Drosophila Cyclin G and epigenetic maintenance of gene expression during development

Cyclins and cyclin-dependent kinases (CDKs) are essential for cell cycle regulation and are functionally associated with proteins involved in epigenetic maintenance of transcriptional patterns in various developmental or cellular contexts. Epigenetic maintenance of transcription patterns, notably of Hox genes, requires the conserved Polycomb-group (PcG), Trithorax-group (TrxG), and Enhancer of Trithorax and Polycomb (ETP) proteins, particularly well studied in Drosophila. These proteins form large multimeric complexes that bind chromatin and appose or recognize histone post-translational modifications. PcG genes act as repressors, counteracted by trxG genes that maintain gene activation, while ETPs interact with both, behaving alternatively as repressors or activators. Drosophila Cyclin G negatively regulates cell growth and cell cycle progression, binds and co-localizes with the ETP Corto on chromatin, and participates with Corto in Abdominal-B Hox gene regulation. Here, we address further implications of Cyclin G in epigenetic maintenance of gene expression. We show that Cyclin G physically interacts and extensively co-localizes on chromatin with the conserved ETP Additional sex combs (ASX), belonging to the repressive PR-DUB complex that participates in H2A deubiquitination and Hox gene silencing. Furthermore, Cyclin G mainly co-localizes with RNA polymerase II phosphorylated on serine 2 that is specific to productive transcription. CycG interacts with Asx, PcG, and trxG genes in Hox gene maintenance, and behaves as a PcG gene. These interactions correlate with modified ectopic Hox protein domains in imaginal discs, consistent with a role for Cyclin G in PcG-mediated Hox gene repression. We show here that Drosophila CycG is a Polycomb-group gene enhancer, acting in epigenetic maintenance of the Hox genes Sex combs reduced (Scr) and Ultrabithorax (Ubx). However, our data suggest that Cyclin G acts alternatively as a transcriptional activator or repressor depending on the developmental stage, the tissue or the target gene. Interestingly, since Cyclin G interacts with several CDKs, Cyclin G binding to the ETPs ASX or Corto suggests that their activity could depend on Cyclin G-mediated phosphorylation. We discuss whether Cyclin G fine-tunes transcription by controlling H2A ubiquitination and transcriptional elongation via interaction with the ASX subunit of PR-DUB.


Background
Cyclins and cyclin-dependent kinases (CDKs) are classically described as essential regulators of the cell cycle. However, the discovery of new cyclins, along with extensive studies of cell cycle regulators, have revealed that they are involved in diverse processes ranging from metabolism to stem cell self-renewal, and play key roles in regulation of transcription [1,2]. Indeed, cyclin/CDK complexes orchestrate the transcription cycle by dynamically phosphorylating the heptapeptide repeats which form the Cterminal domain (CTD) of the largest RNA polymerase II (RNA Pol II) subunit [3,4]. Notably, transcriptional elongation requires phosphorylation of the CTD on serine 2 by the positive transcription elongation factor b (P-TEFb), formed by CDK9 and Cyclin T or K [5][6][7].
Cyclins and CDKs can also directly alter gene expression. For example, mammalian Cyclin D1 binds several core transcription factors in regions close to transcription start sites [8], thus participating in regulation of adjacent genes. Furthermore, a cyclin/CDK complex determines the activity of the multi-subunit complex mediator that plays a fundamental role in eukaryotic gene regulation. Indeed, the mediator kinase subunit, composed of Cyclin C/CDK8, MED12, and MED13 in Drosophila, promotes either transcriptional activation or repression depending on the context [9][10][11][12]. Interestingly, MED12 and MED13, encoded by kohtalo (kto) and skuld (skd), were initially identified as suppressors of homeotic phenotypes induced by mutation of Polycomb (Pc), an epigenetic repressor of transcription [13]. However, kto and skd are also involved in epigenetic repression of the Hox gene Ubx during development [14]. These data link the mediator kinase subunit to the epigenetic mechanisms that ensure faithful transmission of chromatin states from mother to daughter cells.
During development, epigenetic maintenance of gene expression patterns is under control of evolutionary conserved proteins encoded by the Polycomb-group (PcG), trithorax-group (trxG), and Enhancer of Trithorax and Polycomb (ETP) genes, whose roles have been particularly well studied for homeotic (Hox) gene regulation in Drosophila [15][16][17][18]. PcG genes are involved in long-term gene repression, whereas trxG genes participate in maintenance of gene activation and counteract PcG action. ETP genes have been mainly characterized in Drosophila. They interact with both trxG and PcG genes, and behave alternatively as repressors or activators of target genes [17][18][19].
ETPs also comprise the evolutionary conserved GAGA factor that interacts with both BAP and PRC complexes in the regulation of Hox genes [35] and the proteins Corto and dorsal switch protein 1 (DSP1) in Drosophila [36,37]. Corto interacts physically with GAGA and DSP1 which are both involved in the recruitment of PcG complexes to Polycomb Response Elements (PRE) [38]. As several ETPs are differently recruited to PREs depending on tissues or developmental stages, it has been proposed that different combinations of ETPs could favor the recruitment of either PcG or TrxG complexes, thus participating in maintenance of silenced or active states [37].
Drosophila Cyclin G was first isolated during a twohybrid screen using the ETP Corto as bait [39]. Cyclin G has two mammalian homologs, Cyclin G1 and G2, whose functions remain elusive. CCNG1 that encodes Cyclin G1 is a direct target of the tumor-suppressor p53 [40]. Its induction following γ-irradiation leads to cell cycle arrest at the G2/M transition, thus allowing DNA damage repair [41]. On the contrary, overexpression of CCNG1 activates the proliferation of colon carcinoma cells [42]. Overexpression of CCNG2 that encodes Cyclin G2 induces cell cycle arrest at the G1/S transition [43]. Like Cyclin G2 [44], Drosophila Cyclin G bears a PEST sequence at its C-terminal extremity. Furthermore, Cyclin G negatively regulates both cell growth and cell cycle progression, preventing G1/S transition and slowing down the S phase [45,46]. Collectively, these characteristics suggest that Drosophila Cyclin G behaves more like Cyclin G2. Cyclin G co-localizes at many sites with the ETP Corto as well as with the PcG protein PH on larval polytene chromosomes, suggesting that it is involved in the control of gene expression [39]. Although a direct role of mammalian G-type cyclins in gene expression has not been reported, overexpression of CCNG1 in human cells induces chromatin relaxation [47]. Genetic interactions between CycG (encoding Drosophila Cyclin G) and corto showed that both genes are involved in regulating expression of the homeotic gene Abdominal-B (Abd-B) in the early pupal epithelium, corto acting as a repressor and CycG as an activator of Abd-B [39]. Furthermore, Cyclin G and Corto bind the iab-7 cis-regulatory element as well as the promoter of Abd-B in embryos [39]. Altogether, these data strongly suggest a role for Cyclin G in regulation of Hox gene expression during development.
In this work, we address the involvement of Cyclin G in epigenetic maintenance of Hox gene expression in Drosophila. We first demonstrate that Cyclin G also interacts with the ETP ASX, and extensively co-localizes with ASX on polytene chromosomes. We next show that CycG genetically interacts with Asx in maintenance of Hox genes. Interestingly, CycG also interacts with several PcG and trxG genes and behaves genetically as a PcG, since loss of CycG enhances PcG-mediated homeotic phenotypes and suppresses trxG-mediated ones, whereas overexpression of CycG has the opposite effect. These genetic interactions were correlated with modifications of ectopic Hox protein domains in imaginal discs, suggesting a role for Cyclin G in PcG-mediated Hox gene repression during development. Surprisingly, we found that Cyclin G largely co-localizes with C-terminal domain of RNA polymerase II (RNA Pol II CTD) phosphorylated on serine 2, suggesting a role in productive transcription. We propose that Cyclin G acts as an activator or a repressor of transcription depending on the developmental stage, the tissue, or the target gene.

Results
Drosophila Cyclin G interacts with the enhancer of Trithorax and Polycomb ASX Cyclin G physically interacts with the ETP Corto in vivo, and the Cyclin G N-terminal domain (amino-acids 1 to 130) is necessary and sufficient for this interaction [39]. Interestingly, a yeast two-hybrid screen to isolate interactors of the ETP ASX also identified Cyclin G as a potential partner. In this screen, the ASX-C terminal domain (ASX-C, amino-acids 1139 to 1668), containing a PHD domain (amino-acids 1634 to 1665), was used as bait against a two-hybrid library of 0 to 12 h embryonic complementary DNAs (cDNAs) [48]. A deleted form of Cyclin G containing only the 275 N-terminal residues interacted strongly with ASX-C in two-hybrid assays (data not shown). These results showed that Cyclin G residues 1 to 275, N-terminal to the cyclin box, were sufficient for the interaction with ASX.
To further characterize the interaction between Cyclin G and ASX, we co-transfected S2 cells with pAct::Myc-AsxC and pAct::FLAG-CycG and performed immunoprecipitation using anti-Myc and anti-FLAG antibodies. FLAG-Cyclin G co-immunoprecipitated with Myc-ASX-C ( Figure 1B), confirming the interaction between Cyclin G and ASX. To determine which Cyclin G region mediates this interaction, we next constructed different vectors expressing FLAG-tagged, truncated forms of Cyclin G ( Figure 1A). Myc-ASX-C co-immunoprecipitated with the 130 N-terminal residues of Cyclin G ( Figure 1C). Hence this domain, which also mediates the interaction with Corto, was sufficient to bind ASX. However, Cyclin G deleted of this N-terminal domain also co-immunoprecipitated with ASX-C ( Figure 1D), indicating that other parts of Cyclin G interact with ASX-C.
To test the interaction between Cyclin G and ASX in more native conditions, we performed immunoprecipitation using protein extracts of third instar larvae overexpressing a Myc-tagged form of Cyclin G (da > Myc-CycG ΔP ) using anti-Myc and anti-ASX antibodies. Myc-Cyclin G co-immunoprecipitated with ASX ( Figure 1E), demonstrating that Cyclin G interacts with endogenous ASX in vivo.
Taken together, these results confirm that Cyclin G interacts with ASX, and show that the Corto-interacting domain, that is, amino-acids 1 to 130 of Cyclin G, is sufficient for this interaction.

Cyclin G mainly localizes on active chromatin
Since Cyclin G physically interacts with two ETPs (Corto and ASX), and since we previously established that Cyclin G co-localizes with Corto at many loci on polytene chromosomes of larval salivary glands [39], we tested whether Cyclin G also shares binding sites with ASX on chromatin. We co-immunostained salivary gland polytene chromosomes of wild-type larvae with anti-Cyclin G and anti-ASX antibodies. We detected an important overlap of Cyclin G and ASX binding sites (Figure 2A), indicating that interaction of Cyclin G with ASX could take place on chromatin.
Furthermore, Cyclin G preferentially localized at DAPI interbands ( Figure 2B), suggesting that this protein mainly binds open chromatin. This observation led us to test whether Cyclin G binding was preferentially coupled with an active or a repressed chromatin state, by examining its co-localization with specific marks on polytene chromosomes. Comparison of Cyclin G binding sites with the distribution of the repressive histone mark H3K27me3 showed that very few Cyclin G-bound loci were positive for H3K27me3 ( Figure 2C). Conversely, Cyclin G co-localized largely with RNA Pol II phosphorylated on serine 2 ( Figure 2D). These results suggest that Cyclin G binds mainly open chromatin and might be associated with genes undergoing transcription.

CycG genetically interacts with Asx
The physical interaction between Cyclin G and ASX as well as their extensive co-localization on chromatin led us to examine genetic interactions between CycG and Asx, using two loss of function alleles of Asx, Asx XF23 , and Asx 3 [49,50]. To address the effects of CycG misregulation on Asx phenotypes, we combined these Asx mutations with CycG loss of function [ubiquitous inactivation by RNA interference using the UAS::dsCycG2 transgene driven by da::Gal4 (da > CycG RNAi)] or gain of function [ubiquitous overexpression of CycG using the UAS::CycG ΔP transgene (da > CycG ΔP encoding Cyclin G deleted of amino-acids 542-566)]. We focused on the effect of CycG misregulation on PcG-like homeotic transformations induced by Asx mutations. Importantly, these homeotic phenotypes were never detected in CycG loss of function or gain of function flies. All genetic interaction data are shown in Additional file 1. PcG genes derived their name from the most conspicuous common phenotype of adult males carrying PcG mutations: ectopic sex combs on posterior legs. Indeed, in Drosophila melanogaster, this male-specific structure, composed of specialized bristles called teeth, is specific to the first tarsal segment of prothoracic legs (L1). The occurrence of sex comb teeth on mesothoracic (L2) and metathoracic legs (L3) thus indicates a partial transformation of L2 and L3 into L1. Males heterozygous for Asx XF23 presented ectopic sex combs on L2 with a lower penetrance than Asx 3 heterozygous males. CycG loss of function significantly enhanced penetrance of L2 sex combs induced by Asx XF23 ( Figure 3A) but did not modify that of Asx 3 ( Figure 3B). Furthermore, CycG gain of function significantly suppressed L2 sex combs induced by Asx 3 ( Figure 3B). To summarize, CycG gain of function opposed an Asx-induced PcG phenotype on L2, whereas CycG loss of function enhanced this phenotype. Therefore, CycG behaved as an enhancer of Asx regarding homeotic transformation of L2 into L1. This suggests that CycG participates with Asx in the maintenance of mesothoracic leg identity.
Maintenance of abdominal segment identity also relies on PcG, trxG, and ETP genes. In wild-type Drosophila males, tergites of the fifth (A5) and sixth (A6) abdominal segments present a uniform dark pigmentation, whereas tergites of more anterior abdominal segments show only a posterior stripe of dark pigmentation. Thirty percent Figure 2 Cyclin G co-localizes with ASX and binds mainly active chromatin. Immunostaining of polytene chromosomes from w 1118 third instar larvae. DNA was stained with DAPI (blue). (A) Cyclin G (green) and ASX (red) co-localize at many sites. Bottom: close-up view of the region framed by a white rectangle. (B) Cyclin G (green) binds DAPI interbands. Right: close-up view of the region framed by a white rectangle. (C) Cyclin G (green) and H3K27me3 (red) are almost completely exclusive. Rare co-localizations are shown with white arrows. (D) Cyclin G (green) co-localizes at many sites with RNA Pol II phosphorylated on serine 2 (red). Bottom: close-up view of the region framed by a white rectangle.
of Asx 3 /+ males presented darkly pigmented patches on the anterior part of the fourth abdominal tergite (A4), indicating a partial transformation of A4 into a more posterior segment, a phenotype observed in some PcG mutants ( Figure 3E). As shown on Figure 3C, this phenotype was not affected by CycG loss of function, but was completely suppressed by CycG gain of function. Interestingly, 70% of Asx 3 /+; da > CycG ΔP males presented small patches lacking dark pigmentation on tergite A5 ( Figure 3D,E). This corresponds to a partial transformation of A5 into segment A4, which is a classical trxG homeotic transformation. Hence, in the abdomen, CycG gain of function not only suppressed the typical PcG phenotype induced by Asx 3 but also shifted it toward a trxG phenotype.
Together, these results reveal that CycG and Asx are engaged in complex genetic interactions and suggest a role for CycG similar to that of a PcG gene in maintenance of segmental identities during development.

CycG loss of function enhances PcG-induced homeotic transformations
The genetic interactions reported above, the physical interaction between Cyclin G and the ETPs ASX and Corto, and the co-localization on chromatin of Cyclin G with ASX as well as with Corto and the PcG protein PH, led us to further examine genetic interactions between CycG and PcG, trxG, or ETP genes. We associated CycG misregulation with alleles of these genes (listed in Additional file 2) reported to induce a visible dosesensitive phenotype, alone or in combination with other PcG, trxG, or ETP mutations. All genetic interaction data are shown in Additional file 1.
Contrary to Asx mutations, mutant alleles of the ETPs corto and Dsp1 induced no PcG-like leg transformation, whether alone or in combination with CycG misregulation. We next analyzed interactions with mxc (multi sex combs) and crm (cramped), genetically classified as PcG genes [51,52], although their products have not been found in PcG complexes so far. mxc G46 /Y males present sex comb teeth on posterior legs [52], indicating a partial transformation into L1 ( Figure 4A). Indeed, 66% of L2 and 11% of L3 (n = 56) carried at least one sex comb tooth (1.4 ± 1.3 teeth on L2 and 0.2 ± 0.5 on L3). Pharate Figure 4 CycG misregulation alters ectopic sex comb phenotypes of an mxc mutant. (A) Expressivity of mesothoracic leg (L2) and metathoracic leg (L3) transformations into prothoracic leg (L1). The wild-type sex comb is marked by a white arrowhead. mxc G46 /Y males present ectopic sex comb teeth on distal L1, on L2 and L3 (black arrowheads). Only phenotypes affecting L2 and L3 were rated. These phenotypes are enhanced by CycG loss of function. (B) Penetrance of the ectopic sex combs phenotype on male legs. CycG loss of function enhances mxc G46 induced ectopic sex combs on L2 and L3, whereas CycG gain of function suppresses ectopic sex combs on both legs. Fisher's exact test, *P < 0.05 **P < 0.0001 (n ≥ 50). mxc G46 /Y males presented similar phenotypes (Additional file 1). As both gain and loss of function of CycG induced mxc G46 male lethality just prior to adult emergence, ectopic sex combs were scored in pharates. CycG loss of function enhanced the expressivity of this phenotype (3.7 ± 2.3 teeth on L2 and 1.3 ± 2.0 on L3, n ≥ 56; t-test, P < 0.0001) ( Figure 4A). Penetrance of the phenotype was also enhanced by CycG loss of function ( Figure 4B). Conversely, CycG gain of function completely suppressed L2 and L3 sex combs of mxc G46 /Y males ( Figure 4B). CycG misregulation mxc G46 males presented no other phenotypes than modified ectopic extra sex combs. These animals died prior to full pigmentation of the abdomen, preventing evaluation of phenotypes observed in adult mxc G46 males [52]. Interactions between CycG and crm gave similar results. Eighty percent of crm 7 /Y flies presented sex combs on L2 and 35% on L3. Males combining crm 7 /Y and CycG loss of function died before the pupal stage, preventing analysis. In the few pharate escapers combining crm 7 /Y and CycG overexpression, leg transformations were significantly suppressed (Additional file 1).
Next, we tested several genes encoding members of PRC2. Neither E(z) 63 /+ nor esc 21 /+ males presented ectopic sex combs, and this was not significantly modified by CycG misregulation. On the other hand, mutants for polyhomeotic (ph) and Polycomb (Pc), encoding PRC1 subunits, both present ectopic sex combs. In ph-p 410 /Y males, 100% of L2 and L3 carried combs, and these phenotypes were significantly suppressed by CycG gain of function. Furthermore, the 20% of L2 and 5% of L3 sex combs of ph 220 /Y males were totally suppressed by CycG gain of function, and significantly enhanced by CycG loss of function (see Additional file 1). Finally, Pc 3 /+ males exhibit sex combs on 100% of L2 and 98% of L3 which were drastically suppressed by CycG gain of function ( Figure 5A).
We also examined other PcG-induced homeotic phenotypes, focusing first on wing to haltere transformations. Ninety-three percent of female and 96% of male Pc 3 /+ wings exhibit a partial wing into haltere transformation of the area close to the posterior wing, inducing deformation of the wing ( Figure 5B,C). CycG loss of function had no effect on this Pc 3 phenotype, whereas CycG gain of function reduced both its strength and its penetrance ( Figure 5B,C). Similar to Asx 3 /+ males, several PcG and ETP mutants present a partial transformation of A4 into A5. A few esc 21 /+ or Dsp1 1 /Y males presented this phenotype, which was not significantly affected by CycG misregulation (Additional file 1). However, 15% of Pc 3 /+ males presented such A4 transformations ( Figure 5D,E), whose strength and penetrance were significantly enhanced by CycG loss of function and completely suppressed by CycG gain of function ( Figure 5D,E). Ninety-seven percent of ph-p 410 males exhibited A4 into A5 transformations that were significantly suppressed by CycG gain of function (Additional file 1). The effect of CycG loss of function was not analyzed in ph-p 410 males because pharates died prior to abdominal pigmentation. In ph 220 /Y males, posterior A4 transformations were not modified by CycG loss of function but significantly suppressed by CycG gain of function (Additional file 1).
All these genetic data point toward a role of CycG similar to that of a PcG gene as PcG loss of function phenotypes were enhanced by CycG loss of function but suppressed by CycG gain of function.
Together, these data show that CycG loss of function tended to decrease segmental identity transformations of trxG loss of function mutants, whereas CycG overexpression increased them. Hence, CycG antagonizes trxG genes in the posterior abdomen, behaving again as a PcG gene.
CycG participates in PcG-dependent repression of Scr in leg imaginal discs Ectopic sex combs of PcG mutants are due to loss of silencing of the homeotic gene Sex combs reduced (Scr) in the second and third pairs of leg imaginal discs during larval development, leading to acquisition of a partial first-leg identity [54]. We therefore investigated the effect of CycG misregulation on expression of the Scr gene that specifies identity of the first thoracic segment in Drosophila melanogaster [55,56]. We monitored the pattern of SCR by immunostaining of leg imaginal discs, the larval structures that differentiate into legs during metamorphosis. In wild-type third instar larvae, cells with high levels of SCR form two semicircles in the L1 imaginal disc territory that gives rise to the anterior tibia and first tarsal segment ( Figure 7A), whereas L2 and L3 imaginal discs present no SCR positive cells. Importantly, neither CycG loss of function nor CycG gain of function male larvae exhibited any detectable alteration in the spatial pattern of SCR (data not shown).
We next looked at SCR distribution in third instar leg discs from three different PcG mutant males (mxc G46 , Pc 3 , ph-p 410 ). We observed ectopic SCR in the anterior compartment of L2 and L3 imaginal discs in all three contexts ( Figure 7A, Additional file 3). In male larvae combining mxc G46 and CycG loss of function, size of this ectopic SCR domain in both L2 and L3 discs was significantly increased ( Figure 7A; Additional file 4). Conversely, in male larvae combining mxc G46 and CycG gain of function, the ectopic SCR domain was almost completely suppressed (Additional file 4). Ectopic SCR domains were also significantly enlarged in L2 and L3 discs of male larvae combining either Pc 3 or ph-p 410 with CycG loss of function and decreased in those combining Pc 3 or ph-p 410 with CycG gain of function (Additional files 3 and 4).
These results indicate that modulation of PcG-induced ectopic sex comb phenotypes by CycG reflects modulation of the SCR pattern in imaginal discs. Interestingly, CycG loss of function enlarged the SCR domain even in ph 410 and Pc 3 individuals who presented a quasi complete penetrance of the ectopic sex comb phenotype in adults. Together, these data strongly suggest that CycG cooperates with PcG genes in epigenetic Scr repression.

CycG represses Pc 3 -induced ectopic Ultrabithorax in wing imaginal discs
CycG gain of function suppressed Pc 3 -induced wing to haltere transformations. This homeotic transformation is caused by ectopic expression of the homeotic gene Ultrabithorax (Ubx) that specifies haltere identity [57]. We therefore analyzed the effect of CycG misregulation on the UBX profile in wing imaginal discs. In wild-type third instar larvae, we observed a high level of UBX in haltere and third leg imaginal discs, whereas no UBX was detected in wing discs ( Figure 7B). This pattern was not altered by CycG misregulation (data not shown). We observed patches of ectopic UBX in the posterior compartment of Pc 3 /+ wing imaginal discs ( Figure 7B), that is, in the region that will be partly transformed into haltere in adults. CycG loss of function had no significant effect on this ectopic UBX profile, whereas CycG gain of function almost completely suppressed it ( Figure 7B, Additional file 4).
These data show that the suppression of Pc 3 -induced wing to haltere transformation by CycG gain of function correlates with a suppression of ectopic UBX in wing imaginal discs, suggesting a role for CycG in PcGdependent repression of Ubx in this tissue.

Discussion
Drosophila Cyclin G participates in control of the cell cycle and in transcriptional regulation [39,45,46,58]. Physical interaction between Cyclin G and the ETP Corto suggests that Cyclin G might be related to PcG and trxG genes involved in maintenance of gene silencing and gene activation, respectively. The present work strengthens this hypothesis, since we demonstrate that Cyclin G directly binds the ETP ASX as well, colocalizes extensively with ASX on chromatin, and genetically interacts with Asx. Furthermore, Cyclin G mostly binds open chromatin in which genes are undergoing transcription. We also addressed the connection between Cyclin G and the PcG/TrxG system by investigating genetic interactions with a broad range of PcG and trxG alleles. We describe strong interactions between CycG misregulation and several PcG and trxG genes, together with modification of ectopic Hox protein expression induced by PcG mutations. Our results link CycG to maintenance of Hox gene expression patterns during development, and sustain a role for Cyclin G in Representative trx E2 /+ (B) and trx E2 /da > CycG ΔP (C) males present, respectively, small and large light-pigmented cuticle patches on A5, denoting a partial transformation of abdominal segment A5 into A4 (white and black arrowheads). Fisher's exact test, **P < 0.0001 (n ≥ 30). A4, A5, A6: abdominal segments 4, 5, and 6. epigenetic maintenance of transcription, a mechanism that is essential for transmission of gene expression patterns to daughter cells.

Cyclin G, a modulator of Enhancers of Polycomb and Trithorax activity?
Drosophila Cyclin G was shown previously to physically interact with the ETP Corto [39]. Here, we demonstrate that this cyclin also binds the ETP ASX. Unlike Corto, which to date has only been detected in arthropods, ASX proteins are conserved in mammals. These proteins exhibit two remarkable domains, the ASX homology domain (ASXH), an amino terminal region containing two consensus sequences for nuclear receptor binding needed for ASX binding to the repressive PR-DUB complex [28], and a PHD domain located at the carboxyterminal end that interacts with DNA. Mouse and human ASXL1 also contain a number of cyclin recognition sites and motifs for phosphorylation by CDKs scattered along the protein, suggesting that they are phosphorylated by a Cyclin/CDK dimer [59]. Indeed, the presence of a cyclin binding motif close to a CDK target sequence is essential for optimal phosphorylation of cyclin/CDK targets [60]. We show here that CycG binds the C-terminal region of ASX (amino-acids 1139 to 1668), a region which contains a cyclin interaction domain (amino-acids 1210 to 1214) as well as a substrate motif for phosphorylation by CDKs (amino-acids 1261 to 1267) [61]. Hence, a Cyclin G/CDK complex might phosphorylate ASX and modulate its activity. Nuclear Corto is highly phosphorylated [62] and Figure 7 CycG misregulation modulates Hox protein profiles of PcG mutants. (A) Anti-SCR immunostainings of third instar larval leg imaginal discs from wild-type (wt), mxc G46 /Y, mxc G46 /Y;da > CycG RNAi or mxc G46 /Y;da > CycG ΔP males. (B) Anti-UBX immunostaining of third instar haltere (H), wing (W), and prothoracic leg (L3) imaginal discs from a wild-type female larva, and anti-UBX staining of wing discs from Pc 3 /+, Pc 3 /+;da > CycG RNAi or Pc 3 /da > CycG ΔP females. Note the reduced ectopic staining in the Pc 3 /da > CycG ΔP female wing disc (arrowhead).
although Corto lacks a canonical cyclin recognition motif, it contains a substrate motif for phosphorylation by CDKs in the Cyclin G interacting region [39]. Interestingly, this CDK target sequence is located in the chromodomain, involved in chromatin binding [63], suggesting that association of Corto with chromatin might be regulated by phosphorylation by a Cyclin G/CDK complex.
Phosphorylations are of paramount importance to regulate PcG and TrxG protein activity through the cell cycle. For example, in mammals, binding of the PcG protein BMI1 to chromatin is cell-cycle regulated and correlates with its phosphorylation status [64]. Furthermore, phosphorylation of PcG proteins EZH2 and SCML2 by cyclin/CDK complexes is regulated through the cell cycle [65,66]. These findings highlight a direct crosstalk between the Polycomb system of cellular memory and the cell-cycle machinery in mammals. Interestingly, fly Cyclin G controls the G1/S phase transition of the cell cycle and interacts with several CDKs (that is, CDK1, 2, 4, 5) [45,46,67,68]. Further investigations will be needed to determine whether a Cyclin G/CDK dimer phosphorylates the ETPs ASX and Corto and modulates their activity during the cell cycle.
Cyclin G interacts with ETP, PcG, and trxG genes in the regulation of homeotic genes ASX is involved in both activation and repression of homeotic genes [32], and this role is conserved by its mammalian homologs ASXL1 and ASXL2 [33,34]. Accordingly, some Asx mutants present both PcG-like and trxG-like transformations. For example, male Asx P1 flies bear partial transformation of abdominal segment A4 into A5, revealed by patches of dark pigmentation on A4, as well as partial transformation of A5 into A4, revealed by patches of unpigmented cuticle into A5 [69]. In a mouse Asxl2 mutant, vertebras present both posterior and anterior transformations corresponding to PcG and trxG phenotypes, respectively [33]. We show here that deregulation of CycG impacts on homeotic phenotypes of an Asx mutant that presents only PcG phenotypes, that is, ectopic sex combs and transformation of A4 into A5. Importantly, neither inactivation nor overexpression of CycG induces PcG or trxG phenotypes per se. Nevertheless, CycG inactivation enhances ectopic sex combs induced by Asx, while CycG overexpression suppresses both PcG-like phenotypes. Combining the Asx mutation with CycG overexpression even leads to a trxG-like transformation of A5 into A4. In agreement with these results, homeotic phenotypes of mutants for PcG genes crm, mxc, Pc, and ph are all enhanced by CycG inactivation and suppressed by CycG overexpression. On the other hand, transformation of abdominal segment A5 into A4 observed in trxG mutants was enhanced (ash1, trx) and even induced (brm) by CycG overexpression. Hence, in both posterior legs and abdominal segments A4 and A5, CycG behaves as an enhancer of PcG genes.
CycG overexpression, as well as CycG inactivation, induce developmental delay (unpublished data) and flies overexpressing CycG suffer from the Minute syndrome [45,46]. In Drosophila melanogaster, the Minute syndrome has been described as a dominant, haploinsufficient phenotype that includes delayed development, short and thin bristles together with poor fertility and longevity [70]. More than 50 Minute loci were genetically identified; 15 of them were characterized molecularly and contain genes encoding ribosomal proteins [71]. Genetic screens designed to isolate new PcG and trxG genes in flies have frequently identified Minute mutants as suppressors of PcG mutations [19,72]. This effect has been considered to be unspecific, since other factors resulting in developmental delays, that is, low temperature, also suppress the ectopic sex comb phenotype of PcG mutants [73]. Although homeotic phenotypes of PcG mutants are suppressed by CycG overexpression, the fact that CycG inactivation associates developmental delay and enhancement of PcG transformations indicates that CycG acts as a specific modifier of these phenotypes. Hence, our data characterize CycG as a bona fide enhancer of Polycomb-group genes, involved in Hox gene regulation.

Cyclin G, a transcriptional activator or repressor?
In agreement with our genetic analyses, we observed that CycG inactivation enlarges ectopic SCR domains in PcG mutant leg imaginal discs, whereas CycG overexpression eliminates ectopic SCR and UBX domains in leg and wing imaginal discs. This suggests that Cyclin G facilitates the maintenance of Scr and Ubx silencing by PcG complexes in these imaginal tissues. Nevertheless, whether this effect is direct or not remains to be determined. Paradoxically, we previously observed that Cyclin G is required for maintenance of Abd-B expression in the epithelium of abdominal segments A5 and A6 in young female pupae [58]. This effect might be direct as Cyclin G was shown to bind the Abd-B promoter and the iab-7 polycomb response element in embryos [39]. Altogether, these data show that Cyclin G is involved in epigenetic regulation of Hox gene expression, acting as a repressor or an activator depending on the tissue, the developmental stage, and the target gene. CycG is thus similar to many other genes encoding maintenance proteins that affect transcription differently depending on the context. For example, Drosophila E(z) is classified as a PcG repressor, but behaves genetically as an ETP [17,19]. Furthermore, brahma (brm) as well as most genes encoding members of the SWI/SNF chromatin remodeling complex Brahma-associated protein (BAP) are classified as trxG activators, but snr1, that encodes a conserved subunit of BAP, participates in cell-type specific transcriptional repression in the developing Drosophila wing [74].
To tackle the molecular mechanisms by which Cyclin G controls transcription, we analyzed its binding to polytene chromosomes in larval salivary glands. Cyclin G binds DAPI interbands indicative of open chromatin, co-localizes largely with RNA Pol II CTD phosphorylated on serine 2, and shows very few overlaps with H3K27me3. All these data point to the presence of Cyclin G on actively transcribed genes. Deregulation of CycG does not induce homeotic transformations per se, but modulates those due to PcG or trxG mutations. Cyclin G might thus preferentially affect 'destabilized' genes. Since Cyclin G co-localizes with DAPI interbands and with RNA Pol II phosphorylated on serine 2, and behaves as a PcG enhancer, its role could possibly be to moderate the expression of active genes.
We also show that Cyclin G extensively overlaps with ASX. As Cyclin G and ASX co-immunoprecipitate, Cyclin G and ASX can be assumed to interact on chromatin. ASX belongs to the repressive PR-DUB complex, which contains the histone deubiquitinase BAP1, also called Calypso in Drosophila [28]. PRC1 ubiquitinates H2A on lysine 119, and this ubiquitin residue can be removed by PR-DUB. Surprisingly, disruption of PR-DUB enzymatic function led to impaired Hox gene repression, as does a shortage of PRC1. Hence, PcG silencing has been proposed to depend on a dynamic equilibrium between H2A ubiquitination by PRC1 and deubiquitination by PR-DUB [28]. An interesting possibility could then be that Cyclin G influences this equilibrium by modulating ASX activity.
In embryonic stem cells, some key developmental genes, called bivalent, are simultaneously stamped by both repressive (H3K27me3) and activating (H3K4me3) histone marks [75]. Although these genes are associated with RNA Pol II CTD phosphorylated on serine 5, they are transcribed at a low level [76]. Ubiquitination of H2A by PRC1 controls this process, and has therefore been suggested to control transcriptional elongation [76,77]. These data raise the exciting possibility that Cyclin G fine-tunes transcription by controlling H2A ubiquitination via interaction with the ASX subunit of PR-DUB.

Conclusions
Our findings highlight a crosstalk between the Polycomb system and Drosophila Cyclin G. The importance and complexity of the interaction between Cyclin G and ASX warrant further investigation. It is tempting to speculate that this interaction regulates transcriptional elongation. Specific points to be explored in the near future include interaction between Cyclin G and ASX in the context of PR-DUB, involvement of Cyclin G in the balance between ubiquitinated and unubiquitinated H2A, and regulation of PR-DUB activity by Cyclin G through the cell cycle.
Alleles of PcG, trxG, or ETP genes used in this study are listed in Additional file 2, and their characteristics are described in [79]. Genetic interactions between these genes and CycG were assessed in trans-heterozygous flies obtained by crossing females heterozygous for a balanced PcG, trxG, or ETP mutation with males either da > CycG RNAi, da > CycG ΔP , or da::Gal4 as a control. All crosses were performed at 25°C, and parents were transferred to new vials every 3 days. Penetrance of homeotic phenotypes affecting legs, wings, or abdomen was determined among the progeny. Phenotypes were assessed by examining 30 flies for each genotype (whenever possible) under a dissecting microscope. Wings and male legs were mounted in Hoyer's medium. Sex combs were counted under a microscope at × 100 magnification. Statistical significance of results was evaluated using t-test and Fisher's exact test on GraphPad Quick-Calcs Web site: http://www.graphpad.com/quickcalcs/ contingency1/ (accessed April 2014).

Cell transfection and protein-protein interactions
The yeast two-hybrid screen performed to find interactors of the ASX-C terminal domain (ASX-C, residues 1139 to 1668) has been described previously [48,80]. Drosophila S2 cells were cultivated at 25°C in Schneider's Drosophila medium supplemented with 10% fetal calf serum and antibiotics. Then 5.10 6 cells were transfected into 25 cm 2 flasks using Effecten® Transfection reagent kit at a 1/10 DNA-Effecten® ratio according to the manufacturer's instructions (Effecten®; QIAGEN, Venlo, Limburg, The Netherlands). Cells were collected 36 or 72 h after transfection depending on the constructs, and total protein extracts were prepared as previously described [45]. Cross-linking was performed by treatment of cells with 1% paraformaldehyde for 10 min on ice prior to protein extraction. For co-immunoprecipitation, 500 μg of total cell extracts were incubated overnight at 4°C with 3 μg of either goat anti-Myc antibody (ab9132, Abcam, Cambridge, UK), mouse monoclonal anti-FLAG antibody (F3165, Sigma, St. Louis, MO, USA), or goat anti-HA as mock antibody (sc-805, Santa Cruz Biotechnology, Santa Cruz, CA, USA). Then, 30 μl of Bio-Adembeads Protein G (Ademtech, Westbury, New York, USA) were then incubated with the cell lysate for 3 h at 4°C. The beads were washed three times in ELB buffer [45] and resuspended in 30 μl of the same buffer. Furthermore, 20 μl of input, 20 μl of supernatant, and half of the beads were used for Western analysis. Immunoprecipitates were detected with rabbit polyclonal anti-Myc antibody (1:5,000; A00172, GenScript) and mouse monoclonal anti-FLAG antibody (1:2,000).
For co-immunoprecipitation in larvae, protein extracts were prepared from da > Myc-CycG ΔP third instar larvae, previously cleared of gut and fat body and treated with 1% paraformaldehyde for 10 min at room temperature. Five micrograms of either goat anti-Myc antibody, sheep anti-ASX N-ter antibody (described in [81]), or goat anti-HA as mock antibody were incubated with 50 μl of Dynabeads® Protein G (Life Technologies, Carlsbad, CA, USA) for 3 h at 4°C. The bead-antibody complexes were washed in ELB buffer and incubated with 1 mg of protein extracts overnight at 4°C. The beads were washed three times in ELB buffer before Western analysis. Immunoprecipitates were detected with goat anti-Myc antibody (1:5,000).

Immunostaining of imaginal discs
For each genotype, at least 15 third instar wandering larvae were dissected and fixed in 3.7% paraformaldehyde for 20 min at room temperature, then immunostained according to [82,83] using rat polyclonal anti-SCR antibody (1:100 [82]) or mouse anti-UBX monoclonal antibody (1:20; FP3.38 [84]). The universal biotinylated antibody (Vector Laboratories, CA, USA) was used at a 1:200 dilution. Staining was performed with VECTASTAIN Elite ABC system (Vector Laboratories, CA, USA) using DAB as substrate (D4418, Sigma, St. Louis, MO, USA). Note that for a given antibody, discs of all genotypes were incubated for the same length of time in DAB. Imaginal discs were mounted in PBS:glycerol (50:50). All pictures were acquired with a QICAM Fast 1394 digital camera, at × 100 magnification. Staining was quantified by calculating the percentage of stained area in the discs using Image J. SCR positive area was measured as a percentage of the total leg disc area, and UBX positive area was evaluated relative to the presumptive wing blade and hinge area of the wing disc. Statistical significance of results was evaluated using t-test.
Additional file 2: Table S2. trxG, PcG and ETP alleles used in this study. These alleles are described in [79].