Nature  Macmillan Publishers Ltd 1998
8
letters to nature
806 NATURE |VOL 391 |19 FEBRUARY 1998
and 10 Na3HP2O7. FV solution also contained 0.2 NaF and 0.1 Na3VO4. Rarely,
irreversible current rundown still occurred with FVPP. The total Na+
concen-
tration of all cytoplasmic solutions was adjusted to 30 mM with NaOH, and pH
was adjusted to 7.0 with N-methylglucamine (NMG) or HCl. PIP2 liposomes
(20­200 nm) were prepared by sonicating 1 mM PIP2 (Boehringer Mannheim)
in distilled water. Reconstituted monoclonal PIP2 antibody (Perspective
Biosystems, Framingham, MA) was diluted 40-fold into experimental solution.
Current­voltage relations of all currents reversed at EK and showed charac-
teristic rectification, mostly owing to the presence of Na+
in FVPP and possibly
alsoresidualpolyamines.Currentrecordspresented(measuredat30 C,-30 mV
holding potential) are digitized strip-chart recordings. Purified bovine
brain Gbg29
was diluted just before application such that the final detergent
(CHAPS) concentration was 5 M. Detergent-containing solution was washed
away thoroughly before application of PIP2, because application of phos-
pholipid vesicles in the presence of detergent usually reversed the effects of
Gbg; presumably, Gbg can be extracted from membranes by detergent plus
phospholipids.
Molecular biology. R188Q mutation was constructed by insertion of the
mutant oligonucleotides between the Bsm1 and BglII sites of pSPORT­
ROMK1 (ref. 11). A polymerase chain reaction (PCR) fragment (amino acids
180­391) from pSPORT­ROMK1 R188Q mutant was subcloned into pGEX-
2T vector (Pharmacia) for expression of R188Q mutant protein of GST­RKC.
The construction, expression and purification of GST­IKC (amino acids 182­
428 of IRK1), GST­GKC (180­462 of GIRK1), GST­IKN (1­86 of IRK1) have
been described21,22
.
In vitro PIP2 binding assay.
3
H-PIP2 in chloroform-methanol (1:1) (American
Radiolabeled Chemicals; 0.4 Ci nM-1
specific activity) was dried under N2
and sonicated in 100 l phosphate buffered saline (PBS) to form pure 3
H-PIP2
liposomes. Purified GST fusion protein (100 nM) was incubated with 3
H-PIP2
(0.2­1 M) and precipitated by glutathione 4B-Sepharose beads. After 1 wash
with PBS, the precipitates were dissolved in SDS gel loading buffer and counted
in a beta-scintillation counter using a window for 3
H. The bound 3
H
radioactivity was typically in the range 2­8% of the total added. For co-
immunoprecipitation, 25% PIP2 or PIP in 75% phosphatidylcholine (PC)
background (30 g PIP2 or PIP (Boehringer Mannheim) and 90 g
phosphatidylcholine (Sigma)), both in chloroform, were dried down together
and sonicated in 300 l PBS to form mixed liposome. GST fusion proteins were
first incubated with 25% PIP2 or PIP liposome (100 M) and PIP2 antibodies
(1:100 dilution) for 2 h and with protein A­Sepharose for a further 30 min.
After one washwith PBS, the immunoprecipitates were separated by 10% SDS­
PAGE, probed with specific antibodies21,22
, and visualized by ECL (Amersham).
Each experiment was performed at least twice with similar results. The relative
amount of immunoreactivity in each lane was quantified by serial dilutions of
sample21
.
Received 6 June; accepted 13 October 1997.
1. McNicholas, C. M., Wang, W., Ho, K., Hebert, S. C. & Giebisch, G. Regulation of ROMK1 K+
channel
activity involves phosphorylation processes. Proc. Natl Acad. Sci. USA 91, 8077­8081 (1994).
2. Fakler, B., Brandle, U., Glowatzki, E., Zenner, H.-P. & Ruppersberg, J. P. Kir2.1 inward rectifier K+
channels are regulated independently by protein kinases and ATP hydrolysis. Neuron 13, 1413­1420
(1994).
3. Kubo, Y., Reuveny, E., Slesinger, P. A., Jan, Y. N. & Jan, L. Y. Primary structure and functional
expression of a rat G-protein-coupled muscarinic potassium channel. Nature 364, 802­806 (1993).
4. Dascal, N. et al. Atrial G protein-activated K+
channel: expression cloning and molecular properties.
Proc. Natl Acad. Sci. USA 90, 10235­10239 (1993).
5. Krapivinsky, G. et al. The G-protein-gated atrial K+
channel IKACh is a heteromultimer of two inwardly
rectifying K+
-channel proteins. Nature 374, 135­141 (1995).
6. Lesage, F. et al. Molecular properties of neuronal G protein-activated inwardly rectifying K+
channels.
J. Biol. Chem. 270, 28660­28667 (1995).
7. Furukawa, T., Yamane, T., Terai, T., Katayama, Y. & Hiraoka, M. Functional linkage of the cardiac ATP-
sensitive K+
channel to actin cytoskeleton. Pflugers Arch. 431, 504­512 (1996).
8. Hilgemann, D. W. & Ball, R. Regulationof cardiac Na+
, Ca2+
exchange and KATP potassium channels by
PIP2. Science 273, 956­959 (1996).
9. Fukami, K. et al. Antibody to phosphatidylinositol 4,5-bisphosphate inhibits oncogene-induced
mitogenesis. Proc. Natl Acad. Sci. USA 85, 9057­9061 (1988).
10. Kubo, Y., Baldwin, T. J., Jan, Y. N. & Jan, L. Y. Primary structure and functional expression of a mouse
inward rectifier potassium channel. Nature 362, 127­133 (1993).
11. Ho, K. et al. Cloning and expression of an inwardly rectifying ATP-regulated potassium channel.
Nature 362, 31­38 (1993).
12. Sui, J. L., Chan, K. W. & Logothetis, D. E. Na+
activation of the muscarinic K+
channel by a G-protein-
independent mechanism. J. Gen. Physiol. 109, 381­390 (1996).
13. Chan, K. W. et al. A recombinant inwardly rectifying potassium channel coupled to GTP-binding
proteins. J. Gen. Physiol. 107, 381­397 (1996).
14. Zhang, X., Jefferson, A. B., Auethavekiat, V. & Majerus, P. W. The protein deficient in Lowe syndrome
is a phosphatidylinositol-4,5-bisphosphate 5-phosphatase. Proc. Natl Acad. Sci. USA 92, 4853­4856
(1995).
15. Fukami, K., Endo, T., Imamura, M. & Takenawa, T. a-Actinin and vinculin are PIP2-binding proteins
involved in signaling by tyrosine kinase. J. Biol. Chem. 269, 1518­1522 (1994).
16. Fan, Z. & Makielski, J. C. Anionic phospholipids activate ATP-sensitive potassium channels. J. Biol.
Chem. 272, 5388­5395 (1997).
17. Schacht, J. Inhibition by neomycin of polyphosphoinositide turnover in subcellular fractions of
guinea-pig cerebral cortex in vitro. J. Neurochem. 27, 1119­1124 (1976).
18. Kim, J., Mosior, M., Chung, L. A., Wu, H. & McLaughlin, S. Binding of peptides with basic residues to
membrane containing acidic phospholipids. Biophys. J. 60, 135­148 (1991).
19. Harlan, J. E., Yoon, H. S., Hajduk, P. J. & Fesik, S. W. Structural characterization of the interaction
between a pleckstrin homology domain and phosphatidylinositol 4,5-bisphosphate. Biochemistry 34,
9859­9864 (1995).
20. Reuveny, E. et al. Activation of the cloned muscarinic potassium channel by G protein bg subunits.
Nature 370, 143­146 (1994).
21. Huang, C.-L., Slesinger, P. A., Casey, P. J., Jan, Y. N. & Jan, L. Y. Evidence that direct binding of Gbg to
the GIRK1 protein-gated inwardly rectifying K+
channel is important for channel activation. Neuron
15, 1133­1143 (1995).
22. Huang, C.-L., Jan, Y. N. & Jan, L. Y. Binding of Gbg to multiple regions of G protein-gated inward
rectifier K+
channels. FEBS Lett. 405, 291­298 (1997).
23. Krapivinsky, G., Krapivinsky, L., Wickman, K. & Clapham, D. E. Gbg binds directly to the G protein-
gated K+
channel, IKACh. J. Biol. Chem. 270, 29059­29062 (1995).
24. Janmey, P. A. Phosphoinositides and calcium as regulators of cellular actin assembly and disassembly.
Annu. Rev. Physiol. 56, 169­191 (1994).
25. Penniston, J. T. Plasma membrane Ca2+
-pumping ATPases. Ann. NY Acad. Sci. 402, 291­303 (1982).
26. Pitcher, J. A., Touhara, K., Payne, E. S. & Lefkowitz, R. J. Pleckstrin homology domain-mediated
membrane association and activation of the b-adrenergic receptor kinase requires coordinate
interaction with Gbg and lipid. J. Biol. Chem. 270, 11707­11710 (1995).
27. Tagliaalatela, M., Wible, B. A., Caporaso, R. & Brown, A. M. Specification of the pore properties by the
carboxyl terminus of inward rectifying K+
channels. Science 264, 844­847 (1994).
28. Clapham, D. E. & Neer, E. J. New roles for G protein bg-dimers in transmembrane signaling. Nature
365, 403­406 (1993).
29. Casey, P. J., Graziano, M. P. & Gilman, A. G. G protein bg subunits from bovine brain and retina:
equivalent catalytic support of ADP-ribosylation of a subunit by pertussis toxin but differential
interactions with Gsa. Biochemistry 28, 611­616 (1989).
Acknowledgements. We thank E. Phan for technical assistance; I. Bezprozvanny, C. Dessauer, D. Logo-
thetis, C.-C. Lu, O. Moe, S. Muallem and H. Yin for discussions and advice; L. Jan for GIRK1 and ROMK1
antibodies; C. Dessauer and A. Gilman for Gai1; P. Casey for Gbg; and R. Alpern for support and
encouragement. This work was supported by grants from the NKF of Texas (C.L.H.) and from the AHA
and NIH (D.W.H.).
Correspondence and requests for materials should be addressed to C.L.H. (e-mail: chuan1@mednet.
swmed.edu).
Potent and specific
genetic interference by
double-stranded RNA in
Caenorhabditis elegans
Andrew Fire*, SiQun Xu*, Mary K. Montgomery*,
Steven A. Kostas*, Samuel E. Driver & Craig C. Mello
* Carnegie Institution of Washington, Department of Embryology,
115 West University Parkway, Baltimore, Maryland 21210, USA
 Biology Graduate Program, Johns Hopkins University,
3400 North Charles Street, Baltimore, Maryland 21218, USA
 Program in Molecular Medicine, Department of Cell Biology,
University of Massachusetts Cancer Center, Two Biotech Suite 213,
373 Plantation Street, Worcester, Massachusetts 01605, USA
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Experimental introduction of RNA into cells can be used in
certain biological systems to interfere with the function of an
endogenous gene1,2
. Such effects have been proposed to result
from a simple antisense mechanism that depends on hybridiza-
tion between the injected RNA and endogenous messenger RNA
transcripts. RNA interference has been used in the nematode
Caenorhabditis elegans to manipulate gene expression3,4
. Here we
investigate the requirements for structure and delivery of the
interfering RNA. To our surprise, we found that double-stranded
RNA was substantially more effective at producing interference
than was either strand individually. After injection into adult
animals, purified single strands had at most a modest effect,
whereas double-stranded mixtures caused potent and specific
interference. The effects of this interference were evident in
both the injected animals and their progeny. Only a few molecules
of injected double-stranded RNA were required per affected cell,
arguing against stochiometric interference with endogenous
Nature  Macmillan Publishers Ltd 1998
8
letters to nature
NATURE |VOL 391 |19 FEBRUARY 1998 807
mRNA and suggesting that there could be a catalytic or amplifica-
tion component in the interference process.
Despite the usefulness of RNA interference in C. elegans, two
features of the process have been difficult to explain. First, sense and
antisense RNA preparations are each sufficient to cause
interference3,4
. Second, interference effects can persist well into the
next generation, even though many endogenous RNA transcripts
are rapidly degraded in the early embryo5
. These results indicate a
fundamental difference in behaviour between native RNAs (for
example, mRNAs) and the molecules responsible for interference.
Wesought to test thepossibility that this contrast reflects anunderlying
difference in RNA structure. RNA populations to be injected are
generally prepared using bacteriophage RNA polymerases6
. These
polymerases, although highly specific, produce some random or
ectopic transcripts. DNA transgene arrays also produce a fraction of
aberrant RNA products3
. From these facts, we surmised that the
interfering RNA populations might include some molecules with
double-stranded character. To test whether double-stranded character
might contribute to interference, we further purified single-stranded
RNAs and compared interference activities of individual strands with
the activity of a deliberately prepared double-stranded hybrid.
The unc-22 gene was chosen for initial comparisons of activity.
unc22 encodes an abundant but nonessential myofilament pro-
tein7­9
. Several thousand copies of unc-22 mRNA are present in each
Table 1 Effects of sense, antisense and mixed RNAs on progeny of injected animals
Gene segment Size
(kilobases)
Injected RNA F1 phenotype
...................................................................................................................................................................................................................................................................................................................................................................
unc-22 unc-22-null mutants: strong twitchers7,8
unc22A* Exon 21­22 742 Sense Wild type
Antisense Wild type
Sense  antisense Strong twitchers (100%)
unc22B Exon 27 1,033 Sense Wild type
Antisense Wild type
Sense  antisense Strong twitchers (100%)
unc22C Exon 21­22 785 Sense  antisense Strong twitchers (100%)
...................................................................................................................................................................................................................................................................................................................................................................
fem-1 fem-1-null mutants: femal (no sperm)13
fem1A Exon 10 531 Sense Hermaphrodite (98%)
Antisense Hermaphrodite ( 98%)
Sense  antisense Female (72%)
fem1B Intron 8 556 Sense  antisense Hermaphrodite ( 98%)
...................................................................................................................................................................................................................................................................................................................................................................
unc-54 unc-54-null mutants: paralysed7,11
unc54A Exon 6 576 Sense Wild type (100%)
Antisense Wild type (100%)
Sense  antisense Paralysed (100%)§
unc54B Exon 6 651 Sense Wild type (100%)
Antisense Wild type (100%)
Sense  antisense Paralysed (100%)§
unc54C Exon 1­5 1,015 Sense  antisense Arrested embryos and larvae (100%)
unc54D Promoter 567 Sense  antisense Wild type (100%)
unc54E Intron 1 369 Sense  antisense Wild type (100%)
unc54F Intron 3 386 Sense  antisense Wild type (100%)
...................................................................................................................................................................................................................................................................................................................................................................
hlh-1 hlh-1-null mutants: lumpy-dumpy larvae16
hlh1A Exons 1­6 1,033 Sense Wild type ( 2% lpy-dpy)
Antisense Wild type ( 2% lpy-dpy)
Sense  antisense Lpy-dpy larvae ( 90%)k
hlh1B Exons 1­2 438 Sense  antisense Lpy-dpy larvae ( 80%)k
hlh1C Exons 4­6 299 Sense  antisense Lpy-dpy larvae ( 80%)k
hlh1D Intron 1 697 Sense  antisense Wild type ( 2% lpy-dpy)
...................................................................................................................................................................................................................................................................................................................................................................
myo-3-driven GFP transgenes
myo-3::NLS::gfp::lacZ Makes nuclear GFP in body muscle
gfpG Exons 2­5 730 Sense Nuclear GFP­LacZ pattern of parent strain
Antisense Nuclear GFP­LacZ pattern of parent strain
Sense  antisense Nuclear GFP­LacZ absent in 98% of cells
lacZL Exon 12­14 830 Sense  antisense Nuclear GFP­LacZ absent in 95% of cells
myo-3::MtLS::gfp Makes mitochondrial GFP in body muscle
gfpG Exons 2­5 730 Sense Mitochondrial-GFP pattern of parent strain
Antisense Mitochondrial-GFP pattern of parent strain
Sense  antisense Mitochondrial-GFP absent in 98% of cells
lacZL Exon 12­14 830 Sense  antisense Mitochondrial-GFP pattern of parent strain
...................................................................................................................................................................................................................................................................................................................................................................
Each RNA was injected into 6­10 adult hermaphrodites (0:5 106
­1 106
molecules into each gonad arm). After 4­6 h (to clear prefertilized eggs from the uterus), injected animals were
transferred and eggs collected for 20­22 h. Progeny phenotypes were scored upon hatching and subsequently at 12-24-h intervals.
* to obtain a semiquantitative assessment of the relationship between RNA dose and phenotypic response, we injected each unc22A RNA preparation at a series of different concentrations
(see figure in Supplementary information for details). At the highest dose tested (3:6 106
molecules per gonad), the individual sense and antisense unc22A preparations produced some
visible twitching (1% and 11% of progeny, respectively). Comparable doses of double-stranded unc22A RNA produced visible twitching in all progeny, whereas a 120-fold lower dose of
double-stranded unc22A RNA produced visible twitching in 30% of progeny.  unc22C also carries the 43-nucleotide intron between exons 21 and 22.  fem1A carries a portion (131
nucleotides) of intron 10. § Animals in the first affected broods (layed 4­24 h after injection) showed movement defects indistinguishablefrom those of unc-54-null mutants. A variable fraction
of these animals (25%­75%) failed to lay eggs (another phenotype of unc-54-null mutants), whereas the remainder of the paralysed animals did lay eggs. This may indicate incomplete
interference with unc-54 activity in vulval muscles. Animals from later broods frequently show a distinct partial loss-of-function phenotype, with contractility in a subset of body-wall muscles.
k Phenotypes produced by RNA-mediated interference with hlh-1 included arrested embryos and partially elongated L1 larvae (the hlh-1-null phenotype). These phenotypes were seen in
virtually all progeny after injection of double-stranded hlh1A and in about half of the affected animals produced after injection of double-stranded hlh1B and double-stranded hlhlC. A set of
less severe defects was seen in the remainder of the animals produced after injection of double-stranded hlh1B and double-stranded hlh1C). The less severe phenotypes are characteristic
of partial loss of function of hlh-1 (B. Harfe and A.F., unpublished observations).  the host for these injections, strain PD4251, expresses both mitochondrial GFP and nuclear GFP­LacZ (see
Methods). This allows simultaneous assay for interference with gfp (seen as loss of all fluorescence) and with lacZ (loss of nuclear fluorescence). The table describes scoring of animals as
L1 larvae. Double-stranded gfpG caused a loss of GFP in all but 0­3 of the 85 body muscles in these larvae. As these animals mature to adults, GFP activity was seen in 0­5 additional body-
wall muscles and in the 8 vulval muscles. Lpy-dpy, lumpy-dumpy.
Nature  Macmillan Publishers Ltd 1998
8
letters to nature
808 NATURE |VOL 391 |19 FEBRUARY 1998
striated muscle cell3
. Semiquantitative correlations between unc-22
activity and phenotype of the organism have been described8
:
decreases in unc-22 activity produce an increasingly severe twitch-
ing phenotype, whereas complete loss of function results in the
additional appearance of muscle structural defects and impaired
motility.
Purified antisense and sense RNAs covering a 742-nucleotide
segment of unc-22 had only marginal interference activity, requiring
a very high dose of injected RNA to produce any observable effect
(Table 1). In contrast, a sense­antisense mixture produced highly
effective interference with endogenous gene activity. The mixture
was at least two orders of magnitude more effective than either
single strand alone in producing genetic interference. The lowest
dose of the sense­antisense mixture that was tested, 60,000
molecules of each strand per adult, led to twitching phenotypes in
an average of 100 progeny. Expression of unc-22 begins in embryos
containing 500 cells. At this point, the original injected material
would be diluted to at most a few molecules per cell.
The potent interfering activity of the sense­antisense mixture
could reflect the formation of double-stranded RNA (dsRNA) or,
conceivably, some other synergy between the strands. Electrophoretic
analysis indicated that the injected material was predominantly
double-stranded. The dsRNA was gel-purified from the annealed
mixture and found to retain potent interfering activity. Although
annealing before injection was compatible with interference, it was
not necessary. Mixing of sense and antisense RNAs in low-salt
concentrations (under conditions of minimal dsRNA formation) or
rapid sequential injection of sense and antisense strands were
sufficient to allow complete interference. A long interval ( 1 h)
between sequential injections of sense and antisense RNA resulted
in a dramatic decrease in interfering activity. This suggests that
injected single strands may be degraded or otherwise rendered
inaccessible in the absence of the opposite strand.
A question of specificity arises when considering known cellular
responses to dsRNA. Some organisms have a dsRNA-dependent
protein kinase that activates a panic-response mechanism10
. Con-
ceivably, our sense­antisense synergy might have reflected a non-
specific potentiation of antisense effects by such a panic mechanism.
This is not the case: co-injection of dsRNA segments unrelated to
unc-22 did not potentiate the ability of single unc-22-RNA strands
to mediate inhibition (data not shown). We also investigated
whether double-stranded structure could potentiate interference
activity when placed in cis to a single-stranded segment. No such
potentiation was seen: unrelated double-stranded sequences located
5 or 3 of a single-stranded unc-22 segment did not stimulate
interference. Thus, we have only observed potentiation of inter-
ference when dsRNA sequences exist within the region of homology
with the target gene.
The phenotype produced by interference using unc-22 dsRNA
was extremely specific. Progeny of injected animals exhibited
behaviour that precisely mimics loss-of-function mutations in
unc-22. We assessed target specificity of dsRNA effects using three
additional genes with well characterized phenotypes (Fig. 1, Table
1). unc-54 encodes a body-wall-muscle heavy-chain isoform of
myosin that is required for full muscle contraction7,11,12
; fem-1
encodes an ankyrin-repeat-containing protein that is required in
hermaphrodites for sperm production13,14
; and hlh-1 encodes a C.
elegans homologue of myoD-family proteins that is required for
proper body shape and motility15,16
. For each of these genes,
injection of related dsRNA produced progeny broods exhibiting
G
fem-1
unc-54
hlh-1
gfp fusions
unc-22
5.0kb
1.0kb
AB
1.0kb
A B
C
F D E
A
B C
D
A
BC
gfpNLS
G L
lacZmyo-3 5'
gfpmyo-3 5' MtLS
Mitochondrial
Nuclear
Figure 1 Genes used to study RNA-mediated genetic interference in C. elegans.
Intron­exon structure for genes used to test RNA-mediated inhibition are shown
(grey and filled boxes, exons; open boxes, introns; patterned and striped boxes,
5 and 3 untranslated regions. unc-22. ref. 9, unc-54, ref.12, fem-1, ref.14, and hlh-1,
ref.15). Each segment of a gene tested for RNA interference is designated with the
name of the gene followed by a single letter (for example, unc22C). These
segments are indicated by bars and upper-case letters above and below each
gene. Segments derived from genomic DNA are shown above the gene; seg-
ments derived from cDNA are shown below the gene. NLS, nuclear-localization
sequence; MtLS, mitochondrial localization sequence.
a
b
c
d
e
f
g
h
i
Control RNA (ds-unc22A) ds-gfpG RNA ds-lacZL RNA
AdultL1Adult
Figure 2 Analysis of RNA-interference effects in individual cells.Fluorescence
micrographs show progeny of injected animals from GFP-reporter strain PD4251.
a­c, Progeny of animals injected with a control RNA (double-stranded
(ds)-unc22A). a, Young larva, b, adult, c, adult body wall at high magnification.
These GFP patterns appear identical to patterns in the parent strain, with
prominent fluorescence in nuclei (nuclear-localized GFP­LacZ) and mitochondria
(mitochondrially targeted GFP). d­f, Progeny of animals injected with ds-gfpG.
Only a single active cell is seen in the larva in d, whereas the entire vulval
musculature expresses active GFP in the adult animal in e. f, Two rare GFP-
positive cells in an adult: both cells express both nuclear-targeted GFP­LacZ and
mitochondrial GFP. g­i, Progeny of animals injected with ds-lacZL RNA:
mitochondrial-targeted GFP seems unaffected, while the nuclear-targeted GFP­
LacZ is absent from almost all cells (for example, see larva in g). h, A typical adult,
with nuclear GFP­LacZ lacking in almost all body-wall muscles but retained in
vulval muscles. Scale bars represent 20 m.
Nature  Macmillan Publishers Ltd 1998
8
letters to nature
NATURE |VOL 391 |19 FEBRUARY 1998 809
the known null-mutant phenotype, whereas the purified single RNA
strands produced no significant interference. With one exception,
all of the phenotypic consequences of dsRNA injection were those
expected from interference with the corresponding gene. The
exception (segment unc54C which led to an embryonic- and
larval-arrest phenotype not seen with unc-54-null mutants) was
illustrative. This segment covers the highly conserved myosin-
motor domain, and might have been expected to interfere with
activity of other highly related myosin heavy-chain genes17
. The
unc54C segment has been unique in our overall experience to date:
effects of 18 other dsRNA segments (Table 1; and our unpublished
observations) have all been limited to those expected from pre-
viously characterized null mutants.
The pronounced phenotypes seen following dsRNA injection
indicate that interference effects are occurring in a high fraction of
cells. The phenotypes seen in unc-54 and hlh-1 null mutants, in
particular, are known to result from many defective muscle cells11,16
.
To examine interference effects of dsRNA at a cellular level, we used
a transgenic line expressing two different green fluorescent protein
(GFP)-derived fluorescent-reporter proteins in body muscle. Injec-
tion of dsRNA directed to gfp produced marked decreases in the
fraction of fluorescent cells (Fig. 2). Both reporter proteins were
absent from the affected cells, whereas the few cells that were
fluorescent generally expressed both GFP proteins.
The mosaic pattern observed in the gfp-interference experiments
was nonrandom. At low doses of dsRNA, we saw frequent inter-
ference in the embryonically derived muscle cells that are present
when the animal hatches. The interference effect in these differ-
entiated cells persisted throughout larval growth: these cells pro-
duced little or no additional GFP as the affected animals grew. The
14 postembryonically derived striated muscles are born during early
larval stages and these were more resistant to interference. These
cells have come through additional divisions (13­14 divisions
versus 8­9 divisions for embryonic muscles18,19
). At high concen-
trations of gfp dsRNA, we saw interference in virtually all striated
body-wall muscles, with occasional lone escaping cells, including
cells born during both embryonic and postembryonic development.
The non-striated vulval muscles, which are born during late larval
development, appeared to be resistant to interference at all tested
concentrations of injected dsRNA.
We do not yet know the mechanism of RNA-mediated inter-
ference in C. elegans. Some observations, however, add to the debate
about possible targets and mechanisms.
First, dsRNA segments corresponding to various intron and
promoter sequences did not produce detectable interference
(Table 1). Although consistent with interference at a post-transcrip-
tional level, these experiments do not rule out interference at the
level of the gene.
Second, we found that injection of dsRNA produces a pro-
nounced decrease or elimination of the endogenous mRNA tran-
script (Fig. 3). For this experiment, we used a target transcript (mex-
3) that is abundant in the gonad and early embryos20
, in which
straightforward in situ hybridization can be performed5
. No endo-
genous mex-3 mRNA was observed in animals injected with a
dsRNA segment derived from mex-3. In contrast, animals into
which purified mex-3 antisense RNA was injected retained sub-
stantial endogenous mRNA levels (Fig. 3d).
Third, dsRNA-mediated interference showed a surprising ability
to cross cellular boundaries. Injection of dsRNA (for unc-22, gfp or
lacZ) into the body cavity of the head or tail produced a specific and
robust interference with gene expression in the progeny brood
(Table 2). Interference was seen in the progeny of both gonad
arms, ruling out the occurrence of a transient `nicking' of the gonad
a b
c d
Figure 3 Effects of mex-3 RNA interference on levels of the endogenous mRNA.
Interference contrast micrographs show in situ hybridization in embryos. The
1,262-nt mex-3 cDNA clone20
was divided into two segments, mex-3A and mex-
3B, with a short (325-nt) overlap (similar results were obtained in experiments with
no overlap between interfering and probe segments). mex-3B antisense or
dsRNA was injected into the gonads of adult animals, which were fed for 24 h
before fixation and in situ hybridization (ref. 5; B. Harfe and A.F., unpublished
observations). The mex-3B dsRNA produced 100% embryonic arrest, whereas
90% of embryos produced after the antisense injections hatched. Antisense
probes for the mex-3A portion of mex-3 were used to assay distribution of the
endogenous mex-3 mRNA (dark stain). four-cell-stage embryos are shown;
similar results were observed from the one to eight cell stage and in the germ
line of injected adults. a, Negative control showing lack of staining in the absence
of the hybridization probe. b, Embryo from uninjected parent (showing normal
pattern of endogenous mex-3 RNA20
). c, Embryo from a parent injected with
purified mex-3B antisense RNA. These embryos (and the parent animals) retain
the mex-3 mRNA, although levels may be somewhat less than wild type. d,
Embryo from a parent injected with dsRNA corresponding to mex-3B; no mex-3
RNA is detected. Each embryo is approximately 50 m in length.
Table 2 Effect of site of injection on interference in injected animals and their progeny
dsRNA Site of injection Injected-animal phenotype Progeny phenotype
...................................................................................................................................................................................................................................................................................................................................................................
None Gonad or body cavity No twitching No twitching
None Gonad or body cavity Strong nuclear and mitochondrial GFP expression Strong nuclear and mitochondrial GFP expression
unc22B Gonad Weak twitchers Strong twitchers
unc22B Body-cavity head Weak twitchers Strong twitchers
unc22B Body-cavity tail Weak twitchers Strong twitchers
gfpG Gonad Lower nuclear and mitochondiral GFP expression Rare or absent nuclear and mitochondiral GFP expression
gfpG Body-cavity tail Lower nuclear and mitochondrial GFP expression Rare or absent nuclear and mitochondrial GFP expression
lacZL Gonad Lower nuclear GFP expression Rare or absent nuclear-GFP expression
lacZL Body-cavity tail Lower nuclear GFP expresison Rare or absent nuclear-GFP expression
...................................................................................................................................................................................................................................................................................................................................................................
The GFP-reporter strain PD4251, which expresses both mitochondrialGFPand nuclear GFP­LacZ, was used for injections. The use of this strain allowed simultaneousassay for interference
with gfp (fainteroverall fluorescence), lacZ (loss of nuclear fluorescence) and unc-22 (twitching). Body-cavity injections into the tail region were carried out to minimize accidental injection of
the gonad; equivalent results have been observed with injections into the anterior body cavity. An equivalent set of injections was also performed into a single gonad arm. The entire progeny
broods showed phenotypes identical to those described in Table 1. This included progeny of both injected and uninjected gonad arms. Injected animals were scored three days after
recovery and showed somewhat less dramatic phenotypes than their progeny. This could be partly due to the persistence of products already present in the injected adult. After injection of
double-stranded unc22B, a fraction of the injected animals twitch weakly under standard growth conditions (10 out of 21 animals). Levamisole treatment led to twitching of 100% (21 out of 21)
of these animals. Similar effects (not shown) were seen with double-stranded unc22A. Injections of double-stranded gfpG or double-stranded lacZL produced a dramatic decrease (but not
elimination) of the corresponding GFP reporters. In some cases, isolated cells or parts of animals retained strong GFP activity. These were most frequently seen in the anterior region and
around the vulva. Injections of double-stranded gfpG and double-stranded lacZL produced no twitching, whereas injections of double-stranded unc22A produced no change in the GFP-
fluorescence pattern.
Nature  Macmillan Publishers Ltd 1998
8
letters to nature
810 NATURE |VOL 391 |19 FEBRUARY 1998
in these injections. dsRNA injected into the body cavity or gonad of
young adults also produced gene-specific interference in somatic
tissues of the injected animal (Table 2).
The use of dsRNA injection adds to the tools available for
studying gene function in C. elegans. In particular, it should now
be possible functionally to analyse many interesting coding regions21
for which no specific function has been defined. Although the
effects of dsRNA-mediated interference are potent and specific we
have observed several limitations that should be taken into account
when designing RNA-interference-based experiments. First, a
sequence shared between several closely related genes may interfere
with several members of the gene family. Second, it is likely that a
low level of expression will resist RNA-mediated interference for
some or all genes, and that a small number of cells will likewise
escape these effects.
Genetic tools are available for only a few organisms. Double-
stranded RNA could conceivably mediate interference more
generally in other nematodes, in other invertebrates, and, poten-
tially, in vertebrates. RNA interference might also operate in plants:
several studies have suggested that inverted-repeat structures or
characteristics of dsRNA viruses are involved in transgene-
dependent co-suppression in plants22,23
.
There are several possible mechanisms for RNA interference in C.
elegans. A simple antisense model is not likely: annealing between a
few injected RNA molecules and excess endogenous transcripts
would not be expected to yield observable phenotypes. RNA-
targeted processes cannot, however, be ruled out, as they could
include a catalytic component. Alternatively, direct RNA-mediated
interference at the level of chromatin structure or transcription
could be involved. Interactions between RNA and the genome,
combined with propagation of changes along chromatin, have been
proposed in mammalian X-chromosome inactivation and plant-
gene co-suppression22,24
. If RNA interference in C. elegans works by
such a mechanism, it would be new in targeting regions of the
template that are present in the final mRNA (as we observed no
phenotypic interference using intron or promoter sequences).
Whatever their target, the mechanisms underlying RNA inter-
ference probably exist for a biological purpose. Genetic interference
by dsRNA could be used by the organism for physiological gene
silencing. Likewise, the ability of dsRNA to work at a distance from
the site of injection, and particularly to move into both germline
and muscle cells, suggests that there is an effective RNA-transport
mechanism in C. elegans.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Methods
RNA synthesis and microinjection. RNA was synthesized from phagemid
clones by using T3 and T7 polymerase6
. Templates were then removed with two
sequential DNase treatments. When sense-, antisense-, and mixed-RNA popu-
lations were to be compared, RNAs were further purified by electrophoresis on
low-gelling-temperature agarose. Gel-purified products appeared to lack many
of the minor bands seen in the original `sense' and `antisense' preparations.
Nonetheless, RNA species comprising 10% of purified RNA preparations
would not have been observed. Without gel purification, the `sense' and
`antisense' preparations produced notable interference. This interference activ-
ity was reduced or eliminated upon gel purification. In contrast, sense-plus-
antisense mixtures of gel-purified and non-gel-purified RNA preparations
produced identical effects.
Sense/antisense annealing was carried out in injection buffer (ref. 27) at
37 C for 10­30 min. Formation of predominantly double-stranded material
was confirmed by testing migration on a standard (nondenaturing) agarose gel:
for each RNA pair, gel mobility was shifted to that expected for dsRNA of the
appropriate length. Co-incubation of the two strands in a lower-salt buffer
(5 mM Tris-Cl, pH 7.5, 0.5 mM EDTA) was insufficient for visible formation of
dsRNA in vitro. Non-annealed sense-plus-antisense RNAs for unc22B and gfpG
were tested for RNA interference and found to be much more active than the
individual single strands, but twofold to fourfold less active than equivalent
preannealed preparations.
After preannealing of the single strands for unc22A, the single electro-
phoretic species, corresponding in size to that expected for the dsRNA, was
purified using two rounds of gel electrophoresis. This material retained a high
degree of interference activity.
Except where noted, injection mixes were constructed so that animals would
receive an average of 0:5 106
to 1:0 106
RNA molecules. For comparisons
of sense, antisense, and double-stranded RNA activity, equal masses of RNA
were injected (that is, dsRNA was used at half the molar concentration of the
single strands). Numbers of molecules injected per adult are approximate and
based on the concentration of RNA in the injected material (estimated from
ethidium bromide staining) and the volume of injected material (estimated
from visible displacement at the site of injection). It is likely that this volume
will vary several-fold between individual animals; this variability would not
affect any of the conclusions drawn from this work.
Analysis of phenotypes. Interference with endogenous genes was generally
assayed in a wild-type genetic background (N2). Features analysed included
movement, feeding, hatching, body shape, sexual identity, and fertility.
Interference with gfp (ref. 25) and lacZ activity was assessed using C. elegans
strain PD4251. This strain is a stable transgenic strain containing an integrated
array (ccIs4251) made up of three plasmids: pSAK4 (myo-3 promoter driving
mitochondrially targeted GFP); pSAK2 (myo-3 promoter driving a nuclear-
targeted GFP­LacZ fusion); and a dpy-20 subclone26
as a selectable marker.
This strain produces GFP in all body muscles, with a combination of
mitochondrial and nuclear localization. The two distinct compartments are
easily distinguished in these cells, allowing easy distinction between cells
expressing both, either, or neither of the original GFP constructs.
Gonadal injectionwas done as described27
. Body-cavity injections followed a
similar procedure, with needle insertion into regions of the head and tail
beyond the positions of the two gonad arms. Injection into the cytoplasm of
intestinal cells is also effective, and may be the least disruptive to the animal.
After recovery and transfer to standard solid media, injected animals were
transferred to fresh culture plates at 16-h intervals. This yields a series of
semisynchronous cohorts in which it was straightforward to identify pheno-
typic differences. A characteristic temporal pattern of phenotypic severity is
observed among progeny. First, there is a short `clearance' interval in which
unaffected progeny are produced. These include impermeable fertilized eggs
present at the time of injection. Second, after the clearance period, individuals
that show the interference phenotype are produced. Third, after injected
animals have produced eggs for several days, gonads can in some cases
`revert' to produce incompletely affected or phenotypically normal progeny.
Received 16 September; accepted 24 November 1997.
1. Izant, J. & Weintraub, H. Inhibition of thymidine kinase gene expression by antisense RNA: a
molecular approach to genetic analysis. Cell 36, 1007­1015 (1984).
2. Nellen, W. & Lichtenstein, C. What makes an mRNA anti-sense-itive? Trends Biochem. Sci. 18, 419­
423 (1993).
3. Fire, A., Albertson, D., Harrison, S. & Moerman, D. Production of antisense RNA leads to effective and
specific inhibition of gene expression in C. elegans muscle. Development 113, 503­514 (1991).
4. Guo, S. & Kemphues, K. par-1, a gene required for establishing polarity in C. elegans embryos, encodes
a putative Ser/Thr kinase that is asymmetrically distributed. Cell 81, 611­620 (1995).
5. Seydoux, G. & Fire, A. Soma-germline asymmetry in the distributions of embryonic RNAs in
Caenorhabditis elegans. Development 120, 2823­2834 (1994).
6. Ausubel, F. et al. Current Protocols in Molecular Biology (Wiley, New York, 1990).
7. Brenner, S. The genetics of Caenorhabditis elegans. Genetics 77, 71­94 (1974).
8. Moerman, D. & Baillie, D. Genetic organization in Caenorhabditis elegans: fine structure analysis of
the unc-22 gene. Genetics 91, 95­104 (1979).
9. Benian, G., L'Hernault, S. & Morris, M. Additional sequence complexity in the muscle gene, unc-22,
and its encoded protein, twitchin, of Caenorhabiditis elegans. Genetics 134, 1097­1104 (1993).
10. Proud, C. PKR: a new name and new roles. Trends Biochem. Sci. 20, 241­246 (1995).
11. Epstein, H., Waterston, R. & Brenner, S. A mutant affecting the heavy chain of myosin in C. elegans. J.
Mol. Biol. 90, 291­300 (1974).
12. Karn, J., Brenner, S. & Barnett, L. Protein structural domains in the C. elegans unc-54 myosin heavy
chain gene are not separated by introns. Proc. Natl Acad. Sci. USA 80, 4253­4257 (1983).
13. Doniach, T. & Hodgkin, J. A. A sex-determining gene, fem-1, required for both male and
hermaphrodite development in C. elegans. Dev. Biol. 106, 223­235 (1984).
14. Spence, A., Coulson, A. & Hodgkin, J. The product of fem-1, a nematode sex-determining gene,
contains a motif found in cell cycle control proteins and receptors for cell­cell interactions. Cell 60,
981­990 (1990).
15. Krause, M., Fire, A., Harrison, S., Priess, J. & Weintraub, H. CeMyoD accumulation defines the body
wall muscle cell fate during C. elegans embryogenesis. Cell 63, 907­919 (1990).
16. Chen, L., Krause, M., Sepanski, M. & Fire, A. The C. elegans MyoD homolog HLH-1 is essential for
proper muscle function and complete morphogenesis. Development 120, 1631­1641 (1994).
17. Dibb, N. J., Maruyama, I. N., Krause, M. (Author: OK?) & Karn, J. Sequence analysis of the complete
Caenorhabditis elegans myosin heavy chain gene family. J. Mol. Biol. 205, 603­613 (1989).
18. Sulston, J., Schierenberg, E., White, J. & Thomson, J. The embryonic cell lineage of the nematode
Caenorhabditis elegans. Dev. Biol. 100, 64­119 (1983).
19. Sulston, J. & Horvitz, H. Postembyonic cell lineages of the nematode Caenorhabiditis elegans. Dev.
Biol. 82, 41­55 (1977).
Nature  Macmillan Publishers Ltd 1998
8
letters to nature
NATURE |VOL 391 |19 FEBRUARY 1998 811
20. Draper, B. W., Mello, C. C., Bowerman, B., Hardin, J. & Priess, J. R. MEX-3 is a KH domain protein
that regulates blastomere identity in early C. elegans embryos. Cell 87, 205­216 (1996).
21. Sulston, J. et al. The C. elegans genome sequencing project: a beginning. Nature 356, 37­41 (1992).
22. Matzke, M. & Matzke, A. How and why do plants inactivate homologous (trans) genes? Plant Physiol.
107, 679­685 (1995).
23. Ratcliff, F., Harrison, B. & Baulcombe, D. A similarity between viral defense and gene silencing in
plants. Science 276, 1558­1560 (1997).
24. Latham, K. X chromosome imprinting and inactivation in the early mammalian embryo. Trends
Genet. 12, 134­138 (1996).
25. Chalfie, M., Tu, Y., Euskirchen, G., Ward, W. & Prasher, D. Green fluorescent protein as a marker for
gene expression. Science 263, 802­805 (1994).
26. Clark, D., Suleman, D., Beckenbach, K., Gilchrist, E. & Baillie, D. Molecular cloning and character-
ization of the dpy-20 gene of C. elegans. Mol. Gen. Genet. 247, 367­378 (1995).
27. Mello, C. & Fire, A. DNA transformation. Methods Cell Biol. 48, 451­482 (1995).
Supplementary information is available on Nature's World-Wide Web site (http://www.nature.com) or
as paper copy from Mary Sheehan at the London editorial office of Nature.
Acknowledgements. We thank A. Grishok, B. Harfe, M. Hsu, B. Kelly, J. Hsieh, M. Krause, M. Park,
W. Sharrock, T. Shin, M. Soto and H. Tabara for discussion. This work was supported by the NIGMS
(A.F.) and the NICHD (C.M.), and by fellowship and career awards from the NICHD (M.K.M.), NIGMS
(S.K.), PEW charitable trust (C.M.), American Cancer Society (C.M.), and March of Dimes (C.M.).
Correspondence and requests for materials should be addressed to A.F. (e-mail: fire@mail1.ciwemb.edu).
Roleofthehistonedeacetylase
complex in acute
promyelocytic leukaemia
Richard J. Lin*, Laszlo Nagy*, Satoshi Inoue,
Wenlin Shao§, Wilson H. Miller Jr§ & Ronald M. Evans*
* Howard Hughes Medical Institute, and The Salk Institute for Biological
Studies, La Jolla, California 92037, USA
 Graduate Program in Molecular Pathology, University of California San Diego,
School of Medicine, La Jolla, California 92093, USA
§ Lady Davis Institute for Medical Research, McGill University Departments of
Oncology and Medicine, Montreal, Quebec, Canada H3T 1E2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Non-liganded retinoic acid receptors (RARs) repress transcrip-
tion of target genes by recruiting the histone deacetylase com-
plex1­3
through a class of silencing mediators termed SMRTor N-
CoR4,5
. Mutant forms of RAR , created by chromosomal translo-
cations with either the PML (for promyelocytic leukaemia)6­8
or
the PLZF (for promyelocytic leukaemia zinc finger)9,10
locus, are
oncogenic and result in human acute promyelocytic leukaemia
(APL). PML­RAR APL patients achieve complete remission
following treatments with pharmacological doses of retinoic
acids (RA); in contrast, PLZF­RAR patients respond very
poorly, if at all11
. Here we report that the association of these
two chimaeric receptors with the histone deacetylase (HDAC)
complex helps to determine both the development of APL and the
ability of patients to respond to retinoids. Consistent with these
observations, inhibitors of histone deacetylase dramatically
potentiate retinoid-induced differentiation of RA-sensitive, and
restore retinoid responses of RA-resistant, APL cell lines. Our
findings suggest that oncogenic RARs mediate leukaemogenesis
through aberrant chromatin acetylation, and that pharmaco-
logical manipulation of nuclear receptor co-factors may be a
useful approach in the treatment of human disease.
Because both PML­RARa and PLZF­RARa inhibit normal
retinoid signalling6­14
, we reasoned that identification of factors
associated with these proteins might provide mechanistic insights
into their oncogenic functions. PLZF­RARa retains the autono-
mous repression domain, the BTB/POZ (for bric-a`-brac/tramtrack/
broad complex, poxvirus and zinc-finger) domain, from PLZF15
.
Because deletion of this domain abolishes the biological functions
of PLZF­RARa in vivo16,17
, we investigated whether it might
associate directly with components of the nuclear receptor co-
repressor complex1­3
. By using an in vitro interaction assay, we
found that radiolabelled full-length mSin3A and histone deacetylase
1 (HDAC1), but not mSin3B, were specifically retained on matrix-
bound fusion proteins of glutathione S-transferase with the BTB/
POZ domain of PLZF (GST­PLZF; Fig. 1a). Results from a yeast
two-hybrid assay showed that PLZF interacts with all known
components of the co-repressor complex in vivo (Fig. 1b). We
further mapped the PLZF interaction domain in mSin3A to the
paired amphipathic helix 1 (PAH1, residues 112­192) by a mam-
malian two-hybrid assay (Fig. 1c). Finally, using a co-immunopre-
cipitation assay from nuclear extracts of transfected CV1 cells, we
confirmed that PLZF, SMRT, mSin3A and HDAC1 form a complex
in mammalian cells (Fig. 1d). These results, together with the
finding that SMRT interacts with another BTB/POZ oncoprotein,
LAZ3/BCL6 (ref. 18), demonstrated that this family of transcrip-
tional factors recruits histone deacetylases to repress transcription
and implicates histone deacetylases in cellular transformation.
By using a yeast two-hybrid assay, we demonstrated that PLZF­
RARa interacts directly with both SMRT and mSin3A, whereas
PML­RARa interacts only with SMRT19
(Fig. 2B). Most impor-
tantly, we showed using a co-immunoprecipitation assay that
HDAC1 exists in a complex with either PLZF­RARa or PML­
RARa in transfected CV1 cells (Fig. 2C). Furthermore, a similar
assay using nuclear extracts of the NB4 cells established from a
patient with t(15:17) APL20
indicated that endogenous HDAC1 can
be co-precipitated with an anti-PML antibody (Fig. 2D). The
presence of endogenous PML­RARa was confirmed by immuno-
blotting analyses using anti-PML and anti-RARa antibodies (data
Figure 1 Association of co-repressors and HDAC1 with PLZF. a, SDS­PAGE
analysis of 35
S-labelled mSin3A, mSin3B or HDAC1 proteins retained on
immobilized GST­PLZF affinity matrices. I, 20% input; C, GST control; B, bound.
b, Interactions between PLZF and full-length SMRT, mSin3A or HDAC1 in a yeast
two-hybrid assay. c, Interactions between GAL4­DBD fusions of different PAHs of
mSin3A and VP16 fusion of PLZF are analysed by a mammalian two-hybrid assay
in CV1 cells. d, PLZF associates with the histone deacetylase complex in vivo.
CV1 cells were transfected with either vectors only (mock) or plasmids encoding
HA­PLZF, SMRT, mSin3A and HDAC1 (HA­PLZF) and nuclear extracts were
immunoprecipitated (IP) using anti-HA antibodies followed by immunoblotting
analyses using antibodies against SMRT, mSin3A and HDAC1. In lane 1, 100 g
of nuclear extract was applied to ascertain the positions of blotted proteins
(input).