtie2
being added to the total snRNA count. For spliceosome iCLIP with the C6orf10
in vitro splicing substrate, sequence reads were first mapped to the unspliced
substrate and the remaining reads were mapped to the spliced substrate allowing
no mismatches. The nucleotide preceding the iCLIP cDNAs was used to define the
crosslink sites in all analyses.
Mapping of eCLIP reads. For eCLIP sequencing data for all RBPs, we
used GENCODE (GRCh38.p7) genome assembly and the STAR alignment
(v.2.4.2a) using the following parameters from ENCODE pipeline:
STAR --runThreadN 8 --runMode alignReads --genomeDir GRCh38
Gencode v25 --genomeLoad LoadAndKeep --readFilesIn read1, read2,
--readFilesCommand zcat --outSAMunmapped Within –outFilterMultimapNmax
1 --outFilterMultimapScoreRange 1 --outSAMattributes All --outSAMtype
BAM Unsorted –outFilterType BySJout --outFilterScoreMin 10 --alignEndsType
EndToEnd --outFileNamePrefix outfile.
For the PCR duplicates removal, we used a python script ‘barcode collapse
pe.py’ available on GitHub (https://github.com/YeoLab/gscripts/releases/tag/1.0),
which is part of the ENCODE eCLIP pipeline (https://www.encodeproject.org/
pipelines/ENCPL357ADL/).
Normalization of crosslink positions for their visualization in the form of RNA
maps. RNA maps and heatmaps were produced by summarizing the cDNA counts
at each nucleotide using the previously developed RNA maps pipeline15,28
relative to
exon-intron and intron–exon boundaries and BPs on pre-mRNAs. The definition
of intronic start and end positions was based on Ensembl v.75. Only introns longer
than 300 nucleotides were used to draw RNA maps to avoid detection of any RBPs
that recognize 5′SS of introns.
In cases where we compared the relative positions of crosslinking peaks
between RBPs, we regionally normalized the summarized crosslinking of each
RBP relative to the average crosslinking of the same RBP across the region 100
nucleotides upstream and 50 nucleotides downstream of the evaluated splice sites
or BPs. Normalized values were then used to visualize the crosslinking in the form
of RNA maps (Fig. 2 and Supplementary Figs. 5 and 6). The same normalization
was then used to plot heatmaps, by plotting mean values of normalized RNA maps
for each peak in the following regions: peak 4, −29..−23 nucleotides and peak
5, −21..−17 nucleotides relative to BP, peak 6, −11..−5 nucleotides and peak 7,
−3..−1 nucleotides relative to 3′SS. Every RBP was then normalized by the mean
across all the peaks to visualize crosslinking enrichment between the groups on the
same scale across all RBPs (Figs. 6 and 7 and Supplementary Fig. 7).
To assess the role of BP characteristics on spliceosomal RBP assembly (Figs.
4, 6 and 7), we only examined the introns containing the 31,167 BPs that were
Nature Structural & Molecular Biology | www.nature.com/nsmb
Articles NATuRe STRuCTuRAL & MoLeCuLAR BIoLogy
identified both computationally and by iCLIP, which are probably the most
reliable. We divided BPs into ten categories based on BP position or score, and
then normalized the summarized crosslinking of each RBP in each of the ten
BP categories relative to the average crosslinking of the same RBP across the
region 100 nucleotides upstream and 50 nucleotides downstream of all the 31,167
evaluated BPs.
For visualization of spliceosome iCLIP crosslinks along the C6orf10 in vitro
splicing substrate and product (Fig. 2b and Supplementary Fig. 2e) we first
summed the cDNA starts at each nucleotide position and then normalized the
counts by the average number of cDNA starts in the intronic region 101..150
nucleotides relative to the 5′SS of the unspliced substrate. For the unspliced
substrate normalized cDNA counts were logarithmized (log2) and data with
log2(normalized number of cDNA starts) ≥1 were plotted. For the spliced product
normalized cDNA counts were plotted.
Identification and comparison of BPs. It has been shown that the spliceosomal
C complexes harbor a salt-resistant RNP core containing U2, U5 and U6 snRNAs
as well as the splicing intermediates including lariats that withstand treatment
with 1 M NaCl, whereas the spliceosomal B complexes probably dissociated under
high-salt conditions17
. This could explain why the medium purification condition
is more suited than the mild condition to enrich for lariat cDNAs truncating at
position B (Supplementary Fig. 2a). It is conceivable that the medium spliceosome
iCLIP condition strongly enriches spliceosomal C complexes, which are most
effective for lariat detection. In contrast, the mild condition is expected to enrich
additional B complexes that contain large amounts of SF3 components and have
low proportion of lariats, in agreement with the strong enrichment of peaks 4 and
5 (Supplementary Fig. 2a). To identify the maximal diversity of BPs, we therefore
pooled spliceosome iCLIP data produced under mild and medium purification
conditions from Cal51 cells.
To identify BPs we used the spliceosome iCLIP reads that ended precisely at
the ends of introns (we considered only introns that end in the AG dinucleotide)
after removal of the 3′ adapter. We noticed that these reads had a 3.5× increased
frequency of mismatches on the A as the first nucleotide compared to remaining
iCLIP reads (Supplementary Fig. 3a), indicating that these mismatches may have
resulted from truncation at the three-way-junction formed at the BP (Fig. 2c). We
therefore trimmed the first nucleotide from the read if it contained a mismatch
at the first position that corresponded to a genomic adenosine. We then used
spliceosome iCLIP from Cal51 cells to identify all reads that ended precisely at the
ends of introns and defined the position where these reads started and assessed the
random barcode nucleotides that are present at the beginning of each iCLIP read
to count the number of unique cDNAs at each position. The nucleotide preceding
the read start corresponds to the position where cDNAs truncated during the
reverse transcription, and we selected the genomic A that had the highest number
of truncated cDNAs as the candidate BP. If two positions with equal number of
cDNAs were found, we selected the one closer to the 3′SS. Together, this identified
43,637 BPs.
We also attempted to use truncated cDNAs from PRPF8 eCLIP for the
discovery of BPs but found that the number of cDNAs overlapping with intron
ends was much smaller than in spliceosome iCLIP, and was insufficient for BP
discovery. This is probably because of the high amount of nonspecific background
signal in PRPF8 eCLIP, which leads to a lower proportion of cDNAs that align to
the BPs.
The Bedtools Intersect command using option –u was used to compare BP
coordinates from spliceosome iCLIP to the BPs identified in previous studies. We
restricted this comparison to introns where BPs were detected by all three datasets
(iCLIP, RNA-seq and computational prediction).
To define a single ‘computational BP’ per intron, the BP positions
computationally predicted for each intron in hg19 were obtained from http://
bejerano.stanford.edu/labranchor/, and the top-scoring BP in each intron was
used. To define a single ‘RNA-seq BP’ per intron, we used the BP with most lariatspanning
reads in each intron.
Analysis of pairing probability. Computational predictions of the secondary
structure were performed by RNAfold function from Vienna Package (https://
www.tbi.univie.ac.at/RNA/) with default parameters25
. The RNAfold results are
provided in a customized format, where brackets are representing the doublestranded
region on the RNA and dots are used for unpaired nucleotides. We
measured the density of pairing probability by summing the paired positions into
a single vector.
Identification of RBPs overlapping with spliceosomal peaks. For RBP
enrichment in Fig. 3, we used the eCLIP data from the ENCODE consortium16
,
together with available iCLIP experiments from our lab (all listed in Supplementary
Dataset 4), to see if any of the proteins are enriched in the region of spliceosomal
peaks. In total, this included 157 eCLIP samples of 68 RBPs in the HepG2 cell line,
and 89 RBPs in the K562 cell line, and iCLIP samples of 18 RBPs from different
cell lines (Supplementary Dataset 4). Next, we intersected cDNA starts from each
sample to the −100 to +50 nucleotide region relative to the 3′SS and used it as
control for each of the following peaks: Peak 4 (−23..−29 nucleotides relative to
BP), Peak 5 (−21..−17 nucleotides relative to BP), Peak B (−1..1 nucleotides relative
to BP), Peak A (−1..1 nucleotide relative to 5’SS), Peak 6 (−11..−10 nucleotides
relative to 3′SS), Peak 7 (−3..−2 nucleotides relative to 3′SS). The positions of these
peaks were determined based on crosslink enrichments in spliceosome iCLIP.
Statistics. All statistical analyses were performed in the R software environment
(v.3.1.3 and v.3.3.2, https://www.r-project.org).
Reporting Summary. Further information on research design is available in the
Nature Research Reporting Summary linked to this article.
Data availability
The spliceosome iCLIP data generated and analyzed during the current study are
available on EBI ArrayExpress under the accession number E-MTAB-8182 and
are also available in raw and processed format on https://imaps.genialis.com/iclip.
Additional datasets used in this study are listed in Supplementary Dataset 4. Source
data for Fig. 1c are available online. Other data are available upon request.
Code availability
The code to identify BPs from spliceosome iCLIP reads is publicly available at the
GitHub repository (https://github.com/nebo56/branch-point-detection-2).
References
	26.	Gioanni, J. et al. Establishment and characterisation of a new tumorigenic cell
line with a normal karyotype derived from a human breast adenocarcinoma.
Br. J. Cancer 62, 8–13 (1990).
	27.	Blazquez, L. et al. Exon junction complex shapes the transcriptome by
repressing recursive splicing. Mol. Cell 72, 496–509 e9 (2018).
	28.	Chakrabarti, A., Haberman, N., Praznik, A., Luscombe, N. M. & Ule, J. Data
science issues in studying protein–RNA interactions with CLIP technologies.
Annu. Rev. Biomed. Data Sci. 1, 235–261 (2018).
Nature Structural & Molecular Biology | www.nature.com/nsmb