In development, cells acquire identities as a consequence of interactions with each other or the environment, or by asymmetric localization of RNAs or proteins after cell division. These transient events ultimately cause changes in gene expression patterns that define the cell's identity. Maintenance of gene expression patterns is a key component of maintaining cell identity. As cells differentiate in response to intrinsic and extrinsic signals, changes in specific gene expression patterns may also be required. There must be a heritable mark that is stable to DNA synthesis and mitosis that identifies whether a gene is to be expressed or repressed. One of the key problems in biology is to understand the mechanisms of maintenance, and the nature of the heritable mark. Disruption in maintenance of gene expression patterns can have disastrous effects in development, or may result in cancerous transformation (Jacobs and van Lohuizen, 2002; Otte and Kwaks, 2003; Hake et al., 2004; Roberts and Orkin, 2004).
Mechanisms for maintenance must be suitably flexible to take into account the need for changes in patterns of gene expression throughout the process of differentiation (Moazed, 2001; Gaston and Jayaraman, 2003; Smale and Kadonaga, 2003). Underlying DNA sequence is generally too stable to support such flexibility, so maintenance of gene expression patterns through cell divisions must be epigenetic (Cavalli, 2002). Epigenetic regulatory events include mating type silencing in yeast, position-effect variegation, gametic imprinting, and X chromosome inactivation. Epigenetic phenomena share mechanisms, including DNA methylation, modification of histones, use of histone variants, nucleosome remodeling, subnuclear localization, and establishment of specific chromatin structures (Carmo-Fonseca, 2002; Kornberg and Lorch, 2002; Felsenfeld and Groudine, 2003; Vermaak et al., 2003).
The best-studied proteins required for maintenance (maintenance proteins, or MPs) are the Polycomb group (PcG) and trithorax group (trxG) proteins. The corresponding genes were first identified in Drosophila (Slifer, 1942; Lewis, 1947, 1978; Shearn et al., 1971; Garcia-Bellido, 1977; Ingham and Whittle, 1980). It is estimated that there are 30–40 PcG genes in Drosophila (Jurgens, 1985; Landecker et al., 1994) of which 20 have been cloned and characterized (Table 1, and references therein). There are 15 well-characterized trxG genes in Drosophila (Table 2, and references therein).
Table 1. Polycomb Group Gene Families and Conserved Protein Domainsa
PcG gene family
Genes reclassified to the Enhancer of trithorax and Polycomb (ETP) group and/or that are considered to have dual roles in gene activation and repression are underlined. Full gene names are given for the Drosophila genes that represent each gene family; refer to cloning references for other gene name and domain name abbreviations.
Table 2. Selected Trithorax Group (trxG) Gene Families and Conserved Protein Domains
trxG gene family
aGenes reclassified to the Enhancer of trithorax and Polycomb (ETP) group are underlined. Full gene names are given for the Drosophila genes that represent each gene family; refer to cloning references for other gene name and domain name abbreviations.
Most MPs have homologs in species other than Drosophila (Tables 1, 2). Because of genome duplication, typically mammals have at least two homologs. Where tested, MP homologs appear to have similar functions in different species (Mahmoudi and Verrijzer, 2001; Ross and Zarkower, 2003; Zhang et al., 2003). Most MPs have domains of conserved sequence (see Tables 1, 2 and references therein), found in other chromatin proteins, which mediate molecular interactions; examples include the chromodomain, bromodomain, zinc fingers (PHD, RING, or other types), SAM/SPM domain, and WD40 repeats, consistent with a role for MPs in multimeric complexes (see below). Few MPs have enzymatic activity. One PcG protein (Enhancer of zeste (E(z)) and homologs) and two trxG proteins (trithorax (trx) and Abnormal, small, and homeotic 1 (Ash1) and their homologs) are histone lysine methyltransferases (HKMTs), an activity mediated by the conserved Su(var)3-9, Enhancer of zeste, trithorax (SET) domain (reviewed by Marmorstein, 2003). Some MPs interact with proteins or belong to complexes that possess protein modification enzymatic activity. It is noteworthy that many MPs have RING domains, which generally are ubiquitin E3 ligases (Aravind et al., 2003). In Drosophila, mutations of UbcD1/eff, a ubiquitin-conjugating enzyme, enhance the homeotic phenotype of polyhomeotic (ph) mutations (Fauvarque et al., 2001). Recently, a ubiquitin E3 ligase complex specific for histone H2A was purified from human cells and was shown to contain several PcG proteins, including Ring2 that by itself possesses enzymatic activity on oligonucleosome substrates in vitro, mediated by the RING domain (Wang et al., 2004a). In humans, the Polycomb (Pc) homolog Pc2 is required for lysine sumoylation of the carboxy-terminus binding protein (CtBP), a transcriptional corepressor, which is likely due to co-recruitment by Pc2 of CtBP and Ubc9, a small ubiquitin-related modifier (SUMO) E2-conjugating enzyme (Kagey et al., 2003). The SAM domain in the PcG protein SOP2 of Caenorhabditis elegans was also shown recently to interact with UBC9, and sumoylation of SOP2 is required for Hox gene repression in C. elegans (Zhang et al., 2004).
GENETICS OF MPS
With some exceptions, in Drosophila, PcG mutants exhibit posterior homeotic transformations in the abdomen and homeotic transformation of second and third thoracic segments toward more anterior thoracic segments (reviewed by Kennison, 2004). As these phenotypes resemble gain of function mutations in homeotic loci, it was predicted on genetic grounds that PcG genes are required to repress homeotic loci (Lewis, 1978). Subsequently, molecular studies confirmed that homeotic genes are ectopically expressed in PcG mutant embryos, consistent with a role in repression (Struhl and Akam, 1985; McKeon and Brock, 1991; Simon et al., 1992). Mutations in PcG genes enhance each other (Jurgens, 1985; Cheng et al., 1994).
Conversely, mutants of trxG genes exhibit anterior transformations in the abdomen and transformations of the first and third thoracic segments toward the second (Kennison, 1993, 1995). Because the phenotypes of trxG mutations resemble loss of function homeotic gene mutations, it was concluded that trxG genes are required for activation of homeotic loci. Mutations in some trxG genes enhance each other (Shearn, 1989). Most trxG proteins were discovered because they suppress the homeotic transformation phenotypes of PcG proteins (Kennison and Russell, 1987; Kennison and Tamkun, 1988; Shearn, 1989). Mutations in trxG genes reduce homeotic gene expression within their normal domains (Mazo et al., 1990; Breen and Harte, 1993).
The phenotypes of MP mutants in flies and mammals preclude simple assignment into the PcG and trxG groups. In Drosophila, Polycomb mutants exhibit minor anterior transformations in caudal abdominal segments (Denell and Frederick, 1983), which is unexpected because derepression of Abdominal-B (Abd-B) should cause posterior transformations. In mammals, both M33 (Pc homolog; Core et al., 1997; Bel et al., 1998; Katoh-Fukui et al., 1998) and Mixed lineage leukemia (Mll; a trx homolog; Yu et al., 1995, 1998; Hanson et al., 1999) mutants exhibit bidirectional homeotic transformations of the anteroposterior axis, which is unexpected for PcG and trxG gene mutations. On the other hand, compound mutants in mice for the Mll- and B-cell–specific Moloney murine leukemia virus insertion site 1 (bmi1; a Psc homolog) genes do show antagonistic effects on differentiation of the anteroposterior axis as expected for a compound trxG, PcG gene mutation (Hanson et al., 1999). However, compound null homozygous mouse mutants for Melanoma-18 (Mel18) and Bmi1 (both are Psc homologs) show significantly reduced expression of Hoxb6 and Evx1 in the presumptive trunk region of 9.5 dpc embryos, while also showing ectopic expression of other Hox genes as expected for a PcG mutant, suggesting that these genes are indeed needed to maintain both gene activation and repression (Akasaka et al., 2001). Mutants of Ring/Sex combs extra (Sce) in Drosophila show strong posterior homeotic transformations as expected for a PcG gene (Breen and Duncan, 1986; Campbell et al., 1995; Fritsch et al., 2003), whereas both LOF and GOF Ring1a/Ring1 mouse mutants unexpectedly exhibit anterior transformations (del Mar Lorente et al., 2000). One strain of hypomorphic Ring1b/Rnf2 mouse mutant shows mild posterior homeotic transformations (Suzuki et al., 2002).
We believe that the role that indirect effects play in the genetics of MPs has been underappreciated. One possibility that could explain contradictory MP mutant homeotic phenotypes are altered or compensatory cross-regulatory interactions between homeotic cluster/Hox loci in MP mutants (Kennison, 2004); however, this possibility remains to be demonstrated experimentally. Another possibility is that PcG and trxG genes could regulate themselves, as suggested by the observation that PcG proteins bind on polytene chromosomes to regions containing PcG genes (Zink and Paro, 1989; DeCamillis et al., 1992). Strutt and Paro (1997b) showed that a gene near the invected locus was regulated by PcG genes; this gene was subsequently identified as toutatis, a member of the trxG (Fauvarque et al., 2001). ph, Pc, and Posterior sex combs (Psc) all bind to a maintenance element (ME) upstream of ph (Bloyer et al., 2003). Further evidence for cross-regulatory interactions between MP genes comes from a recent study in Drosophila examining expression of PcG genes in PcG mutant embryos, where it was found that three PcG genes, ph, Pc, and Psc, negatively regulate transcript levels of Psc and Suppressor 2 of zeste (Su(z)2) and perhaps Ring/Sce, whereas Additional sex combs (Asx), Polycomblike (Pcl), and Enhancer of Polycomb (E(Pc)) positively regulate Pc and Psc (Ali and Bender, 2004). Cross-regulatory interactions between MP and/or homeotic genes could be the underlying cause of complex phenotypes of MP gene mutants. These possibilities require further investigation, particularly in mammals that possess multiple MP gene homologs.
In Drosophila, MPs regulate numerous other target loci in addition to homeotic genes (Zink and Paro, 1989; DeCamillis et al., 1992). In mice, mutants in several MP genes die early in embryogenesis, before the transcription of homeotic cluster/Hox genes begins at approximately 8.5 dpc (Schumacher et al., 1996; Donohoe et al., 1999; O'Carroll et al., 2001; Ayton et al., 2001; Voncken et al., 2003). Mutations in MP genes upset processes as diverse as position-effect variegation (Sinclair et al., 1983; Yamamoto et al., 2004); telomeric position-effect (Boivin et al., 2003); imprinted and random X-inactivation (Wang et al., 2001; Plath et al., 2003; Silva et al., 2003; Cao and Zhang, 2004) and autosomal locus imprinting (Mager et al., 2003) in mammals; silencing of the germline X chromosome in C. elegans (Xu et al., 2001; Pirrotta, 2002; Bean et al., 2004) and genomic imprinting in plants (Grossniklaus et al., 1998; Hsieh et al., 2003). MPs have also been implicated in control of eye development (Takihara et al., 1997; Gregg et al., 2003; Janody et al., 2004) and chromosome segregation during meiosis and mitosis (Lupo et al., 2001; Balicky et al., 2004). MPs are needed for RNA interference (Dudley et al., 2002; Pal-Bhadra et al., 2002) and are involved in hematopoiesis and oncogenesis (Jacobs and van Lohuizen, 2002; Ema and Nakauchi, 2003; Kleer et al., 2003; Lessard and Sauvageau, 2003; Raaphorst, 2003; Roberts and Orkin, 2004). These examples show the ubiquity of processes requiring MPs. However, it is striking that little is known about MP targets other than homeotic/Hox genes (for exceptions, see Jacobs et al., 1999; Americo et al., 2002; Maurange and Paro, 2002; Bloyer et al., 2003; Chanas et al., 2004; Kirmizis et al., 2004).
Many other examples of contradictory phenotypes of MP mutants, despite common underlying molecular biology, can be discovered in analysis of MP phenotypes in mammalian hematopoiesis and oncogenesis. For example, Bmi1 and Mel18 are both paralogs of Psc yet have opposing roles as an oncogene (Alkema et al., 1993) and tumor suppressor (Kanno et al., 1995), respectively, and they have opposing roles in hematopoietic stem cell self-renewal (Park et al., 2003, 2004; Kajiume et al., 2004). Paradoxically, the trx homolog Mll appears to have a similar role in hematopoietic stem cell renewal as the PcG genes Bmi1 and Retinoic-acid early 28 (Rae28/mph1; a ph homolog; Ohta et al., 2002; Ernst et al., 2004), as all three of these genes are required to maintain hematopoietic stem cell self-renewal activity over time (recently reviewed by Valk-Lingbeek et al., 2004). Confusingly, mutations in the embryonic ectoderm development (eed; an esc homolog) gene lead to hyperproliferative defects in myelo-erythropoiesis and B lymphopoiesis and, thereby, behave oppositely to other MP gene mutations, and yet eed mutants show hypoproliferative defects in T-cell development that are similar to observations of other MP mutants (Lessard et al., 1999; Richie et al., 2002; Lessard and Sauvageau, 2003).
Unless their molecular biology is unexpectedly complex, it seems safest to assume that the molecular function of PcG and trxG genes is conserved between flies and mammals. As noted above, further attention should be paid to indirect effects of MPs, particularly cross-regulation between MPs, which would confuse phenotypic analysis of MP mutants. Additionally, MPs could have different targets in different tissues and may not be obligatorily associated at all targets. We favor the interpretation that there will be many specific MP complexes assembled at specific loci in different developmental stages or tissues (Hodgson et al., 2001; Wang and Brock, 2003), a point made cogently by Otte and Kwaks (2003; and see below). Continued emphasis on molecular analyses in addition to genetic analyses of function and defining the key direct target genes for MPs in different developmental contexts should help to resolve these issues in the future.
MAINTENANCE PROTEIN COMPLEXES
Most characterized MPs are members of multimeric complexes. Some trxG proteins are members of ATP-dependent nucleosome remodeling complexes (Simon and Tamkun, 2002; Langst and Becker, 2004; Sif, 2004). The Drosophila trxG proteins Brahma (Brm), Osa, Moira (Mor), and Snr1, are members of the Brahma Associated Protein (BAP) complex, and (with the exception of Osa) of the PBAP complex that also contains the proteins Polybromo and BAP170 (Mohrmann et al., 2004). Brahma is similar to the core ATPase subunits from the SWI/SNF and RSC complexes of yeast and the related BAF and PBAF complexes, respectively, in humans (Mohrmann et al., 2004). The trxG protein Kismet shows sequence similarity to Brahma and is a member of a yet uncharacterized complex (Daubresse et al., 1999).
The PcG complex PRC1 of Drosophila contains Pc, Ph, Psc, and Ring, as well as non-PcG protein subunits (Shao et al., 1999; Saurin et al., 2001; Levine et al., 2002). Related PRC1-type complexes have also been purified from mammalian cells (Levine et al., 2002; Lavigne et al., 2004; Wang et al., 2004a). Interestingly, PRC1-type complexes have been implicated recently in mediating histone H2B ubiquitination (Wang et al., 2004a). PRC1-type and/or recombinant PRC1 Core complexes (for example, in Drosophila, containing just Pc, Psc, Ph, and Ring) act in vitro to inhibit chromatin remodeling of nucleosome arrays by the human SWI/SNF complex (Shao et al., 1999; Francis et al., 2001), inhibit transcription by RNA polymerase II (King et al., 2002) and link nucleosome arrays in trans (Lavigne et al., 2004). Because the trxG proteins Brm, Mor, and Osa are all members of a SWI/SNF-like complex, it is possible that PRC1 prevents trxG-mediated nucleosome remodeling necessary for transcription to proceed.
Other trxG proteins are members of histone modifying complexes (Petruk et al., 2001; Kouzarides, 2002; Khorasanizadeh, 2004). MLL and the closely related MLL2 are members of distinct HKMT complexes in humans that are very similar to the Complex Proteins Associated with Set1 (COMPASS) HKMT complex of yeast (Miller et al., 2001), except that they each have MENIN (MEN) as an additional subunit (Hughes et al., 2004; Yokoyama et al., 2004). The MEN1 gene is a tumor suppressor, and MENIN is thought to function as both a transcriptional activator and repressor, it associates with RNA polymerase II, binds to Hox loci, and regulates Hox gene expression (Hughes et al., 2004; Yokoyama et al., 2004). Given the evolutionary conservation between the COMPASS, MLL, and MLL2 complexes, it seems likely that related complexes will be isolated in most organisms and that these complexes will have similar functions as HKMTs.
MLL has also been reported to be a subunit of a much larger supercomplex in humans (Nakamura et al., 2002), which perhaps reflects differences in the range of proteins with which MLL (and homologous proteins including Trx) can interact (Rozenblatt-Rosen et al., 1998; Rozovskaia et al., 1999, 2000; Ernst et al., 2001; Marenda et al., 2003; Rudenko et al., 2003; Xia et al., 2003). Trx is a member of the TAC1 complex in Drosophila that contains the histone acetyltransferase (HAT) coactivator CREB binding protein (CBP) and the phosphatase-related protein Sbf1 (Petruk et al., 2001; also see below). However, homologs of CBP and Sbf1 are not found in the COMPASS complex or the MENIN-containing MLL and MLL2 complexes, despite that CBP interacts directly with the C-terminal domain of MLL (Ernst et al., 2001). In Drosophila, CBP interacts directly with another trxG protein, ASH1 (Bantignies et al., 2000). It is likely that CBP interacts with Trx/MLL and ASH1 only under specific conditions at limited loci (Bantignies et al., 2000; Petruk et al., 2001, 2004). Because histone acetylation correlates with gene activity, it is possible that recruitment of the HAT protein CBP to particular loci by trxG proteins confers part of the gene activation function of this group of MPs.
The trxG proteins TRX and ASH1 (but not ASH2 as it lacks a SET domain) are HKMTs. ASH1 and ASH2 belong to distinct protein complexes in Drosophila (Papoulas et al., 1998); however, the yeast HKMT COMPASS complex contains two ASH2 homologs in addition to the TRX homolog SET1 (Nagy et al., 2002), and the human MLL and MLL2 complexes containing MENIN described above also each contain an ASH2 homolog (Hughes et al., 2004; Yokoyama et al., 2004). The molecular function of ASH2 is unknown. Drosophila TRX and murine homolog Mll methylate lysine 4 of histone H3 (H3K4; Petruk et al., 2001; Milne et al., 2002; Nakamura et al., 2002), a histone mark associated with gene activation. ASH1 has a wider specificity, and methylates lysines 4 and 9 of histone H3, and lysine 20 of histone H4 (Beisel et al., 2002; Byrd and Shearn, 2003). Methylation of H3K4 and H3K9 residues was shown to recruit two trxG proteins that are members of the Brm complex, Brm and Mor, and to inhibit binding of PcG proteins; methylation of H3K4 alone also inhibits PcG binding (Beisel et al., 2002). Together, these results suggest that histone methylation induced by trxG proteins might be a heritable mark for transcription, and may prevent binding by PcG proteins (but see below). It also appears possible that interactions between different trxG protein complexes mediating HKMT activity and nucleosome remodeling, respectively, may occur, since direct interaction between trx/MLL and the Brm complex subunit Snr1/INI1 has been reported (Rozenblatt-Rosen et al., 1998; Marenda et al., 2003).
The Drosophila PRC2 PcG complex contains E(z), Extra sex combs (Esc), and Suppressor of zeste 12 (Su(z)12); some variants also contain Pcl (Czermin et al., 2002; Muller et al., 2002). Related complexes exist in mammals, plants, and C. elegans (Cao et al., 2002; Kuzmichev et al., 2002; Pirrotta, 2002; Hsieh et al., 2003; Pires-daSilva and Sommer, 2003). PRC2 contains histone H3 lysine 27 (H3K27) and perhaps H3 K9 HKMT activity, for which the SET domain of E(z) is necessary (Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002, 2004; Muller et al., 2002; Cao and Zhang, 2004). A similar complex to PRC2 in humans, with a different Esc isoform has been described recently and named PRC3, which can also methylate lysine 26 of histone H1 (Kuzmichev et al., 2004). The chromodomain of Pc binds lysine 27 methylated histones with higher affinity than for unmodified histones, suggesting a stepwise recruitment model reminiscent of that demonstrated for recruitment of HP1 after methylation of histone H3K9 by Su(var)3-9 (Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002; Muller et al., 2002). Wang et al. (2004b) also have provided recently evidence for the hierarchical recruitment model that E(z)-mediated methylation of H3K27 enables Pc binding.
PRC2 and PRC1 interact transiently, in early preblastoderm Drosophila embryos but not (at least stably) in subsequent stages (Poux et al., 2001b). If the sole function of PRC2 is to recruit PRC1 as a consequence of H3K27 methylation, it is surprising that, when the PRC1 component Pc is tethered to target loci, mutations in the PRC2 component gene esc abolish silencing (Poux et al., 2001a) and that tethered Pc can also recruit Esc (Poux et al., 2001b). These observations suggest that PRC2 (and likely PRC3) complexes must possess other functions besides Pc recruitment. Breiling et al. (2004) recently provided evidence that binding of the PRC2 component E(z) to target gene promoters depends upon Pc and is continually required for maintenance of gene repression (also see below).
There are reports of histone deacetylases associating with PcG proteins of both PRC1 and PRC2 (van der Vlag and Otte, 1999; Chang et al., 2001). Highly purified PcG complexes lack deacetylase activity, so it is still not clear if histone deacetylation has a role in PcG function. It is possible that HDAC activity at PcG target loci is a necessary preliminary step for PcG-mediated gene repression to occur but is not sufficient for PcG function. In yeast, histone H2B ubiquitination is required for H3 methylation and gene silencing (Dover et al., 2002; Sun and Allis, 2002); however, H2A ubiquitination mediated by a PRC1-type complex in human cells is independent of histone H3K27 methylation, so it remains unclear how H2A ubiquitination contributes to PcG silencing (Wang et al., 2004a). Clarifying the role of “cross-talk” between different histone modifications and their effect on maintenance is a challenge for future studies.
It is clear that we are only beginning to understand the heterogeneity of MP complexes. For example, an interesting recent report shows that human MLL interacts with HPC2 and BMI1 (Xia et al., 2003), but no complex containing these three proteins has been identified so far. Although the data so far are consistent with roles of MPs in histone modification and nucleosome remodeling, it seems likely that there will be further surprises as more MP functions are characterized in the future.
REVISITING ETPS AND THEIR CONTRIBUTION TO MP COMPLEXES
Mutants for two genes in Drosophila originally classified as PcG genes, Additional sex combs (Asx) and E(z), exhibit anterior and posterior transformations simultaneously, suggesting that these two genes are required for both silencing and activation of homeotic genes (Sinclair et al., 1992; LaJeunesse and Shearn, 1996; Milne et al., 1999). A deletion screen covering the Drosophila genome showed that mutations in six PcG genes (Asx, E(z), E(Pc), Psc, Sex combs on midleg (Scm), and Suppressor 2 of zeste (Su(z)2)) surprisingly enhanced homeotic phenotypes of trxG mutations, whereas mutations in other PcG genes suppressed trxG mutations as expected (Gildea et al., 2000). Mutations in all six of the genes listed above also enhance homeotic phenotypes of PcG mutations (Cheng et al., 1994; Campbell et al., 1995). On genetic grounds, Gildea et al. (2000), therefore, reclassified these six genes into the Enhancer of trithorax and Polycomb (ETP) group (Fig. 1A).
The molecular basis for ETP phenotypes is poorly defined. ETP genes could be activators of both PcG and trxG genes. A good candidate for an ETP that activates MP genes is E(Pc) (Sinclair et al., 1998a; Stankunas et al., 1998), which is a subunit of related NuA4 histone acetyltransferase complexes in yeast, C. elegans, and mammals (Fuchs et al., 2001; Boudreault et al., 2003; Carrozza et al., 2003; Ceol and Horvitz, 2004); however, it is not known if MP genes are direct targets of these HAT complexes. E(Pc) mutations do not show homeotic transformation phenotypes on their own (Sato, 1983); therefore, it is likely that E(Pc) does not play a direct role in maintenance processes.
ETP group genes may possess joint roles in maintenance of gene activation and repression (Gildea et al., 2000; Brock and Lohuizen, 2001). There are three promising candidates to be bona fide ETPs in that they directly mediate maintenance of activation as well as repression of target loci, although their status as members of MP complexes remains to be clarified. The GAGA factor (GAF) synergizes with ISWI nucleosome remodeling factors (Tsukiyama et al., 1995) and is thought to establish nucleosome-free regions (Lehmann, 2004). GAF also associates with FACT, a complex needed for transcription of chromatin at homeotic loci (Shimojima et al., 2003). Some GAF isoforms have been identified as subunits in PcG complexes (Hodgson et al., 2001; Faucheux et al., 2003; Mishra et al., 2003; Salvaing et al., 2003). The genetics of Trithoraxlike (Trl), which encodes the GAF and is a member of the trxG, is equivocal. Many authors report that Trl mutations enhance PcG mutations or are needed for silencing (Hagstrom et al., 1997; Strutt and Paro, 1997a; Gildea et al., 2000; Horard et al., 2000; Busturia et al., 2001; Mishra et al., 2003), but Huang et al. (2002) do not see these effects. Bejarano and Busturia (2004) propose that GAF has a role in establishing a chromatin ground state needed for gene regulation by either PcG or trxG. This proposal is supported by the observation that GAF facilitates binding of Pho to a chromatinized template (Mahmoudi et al., 2003).
Another good candidate to be a bona fide ETP is Zeste, which is required for transcriptional activation, is a member of PRC1 (Saurin et al., 2001), and enhances PRC1 function (Mulholland et al., 2003). Zeste is also needed to recruit the Brm complex to target gene regulatory maintenance elements (see below; Dejardin and Cavalli, 2004). However, Zeste has not been identified so far as a subunit of a trxG complex. Batman is a third potential bona fide ETP, as it interacts with GAF, is found at maintenance elements, and is needed for both activation and repression (Faucheux et al., 2003).
Other ETPs do not yet offer a clear molecular explanation of their role within this group. Nothing is known about the molecular biology of Asx or Su(z)2; therefore, whether or not these two proteins are bona fide ETPs remains to be determined. E(z) is a subunit of the PRC2 and PRC3 complexes, and Psc and Scm are members of the PRC1 complex. Hence, on molecular grounds, these proteins are poor candidates to be bona fide ETPs, because they have been found exclusively in multimeric PcG complexes. Although it is possible that Psc, Scm, and E(z) are members of novel complexes with uncharacterized functions, we propose that they continue to be classified as members of the PcG (Fig. 1B). We propose that classification of MPs be based exclusively on molecular evidence, as shown in Figure 1B, as this will avoid classification problems caused by indirect genetic effects.
In Drosophila, MPs act through complex DNA elements termed maintenance elements (MEs; Brock and Lohuizen, 2001) or cellular memory modules (Cavalli and Paro, 1998). MEs bind both PcG and trxG proteins in vivo at intermingled but physically separable regions (Tillib et al., 1999) and are required continuously for maintenance (Busturia et al., 1997). In Drosophila MEs range from approximately 100 bp to several kb, can be located near or far from the promoter, and have been identified in homeotic (Simon et al., 1990; Muller and Bienz, 1991; Chan et al., 1994) and nonhomeotic targets (Kassis, 1994, 2002; Maurange and Paro, 2002; Bloyer et al., 2003; Chanas et al., 2004). To date, no one has identified an ME in organisms other than flies, although a recent report has identified an Ezh2 (an E(z) homolog) binding site near the Hoxa9 locus (Cao and Zhang, 2004). A crucial task for the future is to identify and characterize vertebrate MEs.
It is not obvious how MPs are recruited to MEs. Only two MPs, Pleiohomeotic (Pho) and the closely related protein Pleiohomeotic-like (Pho-l), are sequence-specific DNA binding factors (Brown et al., 1998, 2003). It has been reported recently that Pho and Pho-l recruit the PcG protein complex PRC2, which methylates histone H3K27, which in turn recruits the distinct PcG complex PRC1 (Wang et al., 2004b), because Pc binds preferentially to methylated (as opposed to unmethylated) K27 on histone H3 tails in vitro (Wang et al., 2004b; and references therein). Consistent with these observations, the mammalian homolog of Pho, Yin Yang 1 (YY1), associates with both PRC1 and PRC2 components in vivo (Garcia et al., 1999; Satijn et al., 2001). Mutations in pho and pho-l do not prevent recruitment of PcG proteins to MEs at most polytene chromosome sites in vivo (Brown et al., 2003). Wang et al. (2004b) suggest that Pho and Pho-l are required in diploid cells to recruit PRC2 and PRC1 after mitosis but are not required in nonmitotic cells. It seems unlikely that Pho and Pho-l are sufficient to specify an ME, because Pho binding sites are insufficient to reconstitute maintenance in vivo (Strutt and Paro, 1997a; Mohd-Sarip et al., 2002). Forced recruitment of lexA-Pho fusions to reporters is not sufficient for silencing (Poux et al., 2001a).
These observations suggest that other proteins must act to recruit MPs. Candidate proteins include GAF (see above) which binds GAGA sequences (Biggin and Tijian, 1988; Tsukiyama et al., 1994; Katsani et al., 1999) and is normally considered an anti-repressor (Kerrigan et al., 1991; Okada and Hirose, 1998). GAF facilitates PHO binding to chromatinized templates in vitro (Mahmoudi et al., 2003). Another transcription factor with the same recognition sequence as GAF, called Pipsqueak (Lehman et al., 1998), has been identified as a subunit of a complex containing Ph (Hodgson et al., 2001). Zeste recruits Brm and is required to maintain activation (Mulholland et al., 2003; Dejardin and Cavalli, 2004). Americo et al. (2002) have identified many different binding sites in the engrailed ME, consistent with a role for multiple DNA recognition factors at MEs. In a very interesting study, Ringrose et al. (2003) have developed an algorithm to predict the presence of MEs, based on the presence of multiple binding sites for GAF, Pho, and Zeste. Analysis of the predicted MEs identifies three other sequences that may have a role in ME function. Corto interacts with GAF and may recruit PcG proteins to ME (Salvaing et al., 2003). Comparison of the bxd ME from different Drosophila species shows that binding sites for PHO, GAF, and Zeste are conserved, but their relative positions are not conserved (Dellino et al., 2002). If it is true that there are multiple MP complexes, then it might be expected that MEs should also be variable and be bound by different factors.
MEs have many properties of enhancers and silencers, including position and orientation independence. Proteins that bind MEs might also interact with the promoter to form stable chromatin loops that interfere with transcription, as originally proposed by Pirotta and Rastelli (1994). This model is consistent with the observation that TAFs are subunits of PRC1 (Breiling et al., 2001; Saurin et al., 2001) and that PcG proteins bind near promoters (Orlando et al., 1998). There is intriguing structural evidence consistent with this idea. Min et al. (2003) solved the crystal structure of Pc bound to methylated peptide from the histone H3 N terminus. Surprisingly, they found that Pc forms dimers, raising the possibility that Pc dimers link nucleosomes by means of the N-terminal tails of histone H3, leading to a repressive chromatin structure. GAF appears to have a role in stabilizing loops between enhancers and promoters, by means of GAF oligomerization mediated by the POZ domain, both in cis and in trans (Mahmoudi et al., 2002; Lehmann, 2004).
MEs may mediate changes in chromatin structure, facilitated by the proteins bound to the ME. MEs are known to associate in vivo, both intra- and interchromosomally (Lavigne et al., 2004; and references therein), perhaps by mechanisms similar to those involved in enhancer–promoter looping (recently reviewed by Chambeyron and Bickmore, 2004). Two recent reports suggest that MPs mediate interactions between nucleosome arrays, consistent with the idea that MEs nucleate compaction of chromatin structure (Bantignies et al., 2003; Lavigne et al., 2004). MPs do not appear to spread past the ME, arguing that MPs are not themselves essential structural proteins of repressive or active chromatin but that they may bring about chromatin changes. Although different studies have reached somewhat different conclusions (Schlossherr et al., 1994; Boivin and Dura, 1998; Fitzgerald and Bender, 2001), the balance of the evidence suggests that target loci become less accessible to other proteins when repressed by PcG proteins.
DO MPS ACT IN TRANSCRIPTIONAL INITIATION OR ELONGATION?
For many years there has been circumstantial evidence that MPs might have a role in early phases of transcription of target genes. These roles could include recruitment of the basal transcription complex (initiation) or in altering chromatin or protein structure to allow elongation and transcription of RNA. Early chromatin immunoprecipitation (ChIP) experiments showed that both PcG and trxG proteins were present at promoters of homeotic loci in Drosophila and were recruited early in embryogenesis, before the stages showing homeotic gene misexpression in PcG and trxG mutants (Orlando et al., 1998). Similar observations have been made for MLL at Hox loci in mammals (Milne et al., 2002; Nakamura et al., 2002). Many of the PREs predicted by Ringrose et al. (2003) are found near transcription start sites. Attempts to purify PcG proteins showed that TBP-associated factors (TAFs) essential for initiation copurified with proteins of the Drosophila PRC1 complex (Shao et al., 1999; Breiling et al., 2001; Saurin et al., 2001; Wang and Brock, 2003), consistent with a role for PcG proteins at the promoter. Curiously, TAFs do not appear to copurify with the mammalian PRC1 complex (Levine et al., 2002). Intriguingly, recent observations in yeast suggest that different ISWI-type chromatin-remodeling complexes (containing the same ATPase subunit ISW1) regulate distinct phases of transcription: initiation, elongation, and also termination (Morillon et al., 2003). These observations raise the possibility that other nucleosome remodeling complexes containing trxG members may also play distinct and varied roles in initiation and elongation; however, this hypothesis remains to be tested.
As mentioned above, the Set1 HKMT of Saccharomycetes cerevisiae is a member of the COMPASS complex (Miller et al., 2001), which is conserved in mammals (Hughes et al., 2004; Yokoyama et al., 2004). Surprisingly, the COMPASS complex has a key role in elongation (Dover et al., 2002; Krogan et al., 2003; Wood et al., 2003). If the evolutionary conservation between Set1 and Trx homologs holds, it is reasonable to predict that Trx homologs will also have a role in transcriptional elongation in higher eukaryotes.
Three recent studies have strengthened the evidence that MPs have a role in transcriptional elongation in Drosophila. Two of the studies use heat shock loci as model systems. At heat shock loci, the trxG/ETP protein GAF, TFIID, and RNA polymerase II (RNAP II) cooperate to establish a preset structure that allows ready access of heat shock factor (HSF) after stress, e.g., heat induction. The polymerase begins transcription (i.e., initiation) before induction but then pauses between +25 and +50; subsequent binding of HSF upon induction recruits the Mediator complex leading to phosphorylation of the RNAP II C-terminal domain and recruitment of elongation factors that allow transcriptional release of the polymerase so that elongation can proceed (see Dellino et al., 2004, and references therein). In a very elegant study, Smith et al. (2004) showed that Trx (as part of the TAC1 complex, discussed above) is recruited to heat shock loci, is required for full heat shock response at the hsp70 locus, and binds downstream of the hsp70 transcription start site where it methylates histone H3K4. The associated CBP within the TAC1 complex acetylates histones. These data provide persuasive evidence that Trx has a role in transcriptional elongation. It should be noted that trx null mutations do not abolish the heat shock response entirely, so some transcription at hsp70 must be possible in the absence of Trx.
In another study of heat shock loci, Dellino et al. (2004) show that PcG-mediated silencing of the hsp26 promoter does not prevent recruitment of TBP, RNAP II, or heat shock factor (HSF). However, the strand opening that accompanies the initial synthesis of 25–50 nucleotides does not occur in PcG-silenced reporters, suggesting that an early phase of transcription is affected. It could be argued that heat shock promoters represent a special case, because they are examples of preset, paused transcription. However, Wang et al. (2004b) show that Pc binds just downstream of the Ubx promoter in SL2 embryonic cell lines and in imaginal disks, whereas E(z) binds downstream of the promoter in imaginal disks, and methylation of histone H3K27 occurs downstream of the promoter. Using two Drosophila cell lines with differential expression of the homeotic cluster Abd-B gene, Breiling et al. (2004) show that TBP, RNAP II, and elongation factors are recruited to both repressed abd-A and Abd-B promoters as well as to the active Abd-B promoter. However, binding of PcG proteins belonging to both PRC1 and PRC2 was enriched specifically at repressed promoters, which correlates with trimethylation of both histone H3K9 and H3K27 at repressed promoters (Breiling et al., 2004). These data are consistent with a role for PcG proteins in preventing transcriptional elongation at heat shock and homeotic loci and support the hypothesis that MPs generally function in early phases of transcription.
GENETICS OF MAINTENANCE
Contrary to the evidence presented above, the genetics of MPs have been traditionally interpreted to argue against a role for MP in the early stages of transcription; indeed, PcG and trxG genetics formed the basis for the definition of the “maintenance” as opposed to the “initiation” phase of transcriptional regulation of homeotic target loci. For some MP mutations in Drosophila, it is possible to establish germline clones and to examine homozygous mutant embryos that derive from homozygous mutant mothers. This experimental design is necessitated because mothers deposit MP protein and mRNA into the oocyte, so it is important to eliminate maternal products if the null mutant phenotype is to be examined. Struhl and Akam (1985) were the first to examine homeotic gene expression in early esc−/− embryos. Ultrabithorax (Ubx) expression begins at approximately 2.5 hr after egg deposition. In esc−/− embryos derived from esc−/− mothers, spatial regulation of Ubx is normal from 3 to 6 hr after egg deposition, and Ubx only begins to be expressed ectopically after 6 hr. Expression of the segmentation genes that encode the transcription factors that initially define spatial regulation of Ubx begin to decay at approximately 4 hr. Struhl and Akam (1985) therefore concluded that Esc was not needed to choose where the Ubx homeotic gene was expressed, but was instead needed to maintain the stability of the choice once made. Similar results were obtained by Soto et al. (1995), who found that misexpression of the Abd-B homeotic gene is detectable in maternal −/− zygotic −/− embryos of Asx, E(Pc), Pcl, Psc, and Su(z)2 at approximately 5 hr. Hence, a dogma in the field has been that PcG genes maintain but do not initiate repression of homeotic genes during development.
Early in embryogenesis, when the repressors encoded by segmentation genes are present, there is no need for PcG proteins to maintain repression. Only when the repressors decay is the failure to establish a more permanent repression detected. Whereas loss of permanent repression is detectable at approximately 5 hr, the actual establishment of the more permanent repression system could occur much earlier in wild-type embryos. It could also be that detection of ectopic expression of homeotic loci using in situ hybridization or antibodies is relatively insensitive to small changes in homeotic gene expression and cannot be detected for several hours, even though ectopic expression begins earlier. Therefore, the experiments described in the previous paragraph do not permit the conclusion that MPs are not required for initiation.
Conversely, temperature-shift experiments with a temperature-sensitive trx allele show that Trx is required from 0 to 4 hr, consistent with a role for Trx early in embryogenesis, when homeotic gene transcription patterns are being established (Ingham and Whittle, 1980). Other studies suggest a very early role for esc. If an esc transgene under the control of a heat shock promoter is expressed in an esc−/− embryo, only expression between 2 and 4 hr rescued embryonic lethality, and heat shocks between 2 and 3 hr were most effective (Simon et al., 1995). These results suggest that Esc is required very early, a conclusion supported by previous experiments to establish the temperature-sensitive period of esc (Struhl and Brower, 1982). This conclusion is strengthened by experiments in which Esc–lexA fusions were used to establish silencing at a transgene containing lexA binding sites. Only if the fusion was provided between 0 and 2 hr after egg deposition was silencing possible (Poux et al., 2001b). ChIP experiments show that MP proteins are detectable at MEs and at promoters by 3 hr after egg deposition, again consistent with an early role (Orlando et al., 1998).
Genetic experiments show that MPs are required continuously in development. Cells that have been made homozygous mutant for MP genes using either mitotic recombination or inducible expression of recombinase are examined in the adult cuticle or by scoring homeotic gene expression in the imaginal disk. Trx (Ingham, 1985; LaJeunesse and Shearn, 1995) and PcG genes (Beuchle et al., 2001) are required continuously throughout development, consistent with MP proteins being required for maintenance. These experiments demonstrate that MPs are indeed required after the initial patterns of homeotic gene expression in embryonic development have been established, but they do not rule out a role for MPs in transcriptional initiation or elongation.
TRXG PROTEINS MIGHT BE ANTIREPRESSORS OF PCG PROTEINS
PcG and trxG proteins may act antagonistically. If repression is the default state, and subsequent derepression followed by activation must be maintained, then the function of trxG proteins could be to antagonize PcG proteins. This role could be direct, as with the example of Ash1, whose HKMT function (see above) antagonizes Pc binding (Beisel et al., 2002). A similar example has been shown for the Nucleosome Remodeling and Deacetylase (NuRD) repressor complex, which cannot bind to histone H3 tails methylated at K4 (Zegerman et al., 2002). Or it could be indirect, for example, if Trx- or Ash1-mediated methylation of histone H3K4 is permissive for transcription, and PcG proteins cannot inhibit actively transcribed genes. Methylation of histone H3K4 correlates with recruitment of particular ISWI-type nucleosome remodeling complexes, which are required for elongation (Santos-Rosa et al., 2002; Morillon et al., 2003). There is experimental evidence for the idea that PcG proteins cannot repress active genes. In Drosophila, if PcG-lexA fusions are induced at different times in development, stable embryonic repression is only established on genes that are not transcribed at blastoderm (Poux et al., 2001a).
The evidence discussed above supports the idea that Trx and Ash1 are coactivators, because they methylate histone H3K4 and because they recruit the HAT CBP (Petruk et al., 2001; Beisel et al., 2002; Milne et al., 2002; Poux et al., 2002). A recent study directly tests this idea genetically (Klymenko and Muller, 2004). As expected, in trx and ash1 mutants, homeotic gene expression is reduced within their normal expression domains. Unexpectedly, in double mutants carrying PcG and trxG gene mutations, homeotic gene expression levels are normal in the correct expression domains, and homeotic genes are also strongly expressed outside their normal domains. The implication is that loss of homeotic gene expression in trxG mutants comes from failure to antagonize PcG repression rather than from loss of transcription per se. It can also be inferred that, in imaginal disk clones, methylation of histone H3K4 that occurred before clone induction is insufficient to maintain gene activation. Moreover, because transcription levels of homeotic genes are normal in embryos that are maternal −/− zygotic −/−, neither trx nor ash1 are required for transcriptional initiation. Klymenko and Muller (2004) also showed that redundancy between Trx and Ash1 did not explain their results, because ash1; trx double mutants showed robust expression of Abd-B. The only evidence for a role of Trx in transcriptional initiation came from observations that esc; trx double mutants do exhibit reduced homeotic gene expression. Together, the data argue that for regulation of homeotic genes in Drosophila, Trx and Ash1 are primarily antirepressors, rather than being required for initiation of transcription.
In yeast, the trx homolog SET1 is necessary for transcriptional elongation (see above). Similarly, Trx is needed for transcriptional elongation of hsp70 in Drosophila. Yet the clonal analysis and analysis of maternal −/− zygotic −/− Drosophila embryos shows that Trx and Ash1 are largely dispensable for transcriptional initiation and are required continuously for maintenance. The obvious way out of the apparent contradiction is to postulate that the role of trxG proteins in initiation or elongation is to antagonize PcG proteins. Without this antagonism, transcriptional initiation or elongation (i.e., gene activation) does not occur (Fig. 2).
WHAT ACCOUNTS FOR MEMORY?
If gene expression patterns are to be maintained, there must be a heritable mark that is stable to mitosis. There is a report of histone acetylation being stable to meiosis (Cavalli and Paro, 1998, 1999). The most attractive models suggest that changes to transcriptional states brought about by MPs establish the heritable mark (Cavalli and Paro, 1998). However, it is formally possible that MPs bring about memory by a mechanism separate from their direct effect on transcription. Below, we discuss some alternative models.
The simplest model is that MPs themselves are the heritable mark. For this to be true, one would predict that MP binding would be stable throughout the cell cycle. However, the Drosophila PcG proteins Pc, Ph, and Psc and the human Psc homolog BMI1 leave chromosomes in metaphase and do not return until anaphase; this transition correlates with alterations in the phosphorylation status of BMI1 (Buchenau et al., 1998; Voncken et al., 1999; Muchardt et al., 2002). These observations are inconsistent with these particular PcG proteins, all components of PRC1-type complexes, being the mark; however, a small fraction of the total amount of Ph and BMI1 protein within nuclei do remain associated with metaphase chromosomes (Buchenau et al., 1998; Voncken et al., 1999; Muchardt et al., 2002). It is possible that small yet biologically significant levels of particular MPs mark chromatin through the cell cycle. However, the localization of other PcG proteins, including ETPs and those belonging to PRC2-type complexes, throughout the cell cycle remains unknown; thus, it is possible that any of these PcG proteins could serve as the heritable mark. Further investigation of this model, including the significance of MP phosphorylation, is needed.
If it is true that there are no histone demethylases (for discussion, see Kouzarides, 2002), then the stability of histone methylation makes it a good candidate for a mark that can be transmitted through cell divisions. Histone methylation established by MPs, at histone H3K4 or H4K20 in the case of activation and at histone H3K9 or H3K27 in the case of repression, could constitute the heritable mark. In yeast, trimethylation of H3K4 mediated by Set1 is relatively stable and persists even when transcription has ceased, and Set1 is no longer bound (Ng et al., 2003). The yeast experiments suggest that, although H3K4 trimethylation is not immediately reduced in the absence of the Set1 protein, long-term maintenance of this mark still depends on continued transcriptional activity. This finding suggests that there are mutually interdependent feedback mechanisms between immediate maintenance of transcriptional activation (perhaps involving more transient marks such as histone acetylation or ubiquitination) and continued maintenance of the histone trimethyl mark. In the absence of a demethylase, histone replacement might explain loss of histone methyl marks over time.
Recently, it has been suggested that nucleosome deposition after replication may be semiconservative, because the histone chaperone responsible interacts with histone H3-H4 dimers, not tetramers (Korber and Horz, 2004; Tagami et al., 2004). A recent study on histone protein folding suggests that there is dynamic equilibrium of H3-H4 dimers and tetramers (Banks and Gloss, 2004). If both daughter strands receive equal portions of methylated histones in the form of H3-H4 (as well as H2A-H2B) dimers, and there is a templating mechanism to methylate newly deposited unmethylated histones, then histone methylation as the heritable mark would be faithfully propagated. However, Henikoff et al. (2004) have argued that numerous carefully performed experiments over the past 25 years rule out semiconservative nucleosome deposition during replication and demonstrate that histone H3-H4 units are distributed as tetramers, not dimers, during replication. If so, histone modifications would provide a low fidelity mark, as the mark would get diluted with each round of replication, and some daughter strands may at random receive no nucleosomes with histone modifications. This may represent a problem for epigenetic inheritance models involving histone modification as the only heritable mark, as such memory would be sloppy.
An alternative model for marking chromatin other than by histone modification is through incorporation of histone variants. There is increasing evidence that such variants function in specification of active versus silent chromatin states and are assembled at particular loci (Vermaak et al., 2003; for review, see Korber and Horz, 2004). One of the more striking recent discoveries is that different replication-coupled (RC) and replication-independent (RI) nucleosome deposition pathways exist and require different assembly factors and chaperones. RC nucleosome assembly uses all three subunits of chromatin assembly factor-1 (CAF-1) and assembles histone H3, whereas RI assembly uses only one of the CAF-1 subunits plus the histone regulator A (HirA) and assembles the variant histone H3.3 (reviewed in Henikoff et al., 2004). The RI nucleosome assembly pathway appears to be used specifically for active loci (Ahmad and Henikoff, 2002a, b). Histone H3.3 or H2A.Z may mark active transcription, and MacroH2A is associated with the inactive X chromosome. Histone variants may themselves be differentially posttranscriptionally modified and, therefore, carry different marks.
The clonal analysis experiments discussed above argue against a simple form of each of the models above, because MPs are needed continuously in development. Therefore, histone methylation or replacement by histone variants might be necessary to establish a heritable mark, but are themselves insufficient. These observations suggest the attractive possibility that a function of MPs is to continuously replenish histone marks or histone variants. MPs might prevent loss of methylated histones by histone replacement. If the histone code model (Strahl and Allis, 2000; Jenuwein and Allis, 2001) is correct, and nucleosomes are distributed randomly to daughter strands after DNA replication, then the role of MPs might be to methylate unmodified histones. Recent experiments by Breiling et al. (2004) support this idea. When PcG genes encoding PRC1 components are down-regulated in the Drosophila S2 cell line, the PcG-repressed abd-A Hox gene becomes derepressed over several rounds of replication. Loss of binding of the PRC2 component E(z) at the abd-A promoter occurs, which correlates with loss of histone H3K9 and H3K27 di- and trimethylation and an increase in histone H3K4 di/trimethylation. These results strongly support the model that both PRC1 and PRC2 complexes are required continuously for maintenance of gene repression and of repressive histone methylation marks. Similarly, one function for Trx- or Ash1-mediated methylation might be to ensure that histone H3.3 continues to be replaced at active genes rather than be removed. If particular MPs preferentially bind methylated histones (e.g., Pc/PRC1) and can methylate neighboring nucleosomes (e.g., by E(z)/PRC2 HKMT activity), then after each round of replication, nucleosomes could become remethylated regardless of how they were distributed during DNA replication, as long as there were some methylated nucleosomes in the vicinity. This process could be true for canonical core histones, or for histone variants (Fig. 3).
Relatively little attention has been paid to the possibility that the role of MPs in memory is different from their role in gene activation or repression. The obvious place for this additional role would be at DNA synthesis, because the heritable mark must be stable to mitosis, and passed to both daughter strands. It was pointed out many years ago (Almouzni et al., 1991; Wolffe, 1991) that there might be a mechanistic connection between DNA replication and transcription. This idea received important experimental support recently (Danis et al., 2004). Here, we point out some suggestive evidence that is consistent with MPs having a role in DNA synthesis.
Polytene chromosomes are endoreplicated. However, not all parts of the polytene chromosomes replicate to the same extent. Under-replicated regions are visible as constrictions. Makunin et al. (2002) observed that that the constrictions correspond closely to PcG binding sites on polytene chromosomes and that, in Suppressor of underreplication (Su(ur)) mutants, the constrictions disappear. No one has tested whether PcG targets are late-replicating, or whether PcG mutations affect replication timing. Mutations in ash1, ash2, and E(z), have growth defects (i.e., fewer cells) in imaginal disks, suggesting problems in DNA replication (Shearn et al., 1987). Mutations in two PcG genes, cramped and ph (Yamamoto et al., 1997; Lupo et al., 2000) exhibit defects in mitosis. Phenotypic analyses such as these do not rule out indirect effects, such as misregulation of a gene required for mitosis in MP mutants. For instance, the MPs CBX7 (a Pc homolog) and Bmi1 function independently to extend replicative lifespan when overexpressed in mammalian cells, through transcriptional repression of the INK4a cell-cycle regulator locus (Itahana et al., 2003; Gil et al., 2004). However, an intriguing recent report shows that mammalian PcG proteins directly interact with geminin, a replication licensing factor (Luo et al., 2004). Although the function of this interaction is unclear, it raises the intriguing possibility that PcG proteins have a direct role in DNA replication. In yeast, the trx homolog Set1 (discussed above) also associates with double-strand breaks and is required for meiotic DNA synthesis (Sollier et al., 2004). Finally, mutations in the plant PcG gene Medea suppress mutations in Origin of Replication Complex genes in Arabidopsis (Collinge et al., 2004). None of these observations is conclusive, but they suggest that further investigation is required to determine whether MPs have a role in DNA replication, and if this role is necessary for maintenance.
Enormous progress has been made in the past few years to understand the molecular basis of MPs, and the complexes of which they are constituents. Identification of DNA binding activities of Pho, Pho-l, and YY1, and the roles of proteins like GAF, Zeste, and Pipsqueak provide a link to target recognition. The discovery that SET domains are HKMTs suggest mechanisms of activation and repression, and provide a link to maintenance. The recent observations that MPs act at or near transcription start sites and interact with elongation factors gives a new focus to understanding a role for these proteins in transcriptional initiation or elongation.
Yet in other ways, nearly every essential question remains unanswered. No one has defined MEs in organisms other than Drosophila. The study of Ringrose et al. (2003) identifies many putative MEs as associated closely with promoters, rather than being separated as they are in homeotic loci. Perhaps one reason why searches for mammalian MEs have been unsuccessful is the assumption that, similar to homeotic loci in Drosophila, mammalian MEs will behave like enhancers and be separate from promoters. If the organization of homeotic loci of Drosophila is unusual, it might be worth looking for MEs near mammalian promoters. ChIP studies with MLL at Hoxc8 and Hoxa9 support this idea (Milne et al., 2002; Nakamura et al., 2002). Nor has anyone reconstituted a ME from binding sites, suggesting that we are far from understanding ME architecture. Despite the identification of minimal ME DNA fragments on the order of several hundred basepairs in length and algorithms to predict MEs in Drosophila, the key recognition elements and their overall organization within the ME have not yet been determined. This area is a major project for the future.
Although mechanistic predictions of MP functions are possible, they have so far been difficult to test because of the lack of inducible systems that allow studying changes at target loci in the order of minutes rather than hours. Such a system would allow specification of order of assembly of the components of maintenance. The heat shock system appears to offer an excellent model, but doubts will remain about the universality of conclusions derived from heat shock promoters relative to genes expressed in embryogenesis. Embryonic systems have been challenging because of heterogeneous populations of cells, the difficulties of synchronizing embryos, the lethality of most MP mutations, and the problems in obtaining tissues expressing or not expressing target loci, or mutant for MP genes, in amounts suitable for biochemistry. These considerations suggest that, in the future, more work will be carried out in cell lines.
It seems obvious that histone lysine methylation plays a key role in maintenance. The nature of this role will remain obscure until the downstream consequences of histone lysine methylation with respect to the spatial and temporal regulation of target loci are described and understood. A good first start will be to identify what proteins (for example, Pc) are recruited to methylated histones and to determine how chromatin structure is modified as a consequence of this methylation. Emerging in vivo visualization techniques, such as FRET reporters to assess states of histone acetylation (Kanno et al., 2004) and methylation (Lin et al., 2004), and FRAP to analyze protein dynamics (Festenstein et al., 2003; Hager et al., 2004), combined with ChIP should provide insight into MP function and complement standard genetic and biochemical approaches.
Whereas many mechanisms of MP action have been proposed in the past 15 years, none have been eliminated conclusively. It would seem that the easiest model to test in the short term is the looping model using the recently developed chromosome conformation capture (3C) or RNA–TRAP techniques (de Laat and Grosveld, 2003; Dekker, 2003; Chambeyron and Bickmore, 2004; Splinter et al., 2004). Finally, it seems worthwhile to keep the distinction between the effects of MPs on gene expression from their effects on memory. Although it is appealing to assume that gene regulation and memory mediated by MPs have the same underlying mechanism, they may be different. We predict increased interest in the possible roles of MPs in DNA replication.
We thank Jacob Hodgson, Samantha Beck, Sebastien Bloyer, and Tom Milne for helpful comments on the manuscript, and lab members for years of thoughtful conversations. H.W.B. is supported by grants from the Natural Sciences and Engineering Research Council, the Canadian Institute for Health Research (CIHR), and the National Institutes of Health (USA). C.L.F. received a CIHR studentship.