Content uploaded by C. David Allis
Author content
All content in this area was uploaded by C. David Allis on Mar 25, 2015
Content may be subject to copyright.
69. Y. Habu, T. Kakutani, J. Paszkowski, Curr. Opin. Genet.
Dev. 11, 215 (2001).
70. M. Wassenegger, Plant Mol. Biol. 43, 203 (2000).
71. M. A. Matzke, A. J. Matzke, J. M. Kooter, Science 293,
1080 (2001).
72. J. Bender, Trends Biochem. Sci. 23, 252 (1998).
73. E. U. Selker, Cell 97, 157 (1999).
74. M. N. Raizada, M. I. Benito, V. Walbot, Plant J. 25,79
(2001).
75. R. F. Ketting, T. H. Haverkamp, H. G. van Luenen, R. H.
Plasterk, Cell 99, 133 (1999).
76. H. Tabara et al., Cell 99, 123 (1999).
77. B. H. Ramsahoye et al., Proc. Natl. Acad. Sci. U.S.A.
97, 5237 (2000).
78. P. Svoboda, P. Stein, H. Hayashi, R. M. Schultz, De-
velopment 127, 4147 (2000).
79. C. Cogoni et al., EMBO J. 15, 3153 (1996).
80. G. Faugeron, Curr. Opin. Microbiol. 3, 144 (2000).
81. L. Jackson-Grusby et al., Nature Genet. 27, 31 (2001).
82. J. P. Vielle-Calzada, R. Baskar, U. Grossniklaus, Nature
404, 91 (2000).
83. P. S. Springer, D. R. Holding, A. Groover, C. Yordan,
R. A. Martienssen, Development 127, 1815 (2000).
84. J. P. Vielle-Calzada et al., Genes Dev. 13, 2971 (1999).
85. R. Vinkenoog et al., Plant Cell 12, 2271 (2000).
86. S. Adams, R. Vinkenoog, M. Spielman, H. G. Dickinson,
R. J. Scott, Development 127, 2493 (2000).
87. M. Byrne, M. Timmermans, C. Kidner, R. Martienssen,
Curr. Opin. Plant Biol. 4, 38 (2001).
88. E. B. Cambareri, R. Aisner, J. Carbon, Mol. Cell. Biol.
18, 5465 (1998).
89. P. SanMiguel et al., Science 274, 765 (1996).
90. R. Mauricio, Nature Rev. Genet. 2, 370 (2001).
91. P. Cubas, C. Vincent, E. Coen, Nature 401, 157 (1999).
92. R. Martienssen, Curr. Opin. Genet. Dev. 8, 240
(1998).
93. Z. J. Chen, C. S. Pikaard, Genes Dev. 11, 2124 (1997).
94. L. Comai et al., Plant Cell 12, 1551 (2000).
95. H. S. Lee, Z. J. Chen, Proc. Natl. Acad. Sci. U.S.A. 98,
6753 (2001).
96. We thank E. Selker, E. Richards, V. Chandler, S. Kaep-
pler, S. Jacobsen, and J. Bender for communicating
results prior to publication, two anonymous referees
for suggestions for improvement, and our colleagues
for many interesting discussions. R.M. and V.C. re-
ceive grant support from the NSF (DBI 1057338).
REVIEW
Translating the Histone Code
Thomas Jenuwein
1
and C. David Allis
2
Chromatin, the physiological template of all eukaryotic genetic information, is
subject to a diverse array of posttranslational modifications that largely
impinge on histone amino termini, thereby regulating access to the underly-
ing DNA. Distinct histone amino-terminal modifications can generate syner-
gistic or antagonistic interaction affinities for chromatin-associated proteins,
which in turn dictate dynamic transitions between transcriptionally active or
transcriptionally silent chromatin states. The combinatorial nature of histone
amino-terminal modifications thus reveals a “histone code” that considerably
extends the information potential of the genetic code. We propose that this
epigenetic marking system represents a fundamental regulatory mechanism
that has an impact on most, if not all, chromatin-templated processes, with
far-reaching consequences for cell fate decisions and both normal and patho-
logical development.
Genomic DNA is the ultimate template of our
heredity. Yet despite the justifiable excitement
over the human genome, many challenges re-
main in understanding the regulation and trans-
duction of genetic information (1). It is unclear,
for example, why the number of protein-coding
genes in humans, now estimated at ⬃35,000,
only doubles that of the fruit fly Drosophila
melanogaster. Is DNA alone then responsible
for generating the full range of information that
ultimately results in a complex eukaryotic or-
ganism, such as ourselves?
We favor the view that epigenetics, im-
posed at the level of DNA-packaging proteins
(histones), is a critical feature of a genome-
wide mechanism of information storage and
retrieval that is only beginning to be under-
stood. We propose that a “histone code” ex-
ists that may considerably extend the infor-
mation potential of the genetic (DNA) code.
We review emerging evidence that histone
proteins and their associated covalent modi-
fications contribute to a mechanism that can
alter chromatin structure, thereby leading to
inherited differences in transcriptional “on-
off ” states or to the stable propagation of
chromosomes by defining a specialized high-
er order structure at centromeres. Under the
assumption that a histone code exists, at least
in some form, we discuss potential mecha-
nisms for how such a code is “read” and
translated into biological functions.
Throughout this review, we have chosen
epigenetic phenomena and underlying mecha-
nisms in two general categories: chromatin-
based events leading to either gene activation or
gene silencing. In particular, we center our dis-
cussion on examples where differences in “on-
off ” transcriptional states are reflected by dif-
ferences in histone modifications that are either
“euchromatic” (on) or “heterochromatic” (off )
(Fig. 1A). We also point out that, despite many
elegant genetic and biochemical insights into
chromatin function and gene regulation in the
budding yeast Saccharomyces cerevisiae, some
of the heterochromatic mechanisms (e.g., HP1-
based gene silencing) discussed here do not
exist in an obvious form in this organism. Thus,
we will need to pursue other model systems,
such as Schizosaccharomyces pombe, Caeno-
rhabditis elegans, Drosophila, and mice, to
“crack” the histone code.
Chromatin Template and Histone
Code
In the nuclei of all eukaryotic cells, genomic
DNA is highly folded, constrained, and com-
pacted by histone and nonhistone proteins in
a dynamic polymer called chromatin. For
example, chromosomal regions that remain
transcriptionally inert are highly condensed
in the interphase nucleus and remain cytolog-
ically visible as heterochromatic foci or as the
“Barr body,” which is the inactive X chromo-
some in female mammalian cells (2). The
distinct levels of chromatin organization are
dependent on the dynamic higher order struc-
turing of nucleosomes, which represent the
basic repeating unit of chromatin. In each
nucleosome, roughly two superhelical turns
of DNA wrap around an octamer of core
histone proteins formed by four histone part-
ners: an H3-H4 tetramer and two H2A-H2B
dimers (3). Histones are small basic proteins
consisting of a globular domain and a more
flexible and charged NH
2
-terminus (histone
“tail”) that protrudes from the nucleosome. It
remains unclear how nucleosomal arrays con-
taining linker histone (H1) then twist and fold
this chromatin fiber into increasingly more
compacted filaments leading to defined high-
er order structures.
Central to our current thinking is that
chromatin structure plays an important regu-
latory role and that multiple signaling path-
ways converge on histones (4). Although
histone proteins themselves come in generic
or specialized forms (5), exquisite variation is
provided by covalent modifications (acetyla-
tion, phosphorylation, methylation) of the hi-
stone tail domains, which allow regulatable
contacts with the underlying DNA. The en-
zymes transducing these histone tail modifi-
cations are highly specific for particular ami-
no acid positions (6, 7), thereby extending
the information content of the genome past
the genetic (DNA) code. This hypothesis pre-
dicts that (i) distinct modifications of the
1
Research Institute of Molecular Pathology (IMP) at
the Vienna Biocenter, Dr. Bohrgasse 7, A-1030 Vi-
enna, Austria. E-mail: jenuwein@nt.imp.univie.ac.at
2
Department of Biochemistry and Molecular Genetics,
University of Virginia Health Science Center, Char-
lottesville, VA 22908, USA. E-mail: allis@virginia.edu
10 AUGUST 2001 VOL 293 SCIENCE www.sciencemag.org1074
E PIGENETICS
histone tails would induce interaction affini-
ties for chromatin-associated proteins, and
(ii) modifications on the same or different
histone tails may be interdependent and gen-
erate various combinations on any one nu-
cleosome.
Here, we wish to extend this concept for
overall chromosome structure by proposing that
(iii) distinct qualities of higher order chromatin,
such as euchromatic or heterochromatic do-
mains (7), are largely dependent on the local
concentration and combination of differentially
modified nucleosomes (Fig. 1A). We envision
that this “nucleosome code” then permits the
assembly of different epigenetic states (7),
leading to distinct “readouts” of the genetic
information, such as gene activation versus
gene silencing or, more globally, cell prolifer-
ation versus cell differentiation. Any such mod-
el must account for how these epigenetic states
are established, maintained, and stably inherited
through mitosis and meiosis. Although there is
clear evidence for a “cellular memory” (8),
sudden switches in cell fate do occur, leading to
variegating phenotypes. If the histone code hy-
pothesis is correct, at least in part, deciphering
how that code is translated into biological re-
sponse remains an important and nontrivial
challenge. On the basis of current knowledge,
other possibilities are likely to exist, including
less stringent “charge patches” in histone tails
(9).
Clear evidence is beginning to link alter-
ations in chromatin structure to cell cycle
progression, DNA replication, DNA damage
and its repair, recombination, and overall
chromosome stability (10). It also seems like-
ly that the fundamental nature of chromatin-
based epigenetics will have an impact on X
inactivation, imprinting, developmental re-
programming of cell lineages, and the plas-
ticity of stem cells. The implications for hu-
man biology and disease, including cancer
and aging, are far-reaching.
Su(var)s, Histone Methylation, and
Heterochromatin
It is now widely recognized that heritable, but
reversible, changes in gene expression can
occur without alterations in DNA sequence.
Pioneering studies on radiation-induced chro-
mosomal translocations (11) provided some
of the earliest findings that epigenetic “on-
off ” transcriptional states are largely depen-
dent on the position of a gene within an
accessible (euchromatic) or an inaccessible
(heterochromatic) chromatin environment.
This phenomenon, known as position-effect
variegation (PEV), allowed the development
of genetic screens in Drosophila (12) and S.
pombe (13, 14 ) that have identified ⬃30 to
40 loci involved in modifying PEV. Similar
to PEV, mating-type switching in budding
(15) and fission (16) yeast represents another
paradigm for a variegating mechanism where
the location of a gene within a distinct chro-
matin environment, the mat region, dictates
the establishment of an active or a silent
transcriptional state. In particular for S.
pombe, which appears to contain a higher
order chromatin structure more closely re-
sembling that of multicellular eukaryotes, in-
heritance of silenced chromatin domains has
been shown to be remarkably stable during
both mitosis and meiosis (16).
Fig. 1. Models for euchromatic or hetero-
chromatic histone tail modifications. (A)
Schematic representation of euchromatin
and heterochromatin as accessible or
condensed nucleosome fibers containing
acetylated (Ac), phosphorylated (P), and
methylated (Me) histone NH
2
-termini.
(B) Generic model for antagonistic E(var)
and Su(var) gene function in adding eu-
chromatic (EU) or heterochromatic (HET )
modification marks onto a nucleosomal
template. In addition, Su(var)s also func-
tion in removing euchromatic signals and
E(var)s can destabilize the heterochro-
matic state. (C) Examples of combinato-
rial modifications in histone NH
2
-termini
that are likely to represent “imprints” for active or inactive chromatin. Single-letter abbreviations
for amino acid residues: A, Ala; E, Glu; G, Gly; H, His; K, Lys; L, Leu; M, Met; P, Pro; Q, Gln; R, Arg;
S, Ser; and T, Thr. (D) Proposed synergistic (connected arrowheads) or antagonistic (blocked oval
line) modifications in histone H3 and H4 NH
2
-termini. The arrow with the scissors indicates
possible proteolytic cleavage of the H3 NH
2
-terminus.
www.sciencemag.org SCIENCE VOL 293 10 AUGUST 2001 1075
E PIGENETICS
Among the modifier genes identified in the
above model systems, one subclass suppresses
variegation [the Su(var) group] and comprises
gene products such as histone deacetylases
(HDACs), protein phosphatases (PPTases), and
S-adenosylmethionine (SAM) synthetase (17),
as well as chromatin-associated components
that are best characterized by the heterochroma-
tin protein HP1 [Su(var)2-5] (18). In addition to
the Su(var) group of genes, an antagonizing
class of PEV modifiers enhances variegation
[E(var) group] (12) and counteracts the Su(var)-
induced silent chromatin state. Several E(var)
gene products are components of adenosine
triphosphate (ATP)– dependent nucleosome-re-
modeling machines, such as the SWI/SNF and
brahma complexes (19, 20), which increase
overall nucleosome mobility.
Extending these parallels even further,
Su(var) and E(var) gene products contain
several conserved protein domains—the bro-
mo-, chromo-, and SET domains—that are
also shared with two other classes of antag-
onizing chromatin regulators: the Polycomb
(Pc-G) and trithorax (trx-G) groups. The
Pc-G and trx-G genes are important for main-
taining the expression boundaries of the ho-
meotic selector genes and several other key
developmental genes (21, 22), presumably by
modulating the chromatin structure of their
target loci. The bromodomain (23) is found in
SNF2, TAF
II
250, and mammalian trithorax
(HRX/Mll); the chromodomain (24, 25)is
shared between Polycomb and HP1; and the
SET domain (26) is found in Su(var)3-9, in
the Pc-G member E(z), and in trithorax.
These modules have been widely used during
evolution to generate a considerable function-
al diversity among proteins specialized in
modulating chromatin structure.
Histone acetylation (27, 28) and histone
phosphorylation (29) modification systems
have been characterized in detail. A further
class of enzymatic activities that regulate the
site-specific addition of methyl groups to hi-
stones has recently been described. Original-
ly identified as the PEV modifier Su(var)3-9
in Drosophila, homologs from fission yeast
(Clr4) to human (SUV39H1) have been
shown to encode histone methyltransferases
(HMTases) that selectively methylate histone
H3 at Lys
9
(30). The HMTase function in the
Su(var)3-9 family maps to the highly con-
served SET domain but also requires adjacent
Cys-rich regions. Notably, generation of the
H3-Lys
9
methyl epitope induces a hetero-
chromatic affinity for HP1 proteins that rec-
ognize this epigenetic signal through their
chromodomains (31, 32). These results pro-
vide a strong link among Su(var) function,
gene-silencing activity, and the assembly of
heterochromatin (31–35).
By contrast, an enzymatic HMTase func-
tion has not yet been demonstrated for Pc-G
and trx-G proteins. Instead, E(z) has been
associated with a Pc-G complex containing
HDAC activity (36), and trx or HRX have
been shown to interact with components of
chromatin-remodeling machines (37). In gen-
eral terms, Su(var) and Pc-G gene function
would be characterized by transducing the
addition of heterochromatic marks and the
removal of euchromatic marks on the chro-
matin template. Conversely, the antagonizing
activity of E(var) and trx-G gene function
would involve the establishment of euchro-
matic signals (e.g., increased nucleosome
mobility) and destabilize or degrade (see be-
low) heterochromatic “imprints” (Fig. 1B).
Translating the Histone Code
The histone code hypothesis predicts that the
modification marks on the histone tails
should provide binding sites for effector pro-
teins. In agreement with this notion, the bro-
modomain has been the first protein module
to be shown to selectively interact with a
covalent mark (acetylated lysine) in the his-
tone NH
2
-terminal tail (23, 38, 39). In addi-
tion to the proteins discussed above, the bro-
modomain is also present in many transcrip-
tional regulators having intrinsic histone
acetyltransferase (HAT) activity (e.g., GCN5,
PCAF, TAF
II
250). Consistent with the sec-
ond prediction of the histone code (that there
be combinatorial readout), TAF
II
250, which
itself harbors several histone-modifying ac-
tivities, contains two tandem copies of the
bromodomain. In this configuration it pref-
erentially binds diacetylated histone pep-
tides presenting acetyl-lysine moieties that
are appropriately spaced (40). Use of the
Simple Modular Architectural Research
Tool (SMART; http://smart.embl-heidel-
berg.de) indicates that there are ⬃75 bro-
modomain-containing proteins in humans.
Several of these proteins, such as human
poly-bromodomain protein 1, exhibit many
copies (six) of regularly spaced bromodo-
mains, which could conceivably bind to a
specific combination of acetyl groups pre-
sented on one or several histone tails.
Chromodomains, on the other hand, ap-
pear to be targeting modules for methylation
marks. The chromodomain of HP1 is highly
selective for methylated H3 at Lys
9
, and little
if any binding is observed with H3 peptides
containing a methylated Lys
4
position (32).
Thus, although chromodomains are highly
conserved, it seems likely that not all chro-
modomains—nor their methyl targets— be-
have similarly. In support, chromodomain
swapping experiments have not uniformly
indicated functional conservation in silencing
assays (41, 42). Interestingly, Su(var)3-9
HMTase family members also contain a chro-
modomain, whose integrity is critical for si-
lencing in vivo (33, 43). Several repressive
chromatin-remodeling complexes comprise
components such as the Mi-2/CHD ATPase
subunit of the NuRD complex (44 ), which
harbors two chromodomains and might con-
ceivably recognize dimethylated histone tails
in a manner analogous to double bromodo-
mains. In this regard, we note that Lys
9
and
Lys
27
in the H3 tail are embedded in similar
sequence motifs, and both positions are “hot
spots” for methylation by the SET domain–
containing HMTase G9a (45).
Finally, a hallmark property of all HP1
proteins is the combination of a chromodo-
main with a chromoshadow domain that are
separated by a short but variable hinge re-
gion. Because the chromoshadow domain of
HP1 appears to self-dimerize in solution (46,
47), it is tempting to infer that full-length
HP1 may assemble intermolecular chromo-
domains, thereby generating a bifunctional
cross-linker that is likely to stabilize the more
rigid higher order structure of heterochroma-
tin (35, 48).
Combinations and Switches
The above examples provide support for mod-
ification-induced recruitment of chromatin-as-
sociated proteins to acetylated and methylated
histone NH
2
-termini (Fig. 2A), and it is likely
that other modules exist that specifically recog-
nize phosphorylation marks. Consistent with
the second prediction of the histone code hy-
pothesis, all four NH
2
-termini of the core his-
tones contain short “basic patches” that often
comprise acetylation, phosphorylation, and
methylation marks in close proximity on one
individual tail (4). All three of these modifica-
tions can be found both in active or silenced
chromatin regions, which raises the question of
how combinatorial specificity is used in defin-
ing an imprint for euchromatin or heterochro-
matin (Fig. 1, A and C).
Some evidence is emerging about a pos-
sible combinatorial code. For example, the
histone H3 NH
2
-terminus appears to exist in
two distinct modification states that are likely
to be regulated by a “switch” between Lys
9
methylation and Ser
10
phosphorylation (Fig.
1D). Ser
10
phosphorylation inhibits Lys
9
methylation (30) but is synergistically cou-
pled with Lys
9
and/or Lys
14
acetylation dur-
ing mitogenic and hormonal stimulation in
mammalian cells (49 –51). In this phos-
phorylated-acetylated state, the modified H3
tail marks transcriptional activation (Fig. 1C).
H3 phosphorylation is also important for mi-
totic chromosome condensation (52), where
it may be linked to other secondary signal(s)
such as the nucleosomal incorporation of the
pericentric H3 analog Cenp-A (53). Con-
versely, aberrant Lys
9
methylation antagoniz-
es Ser
10
phosphorylation, leading to mitotic
chromosome dysfunction (30, 54 ). Further,
deacetylation of Lys
14
in H3 (33) is required
to facilitate subsequent Lys
9
methylation by
the Clr4 HMTase, again highlighting an or-
dered interplay to establish distinct histone
10 AUGUST 2001 VOL 293 SCIENCE www.sciencemag.org
1076
E PIGENETICS
tail modifications. Although the single H3-
Lys
9
methyl epitope appears sufficient to re-
cruit HP1 to heterochromatic regions, acety-
lation of Lys
12
in H4 is another repressive
mark (55) that may help to reinforce a silent
chromatin state (Fig. 1C).
The SUV39H1 HMTase also displays
weak activity toward histone H1 (30), and
this is likely to involve methylation of Lys
26
(56). RNA interference (RNAi) for an H1
variant was recently shown to phenocopy
silencing and proliferation defects in the C.
elegans germ line (57 ). These phenotypes are
similar to those seen in mes-2 mutants. Mes-2
is a homolog of the SET domain–containing
E(z) member of the Pc-G group (58). Su-
(var)3-9 (59) and a few other Su(var) genes,
such as E(Pc)(60), have also been shown to
enhance Pc-G– dependent homeotic transfor-
mations (60, 61). Is there a possible mecha-
nistic link between Su(var) and Pc-G func-
tion? Because the Polycomb protein contains
a chromodomain, the dual methylation of
Lys
26
in H1 and of Lys
9
in H3 could con-
ceivably provide a combinatorial signal to
recruit a Pc-G protein complex to develop-
mentally regulated target loci (Fig. 2C).
Collectively, these observations indicate
that one histone modification can influence an-
other in either a synergistic or an antagonistic
way (Fig. 1D), providing a mechanism to gen-
erate and stabilize specific imprints. During
development, stem cell divisions are often char-
acterized by one daughter cell that continues to
proliferate while the other daughter cell starts to
differentiate. Could the proposed “Lys
9
/Ser
10
”
switch or the discussed synergisms provide
an early clue about a more general mecha-
nism for how these cell fates are chosen and
maintained? Do other histone tails or entire
nucleosomes contain similar switches, and to
what extent has this theme been used in other
nonhistone proteins?
Turning the Histone Code Upside
Down
Although HP1 and H3-Lys
9
methylation are
mainly associated with heterochromatic re-
gions, HP1 also interacts with a variety of
transcriptional coactivators involved in gene
regulation in euchromatin (17, 25). Likewise,
whereas histone hypoacetylation correlates
most often with transcriptionally silent chro-
matin domains, acetylation of Lys
12
in H4
has been reported to be a hallmark property
of heterochromatin in organisms ranging
from yeast to flies (7, 55). Also counterintui-
tive are the findings that mutations in the
HDAC Rpd3 are enhancers rather than sup-
pressors of PEV (62). These observations
suggest that not all histone methylation marks
correspond with gene silencing, and that
some histone acetylation events may repress
rather than stimulate the readout of the genet-
ic information.
Indeed, methylation of Lys
4
in H3 occurs
in transcriptionally active macronuclei of Tet-
rahymena and appears to be a euchromatic
imprint in a wide range of organisms (63). In
addition, several arginine-directed HMTases,
such as the steroid receptor coactivators
CARM1 and PRMT1, methylate selective ar-
ginine positions in H3 and H4 NH
2
-termini
and induce synergistic transcriptional activa-
tion from transiently transfected reporter
constructs (64, 65). In vivo evidence that
histones are physiological targets of these
coactivators is beginning to emerge (66, 67 ).
Assuming that euchromatic methylation
marks exist (Fig. 1C), we predict that chro-
modomain-containing, positive regulators
may be recruited to their target loci in much
the same way that Su(var)3-9 –catalyzed H3-
Lys
9
methylation triggers the recruitment of
HP1 to heterochromatin.
There are several intriguing candidates for
such positively acting methyl-docking part-
ners. The chromodomain-containing HAT,
Esa1, is the only known essential HAT in S.
cerevisiae (27) and represents the catalytic
subunit of the NuA4 HAT complex, which
has been linked to transcriptional activation
and nucleosome remodeling in yeast and flies
(68, 69). Because Esa1 displays robust in
vitro acetylation activity toward Lys
5
in H4
(70, 71), it is possible that Arg
3
methylation
in H4, catalyzed by the PRMT1 HMTase (66,
Fig. 2. Translating the “histone code.” (A) Described protein modules of histone-modifying
enzymes that have been shown to interact with site-specific methylation (chromodomain) or
acetylation (bromodomain) marks in histone NH
2
-termini. A protein module that would selectively
recognize phosphorylated positions is currently not known. Abbreviations: HMT, histone methyl-
transferase; HAT, histone acetyltransferase; HDM, histone demethylase; PPTase, protein phospha-
tase; HDAC, histone deacetylase. (B) Proposed histone tail interactions for a “reversed” histone
code, showing a chromodomain-containing HAT (e.g., Esa1) and part of a nucleosome-remodeling
complex that may comprise a bromodomain-containing, inactive HMTase (dashed lettering), such
as the trx-G protein HRX. (C) Possible functional interactions between Su(var) and Pc-G proteins
or between histone- and DNA-methylating enzymes that could be induced or stabilized by
site-selective combinations of methylation marks.
www.sciencemag.org SCIENCE VOL 293 10 AUGUST 2001 1077
E PIGENETICS
67), might play a role in recruiting Esa1 to
active chromatin regions (Fig. 2B). Another
chromodomain-containing HAT, Mof, has
been shown to display strong selectivity for
acetylation of Lys
16
in H4, a hallmark mod-
ification correlated with the doubling of tran-
scriptional up-regulation observed on the
male X chromosome in Drosophila (7). The
chromodomain of Mof has been suggested to
bind RNA (72), raising the possibility that
association with RNA—or even with meth-
ylated RNA—may contribute to the recruit-
ment of Mof-containing complexes, which
also include another chromodomain compo-
nent, Msl3. Because Lys
20
in H4 is a well-
documented methylation site (56), it is con-
ceivable that this methylation mark may be
involved in stabilizing the fly dosage com-
pensation complex, thereby facilitating Mof-
dependent acetylation of adjacent Lys
16
.
According to these views, appropriate
methylation mark(s) would dictate the re-
cruitment of different chromodomain-con-
taining complexes, which in turn contribute
to gene activation or gene silencing. It re-
mains an intriguing, but undocumented, pos-
sibility that distinct histone methylation
marks may also interfere with the association
of repressive chromatin complexes, in much
the same way that nearby modifications may
influence bromodomain recognition and
binding (39). Finally, the molecular func-
tion(s) of the bromodomain-containing HRX
and SNF2 proteins are characterized by tran-
scriptional stimulation and nucleosome re-
modeling. HRX also contains a SET domain
that appears to be catalytically inactive (30)
but has been shown to interact with a SWI/
SNF subunit (37), suggesting that some re-
modeling complexes could transiently incor-
porate a “mute” HMTase (Fig. 2B). Thus,
intrinsically impaired HMTase function in
HRX could preclude methylation-dependent
binding of repressor proteins, thereby rein-
forcing an activated chromatin state. It there-
fore seems plausible that the activities of
several E(var) and trx-G proteins may be
facilitated by the recruitment to transcription-
ally positive histone tail modifications and by
subsequently antagonizing the establishment
of negative marks.
Transient Versus “Stable” Epigenetic
Imprints
Given that histone methylation is linked with
both euchromatic and heterochromatic states,
how stable is this histone modification? On
the basis of thermodynamic principles alone,
methyl groups, in particular methyl-lysine,
have a considerably lower turnover than do
acetyl or phosphoryl groups. The latter two
modifications can be removed from histone
tails by the activity of HDACs or phospha-
tases (29, 73), whereas histone demethylases
(HDMases) have yet to be characterized. If
HDMases do not exist, histone lysine meth-
ylation would be a nearly perfect long-term
epigenetic mark for maintaining chromatin
states. In contrast to DNA methylation—
where the methylated imprint can be removed
by nucleotide excision followed by repair—
DNA replication and semiconservative nu-
cleosome distribution appears as the sole
means to “dilute” histone lysine methylation
below a critical threshold level.
Another potential mechanism for remov-
ing methylation marks from histone tails is
proteolytic processing. Histone NH
2
-termini
are exposed and labile to proteolysis (56 ),
and portions of certain histone tails are
known to be clipped at precise stages in the
cell cycle (74) or at specific stages of devel-
opment (75). For example, in Tetrahymena,
the first six amino acids are removed from the
NH
2
-terminus of H3 in transcriptionally si-
lent micronuclei, but not in transcriptionally
active macronuclei. H3 is ubiquitinated at
specific stages of mouse spermatogenesis
(76), and H3 is also degraded at a low level
in many organisms in what is most often
assumed to be uncontrolled proteolysis oc-
curring during isolation. Ubiquitin-based pro-
tein processing, as opposed to degradation,
can occur (77 ). Conserved lysines in the
COOH-terminal tails of histones H2A and
H2B are also subjected to monoubiquitina-
tion in a pathway that seems not to be tied to
histone turnover (78). Further, the TAF
II
250-
mediated monoubiquitination of H1 has been
shown to correlate with transcriptional stim-
ulation (79). Whether ubiquitination may be
linked to the proteolytic removal of more
stable methylation marks in histone tails—or
whether, in certain cases, it could even rep-
resent a synergistic signal for their addition—
is not known, but remains an intriguing pos-
sibility (Fig. 3). A putative ubiquitin-specific
protease is encoded by an E(var) gene in
Drosophila (80), and the DNA repair and
histone-ubiquitinating rph6 protein has been
implicated in post-replication remodeling of
the chromatin structure at the silent mating-
type loci in fission yeast (81). Similarly, SIR-
dependent gene silencing in S. cerevisiae also
appears to be coregulated by a de-ubiquitinat-
ing enzyme (82).
The extent to which male versus female
genomes are marked differentially by histone
methylation is not known, but it seems likely
that imprinting mechanisms may well use
epigenetic marks outside of DNA methyl-
ation. Nearly complete removal of histones
from the genome is known to occur during
vertebrate spermatogenesis and other special-
ized developmental situations (83). Bulk dis-
placement of histones during spermatogene-
sis would provide a means to “erase” poten-
tial male marks in the germ line, allowing the
reprogramming of developmental imprints.
Fig. 3. A proteolytic model to remove “stable” methylation marks from histone H3. Abbreviations:
Ub, ubiquitin-conjugating activity; Ub protease, ubiquitin-directed proteolytic activity. Depending
on the chromatin environment and/or the nature of the ubiquitin signal, a methylated H3
NH
2
-terminus may be removed by proteolytic processing (left; see also Fig. 1D), or the entire H3
molecule may be degraded (right).
10 AUGUST 2001 VOL 293 SCIENCE www.sciencemag.org1078
E PIGENETICS
Immortal Chromatin
The importance of chromatin in the informa-
tion storage and decoding processes of the
eukaryotic genome is reinforced by the
growth in our knowledge about covalent
modifications of histone proteins, and about
the enzyme systems that transduce or remove
these imprints. Moreover, histone modifica-
tions may also be a “sensor” of the metabolic
state of the cell. For example, the Sir2 en-
zyme uses an essential metabolic cofactor
(nicotinamide adenine dinucleotide) to regu-
late the activity of a family of silencing-
associated HDACs (84 ). Will HDMases be
uncovered only when the correct cofactor,
itself possibly a direct product from interme-
diary carbon metabolism, is added to the test
reactions? The lessons learned from the Sir2
paradigm lead to an attractive new concept:
Because chromatin is the physiological tem-
plate of eukaryotic cells, are genomes pro-
grammed to “open” and “close” on demand
by enzyme complexes that evolved to re-
spond directly to metabolic cues? If correct,
we anticipate that further insights will be
gained as we systematically investigate chro-
matin changes during different physiological
or pathological states.
To what extent does a histone code link
directly to our genetic code, or are these
codes separate indexing mechanisms? Will
we find evidence of interdependence between
histone methylation and DNA methylation,
similar to the interplay between histone
deacetylation and DNA methylation (44)?
Intriguingly, a “chromo-methylase” has re-
cently been described in Arabidopsis that
combines a chromodomain with a DNA
methylating activity (85), and one member of
the SET domain family contains a methyl
CpG binding motif (35) (Fig. 2C). Histone
methylation may also help to explain poorly
understood chromatin effects where deacety-
lase inhibitors and/or 5-aza-cytosine fail to
cause reversal of previously silent genomic
regions (86). Indeed, transcription of many
genes is regulated by histone acetylation in
organisms (e.g., in yeast and flies) that exhib-
it little DNA modification. Further, X chro-
mosome inactivation in mammals correlates
with hypoacetylation of histones, except for a
few X-linked loci that escape this silencing
mechanism (87). In addition, in some branch-
es of mammalian evolution (e.g., marsupials),
no allele-specific DNA methylation has been
observed. Could histone methylation be one
of the conserved mechanisms substituting for
the apparent absence of DNA methylation in
these organisms, and to what extent is the
inactive X chromosome hypoacetylated (88)
because it may be hypermethylated at distinct
histone NH
2
-termini?
How far will epigenetics go past transcrip-
tional effects? Emerging evidence indicates
that programmed DNA rearrangements (89),
imprinting phenomena (90), germ line si-
lencing (57 ), developmentally cued stem
cell divisions (91), and overall chromo-
some stability and identity (52, 92) are all
influenced by epigenetic alterations of the
underlying chromatin structure. In keeping
with the distinct qualities of accessible and
inaccessible nucleosomal states, could it be
that “open” (euchromatic) chromatin repre-
sents the underlying principle that is syn-
onymous for the character of progenitor,
immortal, and young cells? Conversely, is
“closed” (heterochromatic) chromatin the
reflection of a developmental “memory”
that stabilizes lineage commitment and
gradually restricts the self-renewal poten-
tial of our somatic cells? As pointed out by
others (93), epigenetics imparts a funda-
mental regulatory system beyond the se-
quence information of our genetic code and
emphasizes that “Mendel’s gene is more
than just a DNA moiety.”
References and Notes
1. D. Baltimore, Nature 409, 814 (2001).
2. S. W. Brown, Science 151, 417 (1966).
3. K. Luger, A. W. Mader, R. K. Richmond, D. F. Sargent,
T. J. Richmond, Nature 389, 251 (1997).
4. P. Cheung, C. D. Allis, P. Sassone-Corsi, Cell 103, 263
(2000).
5. A. P. Wolffe, D. Pruss, Trends Genet. 12, 58 (1996).
6. B. D. Strahl, C. D. Allis, Nature 403, 41 (2000).
7. B. M. Turner, Bioessays 22, 836 (2000).
8. R. Paro, Trends Genet. 11, 295 (1995).
9. Y. Dou, M. A. Gorovsky, Mol. Cell 6, 255 (2000).
10. A. P. Wolffe, D. Guschin, J. Struct. Biol. 129, 102
(2000).
11. H. J. Muller, Proc. Int. Congr. Genet. 1, 213 (1932).
12. G. Reuter, P. Spierer, Bioessays 14, 605 (1992).
13. G. Thon, A. J. S. Klar, Genetics 131, 287 (1992).
14. R. C. Allshire, J.-P. Javerzat, N. J. Readhead, G. Cran-
ston, Cell 76, 157 (1994).
15. L. Pillus, M. Grunstein, in Chromatin Structure and
Gene Expression, S. C. R. Elgin, Ed. (IRL Press, New
York, 1995), pp. 123–146.
16. S. I. S. Grewal, A. J. S. Klar, Cell 86, 95 (1996).
17. L. L. Wallrath, Curr. Opin. Genet. Dev. 8, 147 (1998).
18. J. C. Eissenberg et al., Proc. Natl. Acad. Sci. U.S.A. 87,
9923 (1990).
19. T. Tsukinama, C. Wu, Curr. Opin. Genet. Dev. 7, 182
(1997).
20. A. J. Kal, T. Mahmoudi, N. B. Zak, C. P. Verrijzer, Genes
Dev. 14, 1058 (2000).
21. R. Paro, P. Harte, in Epigenetic Mechanisms of Gene
Regulation, V. E. A. Russo, R. A. Martienssen, A. D.
Riggs, Eds. (Cold Spring Harbor Laboratory Press, Cold
Spring Harbor, NY, 1996), pp. 507–528.
22. M. van Lohuizen, Curr. Opin. Genet. Dev. 9, 355
(1999).
23. C. Dhalluin et al., Nature 399, 491 (1999).
24. R. Paro, D. S. Hogness, Proc. Natl. Acad. Sci. U.S.A.
88, 263 (1991).
25. D. A. Jones, F. G. Cowell, P. B. Singh, Bioessays 22, 124
(2000).
26. B. Tschiersch et al., EMBO J. 13, 3822 (1994).
27. S. Y. Roth, J. M. Denu, C. D. Allis, Annu. Rev. Biochem.
70, 81 (2001).
28. J. A. Johnson, B. M. Turner, Semin. Cell Dev. Biol. 10,
179 (1999).
29. J.-Y. Hsu et al., Cell 102, 279 (2000).
30. S. Rea et al., Nature 406, 593 (2000).
31. M. Lachner, D. O’Carroll, S. Rea, K. Mechtler, T. Jenu-
wein, Nature 410, 116 (2001).
32. A. J. Bannister et al., Nature 410, 120 (2001).
33. J.-I. Nakayama, J. C. Rice, B. D. Strahl, C. D. Allis, S. I. S.
Grewal, Science 292, 110 (2001); published online 15
March 2001 (10.1126/science.1060118).
34. J. C. Rice, C. D. Allis, Curr. Opin. Cell Biol. 13, 263
(2001).
35. T. Jenuwein, Trends Cell Biol. 11, 266 (2001).
36. J. van der Vlag, A. P. Otte, Nature Genet. 23, 474
(1999).
37. O. Rozenblatt-Rosen et al., Proc. Natl. Acad. Sci.
U.S.A. 95, 4152 (1998).
38. F. Winston, C. D. Allis, Nature Struct. Biol. 6, 601
(1999).
39. D. J. Owen et al., EMBO J. 19, 6141 (2000).
40. R. H. Jacobson, A. G. Ladurner, D. S. King, R. Tjian,
Science 288, 1422 (2000).
41. J. S. Platero, T. Harnett, J. C. Eissenberg, EMBO J. 14,
3977 (1995).
42. G. Wang et al., Mol. Cell. Biol. 20, 6970 (2000).
43. A. V. Ivanova, M. J. Bonaduce, S. V. Ivanov, A. J. S. Klar,
Nature Genet. 19, 192 (1998).
44. J. Ahringer, Trends Genet. 16, 351 (2000).
45. M. Tachibana, K. Sugimoto, T. Fukushima, Y. Shinkai,
J. Biol. Chem. 276, 25309 (2001).
46. S. V. Brasher et al., EMBO J. 19, 1587 (2000).
47. N. P. Cowieson, J. F. Partridge, R. C. Allshire, P. J.
McLaughlin, Curr. Biol. 10, 517 (2000).
48. J. C. Eissenberg, S. C. R. Elgin, Curr. Opin. Genet. Dev.
10, 204 (2000).
49. P. Cheung et al., Mol. Cell 5, 905 (2000).
50. W. S. Lo et al., Mol. Cell 5, 917 (2000).
51. A. L. Clayton, S. Rose, M. J. Barratt, L. C. Mahadevan,
EMBO J. 19, 3714 (2000).
52. Y. Wei, L. Yu, J. Bowen, M. A. Gorovsky, C. D. Allis, Cell
97, 99 (1999).
53. S. G. Zeitlin, C. M. Barber, C. D. Allis, K. E. Sullivan,
J. Cell Sci. 114, 653 (2001).
54. M. Melcher et al., Mol. Cell. Biol. 20, 3728 (2000).
55. B. M. Turner, A. J. Birley, J. Lavender, Cell 69, 375
(1992).
56. K. E. van Holde, Chromatin (Springer-Verlag, New
York, 1989).
57. M. A. Jedrusik, E. Schulze, Development 128, 1069
(2001).
58. R. Holdeman, S. Nehrt, S. Strome, Development 125,
2457 (1998).
59. G. Reuter, personal communication.
60. K. Stankunas et al., Development 125, 4055 (1998).
61. T. Jenuwein, G. Laible, R. Dorn, G. Reuter, Cell. Mol.
Life Sci. 54, 80 (1998).
62. F. De Rubertis et al., Nature 384, 589 (1996).
63. B. D. Strahl, R. Ohba, R. G. Cook, C. D. Allis, Proc. Natl.
Acad. Sci. U.S.A. 96, 14967 (1999).
64. D. Chen et al., Science 284, 2174 (1999).
65. M. R. Stallcup, Oncogene 20, 3014 (2001).
66. B. D. Strahl et al., Curr. Biol, 11, 996 (2001).
67. H. Wang et al., Science 293, 853 (2001).
68. S. Allard et al., EMBO J. 18, 5108 (1999).
69. A. Eisen et al., J. Biol. Chem. 276, 3484 (2001).
70. E. R. Smith et al., Proc. Natl. Acad. Sci. U.S.A. 95,
3561 (1998).
71. A. S. Clarke, J. E. Lowell, S. J. Jacobson, L. Pillus, Mol.
Cell. Biol. 19, 2515 (1999).
72. A. Akhtar, D. Zink, P. B. Becker, Nature 407, 405 (2000).
73. J. Taunton, C. A. Hassig, S. L. Schreiber, Science 272,
408 (1996).
74. C. D. Allis, J. K. Bowen, G. N. Abraham, C. V. Glover,
M. A. Gorovsky, Cell 20, 55 (1980).
75. R. Lin, R. G. Cook, C. D. Allis, Genes Dev. 5, 1601
(1991).
76. W. M. Baarends et al., Dev. Biol. 207, 322 (1999).
77. V. J. Palombella, O. J. Rando, A. L. Goldberg, T. Ma-
niatis, Cell 78, 773 (1994).
78. K. Robzyk, J. Recht, M. A. Osley, Science 287, 501
(2000).
79. A.-D. Pham, F. Sauer, Science 289, 2357 (2000).
80. S. Henchoz, F. De Rubertis, D. Pauli, P. Spierer, Mol.
Cell. Biol. 16, 5717 (1996).
81. J. Singh, V. Goel, A. J. S. Klar, Mol. Cell. Biol. 18, 5511
(1998).
82. D. Moazed, A. Johnson, Cell 86, 667 (1996).
83. R. Ballhorn, S. Weston, C. Thomas, A. J. Wyrobek, Exp.
Cell Res. 150, 298 (1984).
84. S. Imai, C. M. Armstrong, M. Kaeberlein, L. Guarente,
Nature 403, 795 (2000).
85. A. M. Lindroth et al., Science 292, 2077 (2001);
published online 10 May 2001 (10.1126/sci-
ence.1059745).
86. A. El-Osta, A. P. Wolffe, Gene Expr. 9, 63 (2000).
www.sciencemag.org SCIENCE VOL 293 10 AUGUST 2001 1079
E PIGENETICS
87. C. M. Disteche, Trends Genet. 11, 17 (1995).
88. P. Jeppesen, B. M. Turner, Cell 74, 281 (1993).
89. D. R. Roth, S. Y. Roth, Cell 103, 699 (2000).
90. R. I. Gregory et al., Mol. Cell. Biol, 21, 5426 (2001).
91. G. Cavalli, R. Paro, Science 286, 955 (1999).
92. A. H. F. M. Peters, D. O’Carroll, T. Jenuwein,
unpublished data.
93. A. S. J. Klar, Trends Genet. 14, 299 (1998).
94. We thank G. Reuter and members of our laborato-
ries for allowing us to cite unpublished observa-
tions. We are particularly grateful to S. Rea for his
assistance in preparing the figures. Supported by
the IMP through Boehringer Ingelheim and by
grants from the Austrian Research Promotion Fund
and the Vienna Economy Promotion Fund ( T.J.),
and by NIH grant GM53512 and an NIH MERIT
award (C.D.A.).
This article is dedicated to the memory of Alan
Wolffe, an inspirational leader to all of us who
have pondered the mysteries of chromatin and
gene regulation.
VIEWPOINT
RNA: Guiding Gene Silencing
Marjori Matzke,
1
* Antonius J. M. Matzke,
1
Jan M. Kooter
2
In diverse organisms, small RNAs derived from cleavage of double-strand-
ed RNA can trigger epigenetic gene silencing in the cytoplasm and at the
genome level. Small RNAs can guide posttranscriptional degradation of
complementary messenger RNAs and, in plants, transcriptional gene si-
lencing by methylation of homologous DNA sequences. RNA silencing is a
potent means to counteract foreign sequences and could play an impor-
tant role in plant and animal development.
RNA silencing is a new field of research that
has coalesced during the last decade from inde-
pendent studies on various organisms. Scien-
tists who study plants and fungi have known
since the late 1980s that interactions between
homologous DNA and/or RNA sequences can
silence genes and induce DNA methylation (1).
The discovery of RNA interference (RNAi) in
Caenorhabditis elegans in 1998 (2) focused
attention on double-stranded RNA (dsRNA) as
an elicitor of gene silencing, and indeed, many
gene-silencing effects in plants are now known
to be mediated by dsRNA (3). RNAi is usually
described as a posttranscriptional gene-silenc-
ing phenomenon in which dsRNA triggers deg-
radation of homologous mRNA in the cyto-
plasm (4). However, the potential for nuclear
dsRNA to enter a pathway leading to epigenetic
modifications of homologous DNA sequences
and silencing at the transcriptional level should
not be discounted. Although the nuclear aspects
of RNA silencing have been studied primarily
in plants, there are hints that similar RNA-
directed DNA or chromatin modifications
might occur in other organisms as well. Here
we adopt a broad definition of RNA silencing
that encompasses effects in the cytoplasm and
the nucleus, and consider their possible devel-
opmental roles and evolutionary origins.
RNA Guiding Homologous RNA
Degradation
Although they may differ in detail, RNAi in
animals and the related phenomena of post-
transcriptional gene silencing (PTGS) in
plants and quelling in Neurospora crassa re-
sult from the same highly conserved mecha-
nism, indicating an ancient origin (5–10). The
basic process involves a dsRNA that is pro-
cessed into shorter units that guide recogni-
tion and targeted cleavage of homologous
mRNA. dsRNAs that trigger PTGS/RNAi
can be made in the nucleus or cytoplasm in a
number of ways, including transcription
through inverted DNA repeats, simultaneous
synthesis of sense and antisense RNAs, viral
replication, and the activity of cellular or viral
RNA– dependent RNA polymerases (RdRP)
on single-stranded RNA templates (Fig. 1). In
C. elegans, dsRNAs can be injected or intro-
duced simply by soaking the worms in a
solution containing dsRNA or feeding them
bacteria expressing sense and antisense RNA
(10).
Genetic and biochemical approaches are be-
ing used to dissect the mechanism of PTGS/
RNAi. Putative RdRPs, putative helicases, and
members of the PAZ/Piwi family are some of
the common proteins identified in genetic
screens in N. crassa, C. elegans, and Arabidop-
sis (3, 5, 8, 10). Although these proteins provide
clues about dsRNA synthesis and processing,
the most detailed insight into the two-step RNA
degradation process has come from biochemi-
cal experiments with cytoplasmic extracts from
Drosophila (11–15) (Fig. 1). The first step in-
volves a dsRNA endonuclease [ribonuclease III
(RNase III)–like] activity that processes dsRNA
into sense and antisense RNAs 21 to 25 nucle-
otides (nt) long. These small interfering RNAs
(siRNAs), which were first described in a plant
system (16 ), are generated in Drosophila by an
RNase III–type protein termed Dicer. Orthologs
of Dicer, which contains a helicase, dsRNA
binding domains, and a PAZ domain, have
been identified in Arabidopsis, C. elegans,
mammals, and Schizosaccharomyces pombe
(15). In the second step, the antisense siRNAs
produced by Dicer serve as guides for a differ-
ent ribonuclease complex, RISC (RNA-induced
silencing complex), which cleaves the homolo-
gous single-stranded mRNAs. RISC from Dro-
sophila extracts cofractionates with siRNAs
that guide sequence-specific mRNA cleavage
(12). RISC cuts the mRNA approximately in
the middle of the region paired with antisense
siRNA (14) (Fig. 1), after which the mRNA is
further degraded. Although most protein com-
ponents of RISC have not yet been identified,
they might include an endonuclease, an exonu-
clease, a helicase, and a homology-searching
activity (6, 10). A candidate for a 3⬘,5⬘-exonu-
clease is C. elegans MUT7, an RNase D–like
protein recovered in a screen for RNAi mutants
(10). Another component of RISC is a protein
of the PAZ/Piwi family (17), which could in-
teract with Dicer through their common PAZ
domains (18) to incorporate the siRNA into
RISC (17). Genes encoding members of the
PAZ/Piwi family (Arabidopsis: AGO1; N.
crassa: QDE2; C. elegans: RDE1), which are
homologous to the translation factor eIF2C,
have been shown to be required for PTGS/
RNAi in several mutant screens (3, 5, 8, 10).
A putative RdRP was the first cellular pro-
tein shown to be required for PTGS/RNAi in
genetic screens (N. crassa: QDE1; C. elegans:
Ego1; Arabidopsis: SGS2/SDE1) (3, 5, 8, 10),
but its exact role is unclear and the predicted
enzyme activity remains to be established. This
protein might be dispensible when large
amounts of dsRNA are produced from trans-
genes or when viral RdRPs are present (5).
RdRP might be needed only when dsRNA is
synthesized to initiate silencing—for example,
from “aberrant” sense RNAs that are prema-
turely terminated or processed improperly (19).
RISC-cleaved mRNAs may also be used as
templates and converted into dsRNA, increas-
ing the level of siRNAs and enhancing PTGS/
RNAi (Fig. 1).
Putative helicases are another class of en-
zyme found repeatedly in mutant screens (N.
crassa: QDE3; C. elegans: SMG-2; Chlamy-
domas: MUT6; Arabidopsis: SDE3) (3, 5, 8,
10). Those recovered so far are not highly
related and have not yet been characterized
biochemically. A DNA helicase (QDE3) and
members of two RNA helicase superfamilies
(MUT6 and SMG2/SDE3, respectively) have
1
Institute of Molecular Biology, Austrian Academy of
Sciences, A-5020 Salzburg, Austria.
2
Department of
Developmental Genetics, Vrije Universiteit, Amster-
dam, Netherlands.
*To whom correspondence should be addressed. E-
mail: mmatzke@imb.oeaw.ac.at
10 AUGUST 2001 VOL 293 SCIENCE www.sciencemag.org1080
E PIGENETICS
