ABSTRACT
In Drosophila, ubiquitous expression of a short Cyclin G isoform generates extreme developmental noise estimated by fluctuating asymmetry (FA), providing a model to tackle developmental stability. This transcriptional cyclin interacts with chromatin regulators of the Enhancer of Trithorax and Polycomb (ETP) and Polycomb families. This led us to investigate the importance of these interactions in developmental stability. Deregulation of Cyclin G highlights an organ intrinsic control of developmental noise, linked to the ETP-interacting domain, and enhanced by mutations in genes encoding members of the Polycomb Repressive complexes PRC1 and PR-DUB. Deep-sequencing of wing imaginal discs deregulating CycG reveals that high developmental noise correlates with up-regulation of genes involved in translation and down-regulation of genes involved in energy production. Most Cyclin G direct transcriptional targets are also direct targets of PRC1 and RNAPolII in the developing wing. Altogether, our results suggest that Cyclin G, PRC1 and PR-DUB cooperate for developmental stability.
INTRODUCTION
Developmental stability has been described as the set of processes that buffer disruption of developmental trajectories for a given genotype within a particular environment (Palmer, 1994). In other words, developmental stability compensates the random stochastic variation of processes at play during development. Many mechanisms working from the molecular to the whole organism levels contribute to developmental stability (Nijhout and Davidowitz, 2003). For example, chaperones, such as heat-shock proteins, participate in developmental stability in a large variety of developmental processes by protecting misfolded proteins from denaturation (Feder and Hofmann, 1999; Queitsch et al., 2002; Rutherford et al., 2007). In Drosophila melanogaster, adjustment of cell growth to cell proliferation is essential to developmental stability by allowing to achieve a consistant organ size (e.g. wing size) in spite of variation in cell size or cell number (Debat et al., 2011; Debat and Peronnet, 2013).
Developmental noise, the “sum” of the stochastic part of each developmental process, can be observed macroscopically for morphological traits. In bilaterians, quantification of departure from perfect symmetry, the so-called fluctuating asymmetry (FA), is the most commonly used index to estimate developmental noise (Van Valen, 1962; Palmer and Strobeck, 1992). Indeed, the two sides of bilaterally symmetrical traits are influenced by the same genotype and environmental conditions, and differences between them are thus only due to developmental noise. The use of FA as an index of developmental noise makes analysis of the mechanistic and genetic bases of developmental stability compatible with custom genetic and molecular approaches of developmental biology.
The evolutionary role of developmental stability is subject to many speculations (e.g. Dongen, 2006) as its genetic bases remain unclear (for reviews see Leamy and Klingenberg, 2005; Debat and Peronnet, 2013). Experiments showing the role of Hsp90 in buffering genetic variation led to the idea that developmental stability could be ensured by specific genes (Rutherford and Lindquist, 1998; Milton et al., 2003; Debat et al., 2006; Yeyati et al., 2007; Sangster et al., 2008). On the other hand, both theory and experiments show that complex genetic networks can become intrinsically robust to perturbations, notably through negative and positive feedbacks, suggesting that the topology of gene networks is of paramount importance for developmental stability (Barabasi and Albert, 1999; Siegal and Bergman, 2002; Newman, 2003; Kitano, 2004). Several authors have further suggested that hubs, i.e. the most connected genes in these networks, might be particularly important for developmental stability (Rutherford et al., 2007; Levy and Siegal, 2008).
In Drosophila, mutants for dlLP8 and hid, two genes involved in the control of systemic growth and apoptosis respectively, have been reported to display high FA as compared to wild type flies from the same genetic background (Garelli et al., 2012; Colombani et al., 2012; Neto-Silva et al., 2009), suggesting that these genes are important for developmental stability. Two studies have scanned the Drosophila genome for regions involved in developmental stability using FA as an estimator of developmental noise (Breuker et al., 2006; Takahashi et al., 2011). Several deletions increased FA but the genes responsible for this effect inside the deletions were not identified. Nevertheless, these studies confirm that the determinism of developmental stability could well be polygenic, as suggested by Quantitative Trait Loci analyses in mouse (Leamy et al., 2002; Leamy et al., 2005; Leamy et al., 2015). Together, these data reinforce the idea that developmental stability depends on gene networks.
We have shown that the gene Cyclin G (CycG) of Drosophila melanogaster,which encodes a protein involved in transcriptional regulation and in the cell cycle, is important for developmental stability (Salvaing et al., 2008a; Faradji et al., 2011; Debat et al., 2011; Dupont et al., 2015). Indeed, ubiquitous expression of a short Cyclin G version lacking the C-terminal PEST-rich domain (CycGΔP) generates a very high FA in several organs, notably in the wing. Interestingly, FA induced by CycGΔP expression correlates with high variability in cell size and loss of correlation between cell size and cell number, suggesting that the noisy process would somehow be connected to cell cycle related cell growth (Debat et al., 2011). Hence, CycG deregulation provides a convenient sensitized system to tackle the impact of cell growth variability on developmental stability.
We previously showed that CycG encodes a transcriptional cyclin and interacts with genes of the Polycomb-group (PcG), trithorax-group (trxG), and Enhancer of Trithorax and Polycomb (ETP) families (Salvaing et al., 2008a; Salvaing et al., 2008b; Dupont et al., 2015). These genes encode evolutionary conserved proteins assembled into large multimeric complexes that bind chromatin. They ensure maintenance of gene expression patterns during development (for recent reviews see Grossniklaus and Paro, 2014; Kingston and Tamkun, 2014; Geisler and Paro, 2015). PcG genes are involved in long-term gene repression, whereas trxG genes maintain gene activation and counteract PcG action. ETP genes encode cofactors of both trxG and PcG genes, and behave alternatively as repressors or activators of target genes (Gildea et al., 2000; Grimaud et al., 2006; Beck et al., 2010). More recently, we discovered that CycG behaves as an Enhancer of Polycomb regarding homeotic gene regulation suggesting that it is involved in the silencing of these genes (Dupont et al., 2015). Importantly, Cyclin G physically interacts with the ETP proteins Additional Sex Comb (ASX) and Corto via its N-terminal ETP-interacting domain, and co-localizes with them on polytene chromosomes at many sites (Salvaing et al., 2008a; Dupont et al., 2015). Hence, Cyclin G and these ETPs might share many transcriptional targets and might in particular control cell growth via epigenetic regulation of genes involved in growth pathways.
Here, we investigate in depth the role of CycG in developmental stability. We first show that localized expression of CycGΔP in wing imaginal discs is necessary and sufficient to induce high FA of adult wings. Furthermore, this organ-autonomous effect increases when the ETP-interacting domain of Cyclin G is removed. We show that several mutations for PcG or ETP genes, notably those encoding members of the PRC1 and PR-DUB complexes, substantially increase CycG-induced FA. Next, we report analysis of the transcriptome of wing imaginal discs expressing CycGΔP by RNA-seq and find that transcriptional deregulation of genes involved in translation and energy production correlates with high FA of adult wings. By ChIP-seq, we identify Cyclin G binding sites on the whole genome in wing imaginal discs. Strikingly, we observe a significant overlap with genes also bound by ASX, by the Polycomb Repressive complex PRC1, and by RNAPolII in the same tissue. We identify a subnetwork of 222 genes centred on Cyclin G showing simultaneous up-regulation of genes involved in translation and down-regulation of genes involved in mitochondrial activity and metabolism. Taken together, our data suggest that Cyclin G and the Polycomb complexes PRC1 and PR-DUB cooperate in sustaining developmental stability. Precise regulation of genes involved in translation and energy production might be important for developmental stability.
RESULTS
Expression of CycGΔP in wing precursors is necessary and sufficient to induce high wing FA
We previously reported that expression of CycG deleted of the PEST-rich C-terminal domain (amino-acids 541 to 566) (CycGΔP) under control of ubiquitous drivers (da-Gal4 or Actin-Gal4) generated extremely high FA, notably in wings (Debat et al., 2011) (Figure 1). The strength of this effect was unprecedented in any system or trait. Expression of CycGΔP thus provides a unique tool to investigate developmental stability in depth. To determine whether wing FA was due to local or systemic expression of CycGΔP, we tested a panel of Gal4 drivers specific for wing imaginal discs or neurons. A brain circuit which relays information for bilateral growth synchronization was recently identified (Vallejo et al., 2015). It notably involves a pair of neurons expressing the dILP8 receptor that connects with the insulin-producing cells (IPCs) and the prothoracicotropic hormone (PTTH) neurons. This circuit was particularly appropriate to test the existence of a remote effect of CycGΔP expression in generating high FA in the wing. Expression of CycGΔP in this circuit (using dilp3-, NPF-, pdf-, per-, phm- and R19B09-Gal4 drivers) did not increase FA of adult wings (Figure 2 and Table 1). Furthermore, expression of CycGΔP in cells of the future wing hinge using the ts-Gal4 driver did not affect wing FA either. By contrast, expressing CycGΔP with 5 different wing pouch drivers (nub-, omb-, rn-, sd- and vg-Gal4) induced high FA. We thus concluded that CycGΔP-induced wing FA was due to an intrinsic response of the growing wing tissue.
The Cyclin G ETP interacting domain sustains developmental stability
The 566 amino-acid Cyclin G protein exhibits 3 remarkable domains: the ETP-interacting domain (amino-acids 1 to 130) that physically interacts with the ETPs Corto and ASX, a cyclin domain (amino-acids 287 to 360) that presents high similarity with the cyclin domain of vertebrate G-type cyclins, and a PEST-rich domain (amino-acids 541 to 566) (Salvaing et al., 2008a; Faradji et al., 2011; Dupont et al., 2015). To test whether the interaction with ETPs (and thus transcriptional regulation by Cyclin G) could be important to control FA, we generated new transgenic lines enabling to express different versions of the CycG cDNA: CycGFL (encoding the full-length protein), CycGΔE (encoding an ETP-interacting domain deleted protein), CycGΔP (encoding a PEST domain deleted protein), and CycGΔEΔP (encoding an ETP-interacting and PEST domain deleted protein). In order to express these different cDNAs at the same level and compare the amounts of FA induced, all transgenes were integrated at the same site using the PhiC31 integrase system (at position 51C on the second chromosome). Expression of these transgenic lines was ubiquitously driven by da-Gal4. We first confirmed that expression of CycGΔP induced very high FA as compared to yw and da-Gal4/+ controls. Furthermore, expression of CycGFL also significantly increased FA, although to a much lesser extent. Interestingly, expression of either CycGΔE or CycGΔEΔP significantly increased FA as compared to CycGFL or CycGΔP, respectively (Figure 3 and Table 2). These results show that the ETP interacting domain tends to limit Cyclin G-induced FA and suggest that the interaction between Cyclin G and chromatin regulators sustains developmental stability.
CycG and PcG or ETP genes interact for developmental stability
We next addressed genetic interactions between CycG and PcG or ETP genes for developmental stability. The alleles used are listed in Table 3. FA of flies heterozygous for PcG and ETP loss of function alleles was not significantly different from that of control flies. However, when combined with a da-Gal4, UAS-CycGΔP chromosome, many of these mutations significantly increased wing FA as compared to da-Gal4, UAS-CycGΔP flies (Figure 4 and Table S1). This was notably the case for alleles for PRC1 and PR-DUB encoding genes, the PcG genes Sex comb extra (Sce1, Sce33M2 and SceKO4), calypso (caly1 and caly2), Sex comb on midleg (ScmD1), Polycomb (Pc1), and polyhomeotic (ph-p410 and ph-d401ph-p602). No modification of CycGΔP-induced FA was observed with the Psc1 allele. However, this allele has been described as a complex mutation with both loss and gain of function features (Adler et al., 1989).
Opposite effects were observed for different alleles of ETPs Asx and corto. Asx22P4 increased da-Gal4, UAS-CycGΔP FA whereas AsxXF23 decreased it. AsxXF23 behaves genetically as a null allele but has not been molecularly characterized (Simon et al., 1992), whereas the Asx22P4 allele does not produce any protein and thus likely reflects the effect of loss of ASX (Scheuermann et al., 2010). Similarly, the cortoL1 allele increased CycGΔP-induced FA whereas the corto420 allele had no effect. In order to characterize these corto alleles, we combined them with the Df(3R)6-7 deficiency that uncovers the corto locus, amplified the region by PCR and sequenced it. The corto420 allele corresponds to a substitution of 14,209 nucleotides starting at position −59 upstream of the corto Transcriptional Start Site (TSS) by a 30-nucleotide sequence. Hence, this allele does not produce any truncated protein. By contrast, cortoL1 corresponds to a C towards T substitution that introduces a stop codon at position +73 downstream the TSS, generating a 24 amino-acid polypeptide. cortoL1 might then behave as a dominant-negative mutation. Lastly, no modification of CycGΔP-induced FA was observed for E(z)63 and esc21.
Interestingly, Asx and caly encode proteins of the Polycomb Repressive complex PR-DUB whereas Pc, ph, Sce and Scm encode proteins of the Polycomb Repressive complex PRC1, and E(z) and esc encode proteins of the Polycomb Repressive complex PRC2. Taken together, these results indicate that Cyclin G interacts with the Polycomb complexes PRC1 and PR-DUB, but not with PRC2, for developmental stability.
Expression of CycGΔP or CycGΔEΔP does not modify the bulk of H2AK118ub
Cyclin G binds polytene chromosomes at many sites and co-localizes extensively with PH and ASX suggesting a potential interaction with the PRC1 and PR-DUB complexes on chromatin (Salvaing et al., 2008a; Dupont et al., 2015). The two genes Sce and caly encode antagonistic enzymes of the PRC1 and PR-DUB complexes, respectively. SCE, aka dRing, ubiquitinates histone H2A on lysine 118 (H2AK118ub) whereas Calypso, aka dBap1, is the major deubiquitinase of the same H2A residue (Scheuermann et al., 2010; Scheuermann et al., 2012). To investigate whether Cyclin G was related to these ubiquitin ligase/deubiquitinase activities, we immunostained polytene chromosomes from w1118 larvae with anti-Cyclin G and anti-human H2AK119ub antibodies (homologous to Drosophila H2AK118ub) (Pengelly et al., 2015). Cyclin G and H2AK118ub co-localized extensively on chromosome arms suggesting that Cyclin G transcriptional activity might somehow be connected to the presence of this histone mark (Figure S1A). However, when either CycGΔP or CycGΔEΔP was expressed in the posterior compartment of wing imaginal discs using the en-Gal4 driver, the global amount of H2AK118ub was not markedly modified (Figure S1B, Figure S1C). We thus concluded that high FA was not related to a global perturbation of H2AK118 ubiquitination level.
Cyclin G controls the expression of genes involved in translation and energy production
Cyclin G controls the transcription of the homeotic gene Abdominal-B and more specifically behaves as an Enhancer of PcG gene in the regulation of homeotic gene expression (Salvaing et al., 2008b; Dupont et al., 2015). However, the high number of Cyclin G binding sites on polytene chromosomes suggests that this cyclin has many other transcriptional targets. We thus hypothesized that the high FA induced by expression of CycGΔP might be related to the deregulation of Cyclin G transcriptional targets. To further address the role of Cyclin G in transcriptional regulation, we deep-sequenced the transcripts from wing imaginal discs of da-Gal4, UAS-CycGΔP/+ and da-Gal4/+ third instar larvae. Sequence reads were aligned with the Drosophila melanogaster genome to generate global gene expression profiles. We performed differential analyses to obtain expression changes for da-Gal4, UAS-CycGΔP/+ as compared to the da-Gal4/+ control. With an adjusted p-value threshold of 0.05, we retrieved 530 genes whose expression was significantly different between the two genotypes (Table S2). Surprisingly, expression of CycG was only weakly induced in da-Gal4, UAS-CycGΔP/+ imaginal discs as compared to da-Gal4/+ imaginal discs (1.3 fold). In order to test the hypothesis that Cyclin G could directly or not, induce its own repression, we designed primers in the 3’UTR to measure expression of the endogenous CycG gene. Indeed, expression of endogenous CycG was significantly decreased when CycGΔP was expressed (Figure 5A and Table S3). Among the 530 genes deregulated in da-Gal4, UAS-CycGΔP/+ imaginal discs, 216 were up-regulated and 314 down-regulated. Analysis of Gene Ontology (GO) revealed that up-regulated genes were enriched in the categories cytoplasmic translation and translational initiation whereas down-regulated genes were enriched in the category mitochondrial respiratory chain complex (Figure 5B and Table S4). By RT-qPCR, we verified that several ribosomal protein genes (RpL15, RpL7 and Rack1) were over-expressed in da-Gal4, UAS-CycGΔP/+ imaginal discs (Figure 5C and Table S5).
In conclusion, CycG-induced fluctuating asymmetry correlates with activation of genes involved in translation and repression of genes involved in energy production.
Cyclin G binds the Transcriptional Start Sites of many genes also bound by PRC1 and ASX
In order to determine the direct transcriptional targets of Cyclin G, we analysed by ChIP-seq the genome-wide binding sites of Cyclin G in +/ UAS-Myc-CycGΔP; da-Gal4/+ imaginal discs. 889 genes with significant peaks at the transcriptional start site (TSS) were recovered (Table S6 and Figures 6A and 6B). ChIP-qPCR analysis of Cyclin G binding on RPL7, RPL5, and Rack1 confirmed that Cyclin peaked on the TSS of these genes and decreased on the gene body (Figure 6C and Table S7). Furthermore, Cyclin G bound its own TSS almost significantly. We then analysed the binding of Cyclin G on its own gene by ChIP-qPCR and verified the presence of Cyclin G on its TSS (Figure 6C and Table S7). As endogenous CycG was down-regulated when CycGΔP was expressed, this suggests that Cyclin G represses its own promoter.
The 889 Cyclin G-bound genes were enriched in GO categories cytoplasmic translation and protein phosphorylation (Figure 6D). Comparison of the 530 genes deregulated in imaginal discs expressing CycGΔP with the 889 genes presenting a peak at the TSS showed that only 62 genes were both deregulated (39 up- and 23 down-regulated) and bound by Cyclin G (Table S8). Strikingly, the 39 up-regulated genes were significantly enriched in the GO category translation (GO:0002181~cytoplasmic translation, 14 genes, enrichment score: 11.84, adjusted p-value 2.07E-16).
Using published datasets, we analysed the correlation between regions bound by Cyclin G in +/UAS-Myc-CycGΔP; da-Gal/+ imaginal discs and those bound by PRC1, PR-DUB or RNAPolII, or enriched in H3K27me3, in wild type wing imaginal discs (Table S9). Cyclin G-bound regions were significantly exclusive from H3K27me3, corroborating polytene chromosome immunostainings (Dupont et al., 2015). The same comparisons were performed gene-wise and gave the same results. Notably, 80% of Cyclin G-bound genes were bound by RNAPolII (Figure 7). Considering RNAPolII as a proxy for transcriptional activity, we concluded that Cyclin G-bound genes were located in open chromatin and were either paused or transcribed. However, Cyclin G-bound genes were also significantly enriched in PRC1 target genes. Given that PRC1 has the ability to block transcriptional initiation (Dellino et al., 2004), it suggests that Cyclin G-bound genes were most probably paused. Cyclin G also shared many target genes with ASX but, though ASX and Calypso belong to the PR-DUB complex, Cyclin G did not share binding sites with Calypso. This indicates either that the interaction between Cyclin G and ASX destabilizes the PR-DUB complex or that it takes place outside PR-DUB.
Cyclin G is central in the wing imaginal disc network
These genome-wide analyses indicate that Cyclin G coordinates the expression of genes involved in translation and energy production. However, only a few Cyclin G-bound genes were deregulated in da-Gal4, UAS-CycGΔP/+ imaginal discs. To better understand how Cyclin G orchestrates target gene expression, we developed a systems biology approach. We first built an interactome based on genes expressed in control da-Gal4/+ wing imaginal discs (with a cutoff of 10 reads). Edges corresponding to protein-protein interactions (PPI) and transcription factor-gene interactions (PDI) were integrated into this interactome through DroID (Murali et al., 2011). The resulting wing imaginal disc interactome, further called the WID network, was composed of 9,966 nodes (proteins or genes) connected via 56,133 edges (interactions) (WID.xmml file). We then examined the position of Cyclin G in this network. Betweenness centrality - i.e. the total number of non-redundant shortest paths going through a certain node – is a measure of centrality in a network (Yu et al., 2007). A node with a high betweenness centrality could control the flow of information across the network (Yamada and Bork, 2009). With 8.32E-03, Cyclin G had one of the highest value of betweenness centrality of the network, ranking at the 30th position among the 9,966 nodes. This suggests that Cyclin G represents a hub in the WID network.
In order to isolate a connected component of the WID network that showed significant expression change when CycGΔP is expressed, we introduced the expression matrix describing expression of the 530 significantly deregulated genes in the WID network. We next used JactiveModules to identify sub-networks of coderegulated genes (Ideker et al., 2002). A significant sub-network of 222 nodes and 1069 edges centred on Cyclin G was isolated (Z score 48.53). This sub-network was laid out according to functional categories (Figure 8A, CycG_subnetwork.xmml). Four modules composed of genes respectively involved in transcription, mitochondrial activity, translation, and metabolism, were found to be highly connected to Cyclin G. Strikingly, the “translation” module was mainly composed of genes up-regulated in da-Gal4, UAS-CycGΔP/+ wing imaginal discs. On the contrary, the “mitochondrion” and “metabolism” modules were mainly composed of genes down-regulated in da-Gal4, UAS-CycGΔP/+ wing imaginal discs. Hence, high fluctuating asymmetry of da-Gal4, UAS-CycGΔP/+ flies correlated positively with the expression of genes involved in translation and negatively with the expression of genes involved in energy production and metabolism. Interestingly, Cyclin G-bound genes in this sub-network were enriched in genes bound by the PRC1 proteins PC, PH and PSC, as well as by RNAPolII, and to a lesser extent by ASX (Figure 8B).
DISCUSSION
The CycG gene of Drosophila melanogaster encodes a cyclin involved in transcriptional control, cell growth and the cell cycle (Salvaing et al., 2008; Faradji et al., 2011). Mild overexpression of a cDNA encoding Cyclin G deleted of a short C-terminal sequence potentially involved in Cyclin G degradation (a PEST-rich domain; da-Gal4, UAS-CycGΔP/+) induces high fluctuating asymmetry (FA), notably of wings (Debat et al., 2011). Under laboratory conditions (i.e. low environmental variation combined with near isogenic lines), this FA should mainly result from developmental noise (Debat and Peronnet, 2013). Thus, da-Gal4, UAS-CycGΔP flies provide a unique tool to investigate the genetic bases of developmental stability. Cyclin G interacts physically with two chromatin regulators of the Enhancers of Trithorax and Polycomb family (ETP), and genetically with many Polycomb-group (PcG) and trithorax-group (trxG) genes (Dupont et al., 2015). This prompted us to re-examine CycG-induced developmental stability, notably by testing the effect of chromatin regulator mutations, and to investigate deeply the role of Cyclin G in transcriptional regulation.
Cyclin G maintains developmental stability through an organ-autonomous process that involves the PRC1 and PR-DUB complexes
In Drosophila very few mutations have been shown to induce an abnormally high FA. Among them are mutations of the gene encoding the Drosophila insulin-like peptide 8 (Dilp8). Dilp8 participates in systemic coordination of growth. Being produced in growing tissues, it is secreted into the haemolymph and regulates hormone production via a well-identified neuronal circuit (Parker and Shingleton, 2011; Garelli et al., 2012; Colombani et al., 2012). Notably, the neurons that produce Lgr3, the Dilp8 receptor, have been identified, and inactivation of Lgr3 in these neurons also induces high FA. We investigated here the role of CycG in this process by deregulating it in the different modules of the circuit. CycG-induced wing FA only occured when the deregulation was local, i.e. in wing imaginal discs. More particularly, deregulation of CycG in the Lgr3 neurons did not increase FA. We cannot exclude that Cyclin G induces expression of a systemic factor that is dumped into the haemolymph. However, neither Dilp8 nor any other insulin-like peptide gene were found deregulated in da-Gal4, UAS-CycGΔP wing imaginal discs. Altogether, these observations suggest that CycG maintains developmental stability through an autonomous mechanism which would not involve the systemic Dilp8/Lgr3 pathway. Such a mechanism recalls Garcia-Bellido’s Entelechia model which proposes that local interactions between wing imaginal disc cells, or populations of these cells, orchestrate their own proliferation in order to generate an adult organ of constant size and shape, independently of global cues (García-Bellido and García-Bellido,1998; García-Bellido 2009).
Expression of Cyclin G deleted of the ETP interacting domain doubles FA as compared to expression of Cyclin G with this domain, irrespective of whether the PEST domain is present or not. Hence, the interaction between Cyclin G and chromatin regulators might somehow participate in developmental stability. To test this hypothesis, we combined the da-Gal4, UAS-CycGΔP chromosome and ETP or PcG mutations. We observed that mutations of the PRC1 and PR-DUB encoding genes strongly increase FA. Moreover, many of the genes that are bound by Cyclin G in wing imaginal discs are also bound by PRC1 and by ASX. Altogether these observations suggest that transcriptional regulation of target genes shared by Cyclin G, PRC1 and ASX is of paramount importance for developmental stability. We did not observe any significant overlap between Cyclin G-bound genes and binding sites for Calypso, the second component of PR-DUB. Yet, caly mutations strongly increase CycG-induced FA. Thus, the role of PR-DUB in this context remains to be clarified.
PRC1 and PR-DUB contain antagonistic enzymes (SCE/dRing and Calypso) that respectively ubiquitinates and deubiquitinates H2A on lysine 118 in Drosophila (lysine 119 in human). Cyclin G co-localizes extensively with H2AK118ub on polytene chromosomes. However, no modification in the global level of H2AK118 ubiquitination was detected in tissues where Cyclin G isoforms were expressed. It was recently shown that canonical PRC1 accounts for only a small fraction of global H2AK118ub, most of this ubiquitination being due to L(3)73Ah, a homolog of mammalian PCGF3 (Lee et al., 2015). Altogether, our data suggest that H2AK118ub is not involved in developmental stability and rather support the importance of the interaction between Cyclin G and canonical PRC1 in this process. It is tempting to speculate that PRC1 and PR-DUB are partners in the ubiquitination/deubiquitination of an unknown protein important for developmental stability.
Regulation of growth during the cell cycle might be a factor of developmental stability
Additional evidence further connects developmental stability to growth regulation during the cell cycle. Indeed, CycG-induced developmental noise is associated with high variance in cell size along with loss of correlation between cell size and cell number (Debat et al., 2011). As Cyclin G is involved in the control of growth in G1 phase of the cell cycle (Faradji et al., 2011), this supports the hypothesis that a mechanism linked to the regulation of cell cycle-dependent growth is essential for developmental stability (Debat et al., 2011). The fact that genes deregulated in wing imaginal discs deregulating CycG are involved mainly in translation, energy production and metabolism, strengthens this hypothesis.
It was shown that promoters of actively transcribed genes, notably GAPDH and several ribosomal protein genes, are bookmarked by ubiquitination during mitosis (Arora et al., 2012; Arora et al., 2015). This mechanism would allow postmitotic resumption of their transcription at the very beginning of the G1 phase. Ubiquitination of these genes correlates with active histone marks such as H3K4me3 and H3K36me3 but not with the repressive histone mark H3K27me3. The enzymes responsible for this ubiquitination are the vertebrate PSC homolog BMI1, and Ring1A, one of the SCE/dRing homologs (Arora et al., 2015). In vertebrates, the major PRC1 component that catalyzes H2A ubiquitination is not Ring1A but its homolog Ring1B suggesting that the role of BMI1 and Ring1A in molecular bookmarking are independent of PRC1, and that BMI1 and Ring1A ubiquitinate another chromatin protein (Arora et al., 2015). In Drosophila, this role might be played by PRC1 and SCE/dRing. Cyclin G is exclusive of H3K27me3, and binds the promoter of many ribosomal protein genes (Dupont et al., 2015 and the present work). Furthermore, CycG deregulation impairs G1 phase progression and cell growth (Faradji et al., 2011). Lastly, the highest FA is observed when Cyclin G lacking the PEST domain, a potential ubiquitination site, is expressed. Hence, an exciting hypothesis would be that Cyclin G is ubiquitinated by PRC1, or PSC and SCE/dRing outside PRC1, thus releasing the transcriptional standby of active genes at the end of mitosis. In agreement with this, we found that genes involved in metabolism and mitochondrial activity are down-regulated in the CycGΔP context. However, we observed at the same time that ribosomal protein genes are up-regulated which should rather promote growth. This foreshadows a complex relationship between Cyclin G and the PRC1 and PR-DUB complexes in the cell cycle-dependent regulation of these genes and appeals to the use of a more integrative, systems biology, approach.
Fine-tuned regulation of genes involved in translation, metabolism and mitochondrial activity is necessary for developmental stability
Cyclin G appears central in a small regulatory sub-network that connects genes involved in metabolism, mitochondrial activity and translation. Besides, many of Cyclin G’s direct transcriptional targets in this network are also targets of PRC1 and RNAPolII, and to a lesser extent of ASX. Interestingly, it was recently shown by a large scale analysis of the Drosophila wing imaginal disc proteome that wing size correlates with some basic metabolic functions, positively with glucose metabolism and negatively with mitochondrial activity, but not with ribosome biogenesis (Okada et al., 2016). In agreement with this, we report here that many genes involved in basic metabolism, such as for example Gapdh1, Gapdh2 or Jafrac1, are down-regulated in the CycGΔP context, which also agrees with the small mean size of CycGΔP flies, organs and cells. However, while mitochondrial genes are negatively regulated, ribosomal biogenesis genes are simultaneously positively regulated. Although transcriptome variations are probably not a direct image of proteome variations, our data suggest that robustness of wing size correlates with the fine-tuning of these key functions relative to each other.
Noisiness of gene expression as a source of developmental noise
Cyclin G, PRC1 and PR-DUB are mainly involved in the regulation of transcription. An exciting hypothesis would be that alteration of developmental stability is due to the noisy transcription of their shared targets. CycG-induced high FA is associated with high variability of cell size, that might be due to variability in expression of target genes which are mainly involved in growth control. Phenotypic variations in isogenic populations of both prokaryotic and eukaryotic cells may indeed result from stochastic gene expression mechanisms (McAdams and Arkin, 1997). An increasing corpus of data suggests that the process of gene regulation per se can strongly affect variability in gene expression among adjacent cells (for a review see Sanchez et al., 2013). Transcriptional noise may arise at all steps of transcription. For example, the architectural features of promoters have clear effects on mRNA and protein fluctuations in a population of genetically identical cells (Sanchez et al., 2013). RNA polymerase II pausing during elongation is also a source of transcriptional noise (Rajala et al., 2010). In particular, H3K36 methylation, that is related to transcriptional elongation, prevents spurious cryptic transcription from within the gene body (Venkatesh et al., 2012). Recently, activity of the Polycomb complex PRC2 was shown to be important to prevent spurious transcription of inactive genes and to suppress pervasive transcription of intergenic regions (Lee et al., 2015). Mutations of E(z) and esc that encode two PRC2 members had no effect on CycG-induced FA. Dysfunction of PRC2-dependent spurious transcription control is thus unlikely to be the cause of any CycG-induced developmental noise. Nevertheless, a similar but weaker effect on intergenic transcription was attributed to PRC1 (Lee et al., 2015). The binding of Cyclin G on many TSS is rather in favor of a role in limiting noisy initiation of transcription. Interestingly, in several cases, noise in gene expression specifically concerns a subset of genes (Weinberger et al., 2012). For example, H3K36 methylation hinders cryptic transcription in a subclass of genes involved in longevity in S. cerevisiae and C. elegans (Sen et al., 2015). It is thus tempting to speculate that cooperation between Cyclin G and the PRC1 and PR-DUB complexes is important to prevent spurious transcription of genes involved in growth in the broad sense. It will be very interesting to address these points in the future.
MATERIAL AND METHODS
Plasmids
The pPMW-attB plasmid was built as follows: Gateway® vector pPMW (Invitrogen, a gift from T. Murphy; (Huynh and Zieler, 1999) was linearized by digestion with NsiI; the attB sequence was amplified from pUASTattB (Bischof et al., 2007) using primers attB-NsiIF and attB-NsiIR (Table S10) and the PCR product was digested with NsiI; the digested PCR product and the linearized plasmid were ligated and sequenced. This plasmid was deposited at Addgene (plasmid # 61814).
The full-length CycG cDNA (CycGFL, encoding the 566 amino-acid protein) was amplified from S2 cell cDNAs using primers CycGnF and CycGnR. cDNAs encoding truncated forms of Cyclin G (CycGΔP, Cyclin G deleted of the putative PEST domain corresponding to amino-acids 542 to 566;CycGΔE, Cyclin G deleted of the ETP-interacting domain corresponding to amino-acids 1 to 130; CycGΔEΔP, Cyclin G deleted of both domains) were amplified from the full-length CycG cDNA using primers CycGnF and CycG541R, CycG130F and CycGnR, and CycG130F and CycG541R, respectively (Table S10 and Dupont et al., 2015). The PCR products were cloned into pENTR/D-TOPO® (Invitrogen), transferred into pPMW-attB and the resulting plasmids pPMW-attB-CycGFL, pPMW-attB-CycGΔP, pPMW-attB-CycGΔE, pPMW-attB-CycGΔEΔP were sequenced.
Drosophila melanogaster strains and genetics
Flies were raised on standard yeast-cornmeal medium at 25°C.
Myc-CycG transgenic lines were obtained by PhiC31-integrase mediated insertion into strain y1M{vas-int.Dm}ZH-2Aw*;M{3xP3-RFP.attP’}ZH-51C (stock BL-24482). Plasmids pPMW-attB-CycGFL, pPMW-attB-CycGΔP, pPMW-attB-CycGΔE and pPMW-attB-CycGΔEΔP were injected into embryos, G0 adults were back-crossed to yw, and G1 transformants were crossed to yw again to obtain G2 transformants (BestGene Inc.). Transformants were individually crossed with yw; Sp/CyO, and the curly wing siblings were crossed with each other. Homozygous transgenic lines were then obtained by crossing 5 females and 5 males. The resulting lines were named UAS-Myc-CycGFL, UAS-Myc-CycGΔP, UAS-Myc-CycGΔE and UAS-Myc-CycGΔEΔP.
To estimate fluctuating asymmetry (FA), five replicate crosses were performed for each genotype, wherein 6 females carrying a Gal4 driver were mated with 5 males carrying a CycG transgene. Parents were transferred into a new vial every 48 h (three times) then discarded. 30 females were then sampled from the total offspring. Gal4 drivers used were daughterless-Gal4 (da-Gal4) (Wodarz et al., 1995), nubbin-Gal4 (nub-Gal4), optomotor-blind-Gal4 (omb-Gal4), rotund-Gal4 (rn-Gal4), scalloped-Gal4 (sd-Gal4), teashirt-Gal4 (tsh-Gal4), vestigial-Gal4 (vg-Gal4) (from the Bloomington Drosophila stock center), and Insulin-like peptide 3-Gal4 (dILP3-Gal4), neuropeptide F-Gal4 (NPF-Gal4), Pigment-dispersing factor-Gal4 (Pdf-Gal4), period-Gal4 (per-Gal4), phantom-Gal4 (phm-Gal4), Prothoracicotropic hormone-Gal4 (Ptth-Gal4), R10B09-Gal4, kind gifts from Dr Maria Dominguez’s lab (Ferres-Marco et al., 2006).
The da-Gal4, UAS-CycGΔP third chromosome, obtained by recombination of da-Gal4 with the original UAS-CycGΔP transgene (RCG76), was used to test genetic interactions between CycG and several PcG or ETP mutations (Dupont et al., 2015). Alleles used are described in (Soto et al., 1995; Beuchle et al., 2001;Salvaing et al., 2006; Gaytán de Ayala Alonso et al., 2007; Fritsch et al., 2003; Gutiérrez et al., 2012) (Table 3). To estimate FA, 5 crosses were performed for each genotype, wherein 6 PcG or ETP mutant females were mated with 5 da-Gal4, UAS-CycGΔP males, the parents were transferred into a new vial every 48 h (three times) then discarded. 30 females combining the PcG or ETP mutation and da-Gal4 UAS-CycGΔP were sampled from the offspring. PcG and ETP mutant females were crossed with da-Gal4 in parallel to measure FA of heterozygous mutants.
Morphometrics
Right and left wings of sampled females were mounted on slides, dorsal side up, in Hoyer’s medium. Slides were scanned with a Hamamatsu Nanozoomer Digital Slide scanner, running the Nanozoomer software with a 20x objective and an 8-bit camera. Wing pictures were separately exported into tif format using NDP.view and the 5x lens. All wings were oriented with the hinge to the left. Image J was used to digitize 15 landmarks or only landmarks 3 and 13 when indicated (Figure 1B). Analysis of size FA was performed as described previously (Debat et al., 2011) using the FA10 index as FA estimator, i.e. FA corrected for measurement error, directional asymmetry and inter-individual variation (Palmer and Strobeck, 1992). For all genotypes, the interaction individual*side was significant, indicating that FA was larger than measurement error.
RNA-seq experiments and RT-qPCR validations
Wing imaginal discs from da-Gal4/UAS-CycGΔP and da-Gal4/+ third instar female larvae were dissected, and total RNAs were extracted as previously described except that 150 discs homogenized by pipetting were used for each extraction (Coléno-Costes et al., 2012). Three biological replicates were generated for each genotype. Library preparation and Illumina sequencing were performed at the Ecole Normale Supérieure Genomic Platform (Paris, France). Messenger (polyA+) RNAs were purified from 1 μg of total RNA using oligo(dT). Libraries were prepared using the strand specific RNA-Seq library preparation TruSeq Stranded mRNA kit (Illumina). Libraries were multiplexed by 6 on 2 flowcell lanes. A 50 bp single read sequencing was performed on a HiSeq 1500 device (Illumina). A mean of 38.1 ± 4.8 million reads was obtained for each of the 6 samples (Table S11). They were aligned with the Drosophila melanogaster genome (dm6, r6.07) using TopHat 2 (v2.0.10) (Kim et al., 2013). Unambiguously mapping reads (a mean of 24.9 ± 4.9 million reads) were then assigned to genes and exons described by the Ensembl BDGP5 v77 assembly, by using the “summarizeOverlaps” function from the “GenomicAlignments” package (v 1.2.2) in “Union”mode (Lawrence et al., 2013). Library size normalization and differential expression analysis were both performed with DESeq 2 (v 1.6.3) and genes with adjusted p-value below 0.05 were retained as differentially expressed (Love et al., 2014). Gene Ontology analysis was performed using DAVID (Huang et al., 2009; Huang et al., 2009).
For RT-qPCR validations, RNAs were treated with Turbo DNAse (Ambion), and cDNA were synthesized with SuperScript II Reverse transcriptase (Invitrogen) using random primers. RT-qPCR experiments were carried out in a CFX96 system (Bio-Rad) using SsoFast EvaGreen Supermix (Bio-Rad). Three biological replicates were performed for each genotype. Expression levels were quantified with the Pfaffl method (Bustin et al., 2009). The geometric mean of two reference genes, Lamin (Lam) and rasputin (rin), the expression of which did not vary when CycGΔP was expressed, was used for normalization (Vandesompele et al., 2002). Sequences of primer couples are listed in Table S10.
An interactome was built using Cytoscape (v 2.8.3) and the DroID plugin (v 1.5) to introduce protein-protein and transcription factor-gene interactions (Murali et al., 2011). The jActiveModules plugin (v 2.23) was used to find sub-networks of coderegulated genes in the interactome by using “overlap threshold” 0.8, “score adjusted for size”, and “regional scoring” (Ideker et al., 2002).
ChIP-seq experiments and ChIP-qPCR validations
Wing imaginal discs from +/UAS-Myc-CycGΔP;; +/da-Gal4 and +/da-Gal4 third instar female larvae were used for chromatin immunoprecipitation (ChIP).
For ChIP-seq experiments, 600 wing imaginal discs were dissected (taking one disc per larva) in Schneider medium and aliquoted per 50 in 1.5 mL microtubes on ice. The 12 microtubes were treated as described in (Coléno-Costes et al., 2012) with minor modifications. Discs were fixed at 22°C. 12 sonication cycles were performed (Diagenode Bioruptor sonifier; cycles of 30’’ ON, 30’’ OFF, high power). After centrifugation, the 12 supernatants were pooled, homogenized, and 2% were removed (Input). The remaining fragmented chromatin was redistributed into 12 tubes and each tube was adjusted to 1 mL with 140 mM NaCl, 10 mM Tris-HCl pH 8.0, 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 0.1% BSA, Roche complete EDTA-free protease inhibitor cocktail). For immunoprecipitation, 3 μg of anti-Myc antibody (Abcam 9132) were added per tube. The of water. Two biological replicates were performed.
Library preparation and Illumina sequencing were performed at the Ecole normale superieure Genomic Platform (Paris, France). Libraries were prepared using NEXTflex ChIP-Seq Kit (Bioo Scientific), using 38 ng of IP or Input DNA. Libraries were multiplexed by 10 on one flowcell run. A 75 bp single read sequencing was performed on a NextSeq 500 device (Illumina). Reads were filtered by the “fastq_quality_filter” command from the “fastx-Toolkit” package (http://hannonlab.cshl.edu/fastx_toolkit/), using a threshold of 90% bases with mapping quality ≥ 20. A mean of 55.6 ± 15.2 million reads was obtained for each of the 4 samples (Table S12). Reads that successfully passed the filtering step were aligned to the Drosophila genome (dm6, r6.07) using Bowtie 2 (http://bowtie-bio.sourceforge.net/bowtie2/) (v2.1.0) with default parameters (Langmead and Salzberg, 2012). Peaks were called by MACS2 (v2.1.0) by comparing each ChIP to its input library, with fragment size fixed at 110 bp and otherwise default parameters (Zhang et al., 2008). Peak reproducibility between the two biological replicates was then analysed with the IDR method (https://www.encodeproject.org/software/idr/) (Li et al., 2011). Briefly, an IDR score was assigned to each peak by the “batch-consistency-analysis” function, using the recommended parameters for MACS peaks (peak ranking based on p-value). Peaks below the 0.05 threshold were considered reproducible. The overlapping reproducible peaks from both replicates were fused using the BEDtools suite “merge” function (Quinlan and Hall, 2010), resulting in the final list of peaks kept for subsequent analysis. Cyclin G-bound genes were defined as genes from the genome annotation file (dm6, r6.07) which overlapped at least one of these Cyclin G peaks, as obtained by the BEDtools suite “intersect” function (Quinlan and Hall, 2010).
For ChIP-qPCR validations, ChIP were performed similarly with the anti-Myc antibody. Rabbit IgG (Diagenode) were used as a negative control (mock). qPCR experiments were carried out in a CFX96 system (Bio-Rad) using SsoFast EvaGreen Supermix (Bio-Rad). Sequences of primer couples are listed in Table S10. Data were normalized against Input chromatin.
Heatmaps and aggregation plots of Cyclin G signal over gene bodies and Transcription Start Sites (TSS) were generated using the ngsplot package. (https://github.com/shenlab-sinai/ngsplot) (Shen et al., 2014). Some genes with spurious signal (such as genes from the histone complex) were excluded from the analysis based on signal uniformity over the full length of the gene (cumulative derivative of Cyclin G signal over gene length = 0).
Data access
High-throughput sequencing data have been submitted to GEO. Accession numbers for RNA-seq data: GSE99462, GSM2644389, GSM2644390, GSM2644391, GSM2644392, GSM2644393, GSM2644394. Accession number for ChIP-seq data: GSE99461, GSM2644385, GSM2644386, GSM2644387, GSM2644388.
Genomic association
Genomic loci enriched for Polycomb (PC), Posterior Sex Comb (PSC), Polyhomeotic (PH), RNA Polymerase II (RNAPolII) and H3K27me3 in wild type imaginal discs of third instar larvae were retrieved from GEO (GSE42106) (Schaaf et al., 2013) (H3K27me3_WholeWingDisc GSM1032567, PcRJ_AnteriorWingDisc GSM1032571, PcRJ_PosteriorWingDisc GSM1032574, Ph_WholeWingDisc GSM1032576, PolII_WholeWingDisc GSM1032577, Psc_WholeWingDisc GSM1032578. Binding sites for PC in the whole wing disc were defined as the overlap between PC binding sites in the anterior and posterior wing disc compartment, as obtained by the BEDtools “intersect” function. For ASX and Calypso, the bed files were a kind gift from Dr. Jürg Müller (Scheuermann et al., 2010). The mappability file for dm6 genome with 25 nt reads (the smallest size in the compared data) was generated using the Peakseq code (http://archive.gersteinlab.org/proj/PeakSeq/Mappability_Map/Code/). The overall size of the mappable genome was used as the effective genome size for the GAT software (https://github.com/AndreasHeger/gat) to assess the significance of the overlap between peaks of Cyclin G and other factors (Heger et al., 2013). As GAT performs a two-tailed test, it displays low p-values both for significant overlap and exclusion (as between Cyclin G and H3K27me3).
Gene overlap significance assessment was made as follows: under the null hypothesis, genes that are enriched for ASX, Calypso, PC, PSC, PH, RNAPolII or H3K27me3 in wild type imaginal discs of third instar larvae should not exhibit any bias towards Cyclin G targets. Thus, the overlap between n enriched genes and K Cyclin G targets genes should be explained by random sampling without replacement of n genes within the total amount N of Drosophila melanogaster genes. The amount of overlap under the null hypothesis X follows a hypergeometric law: X~HY(K, N, n). The significance of the observed overlap k was computed as the probability of observing X higher or equal to k under the null hypothesis: P(X ≥ k).
Acknowledgments
We thank Dr. Emmanuèle Mouchel-Vielh and Dr. Jean-Michel Gibert for stimulating discussions and critical reading of the manuscript, the Bloomington Stock Center for fly strains, Dr. Jürg Müller for the alleles of Asx and caly and for the ASX and Calypso ChIP bed files, Dr. Maria Dominguez and Dr Sergio Juarez-Carreño for the dilp3-Gal4, R19B09-Gal4, npf-Gal4, pdf-Gal4, per-Gal4, ptth-Gal4 and tsh-Gal4 drivers. This work was funded by the Centre National de la Recherche Scientifique (CNRS), Université Pierre et Marie Curie (UPMC), Sorbonne Universités (grant SU-14-R-CDV-05-1 to FP), and Fondation ARC pour la Recherche sur le Cancer (grant PJA20131200314 to FP). The École Normale Supérieure genomic platform was supported by the France Génomique national infrastructure, funded as part of the “Investissements d’Avenir” program managed by the Agence Nationale de la Recherche (contract ANR-10-INBS-09). CAD was funded by a doctoral fellowship from the MESR (Ministère de l’Enseignement Supérieur et de la Recherche). JD was funded by a doctoral fellowship from the MESR and by the Fondation pour la Recherche médicale (FDT20160435164).