Gene silencing is required to stably maintain distinct patterns of gene expression during eukaryotic development and has been correlated with the induction of chromatin domains that restrict gene activity. We describe the isolation of human (EZH2) and mouse (Ezh1) homologues of the Drosophila Polycomb‐group (Pc‐G) gene Enhancer of zeste [E(z)], a crucial regulator of homeotic gene expression implicated in the assembly of repressive protein complexes in chromatin. Mammalian homologues of E(z) are encoded by two distinct loci in mouse and man, and the two murine Ezh genes display complementary expression profiles during mouse development. The E(z) gene family reveals a striking functional conservation in mediating gene repression in eukaryotic chromatin: extra gene copies of human EZH2 or Drosophila E(z) in transgenic flies enhance position effect variegation of the heterochromatin‐associated white gene, and expression of either human EZH2 or murine Ezh1 restores gene repression in Saccharomyces cerevisiae mutants that are impaired in telomeric silencing. Together, these data provide a functional link between Pc‐G‐dependent gene repression and inactive chromatin domains, and indicate that silencing mechanism(s) may be broadly conserved in eukaryotes.
Maintenance of different patterns of gene expression is the underlying mechanism by which cell type identity is inherited in eukaryotes and has been correlated with the organization of chromatin domains that modulate gene activity (Zuckerkandl, 1974). Paradigms for chromatin‐controlled regulation of key developmental loci include silencing of the mating type loci in Saccharomyces cerevisiae (reviewed in Loo and Rine, 1995) and restriction of expression boundaries within the homeotic gene cluster (HOM‐C) in Drosophila (Wedeen et al., 1986; Simon et al., 1993). Moreover, gene expression in eukaryotes is influenced by structural alterations in chromatin, phenomena referred to collectively as position effects. For example, genes placed in the vicinity of yeast telomeres (Gottschling et al., 1990; Nimmo et al., 1994) and centromeres (Allshire et al., 1994), or at cytologically defined heterochromatic positions in the fly (reviewed in Weiler and Wakimoto, 1995), are severely downregulated. In addition, many genes on the inactive X chromosome and several imprinted genes (for reviews, see Efstradiadis, 1994; Migeon, 1994) are repressed in mammals. Basic chromatin components are known to participate in transcriptional regulation because alterations in dosage (Moore et al., 1979; Han and Grunstein, 1988) and modification (Jeppesen and Turner, 1993) of the core histones affect gene expression. However, more complex levels of control are suggested from genetic screens on position effect variegation (PEV) in Drosophila and S.cerevisiae. These screens have identified distinct classes of genes that appear centrally involved in the regional organization of chromatin domains and the regulation of chromatin‐dependent gene activity.
In contrast to classical transcription factors, many of these chromatin regulators do not seem to bind DNA in a sequence‐specific manner, but rather associate with structurally altered chromatin or are recruited by factors already present at target sequences (for reviews, see Loo and Rine, 1995; Orlando and Paro, 1995; Pirrotta, 1995). Models for the function of chromatin regulators have been proposed to explain variegation and the clonal nature of gene expression within a given cell population (Tartof et al., 1984; Henikoff, 1995; Karpen, 1995; Csink and Henikoff, 1996; Dernburg et al., 1996). Consistent with these models, telomeric silencing in yeast and PEV in Drosophila display an epigenetic component, and it has been suggested that the epigenetic fine‐tuning of chromatin‐controlled gene expression patterns facilitates the generation of cell type diversity in eukaryotes (Zuckerkandl, 1974; Eissenberg et al., 1995; Pillus and Grunstein, 1995).
In Drosophila, ∼120 loci have been described genetically that enhance or suppress position‐dependent gene expression, but only ∼10% of the corresponding genes have been isolated so far (Reuter and Spierer, 1992). Interestingly, several of the known modifiers of PEV share structural motifs with both activators [trithorax or trx‐group (trx‐G)] and repressors [Polycomb or Pc‐group (Pc‐G)] of chromatin‐dependent gene activity within HOM‐C. The biological significance of chromatin‐mediated gene regulation, particularly for controlling mammalian development, has gained further importance by the functional analysis of murine homologues of Drosophila Psc (Pc‐G) and trx. Both of these mammalian homologues appear to be involved in specifying expression boundaries within the murine HOX‐clusters (Alkema et al., 1995; Yu et al., 1995; Akasaka et al., 1996), and their deregulation has been implicated in inducing leukaemia (Gu et al., 1992; Tkachuk et al., 1992; van der Lugt et al., 1992; Alkema et al., 1995).
The structural similarities between gene products that are implicated in controlling gene expression at specific euchromatic locations (HOM‐C) and in the vicinity of heterochromatin (PEV) point to a common mechanism(s) for the regulation of chromatin‐dependent gene activity. However, only a few modifiers of PEV have been shown to induce homeotic transformations (Dorn et al., 1993; Farkas et al., 1994), and a direct structural involvement of Pc‐G genes with heterochromatic regions has not yet been demonstrated. Recently, a 130 amino acid, carboxy‐terminal region of high sequence similarity has been identified that is shared between trx and the Pc‐G gene Enhancer of zeste [E(z)] (Jones and Gelbart, 1993). This carboxy terminus is also conserved in Su(var)3‐9, a dominant suppressor of PEV in Drosophila (Tschiersch et al., 1994). Because of the genetically defined antagonistic relationship between trx and E(z) (Jones and Gelbart, 1993; LaJeunesse and Shearn, 1996), this novel protein domain, termed SET (Tschiersch et al., 1994), may play a role in the assembly of either activating or repressing chromatin complexes, dependent on interactions with accessory proteins from the trx‐ or Pc‐G. Consistent with this notion, E(z) mutants have been shown to reduce immunostaining of some trx‐ and Pc‐G proteins at polytene chromosomes (Rastelli et al., 1993; Platero et al., 1996). Finally, the SET domain is evolutionarily conserved and is present in gene products ranging from yeast to man (Stassen et al., 1995). Based on the above criteria, the SET domain appears to define a new gene family of important chromatin regulators.
Here, we describe the isolation of mammalian homologues of the Drosophila Pc‐G gene E(z) and demonstrate function of the E(z) gene family in mediating gene repression at well‐defined heterochromatic regions in Drosophila and S.cerevisiae chromatin. These data provide a functional link between Pc‐G‐dependent gene silencing and the organization of inactive chromatin domains.
Mammalian E(z) homologues are encoded by two loci
Based on sequence information from the conserved carboxy‐terminal SET domain of Drosophila E(z), we screened a human B‐cell‐specific cDNA library for mammalian homologues (see Materials and methods). Out of 500 000 plaques, 36 primary phages were selected. Five isolates contained partial cDNA fragments of a gene that was recently mapped to human chromosome 17 and which has been named EZH1 (Abel et al., 1996). By contrast, the majority of the isolates contained novel cDNA sequences from a second human gene, which we designated EZH2 (Enhancer of zeste homologue 2). A full‐length human EZH2 cDNA of 2.6 kb encoding a protein of 746 amino acids was obtained by combining DNA fragments from different subclones. Using the human EZH2 cDNA as a probe, we next screened a mouse brain cDNA library for murine homologues of E(z). From the mouse brain library, only cDNAs encoding sequences with highest homology to human EZH1 were found. A 5′ RACE amplification was used to complete the missing 5′ end and, after subcloning, a murine Ezh1 cDNA that encodes a protein of 747 amino acids was obtained. Recently, a murine cDNA homologous to human EZH2 has been isolated by others (Hobert et al., 1996a).
To examine directly the number of mammalian E(z) homologues, we hybridized, under reduced stringency, a DNA blot containing Drosophila, mouse and human DNA with a mixed DNA probe that was derived from the SET domains of the human EZH2 and mouse Ezh1 cDNAs (Figure 1A, top). Whereas only a single fragment is detected in Drosophila DNA, 2–4 fragments hybridized with this mixed SET domain probe in both mammalian DNAs. The fragments could either be correlated to predicted sizes from the genomic murine Ezh1 locus (Figure 1A, bottom) or are exclusively detected by the human EZH2 SET domain probe (data not shown). Moreover, homology searches (Altschul et al., 1990) against the currently available human expressed sequence tags (ESTs) (Bassett et al., 1995) indicated an almost perfect match with either EZH2 or EZH1. Thus, we conclude that human and mouse E(z) homologues are represented by not more than two distinct loci which most likely reflect a gene duplication of a single ancestral locus.
To determine the accurate sizes of the mRNAs encoded by Ezh1 and EZH2, we hybridized an RNA blot containing poly(A)+ RNA from human (BJAB) and mouse (PD31 and J558L) B‐cell‐specific cell lines and from mouse kidney with DNA probes that were derived from the Ezh1 and EZH2 cDNAs (Figure 1B). The Ezh1 probe detected a specific mRNA of ∼4.4 kb which is present in all RNAs analysed and whose size is in good agreement with the 4.0 kb Ezh1 cDNA that lacks ∼200 bp of 3′ untranslated sequences. Rehybridization of the RNA blot with a DNA probe specific for the full‐length EZH2 cDNA identified a mRNA of 2.8 kb which closely matches the size of the EZH2 cDNA in both the human and mouse cell lines, but not in mouse kidney.
Structural conservation of mammalian and fly E(z) proteins
Human EZH1 and murine Ezh1 (G.Mattei, A.Lebersorger and T.Jenuwein, unpublished), and probably also EZH2/Ezh2, are encoded by orthologous loci in mouse and man, and in both comparisons amino acid identities are 98%. By contrast, inverse comparisons of EZH2/EZH1 or Ezh2/Ezh1 only reach an overall amino acid identity of 67%, suggesting that the presumed gene duplication may have occurred at a relatively early stage after the divergence of vertebrate and non‐vertebrate species. Considering the overall cDNA size and sequence comparisons, EZH2/Ezh2 and E(z) appear to be most closely related, whereas EZH1/Ezh1 is the more divergent of the mammalian gene pairs. Over the entire length of the 760 amino acids containing E(z) protein (Jones and Gelbart, 1993), EZH2/Ezh2 is 61% and EZH1/Ezh1 is 55% identical to E(z). Thus, human and mouse homologues of Drosophila E(z) display the highest structural conservation among any of the currently known mammalian and Drosophila chromatin regulators.
Alignment of all four mammalian E(z) proteins with Drosophila E(z) reveals four regions of high sequence identity (Figure 2). In addition to the most highly conserved SET domain (86% identity), an 115 amino acid, cysteine‐rich region immediately precedes the SET domain (68% identity). Most of the cysteine residues, originally noted for E(z) (Jones and Gelbart, 1993), are conserved in the mammalian proteins. However, these cysteine residues (a total of 18) are unusually spaced and do not show any apparent homology to other well‐defined cysteine‐rich regions. In the amino‐terminal half, two stretches, one of 66 amino acids (domain I; 66% identity) and one of 114 amino acids (domain II; 56% identity), are highly related between mammalian and fly E(z) proteins. Domain II also contains six conserved cysteine residues with unusual spacing and is separated from domain I by a stretch of charged amino acids. Finally, the position and sequence of the nuclear localization signal of the E(z) protein (Jones and Gelbart, 1993) is also conserved in all mammalian homologues.
Complementary expression profiles of murine Ezh1 and Ezh2
The differences in the abundance of Ezh1 and Ezh2 transcripts in kidney suggested that both loci may be differentially regulated during mouse development. To analyse the temporal and tissue‐specific expression profile of the two Ezh genes during mouse development, we performed an RNase protection analysis with antisense RNA probes that are specific for the SET domain of Ezh1 or for the amino‐terminal end of Ezh2. Ezh1‐specific transcripts were detected ubiquitously during mouse embryogenesis; however, their relative abundance is gradually upregulated during development and reaches 4‐ to 5‐fold higher levels in kidney, brain and skeletal muscle of adult mice (Figure 3, top panel). A 2‐ to 3‐fold increase in the relative abundance (after correction for differences in the amount of RNA as indicated by the S16 RNA control transcripts) of Ezh1 transcripts was also detected after retinoic acid‐induced in vitro differentiation of embryonic stem cells. By contrast, Ezh1 transcripts remained at a low level in liver, spleen and thymus.
The RNase protection analysis of Ezh1 transcript levels during mouse embryogenesis and ES cell differentiation demonstrates a specific upregulation of Ezh1 gene activity at later stages of development. By contrast, Ezh2 transcript levels are downregulated during progressing development and display a more thymus‐restricted expression (Figure 3, middle panel; Hobert et al., 1996a), resulting in complementary expression profiles of Ezh1 and Ezh2 that may indicate dosage dependence or distinct functions of the encoded proteins. Further, observed splicing variants for EZH2 (data not shown) or Ezh2 (Hobert et al., 1996a) that remove the amino‐terminal half of the conserved cysteine‐rich region or part of the SET domain in EZH1 (Abel et al., 1996) could encode functionally distinct protein isoforms. Alternatively, Ezh1 and Ezh2 may provide similar functions, but because of the presence of two loci, transcript levels may be balanced to protect against negative gene dosage effects which, in analogy to other described Drosophila chromatin regulators (Reuter et al., 1990; Eissenberg et al., 1992; Tschiersch et al., 1994), could interfere with normal gene function.
EZH2 and E(z) enhance PEV in Drosophila
Although Pc‐G‐dependent gene silencing has been correlated with the establishment of repressive chromatin domains (see Introduction), most of the Pc‐G genes do not affect gene repression in the vicinity of heterochromatin (Grigliatti, 1991; G.Reuter, unpublished). However, E(z) mutants have been described recently that reduce immunostaining of several Pc‐G proteins at natural euchromatic positions (Rastelli et al., 1993) and prevent the recruitment of Pc‐G complexes at ectopic heterochromatic locations (Platero et al., 1996). Together with the extremely high degree of evolutionary conservation between mammalian and fly E(z) homologues, these data suggested a crucial role for E(z) proteins in regulating regional gene activities in chromatin.
To demonstrate an involvement of E(z) genes in controlling the organization of repressive chromatin domains, we analysed the potential of EZH2 and E(z) to modify PEV in Drosophila. By P element‐mediated germline transformation, we established nine transgenic fly lines that carry the human EZH2 cDNA under the control of the heat shock promoter hsp70. As a control, we also generated transgenic flies carrying the E(z) cDNA (Jones and Gelbart, 1993) (four lines) or a genomic fragment encoding E(z) sequences (Jones and Gelbart, 1993) (one line). For all of the lines, the chromosomal assignment of the transgene was mapped genetically (see Table I) and the transgenic lines were crossed into In(1)wm4h and In(1)wm4h; Su(var)2‐1 indicator strains (see Materials and methods). These indicator strains carry an inversion [In(1)wm4h] which places the white gene adjacent to pericentric X heterochromatin, resulting in stochastically repressed, variegated patterns of gene expression that can be easily detected as red or white patches in the Drosophila eye. Mutations in the suppressor of variegation Su(var)2‐1, which has been shown to reduce deacetylation of histone H4 (Dorn et al., 1986), derepress heterochromatin‐restricted wm4h expression, leading to an almost wild‐type red eye phenotype.
Visual inspection of progeny of the respective crosses indicated that six out of the nine EZH2 lines and two out of the five E(z) lines induced an increase in the proportion of unpigmented areas in the eyes, therefore classifying both EZH2 and E(z) as dose‐dependent modifiers of PEV (Figure 4). Moreover, after mobilization of the E(z) transgene in a negative line (line D), eight sublines with a significant enhancer effect on PEV could be selected (Table I). PEV enhancement was largely independent of heat shock treatment, and subsequent expression analysis by in situ hybridization demonstrated a low, basal transcription throughout embryonic development of those lines displaying an enhancer effect (data not shown).
To quantify the degree of PEV enhancement of the individual transgenic lines, eye pigments were extracted from the progenies of the respective crosses and pigment absorbance at 480 nm was measured. The result of these quantitations, summarized in Table I, shows that the strongest enhancer effect of EZH2 (line E) is similar to E(z) and results in an ∼10‐fold reduction in the concentration of red eye pigments as compared with non‐transgenic siblings. Thus, the Drosophila Pc‐G gene E(z), which has been implicated in restricting gene‐specific expression boundaries within the chromatin domains of HOM‐C (Jones and Gelbart, 1990; Phillips and Shearn, 1990), is also capable of stabilizing a repressive transcriptional state in the vicinity of constitutive heterochromatin. Moreover, this potential of E(z) to repress heterochromatin‐associated gene activity is conserved in its human homologue EZH2.
EZH2‐mediated enhancement of PEV displays paternal imprinting
In the course of the matings to the indicator strains, we observed that non‐transgenic In(1)wm4h and In(1)wm4h; Su(var)2‐1 male progeny, derived from crosses with most of the transgenic EZH2 or of the transgenic E(z) lines [lines indicated (p) in Table I], also exhibited an ∼2‐ to 3‐fold reduction in the proportion of red eye pigments as compared with unmated In(1)wm4h and In(1)wm4h; Su(var)2‐1 control flies (Table I), although the transgene was no longer present. This enhancer effect on wm4h gene repression was strictly dependent upon previous paternal transmission of the transgene. Although the molecular basis for this imprinting‐like effect is not known, this result suggests that EZH2 and E(z) can participate not only in the induction, but also in the propagation of an altered chromatin structure which represses wm4h gene activity in trans, and is reminiscent of transvection phenomena described for the zeste1–white interaction (Kalisch and Rasmuson, 1974; Jones and Gelbart, 1990; Phillips and Shearn, 1990). In support of this interpretation, most transgenic EZH2/zeste1 heterozygote females display zeste1 gene variegation after heat shock (data not shown). Thus, together with the enhancement of PEV shown above, these data extend the functional conservation between EZH2 and E(z) in regulating repressive transcriptional states in Drosophila chromatin.
EZH2 and Ezh1 restore telomeric silencing in S.cerevisiae
As a second assay to examine involvement of mammalian E(z) genes in mediating the organization of repressive chromatin domains, we analysed the potential of human EZH2 and mouse Ezh1 to restore gene repression in S.cerevisiae mutants that are defective in telomeric silencing. Despite cytogenetic differences between chromatin of yeast and multicellular eukaryotes (Pillus and Grunstein, 1995), telomeric silencing or telomeric position effect (TPE) closely resembles PEV in Drosophila. For example, insertion of a URA3 gene at the subtelomeric region of the left arm of chromosome VII results in stochastic silencing of URA3 expression in many, but not all, cells within a population. If URA3 is repressed, cells survive selection with the uracil analogue 5‐fluoro‐orotic acid (5‐FOA). The degree of telomeric silencing can be quantitated by comparing the relative growth of serial cell dilutions in permissive versus 5‐FOA media, providing a very sensitive assay to detect chromatin‐mediated changes in transcriptional states (Gottschling et al., 1990).
Recently, an S.cerevisiae SET domain gene (SET1) has been identified whose disruption in URA3‐tel marker strains abolishes the ability of the mutant cells to grow on 5‐FOA, indicating complete (>100 000‐fold) derepression of telomeric silencing (Nislow et al., submitted; Figure 5A, top panel). Plasmid‐borne SET1 restored growth on 5‐FOA, demonstrating that the telomeric silencing defect was due to the set1 mutation (Figure 5A, middle panel). In addition to the carboxy‐terminal SET domain (41% amino acid identity), SET1 (1080 amino acids) and EZH2 (746 amino acids) also share several other stretches of moderate sequence similarity that are conserved in the E(z)/EZH gene family (see Figure 2). Using low‐copy (CEN4‐LEU) vectors carrying galactose‐inducible EZH2 and Ezh1 cDNAs or a haemagglutinin‐tagged EZH2 variant (to detect protein levels), we introduced mammalian E(z) homologues into set1Δ indicator strains that contain single (URA3‐tel; LPY1297) or double (URA3‐tel and ADE2‐tel; LPY1727) telomeric reporter genes. Transformants were grown selectively to saturation and plated in serial 10‐fold dilutions onto complete, −LEU and −LEU/5‐FOA media (Figure 5A, bottom panels).
Surprisingly, on average 50% of both EZH2 (16/32) and HA‐EZH2 (17/32) primary transformants were capable of growing at low cell densities on 5‐FOA, indicating complementation of set1Δ and a capacity for human EZH2 to restore telomeric silencing to almost wild‐type levels (Table II). Similarly, Ezh1 (5/12) primary transformants also rescued the set1Δ‐dependent loss in telomeric silencing, although, when compared with permissive growth, only 5% of these Ezh1 transformants did form colonies on 5‐FOA, suggesting that Ezh1 has a weaker potential than EZH2 (30%) or SET1 (75%) to restore set1Δ‐dependent telomeric silencing. In control transformations with the empty CEN4‐LEU vector, seven out of 78 primary colonies exhibited some growth on 5‐FOA; however, these likely suppressor transformants did not grow upon further subcloning (see below).
Replica plating onto uracil‐deficient medium indicated that EZH2/Ezh1 5‐FOA‐resistant colonies are still URA3+ (data not shown). To rule out that rescue of set1Δ‐dependent telomeric silencing by the mammalian E(z) homologues may be the consequence of interference with URA3 transcription, we next analysed repression of the second telomeric reporter gene (ADE2) which had been inserted at the subtelomeric region of the right arm of chromosome V (Singer and Gottschling, 1994). If ADE2 is repressed, cells accumulate a chromogenic intermediate providing a colour assay to detect silenced (pink to red) or active (white) transcriptional states. Whereas the isogenic wild‐type SET1 strain displays a weak pink colour with sectored patches at the edge of the colonies, disruption of SET1 results in predominantly white (derepressed) colonies (Figure 5A, right panel). Complementation of set1Δ by plasmid‐borne SET1 gives rise to transformants that develop a darker pink colour. Importantly, replica transformants of EZH2 and HA‐EZH2 5‐FOA‐resistant colonies display a red colour on complete medium, whereas most of the 5‐FOA‐sensitive transformants remain largely white and unsectored. To quantify repression of the ADE2‐tel marker gene, respective transformants were plated at a density of 200–500 colonies and ∼200 colonies/transformant were inspected under a dissecting microscope for the presence of red‐sectored patches. The result of this quantitation is summarized in Table II and indicates that EZH2 also restores set1Δ‐dependent silencing of a telomere‐proximal ADE2 gene to wild‐type (SET1) levels.
EZH2 and Ezh1 induce epigenetic switching of an active to a repressed transcriptional state
In contrast to SET1, EZH2 and Ezh1 complement a set1Δ‐dependent silencing defect on average in only every other transformant (see Table II), despite the presence of similar HA‐EZH2 protein levels in silenced versus non‐silenced colonies (data not shown). Galactose induction did not increase this ratio of gene silencing (data not shown), indicating that there is not a simple quantitative relationship between the amount of EZH2 or Ezh1 and the extent of silencing. Rather, this observation suggests that the mammalian E(z) proteins participate in an apparent epigenetic mechanism of chromatin‐controlled gene repression. We further examined this epigenetic component by analysing the stability of EZH2/Ezh1‐dependent gene repression. For each of two independent silenced or non‐silenced primary colonies expressing either EZH2, HA‐EZH2 or Ezh1, four subclones were grown selectively to saturation and analysed in a 5‐FOA dilution experiment as described above. The results, shown representatively for HA‐EZH2 subclones in Figure 5B, indicated that telomeric URA3 repression was stably maintained in all of the subclones tested. In contrast, the non‐silent state was metastable and subclones from initially 5‐FOA‐sensitive colonies could switch to a repressed state (Figure 5B, bottom panel, and Table II). This epigenetic switching to URA3 repression was clearly dependent upon the presence of the mammalian E(z) homologues, since neither subclones from a putative suppressor mutation nor control subclones that only contained an empty CEN4‐LEU vector promoted growth on 5‐FOA. Together, these data illustrate the involvement of E(z) genes in inducing repressive chromatin structures and demonstrate function of a mammalian protein to regulate telomeric silencing in S.cerevisiae chromatin.
Our analysis of mammalian homologues of the Drosophila Pc‐G gene E(z) demonstrates a surprising and unprecedented functional conservation of gene repression in eukaryotic chromatin: extra gene copies of human EZH2 enhance PEV in Drosophila and EZH2 or murine Ezh1 restore TPE in yeast mutants that are null for SET1, an S.cerevisiae SET domain gene. Together, these data provide a functional link between the stable maintenance of gene repression mediated by a Pc‐G gene and the organization of inactive chromatin domains. Moreover, since E(z) homologues and SET1 share the SET domain, our results also implicate an important functional role of the SET domain in participating in an evolutionarily conserved mechanism(s) that regulates repressive chromatin states in eukaryotic chromatin.
E(z) homologues mediate gene repression in euchromatin and heterochromatin
E(z) was initially identified as a mutation that increased white gene repression through genetic interactions with the transcription factor zeste (Kalisch and Rasmuson, 1974; Jones and Gelbart, 1990; Phillips and Shearn, 1990). Because E(z) restricts expression boundaries of homeotic selector genes (Jones and Gelbart, 1990; Phillips and Shearn, 1990), it has been classified as a Pc‐G gene (but see also LaJeunesse and Shearn, 1996). More recently, it has been shown that E(z) is also involved in the repression of some of the early acting segmentation/gap genes (Moazed and O'Farrell, 1992; Pelegri and Lehmann, 1994). Together, these observations could indicate recruitment of E(z) protein by bound transcription factors, resulting in the downregulation of targeted promoters in euchromatin. Consistent with this interpretation, LexA fusions encoding several Drosophila Pc‐G proteins have been shown to reduce LexA‐dependent reporter activity in transient transfection studies (Bunker and Kingston, 1994).
E(z) null alleles, on the other hand, have been shown to reduce immunostaining of several Pc‐G proteins in polytene chromosomes (Rastelli et al., 1993; Platero et al., 1996) and to result in a general decondensation of chromatin structure (Rastelli et al., 1993) that is reflected by increased chromosome breakage (Gatti and Baker, 1989). Here, we show that extra gene copies of either Drosophila E(z) or human EZH2 result in the enhancement of heterochromatin‐induced gene inactivation in position effect rearrangements (Figure 4 and Table I). Further, the E(z)5 null and the E(z)1 antimorphic mutants (only one functional gene copy) are weak suppressors of PEV (Table I). This haplo‐suppressor/triplo‐enhancer dosage effect of E(z) (and most likely also of EZH2) is reminiscent of gene products that have been postulated to be key regulators in the organization of chromatin domains (Reuter and Spierer, 1992). Only four Drosophila chromatin regulators that belong to the class of haplo‐suppressors/triplo‐enhancers have been described to date and include Su(var)2‐5 (which encodes the heterochromatin‐associated protein HP‐1) (Eissenberg et al., 1990), Su(var)3‐7 (Reuter et al., 1990), Su(var)3‐9 (Tschiersch et al., 1994), and now E(z). By contrast, most of the Pc‐G genes, including Polycomb itself, do not affect wm4h expression (Grigliatti, 1991; G.Reuter, unpublished).
Possibly, some of the conserved protein domains of E(z) (see Figure 2) may be surfaces for recruitment to selected promoters in euchromatin, whereas others could be involved in remodelling repressive chromatin structures in heterochromatin. Such dual roles in the regulation of both euchromatin and heterochromatin have been demonstrated for the positive chromatin regulators trx‐like (which encodes the nucleosome‐disrupting GAGA transcriptional activator; Farkas et al., 1994) and for the putative transcriptional regulator E(var)3‐93D (Dorn et al., 1993). Moreover, hierarchies of heterochromatin appear to exist—and may also relate to different levels of euchromatic organization—since some suppressors of PEV relieve white gene repression in pericentric but not telomeric heterochromatin in Drosophila (Wallrath and Elgin, 1995).
Immunostaining of several Pc‐G proteins on polytene chromosomes indicated localization at cytological positions that correspond to some modifiers of PEV (DeCamillis et al., 1992; Rastelli et al., 1993), suggesting that Pc‐G genes would only have an indirect effect on heterochromatin, particularly since Pc‐G proteins do not appear to decorate heterochromatic regions. However, recent immunodetection of E(z) protein at polytene chromosomes demonstrated a complex staining pattern that in addition to discrete locations also includes a more general distribution and masked positions within several Pc‐G complexes (Carrington and Jones, 1996). Moreover, recruitment of Pc‐G complexes to ectopic heterochromatic positions has been shown to depend on E(z) (Platero et al. 1996). In fact, since human EZH2 both enhances PEV in Drosophila and restores telomeric silencing in S.cerevisiae (see below), we favour the alternative view that E(z)‐related proteins display multiple functions that may also include a more direct structural involvement in the reorganization of repressive chromatin domains.
Conserved gene silencing from yeast to man
In S.cerevisiae, several silencing factors, namely the Silent Information Regulators SIR2–4, have been shown to be crucial for the maintenance of repressive chromosomal domains at both the silent mating type loci (HM‐loci) and at the telomeres, whereas other modifying proteins have more locus‐specific functions (Aparicio et al., 1991). Variegating gene expression at yeast telomeres is thought to result from competition between transcriptional activators and silencing complexes (Aparicio and Gottschling, 1994), and may act via a spreading mechanism (Renauld et al., 1993; Hecht et al., 1996) reminiscent of PEV in Drosophila pericentric heterochromatin. In addition, SIR3 and SIR4 appear to interact with histones, resulting in the establishment of repressive multimeric chromatin complexes (Hecht et al., 1995) that may be prototypes of restrictive chromatin domains mediated by Drosophila Pc‐G genes. These intriguing parallels between maintenance mechanisms for repressive transcriptional states in single‐celled and multicellular organisms have been extended by the discovery of a SIR2 gene family that is conserved from bacteria to mammals (Brachmann et al., 1995).
The potential of human EZH2 or murine Ezh1 to restore silencing in S.cerevisiae mutants is specific for loss of the endogenous SET1 gene, since EZH2 cannot suppress sir2, sir3 or sir4 single mutants (data not shown). Further, overexpression of EZH2 or SET1 in telomeric marker strains that are wild type for the SIR genes does not interfere with TPE (see Table II). These data demonstrate that the S.cerevisiae SET1 and mammalian E(z) proteins do not bypass a requirement for SIR function, but possibly modulate interactions with accessory components of yeast chromatin. In agreement with this interpretation, the EZH2 SET domain appears to represent a protein interaction surface, and preliminary data from a structure–function analysis of EZH2 indicate that deletion of the SET domain abolishes the ability of EZH2 to rescue telomeric silencing in set1Δ mutants (G.Laible, A.Wolf and T.Jenuwein, unpublished observation). Interestingly, complementation of set1Δ‐dependent telomeric silencing by the mammalian E(z) proteins (Figure 5 and Table II) displays an epigenetic component that is very reminiscent of the EZH2‐ or E(z)–mediated paternal imprinting of PEV in Drosophila (Table I). Together, these results suggest that E(z)‐related proteins may function both in the establishment and propagation of repressed chromatin domains. Recently, a Drosophila homologue of the S.cerevisiae ORC2 gene was observed to rescue mating‐type silencing in a similar epigenetic fashion (Ehrenhofer‐Murray et al., 1995).
The SET protein family of chromatin regulators
Based on our findings of functional conservation between mammalian (EZH2 and Ezh1) and an S.cerevisiae (SET1) SET domain protein, we suggest that the SET domain represents an ancient protein module that is intrinsically involved in chromatin‐controlled gene regulation. Figure 6 summarizes the relationship of representative SET protein family members according to sequence identities within the 130 amino acids comprising the SET domain (but ignoring other structural similarities) that results in a classification of SET domain proteins into four different subgroups. The SET domain of E(z)‐related genes (subgroup I) appears most similar to that of subgroup II genes which include the trithorax homologues trx/HRX, an open reading frame (C26E6.10) in Caenorhabditis elegans and S.cerevisiae SET1. The SET domain genes S.cerevisiae Ezl‐1 and Drosophila ash‐1 (which seems to be related to trx; Tripoulas et al., 1996) appear to be unique, because in both of these genes the SET domain is not located at the carboxy terminus (subgroup III). Finally, SET domains of subgroup IV genes comprise those present in Su(var)3‐9 related proteins and human G9a and KG‐1.
Possibly, additional structural motifs, like the cysteine‐rich cluster (Jones and Gelbart, 1993) in E(z) homologues, putative zinc fingers and A/T‐hooks in trx/HRX (Tkachuk et al., 1992; Ma et al., 1993; Stassen et al., 1995), a PHD‐finger in ash‐1 (Tripoulas et al., 1996), the chromo box in Su(var)3‐9 (Tschiersch et al., 1994) or ankyrin repeats in G9a (Milner and Campbell, 1993), may impart distinct target specificities to SET domain proteins as they function in chromatin‐dependent gene regulation. Moreover, differences in the primary structure of the SET domain itself may also provide distinct specificities to individual SET domains. In agreement with this interpretation and consistent with the more distant relationship of the SET domain in Su(var)3‐9 with respect to EZH2 and SET1, the human Su(var)3‐9 homologue (SUV39H) did not rescue set1Δ‐dependent telomeric silencing in S.cerevisiae (data not shown).
Although we are currently only beginning to understand the molecular functions of SET domain genes, our analysis of mammalian E(z) proteins provides a direct demonstration for conserved gene silencing in eukaryotes and suggests that the SET domain, like the 50 amino acids comprising the chromo box (Messmer et al., 1992; Platero et al., 1995), may be of primary importance in organizing chromatin domains that regulate gene expression in eukaryotic chromatin.
Materials and methods
Molecular cloning of mammalian E(z) homologues
A DNA probe encoding the conserved SET domain of E(z) (Jones and Gelbart, 1993) was amplified by the polymerase chain reaction (PCR) from 1 μg of Drosophila melanogaster DNA (Clontech) using the two primers E(z)‐1910 (5′‐actgaattCGGCTGGGGCATCTTTCTTAAGG) and E(z)‐2280 (5′‐actctagACAATTTCCATTTCACGCTCTATG). A human B‐cell‐specific (BJAB) λgt10 cDNA library was kindly provided by M.Busslinger. A total of 5×105 plaque‐forming units (p.f.u.) were plated and transferred to GeneScreen membranes as described (Sambrook et al., 1989). The membranes were pre‐hybridized for 30 min at 50°C in Church buffer, and hybridized O/N at 50°C with the E(z) SET domain DNA probe. Following reduced stringency washes for 10 min in 2× SSC, 1% SDS at room temperature (RT) and 10 min at 50°C, positive phages were identified by autoradiography. Single, well‐isolated plaques were obtained after two rounds of rescreening.
Mouse cDNA clones were isolated from a Balb/c brain λgt11 cDNA library (Clontech) by screening, under reduced stringency as above, with a 390 bp EcoRI–XhoI cDNA fragment that encodes the EZH2 SET domain. Mouse genomic clones were isolated from a C57Bl/6 cosmid library (Stratagene) by screening with a 498 bp PCR‐amplified DNA probe that encodes amino acids 242–408 of Ezh1. A membrane representing ∼3×104 cosmid colonies (kindly provided by D.Barlow) was hybridized O/N at 65°C in Church buffer and washed under high stringency (10 min washes in 2× SSC, 1% SDS at RT and at 65°C, followed by a 10 min wash in 0.2× SSC, 1% SDS at 50°C). Positive clones were identified on frozen masterplates and rescreened until single colonies could be identified.
The cDNA inserts of recombinant phages or DNA fragments from the genomic cosmid clone were subcloned into the polylinker of pBluescript KS (Stratagene) and sequenced on an automated sequencer (Applied Biosystems) using the dideoxy method. The complete sequence of independent isolates was determined by primer walking. Sequence data were analysed using the University of Wisconsin Genetics Computer Group sequence analysis software package (Devereux et al., 1984). Homology searches were performed with the Blast (Altschul et al., 1990) network service and also included sequence comparisons with mammalian ESTs (Bassett et al., 1995).
Full‐length cDNAs and 5′ RACE
A full‐length human EZH2 cDNA of 2.6 kb was obtained after combining a 0.7 kb SalI–BglII DNA fragment comprising 5′ untranslated and amino‐terminal sequences from clone λ13 with a Bluescript‐based subclone of λ37. Missing 5′ sequences of murine Ezh1 were generated by RACE amplification (Marathon cDNA amplification kit; Clontech) of 1 μg poly(A)+ RNA from the murine B‐cell line J558L. First‐strand cDNA synthesis was primed with a lock‐docking oligo(dT) primer. After second‐strand synthesis, the cDNA was amplified using a 5′ adapter oligo and the gene‐specific primer Ezh1‐200 (5′‐CATGGACGCAATGGCGCTGAAGATCATGTC). 5′ RACE amplification was performed with a nested primer pair. The resultant 0.6 kb PCR product was trimmed with NotI and ScaI, and combined in several subcloning steps with a Bluescript‐based 3.4 kb subclone of λm31. The complete human EZH2 and murine Ezh1 cDNAs were sequenced over their entire length to confirm the authenticity of the encoded sequences.
DNA blot analysis
Mouse genomic DNA was prepared from kidney, brain and heart of young 129/Sv males. Human male placenta DNA (Sigma) and Drosophila melanogaster DNA (Clontech) was purchased from the indicated suppliers. 15 μg of genomic DNA was digested O/N, fractionated in a 0.8% agarose gel, transferred to a GeneScreen membrane and cross‐linked under UV light. Following hybridization O/N at 42°C in a buffer containing 20% formamide, 5× SSC, 1× Denhardt's, 20 mM NaPO4 (pH 6.7), 0.1 mg denatured salmon sperm DNA (Sigma) per millilitre, the membrane was washed at reduced stringency for 10 min at RT in 2× SSC, 0.1% SDS and for 30 min at 65°C in 1× SSC, 0.1% SDS. DNA probes encoding the carboxy‐terminal SET domains of mammalian E(z) homologues were a 390 bp EcoRI–XhoI fragment from the EZH2 cDNA or a 330 bp EcoRI–XhoI fragment from a modified Ezh1 cDNA subclone in which nucleotides 4 bp downstream of the stop codon have been converted to an XhoI site.
In vitro differentiation of embryonic stem cells
A total of 1.5×106 feeder‐independent mouse (129/Sv) embryonic stem cells (D3) were plated onto gelatin‐coated bacterial dishes (150 cm2) in high‐glucose Dulbecco‘s modified Eagle's medium (DMEM), supplemented with 15% fetal calf serum (Boehringer Mannheim), 2 mM l–glutamine (Gibco), 1% non‐essential amino acids (Gibco), 0.1 mM β‐mercaptoethanol and 1% conditioned medium containing leukaemia inhibiting factor (LIF). LIF‐conditioned medium was harvested from LIFcα‐5 producer cells (Smith et al., 1988). Differentiation was induced with 0.1 μM all‐trans retinoic acid (RA) (Sigma). Following induction for 2 days, floating embryoid bodies (EB) were removed and cultivated (in the absence of RA) for another 5 days.
RNA isolation and analysis
Total RNA was isolated by isopycnic centrifugation as described (Sambrook et al., 1989). Poly(A)+ RNA was purified using a PolyATrack kit (Promega). For RNA blot analysis, 5 μg of poly(A)+ RNA were fractionated on a formaldehyde–agarose gel, transferred to a GeneScreen plus membrane in 10× SSC and baked for 2 h at 80°C. The membrane was sequentially hybridized under stringent Church conditions with DNA probes that comprised a 1.9 kb PCR‐amplified fragment encoding amino acids 173–747 (plus 180 bp of 3′ untranslated sequences) of Ezh1, the 2.6 kb full‐length EZH2 cDNA and a 0.3 kb DNA fragment that is specific for a murine β‐actin cDNA (Alonso et al., 1986). RNase protection analysis was carried out as described (Sambrook et al., 1989). 10 μg of total RNA were incubated at 60°C with a riboprobe that specifically protects 320 bp of the Ezh1 SET domain or at 58°C with a riboprobe that protects 270 bp of Ezh2 amino‐terminal sequences. As a control, a riboprobe that protects 156 bp of murine S16 rRNA sequences (Urbanek et al., 1994) was included in the hybridizations. The murine Ezh1 SET domain riboprobe was derived from a 370 bp PCR‐amplified fragment (encoding amino acids 637–743 of Ezh1) that has been subcloned into pGEM‐3Zf (Promega). After linearization with PvuII, antisense RNA was internally labelled by trancription with SP6 RNA polymerase in the presence of [32P]GTP and the full‐length riboprobe of 470 nucleotides was purified from a denaturing urea–polyacrylamide gel. Following a similar strategy, an Ezh2‐specific riboprobe was generated that protects sequences encoding amino acids 4–92 of Ezh2.
Epitope tag and expression plasmids
A NotI site just downstream of the start codon in EZH2 was inserted by replacing a 0.7 kb BamHI–BglII DNA fragment with a PCR product that was generated with a 3′ primer and the modifying oligonucleotide 5′‐ataggattcATAATCATGggcggccgcACTGGGAAGAAATCTGAGAAGGG (start codon underlined). Overlapping oligonucleotides encoding an in‐frame triple haemagglutinin (HA) tag preceding six histidines were ligated, cleaved with NotI and inserted into the NotI‐digested pEZH2 derivative. The resulting epitope‐tagged variant (HA‐EZH2) extends the amino terminus by 44 amino acids.
For Drosophila expression plasmids, EZH2 and E(z) coding sequences were converted by PCR amplification into NotI fragments and inserted into the NotI site of the P element, rosy+ vector pNHT4 (Schneuwly et al., 1987) which directs expression from the hsp70 promoter. The sequence of the primers for human EZH2 was upstream 5′‐ctaagcggccgcAGGCAGTGGAGCCCCGGC and downstream 5′‐attagcggccgCTGGTACAAAACACTTTGC. The NotI primer pair for Drosophila E(z) was 5′‐ataagaatgcggccgcgtcgacTATAAAGTCCCTCGAAGGCATTATGAATAGC and 5′‐ataagaatgcggccgcgtcgacGATTTTCATAAGTAGACTCGTTATC (start and stop codons underlined). The genomic E(z) sequences are expressed from the natural promoter and the respective construct has been described (Jones and Gelbart, 1993).
For yeast expression plasmids, a 0.7 kb BamHI–EcoRI DNA fragment encoding the Gal1‐10 promoters from Gal‐CLB5 (Schwob and Nasmyth, 1993) was inserted into the polylinker of the low‐copy CEN4‐LEU plasmid YCplac111 (Gietz and Sugino, 1988), resulting in plasmid YCplac111‐G1/10. A 2.4 kb SalI–XhoI DNA fragment encoding EZH2 or a 2.5 kb SalI–XhoI DNA fragment encoding HA‐EZH2 were transferred into the SalI site of YCplac111‐G1/10 in an orientation that allows transcription of the cDNAs by the Gal1 promoter. Ezh1 coding sequences were converted by PCR amplification into a 2.3 kb SalI fragment and inserted into the SalI site of YCplac111‐G1/10 in the same orientation as EZH2 or HA‐EZH2. The Ezh1‐specific primer pair was 5′‐tatcccgggtcgaCCATGAGGAAAATGGATATAGCAAGTCCCC and 5′‐tatctcgagtcgactAGGGCTAGACGTCCGTTTCCCTCTCG (start and stop codons underlined). The SET1 control plasmid contained a 3.8 kb Asp718–HindIII genomic fragment inserted into the polylinker of the CEN6‐LEU plasmid pRS315 (Nislow et al., submitted). For all of the fly and yeast expression plasmids, the 5′ and 3′ boundaries of the respective cDNAs were confirmed by sequencing.
All Drosophila stocks were maintained under standard conditions. A description of chromosomes and mutations is found in Lindsley and Zimm (1992). P element‐mediated germline transformation was carried out by standard methods (Rubin and Spradling, 1982), using a ry506 fly strain as recipient.
The enhancer effect of the transgenes on wm4h PEV was quantified by measuring the absorbance (at 480 nm) of red eye pigments in wm4h/Y; T(2;3)apXa Su(var)2‐101 and wm4h/Y; Cy/+;Sb/+ male offspring after crossing wm4h; Cy/T(2;3)apXa Su(var)2‐101/Sb females to +/Y or wm4h/Y males carrying the respective transgenes heterozygous to CyO or TM3, ryRK Sb e balancer chromosomes. By contrast, the loss‐of‐function mutation E(z)5 (Phillips and Shearn, 1990) and the antimorphic mutation E(z)1 (Kalisch and Rasmuson, 1974) increased the concentration of wm4h red eye pigments in both Cy/+; Sb/E(z) and +/T(2;3)apXa Su(var)2‐101/E(z) genotypes, whereas the temperature‐sensitive E(z)12 allele (MY939) and the hypomorphic E(z)15 mutation (NU808) (Phillips and Shearn, 1990) had no effect at 25°C.
Modification of the zeste1–white eye colour by extra copies of EZH2 or E(z) was analysed in transgenic female and male offspring heterozygous for Dp(1;1)z59d15. The antimorphic E(z)1 mutation (Kalisch and Rasmuson, 1974) has been used for comparison in control matings.
Remobilization of the P[hspE(z)ry+] transgene in line D was activated in males carrying the TM3, ryRK Sb e P[(ry+)Δ2‐3] (99B) transposase source. After a cross to wm4h; Cy/T(2;3)apXa Su(var)2‐101/Sb females, exceptional wm4h/Y; +/T(2;3)apXa Su(var)2‐101/P[hspE(z)ry+] male offspring with a significant enhancer effect were selected. In all proven exceptions (eight out of 2329 F1 males), red eye pigment measurements from wm4h/Y; +/T(2;3)apXa Su(var)2‐101/P[hspE(z)ry+] males indicated a significant reduction in the proportion of red eye pigments.
Yeast transformations and 5‐FOA dilution assays
Saccharomyces cerevisiae cells were grown at 30°C in complete (YPD) or selective media. One gram of 5‐FOA (Diagnostic Chemicals) per litre of medium was added to determine resistance to 5‐FOA. Yeast transformations were performed by the lithium–acetate protocol. The yeast strains used in this study are UCC1001 [MATa his3‐Δ200 trp1‐Δ1 leu2‐Δ1 ura3‐52 ade2‐101 lys2‐801 adh4::URA3 (URA3 @ VII‐L), Renauld et al., 1993], LPY1297 (isogenic to UCC1001, except set1Δ::HIS3), LPY1683 (MATa his3‐Δ200 trp1‐Δ63 leu2‐Δ1 ura3‐52 ade2‐101 lys2‐801 adh4::URA3 (URA3 @ VII‐L) DIA5‐1 (ADE2 @ V–R); a derivative of UCC3505, Singer and Gottschling, 1994) and LPY1727 (isogenic to LPY1683, except set1Δ::HIS3)
5‐FOA dilution assays were carried out by growing transformed cells picked from single colonies in −LEU/2% raffinose media for 3–4 days at 30°C until the culture reached saturation. Cells were concentrated by centrifugation, counted in an automatic cell counter (CASY1) and adjusted to a concentration of 200 000 cells/μl. The cell suspension was added to a 96‐well microtitre plate and serial 1:10 or 1:5 dilutions were made by using an eight‐channel pipettor. Dilutions were transferred onto YPD and selective media plates with a stainless steel pin replicator. Colonies were inspected after 2–4 days of incubation at 30°C.
Note added in proof
While this manuscript was in press, the curly leaf mutant was shown to encode a plant member of the E(z) gene family that is involved in regulating homeotic gene expression in Arabidopsis, thereby extending conserved mechanism(s) controlling gene activities in eukaryotic chromatin (Goodrich,J., Puangsomlee,P., Martin,M., Long,D., Meyerowitz,E.M. and Coupland,G. (1997) A Polycomb‐group gene regulates homeotic gene expression in Arabidopsis. Nature, 386, 44–51.
We would like to thank Meinrad Busslinger for the human BJAB cDNA library, Rick Jones and Allen Shearn for the E(z) cDNA and for (E)z mutant fly strains, Dan Gottschling for SIR mutant yeast strains, Ken Abel for DNA sequence information on human EZH1 prior to publication, Gotthold Schaffner for oligonucleotide synthesis, Robert Kurzbauer and Ivan Botto for sequence analysis, and Meinrad Busslinger and Hartmut Beug for helpful comments and critical reading of the manuscript. This work was supported by the IMP and, in part, by the Austrian Industrial Research Promotion Fund. G.R. and R.D. were supported by grants from the Deutsche Forschungsgemeinschaft (DFG‐Re911/1‐3 and DFG‐Do407/1‐2). C.N. was a postdoctoral fellow of the American Cancer Society and research in L.P.'s laboratory was supported by grants from the National Science Foundation (MCB9257685) and the Pew Charitable Trusts (T89‐03508‐044).
- Copyright © 1997 European Molecular Biology Organization