A prominent gene activation role for C-terminal binding protein in mediating PcG/trxG proteins through Hox gene regulation

C-terminal binding proteins (CtBPs) are transcriptional co-repressors that modulate the activity of a wide-spectrum of transcription factors. Although invertebrate genomes, such as Drosophila melanogaster, contain a single CtBP gene, vertebrates possess two homologues, Ctbp1 and Ctbp2, and these genes encode protein products with similar functionally conserved domain structures (Chinnadurai, 2002). Since its initial identification as a C-terminal binding partner of the viral protein E1A (Schaeper et al., 1995), the endogenous role of CtBP has been extensively studied in both invertebrate and vertebrate model systems. In Drosophila, CtBP plays essential roles in embryo development, acting as a co-repressor for many transcriptional repressors, such as Hairy (Poortinga et al., 1998), Krüppel, Knirps, Snail (Nibu et al., 1998a,b) and Brinker (Bi et al., 2018; Hasson et al., 2001; Yang et al., 2013). A large number of studies in mammalian cell lines have also confirmed the co-repressor role of CtBPs (Chinnadurai, 2002). CtBPs physically interact with ZEB (Postigo and Dean, 1999; Shi et al., 2003), histone deacetylases (HDACs; Subramanian and Chinnadurai, 2003) and other enzymes with histone modification activities, such as histone methyltransferases and demethylases (Shi et al., 2003), explaining the contribution of CtBP to transcriptional repression (Stankiewicz et al., 2014). Other studies have shown that human HPC2 (also known as CBX4), a chromo domain-containing protein that specifically recognizes trimethylated histone H3K27 (H3K27me3, a hallmark of repression), interacts with and sumoylates CtBP2 (Kagey et al., 2005, 2003; Sewalt et al., 1999). Although demonstration of a direct functional link is lacking in these studies, cooperation with Polycomb group (PcG) proteins might be an additional means through which CtBP exerts its repressive roles. Indeed, CtBP appears to be required for Pho (also known as YY1)-dependent recruitment of PcG proteins (Basu and Atchison, 2010; Srinivasan and Atchison, 2004), and mouse CtBP2 is necessary for PRC2-mediated silencing in embryonic stem cells (Kim et al., 2015).

CtBPs have also been speculated to play a co-activation role in transcription, but only sporadic studies exist to support this. For example, CtBP is directly required for Wingless (Wg; a fly Wnt ligand) target expression in flies, depending on its state of homo-dimerization (Bhambhani et al., 2011; Fang et al., 2006). Recent studies have also shown that CtBPs might activate genes that promote proliferation, epithelial-to-mesenchymal transition and cancer stem cell self-renewal activity (Dcona et al., 2017) and promote NeuroD1 expression through association with the histone demethylase LSD1 (KDM1A), which removes H3K9me2 to facilitate the H3K9ac modification (Ray et al., 2014). These results indicate that CtBP may be more than just a dedicated co-repressor.

In this paper, we directly tested the role of Drosophila CtBP in the epigenetic regulation of gene expression. We show that reduction of CtBP antagonizes the leg transformation caused by mutations in PcG genes and enhances trxG-dependent phenotypes. We demonstrate using RNA-seq that CtBP is required for the derepression of more than 100 PcG target genes highly enriched by homeobox transcription factors. Furthermore, we find that CtBP interacts with many trxG proteins and potentially aids in their recruitment to Hox gene loci when derepressed. These results unveil a broader activation role for CtBP function. As CtBP interacts with both PcG and trxG proteins and functionally contributes to both transcriptional activation and repression, we propose that CtBP may be important for maintaining the balance between repressive and active epigenetic players, possibly by providing a platform for a dynamic transcriptional-epigenetic switch in gene regulation.

To directly test the role of CtBP in the context of epigenetics, we examined the potential dose-dependent modifier function of CtBP, as previously reported (Kennison and Tamkun, 1988), with the rationale that the heterozygous CtBP phenotype in later developmental stages might be masked by its enormous maternal contribution and its essential role in the regulation of pair-rule gene expression (Nibu et al., 1998b; Poortinga et al., 1998). Heterozygous Pc mutant males often display extra sex combs in the second and/or third legs, commonly referred to as sex comb or leg transformations, as do mutants of other PcG genes, such as Sex comb on midleg (Scm) (Fig. 1A). In light of the several studies suggesting CtBP as a helper in PcG silencing (Basu and Atchison, 2010; Sewalt et al., 1999; Srinivasan and Atchison, 2004), we hypothesized that loss of CtBP would enhance the PcG sex comb transformations. Surprisingly, we found that this was not the case.

We phenotypically compared the heterozygous Pc (Pc³/+ and Pc¹/+) mutant flies with flies with heteroallelic combinations of Pc and one of three CtBP alleles (CtBP^32P, CtBP^De10 or CtBP^Df) by a semi-quantitative measuring of sex comb transformations, such that they were categorized as wild type (i.e. no extra sex comb), moderate (extras only on the second legs) and strong (extras appeared on both second and third legs) transformations (Fig. 1A). Heterozygotic CtBP mutant flies appeared to be completely normal, fertile and with absolutely no leg transformation (Fig. 1C,D). However, we found that heteroallelic combinations of Pc and CtBP always caused less strong leg transformations than the heterozygotes of two Pc alleles, Pc³ and Pc¹ (a less strong Pc allele); the same was observed for Scm^D1, which produced much weaker leg transformations than Pc (Fig. 1C). A probable more precise evaluation of the severity of leg transformation by extra sex comb teeth counting resulted in quite similar outcomes (Fig. 1D). The alleviation of the sex comb transformation of Pc and other PcG genes by reduction of CtBP indicates that CtBP may specifically antagonize PcG gene function. The stronger CtBP allele CtBP^De10 had greater alleviation effects suggests that CtBP may function in a gene dosage-dependent manner (Fig. 1C-E).

Given these results, we next asked whether and how CtBP genetically interacts with trxG genes. The prototype of trxG genes, trithorax (trx), has been extensively studied at the genetic level (Breen, 1999; Ingham, 1983), and loss of trx results in a variety of developmental defects including homeotic transformations of thoracic and abdominal segments (Breen, 1999; Karch et al., 1985; Singh and Mishra, 2014). Developmental defects have also been studied for other trxG genes, such as ash1, ash2 and brm (Gindhart and Kaufman, 1995; Gutiérrez et al., 2003; Shearn, 1989). We used several previously published mutant alleles of trxG, including trx (trx¹ and trx^E2), ash1 (ash1^B1) and brm (brm²), to explore the genetic interaction between CtBP and trxG (Gindhart and Kaufman, 1995). Heterozygosity of these alleles was very rarely sufficient for a discernable haltere-to-wing transformation; however, they produced abdominal tergite boundary defects indicative for enhancement of Pc function (Du et al., 2016) at a considerable rate (Fig. 1B). We found that heteroallelic combinations of these mutations with CtBP alleles remarkably enhanced such tergite defects, even though none of them was seen in heterozygous CtBP mutants (Fig. 1F). Together, these data led us to the conclusion that CtBP genetically antagonizes PcG and enhances trxG gene functions in a gene-specific and dose-dependent manner.

Pc is known to have heterozygous wing phenotypes (Castelli-Gair and García-Bellido, 1990) and, as such, both Pc¹ and Pc³ heterozygotes produce adult wing defects, with the strongest phenotypes displaying severe curvature, indicating a partial transformation to metathorax, and moderate ones bearing clear notches in the posterior wing edges (Fig. 2A-C). Although heterozygotic CtBP mutants exhibited completely normal wings, heteroallelic combination with a strong CtBP allele, CtBP^De10, significantly suppressed the Pc wing phenotypes (Fig. 2D), similar to what was observed in leg transformations (Fig. 1C,D).

As previously published (Castelli-Gair and García-Bellido, 1990; Dupont et al., 2015), Pc³ is heterozygously sufficient for ectopic posterior wing pouch expression of Ultrabithorax (Ubx), presumably being derepressed by loss of Pc (Fig. 2E,G). As expected, heterozygotic CtBP^De10 caused no ectopic Ubx in wing discs (Fig. 2F), but it did markedly diminish the Ubx expression when heteroallelically combined with Pc³ (Fig. 2G,H). Semi-quantifications of Ubx intensity indicated that those changes were significant (Fig. 2Y). We observed a similar pattern using the antibody FP6.87, which cross-recognized both Abdominal-A (Abd-A) and Ubx (Fig. 2I-L,Z), although we were unable to determine whether abd-A, a Hox gene member proximal to Ubx in the Bithorax complex (BX-C), would also be ectopically expressed in Pc³ heterozygotes. The other Ubx-neighboring, but more distant, Hox member Abdominal-B (Abd-B) remained silent (Fig. 2M-P), whereas Antennapedia (Antp), known to have a strong wing disc expression, was not affected by the reduction of either Pc or CtBP or both (Fig. 2Q-T). Pc protein levels were slightly lower in Pc³ heterozygous wing discs and were not affected by the reduction of CtBP (Fig. 2U-X), which we confirmed using RT-qPCR (Fig. S1A). To provide a possible explanation of the notch phenotypes seen in adult wings (Fig. 2B), we stained wing discs for Distal-less (Dll), a Wg target important for wing development. We found an obvious correlation between the ectopic Ubx expression and the reduction of Dll in the posterior wing pouch (Fig. S2). We therefore demonstrated that CtBP is required for the derepression of Ubx (and possibly abd-A) caused by loss of Pc, which in turn negatively influences the Wg signaling pathway and results in the defective wings. These data indicate that CtBP may antagonize Pc function at the level of gene expression.

To further explore the role of CtBP on PcG target genes and to extend our genetic findings to a genome-wide level, we used a cell model of PcG target derepression as previously reported (Fang et al., 2009; Schwartz et al., 2006). Briefly, we used combinational RNAi for PcG and CtBP in Drosophila Kc167 (Kc) cells and observed the differential expression of PcG target genes using RNA-seq. As expected, double RNAi of polyhomeotic proximal (ph-p) and Enhancer of zeste [E(z)], both of which encode core components of polycomb repressive complexes (PRCs), caused expression and/or derepression of many genes, or PcG repressed genes, of which we selected 381 genes for further analyses based on their expression changes being significantly increased by more than twofold (Fig. S3). Co-knockdown of CtBP with ph-p and E(z) caused different effects on gene expression and clustered these PcG repressed genes into three categories (Fig. 3A; Table S2; see also Materials and Methods for detail). Strikingly, we found that CtBP is required for the expression of 153 (∼40%) of the PcG repressed genes and we grouped them as CtBP co-activated genes (Fig. 3A). For genes for which expression levels are not significantly affected (143, ∼38%) or enhanced (85, ∼22%) by co-depletion of CtBP, we created CtBP unaffected genes and CtBP co-repressed genes, respectively (Fig. 3A). We next examined these three groups of genes using DAVID Functional Category analysis (Huang et al., 2009) and found that the CtBP co-activated gene group was highly enriched with transcription factors and developmental proteins and, most prominently, the homeobox transcription factors (Fig. 3B). The group of genes not regulated by CtBP was enriched by transcription factors as well but with much less confidence (Fig. 3B). In sharp contrast, CtBP co-repressed genes were bereft of any significant functional enrichment. Consistent with the DAVID analysis, Gene Ontology (GO) detailed CtBP co-activated genes with respect to both molecular function and biological process, and highlighted the gene function enrichment of transcription and development (Fig. 3C,D). These results suggest that the prominent function of CtBP in the context of PcG relief may be gene activation.

An important question to answer is whether CtBP might act the same fashion for direct PcG targets. It is almost certain that PcG may not repress all of the 381 genes directly and it will be a difficult task to define or identify the bona fide Pc targets by simple rules. However, we suggest that the combination of the genome-wide profile of Pc and the derepression could be a good indicator. Therefore, we assumed that the intersection of 73 genes from the 381 used in Fig. 3 and the 333 with strong PcG binding sites (Schwartz et al., 2006) were most likely to be direct PcG targets (Fig. 4A; Table S3). Interestingly, cluster analysis categorized most of these PcG targets (65, ∼89%) as CtBP-associated PcG targets, whereas only a handful of them (8, ∼11%) were CtBP-unassociated PcG targets (Fig. 4B; Table S3). Importantly, DAVID Functional Category analysis generated a similar functional enrichment chart of the 65 CtBP-associated PcG targets with even higher confidences of homeobox genes (Fig. 4C), and about two thirds (49, ∼67%) of the 73 PcG targets were DNA-binding proteins (Fig. 4C). Therefore, it is most likely that CtBP is highly and selectively required for the activation of PcG targets, in particular, the homeobox transcription factors, including Hox genes. The fact that CtBP knockdown affects the derepression of PcG genes in both directions suggested that the CtBP effects are unlikely to be caused by the fluctuations of RNAi efficiency. Nevertheless, we have confirmed the knockdown levels for every gene subjected to RNAi in each experiment by RT-qPCR and/or western blots (Fig. S1B-F). In addition, we used RT-qPCR to verify the effect of CtBP on the derepression of four Hox genes, Antp, Ubx, abd-A and Abd-B (Fig. 4D-G). Importantly, knockdown of Pc alone was sufficient to derepress these Hox genes, and we observed similar subdued derepression effects with the addition of CtBP RNAi (Fig. 4H-K). At the same time, knockdown of CtBP alone exerted minimal effects on expression of these genes (Fig. 4D-K). These results indicate that CtBP plays a modulatory activation role in the dynamic regulatory processes of PcG target genes, such as derepression, not through modulating any specific PRC component. We thus unveil a previously unsuspected prominent transcriptional activation function for CtBP in the context of epigenetics.

To examine whether CtBP activates PcG target genes directly or indirectly, we performed chromatin immunoprecipitation (ChIP)-qPCR assays at several Hox gene regulatory sites (Beisel et al., 2007; Müller and Bienz, 1991; Orlando et al., 1998; Papp and Müller, 2006; Schwartz et al., 2006; Simon et al., 1993; Zink and Paro, 1995), including Polycomb response elements (PREs) and/or proximal promoter regions (promoter) in both the Antennapedia gene complex (ANT-C) and BX-C, and an intergenic site for the negative binding of Pc (Fig. 5A). CtBP RNAi alone caused dramatic decreases in CtBP ChIP signals at Antp and Ubx sites, indicating physically direct occupation of CtBP (Fig. 5B; Table S4-1), and it is worth noting that PcG RNAi caused little change of CtBP binding at the three PREs but lowered binding at the proximal promoter sites (Fig. 5B; Table S4-1). These results suggest that CtBP might preferentially stay at the PREs rather than the proximal promoters upon the relief of PcG repression. Knockdown of CtBP alone resulted in a reduced, but not significantly, Pc occupation along with the related H3K27me3 histone mark (Fig. 5C,D; Tables S4-2,S4-3), which could be consistent with the published studies (Basu and Atchison, 2010; Srinivasan and Atchison, 2004).

For PcG RNAi cells, the impact of CtBP was more dramatic. As expected, PcG RNAi resulted in significantly lower Pc binding and H3K27me3 levels, with higher levels of CBP and UTX bindings and H3K27ac modifications (Fig. 5), consistent with the derepression of Ubx and Antp (Fig. 4D-K). However, when PcG RNAi was combined with CtBP knockdown, these repression and activation marks were abolished (Fig. 5C-G; Table S4), indicating that reduction of CtBP completely prevented the repression/activation mark switch and provides a molecular explanation that CtBP is required for the derepression of PcG targets, such as Ubx and Antp. We were intrigued with two seemingly puzzling observations that when both E(z) and ph-p, which encode the catalytic subunit for H3K27me3 and a Pc-interacting key component of PRC1, respectively, were knocked down, co-knockdown of CtBP maintained high levels of Pc binding and H3K27me3 levels (Fig. 5C,D; Tables S4-2 and S4-3). One possible explanation could be that CtBP helped the recruitment of UTX, the major H3K27me3 demethylase, to the derepressed Hox sites. Therefore, loss of CtBP might cause the stabilization of the H3K27me3 mark, which in tune recruits Pc through the Chromo domain. In other words, CtBP might facilitate the removal of the repressive proteins or histone marks upon the derepression of the Hox genes. This might be mechanistically different from the result that loss of CtBP caused moderate reductions of Pc and a subtle decrease of H3K27me3 level without significant difference, when the Hox genes were barely expressed and the UTX levels were kept low (Fig. 5C,D; Tables S4-2 and S4-3).

To further explore this matter, we carried out co-immunoprecipitation (co-IP) experiments to show that CtBP indeed physically interacted with both UTX and CBP (Fig. 5H; Fig. S4). Reciprocal co-IPs confirmed these protein-protein interactions (Fig. 5I,J; Fig. S4). We also used ethidium bromide (EB) treatment to eliminate the probable DNA mediated interactions (Fig. 5H-J). These results indicate that CtBP interacts with many proteins to affect gene activation and support a model that CtBP is required for UTX and CBP chromatin recruitment to promote the molecular switch between H3K27me3 and H3K27ac.

Taken together, these data indicate that CtBP might play a role in the orchestration of epigenetic regulators and histone modifications in the course of dynamic switching between repression and activation of Hox genes.

We next asked whether CtBP had direct impact on the transcription initiation and elongation in the context of Hox gene regulation by PcG/trxG. For this purpose, we tested the chromatin bindings of RNA Polymerase II [with αC-terminal domain (CTD)], as well as its phosphorylated forms at Ser2 (αSer2-P), which are required for gene initiation and elongation, respectively (Komarnitsky et al., 2000; Samkurashvili and Luse, 1998). The ChIP results by αCTD and αSer2-P showed different influences of the reduction of PcG and CtBP functions. PcG knockdown caused remarkable increase of αCTD signals in the proximal promoter regions (Fig. 6A; Table S4-7) and αSer2-P signals in both PREs and promoters of Antp and bxd (Fig. 6B; Table S4-8), indicating that the relief of PcG repression automatically promoted the initiation and elongation of transcription and were consistent with the prominent derepression of the PcG targets (Figs 3 and 4). Apparently, co-knockdown of CtBP diminished such increases, suggesting a requirement of CtBP in both transcription initiation and elongation (Fig. 6A,B; Table S4). Furthermore, the massive elevations of αSer2-P signals in PREs in derepressed Hox loci emphasized the significant regulatory role of PREs, and were consistent with our results that CtBP might preferentially operate on PREs (Figs 5B and 6A; Table S4).

Indeed, we found physical interactions between CtBP and the phosphorylated form of RNA polymerase II (Ser2-P) (Fig. 6C). In addition, CtBP interacted with Fs(1)h (Fig. 6C), the fly homologue of BRD4, a candidate kinase of RNA Pol II (Steffen and Ringrose, 2014). By confirming the interactions between RNA Pol II Ser2-P, Fs(1)h, CBP and UTX (Fig. 6D-F, Fig. S4), our results suggested that CtBP may be directly required to achieve active transcription of Hox genes by interacting with epigenetic coactivators (i.e. CBP and UTX) and the core transcriptional machinery including RNA Pol II Ser2-P and Fs(1)h.

For over two decades, CtBP has been extensively studied in both invertebrate and mammalian model systems. CtBP physically interacts with a large number of sequence-specific DNA binding proteins and mediates their transcriptional activities. Although the prevailing model is that CtBP functions as a co-repressor for these DNA binding factors, its major biological function in vivo remains controversial, particularly during the course of animal development. In Drosophila, genetic studies have shown that CtBP may both enhance and repress gene expression in a transcription factor-dependent manner (Nibu et al., 1998b; Phippen et al., 2000; Poortinga et al., 1998). In mice, the co-repressor role of CtBP does not sufficiently explain the deficits found in both CtBP1 and CtBP2 mutant mice. Rather, the study suggests that CtBPs may contribute equally to gene repression and activation (Hildebrand and Soriano, 2002). Thus, the question remains: which one is more important for the in vivo function of CtBP, repression or activation?

Answering this question is not an easy task owing to the apparent pleiotropic nature of CtBP. In Drosophila, loss of CtBP severely disrupts the embryonic body plan and inhibits early embryonic regulators, including many of the pair-rule genes and potentially some gap genes (Nibu et al., 1998a; Poortinga et al., 1998), preventing further investigation on its role in later developmental stages. In particular, it is quite difficult to dissociate the direct action of CtBP on Hox from its role on pair-rule genes, which in turn affect Hox expression. This scenario might help to explain in part why genetic studies in mouse are not particularly informative (Hildebrand and Soriano, 2002).

To circumvent this difficulty, we examined the role of CtBP in the epigenetic context. We found that CtBP suppresses PcG homeotic transformations and enhances those of trxG (Fig. 1). Therefore, CtBP may function as a dose-dependent modifier of the major epigenetic regulators, opposing its co-repressor role in fly development. Pc transformations have at least partially known gene bases, such as those we demonstrated with ectopic expression of Ubx in the wing discs (Fig. 2) and Sex combs reduced in leg discs shown by other studies (Pattatucci and Kaufman, 1991). Therefore, the fact that CtBP suppresses Pc transformations may suggest or reflect an altered embryonic Hox pattern that was too delicate to be observed. Importantly, the heteroallelic combination strategy that we employed in this study allowed us to dissociate the role of CtBP in earlier embryonic development from later stages and led to the finding that activation may be a prominent function for CtBP during the course of development. Although it is hard to tell whether CtBP acts directly to modulate PcG and trxG functions in these in vivo experiments, it is most likely that CtBP may function in a parallel temporal fashion with Pc, as neither altered expression of pair-rule genes, such as even skipped and fushi tarazu, nor any defect in adult flies was discernable for heterozygous CtBP alleles.

In agreement with our in vivo studies, we have demonstrated, using RNA-seq analyses in embryonic-derived Kc cells, that CtBP plays an essential activation role in the regulation of most direct PcG targets, having the highest enrichment of homeobox transcription factors (Figs 3 and 4). In sharp contrast, CtBP, in conjunction with PcG, co-represses a relatively smaller proportion of genes with diverse functions, among which only a handful may potentially direct PcG targets. These findings might suggest that CtBP has a prominent activation role that is highly selective for developmental transcription factors, including Hox genes, whereas the repressive role of CtBP is unspecific and possibly irrelevant with PcG repression. Using ChIP and co-IP assays, we further provide evidence that CtBP physically interacts with CBP and UTX, helping to recruit these histone modifiers to several well-characterized regulatory sites in Hox loci. Accordingly, CtBP is required to coordinate the switch from a repressive H3K27me3 form to an active H3K27ac mark (Fig. 5). For the sake of inspiration, we would imagine a wider function of CtBP in rewiring epigenetic networks composed by various kinds of histone modifications and other marks. For example, it has been proposed that H3K27ac signal coincides with H3K4me3, which constitutes a key part of the molecular mechanism of antagonizing Pc silencing (Tie et al., 2009). Indeed, our results might allow us to favor a model in which CtBP constitutes part of a large complex including many epigenetic writers and erasers, such as CBP and UTX, closely working with the core transcriptional machinery centered by RNA Pol II (Figs 5 and 6). Therefore, CtBP may play a profound role in PcG/trxG-mediated Hox gene regulation.

Although many biochemical studies have indicated that CtBPs are mainly associated with repressive histone modifiers (Shi et al., 2003), it is noteworthy that these studies were mostly performed in cell lines with a static gene expression status and so may not properly reflect the dynamic changes of transcriptional programs in vivo. Furthermore, the previously identified CtBP-interacting proteins are not dedicated transcriptional repressors/co-repressors. As such, there are examples illustrating the requirement of LSD1 and some of the HDACs for transactivation (Ray et al., 2014; Wirén et al., 2005). However, CtBP has been reported by others to interact with proteins for gene activation. For example, it has been documented in detail that CtBP1 directly binds to the bromodomain of p300 (Kim et al., 2005), coincident with our findings that CtBP interacts with CBP in flies (Fig. 5H,J).

Together, our findings highlight a direct and prominent activation role of CtBP in early fly development stages, probably no later than the determination of segment identities when Pc and Hox genes are acting. Mechanistically, CtBP could play its coactivator role in the dynamic switch or establishment of proper chromatin structures for prolonged gene activation by recruiting histone modifiers such as CBP and UTX. Alternatively, CtBP might play roles in both activation and repression, functioning in the balance between PcG and trxG actions (Piunti and Shilatifard, 2016). For example, it could be that CtBP activates in PREs or enhancers but represses at promoters, as shown by the unchanged CtBP occupation upon PcG depletion (Fig. 5B). Significantly, unlike previous findings that CtBP only activates genes in exceptional cases, we have provided evidence that CtBP directly activates gene expression in the major developmental processes, such as the precise patterning of Hox genes, indicating that activation is probably an important role for CtBP in the context of development. We have thus uncovered a gene co-activation role for CtBP in vivo that might have been previously masked due to its impact on the earlier developmental regulators.

It will be intriguing to know whether CtBP activates PcG targets beyond Hox genes or an opposite mechanism exists, as in mouse ESC differentiation, in which CtBP2 couples with NuRD-mediated deacetylation (18) for our identified CtBP-unassociated PcG targets. In mouse, the loss of CtBP2 causes incomplete ESC differentiation, a process known to be dependent upon gene activation (Betschinger et al., 2013; Tarleton and Lemischka, 2010). Our finding could provide an alternative interpretation, if the prominent gene activation role of CtBP were conserved in mouse. Given the complexity of the role of CtBP in cancers (Ding et al., 2020; Wu et al., 2021), our findings might provide additional data that can be used towards the development of novel strategies based on CtBP as a therapeutic target.

All fly lines were cultured on standard medium at 25°C. Three CtBP alleles were used in this study: the strong CtBP^87De−10 allele and the less strong CtBP⁰³⁴⁶³ allele (also known as CtBP^P1590), which we referred to as CtBP^De10 and CtBP^32P, have been previously described (Poortinga et al., 1998), and a CtBP deficiency [Df (3R) Exel8157] line (CtBP^Df) from the Exelixis collection (Parks et al., 2004), which has nearly the entire coding region of CtBP removed. PcG/trxG alleles, including Pc¹, Pc³, Scm^D1, trx¹, trx^E2, brm² and ash1^B1, have been previously described (Duncan, 1982; Gindhart and Kaufman, 1995; Kennison and Tamkun, 1988; Lewis, 1978): Pc³ is an extreme antimorphic allele, whereas Pc¹ is a less extreme amorphic allele of Pc. All stocks were from the Bloomington Drosophila Stock Center and we used w¹¹¹⁸ as the wild-type control allele. All the crosses were carried out at 25°C.

To test the genetic interactions between CtBP and the PcG/trxG genes, male CtBP alleles over TM6B balancer or w¹¹¹⁸ were crossed to virgin PcG/trxG alleles also over a TM6B balancer to generate heterozygotes of PcG/trxG mutants or heteroalleotic combinations of CtBP alleles over PcG/trxG ones. For semi-quantitative assay of sex comb transformations, adult male progenies were assessed for extra sex combs in the second and third legs. Posterior wing phenotypes and abdominal segment defects were scored as previously described (Castelli-Gair and García-Bellido, 1990; Du et al., 2016). For wing phenotype observation, adult wings were dissected in isopropanol and mounted to onto glass slides with Euparal mounting media (BioQuip). All images were acquired using a MZ16F stereomicroscope equipped with a QImaging digital camera (Leica).

Rabbit polyclonal antibodies against CtBP (Fang et al., 2006), UTX (Zhang et al., 2013), Pc, CBP and Fs(1)h (Tie et al., 2016) have been previously described. Mouse monoclonal anti-Ubx (FP3.38) was kindly supplied by Dr White (White and Wilcox, 1984). Mouse monoclonal anti-Ubx/Abd-A (FP6.87), anti-Abd-B (1A2E9) and anti-Antp (8C11) were obtained from the Developmental Studies Hybridoma Bank. Rabbit polyclonal anti-Dll was kindly provided by Sean Carroll (Panganiban et al., 1995). Rabbit anti-H3K27ac (ab4729), mouse anti-H3K27me3 (ab6002), mouse anti-RNA Pol II (8WG16, ab817) and rabbit anti-RNA Pol II Ser2-P (ab5095) were from Abcam. Note that the monoclonal antibody 8WG16 (αCTD) primarily recognizes the non-phosphorylated CTD repeat within the Rpb1 subunit, which may be used as an indication of the initiation of gene transcription. The rabbit anti-RNA Pol II Ser2-P (αSer2-P, ab5095, Abcam) specifically recognizes the Ser2 phosphorylation and reflects the elongation process of transcription. Rabbit anti-IgG, mouse anti-IgG and mouse anti-tubulin were purchased from Sigma-Aldrich.

Immunostaining of wing imaginal discs were carried out as previously described (Fang et al., 2006) with anti-Ubx (1:20), anti-Ubx/Abd-A (1:100), anti-Abd-B (1:100), anti-Antp (1:20), anti-Pc (1:200) and anti-Dll (1:100). All images were collected using an Axiophot (Zeiss) coupled to a Carl Zeiss LSM 710 confocal apparatus (Zeiss). For semi-quantification of immunostaining in the wing discs, single color channels were processed into inversed and grayscale mode in Photoshop (Adobe, v11.0) and the mean gray values were measured for area with signal and subtracted by one without signal (in the anterior compartment) using ImageJ (National Institutes of Health, v1.46), and served as index for quantification.

Kc cells were cultured and RNAi treated as previously described (Fang et al., 2006). For western blot analyses, anti-CtBP (1:1000), anti-Pc (1:1000), anti-UTX (1:1000), anti-CBP (1:1000), αSer2-P (1:2000), anti-Fs(1)h (1:1000) and anti-tubulin (1:10,000) were used. Co-IP assay was performed according to standard procedures with anti-CtBP (10 μl), anti-CBP (10 μl), anti-UTX (10 μl), anti-Fs(1)h (10 μl) or normal IgG (5 μg). An initial protein-protein crosslinking step, incubating cells in a 5 mM dimethyl 3,30-dithio-bis(propionimidate) dihydrochloride (DTBP, Sigma-Aldrich) solution for 30-60 min on ice, was included in the co-IP experiments between CtBP and UTX, CBP, as well as Fs(1)h. ChIP assays were performed essentially as previously described (Fang et al., 2006). Approximately three million cells and anti-CtBP (10 μl), anti-Pc (2 μl), anti-CBP (2 μl), anti-UTX (2 μl), anti-H3K27me3 (2 μl), anti-H3K27ac (2 μl), anti-RNA Pol II (αCTD, 2 μl), anti- RNA Pol II Ser2-P (αSer2-P, 2 μl) antibodies or normal IgG (5 μg) were used per ChIP assay. All primers for dsRNA synthesis and PCR are listed in Table S1.

Total RNA was extracted from cells or wing imaginal discs (∼20 discs and ∼1 μg total RNA yielding) using Trizol total RNA isolation reagent (Invitrogen). cDNA was synthesized with oligo-(dT)₁₈ using the Superscript III First Strand Synthesis System (Invitrogen) according to the manufacturer's instructions. qPCR amplification was then performed on iCycler iQ™ Real time PCR Detection System (Bio-Rad) with FastStart Universal SYBR green Master Mix (Roche Applied Science). The Ct values were determined by the absolute quantification method using standard curves. For all RT real-time quantitative PCR (RT-qPCR) experiments, the expression of genes was normalized to β-Tubulin at 56D. At least three independent experiments were performed for each assay. RT-qPCR primers for the respective genes are listed in Table S1.

Total RNA was extracted from RNAi-treated Kc cells using Trizol total RNA isolation reagent (Invitrogen). RNA-seq data were obtained by the service provided by BGI from their deep sequencing platform build on the Illumina HiSeqTM 2000. RNA-seq data have been deposited in Gene Expression Omnibus under accession number GSE88807. The clean reads were then subject to further analysis based on the RNA-seq pipeline, using the open source Bowtie, TopHat, CuffLinks and CummeRbund software suite, built in our lab (Trapnell et al., 2012). Briefly, reads were mapped to the Drosophila melanogaster genome (BDGP5.25, the fruit fly iGenome package, Ensemble build) using Bowtie (v2.2.4) (Langmead and Salzberg, 2012) and the splice junctions were determined from the mapped reads using TopHat (v2.0.13) (Kim et al., 2013). Cufflink (v 2.2.1) (Trapnell et al., 2012) was used for calculating expression levels in fragments per kilobase of exon per million fragments mapped (FPKM), data visualization and measuring expression levels. The Cuffdiff program within Cufflinks was used to test for statistically significant differences in transcript expression between different RNAi-treated cells. We restricted our analysis to 11,282 protein-coding genes across all nine samples. The downstream statistical analyses and plots were performed in R (v3.1.1; http://www.r-project.org/). We then selected 381 PcG target genes for further analysis, using the rule that expressional fold increase ≥2 and P≤0.05 (unpaired two-tailed Student's t-test) (Fig. S3; Table S2).

For bioinformatics analyses, heatmaps and cluster analyses were generated in R with the function pheatmap of the g plots package (Liquet et al., 2012). Supervised hierarchical clustering was performed according to Euclidean distance using Ward's Method (Murtagh and Legendre, 2011). GO analysis was conducted using DAVID (Huang et al., 2009), Cytoscape (v3.4.0) and a BiNGO plugin (v3.0.3) (Maere et al., 2005; Saito et al., 2012) with chosen lists of genes based on the clustering describe above. We used the hyper-geometric test and the Benjamini–Hochberg FDR correction method (Benjamini Y, 1995), and a 0.001 significance level due to the high proportion of associated GO terms.

For statistical analysis, experiments were performed at least three times. Data was presented as the mean± s.e.m. Statistical analyses, except those in bioinformatics, were performed using GraphPad Prism in which proper statistical methods and the α value were applied and indicated in each figure. The statistical analysis for ChIP-qPCR in Figs 5 and 6 was performed using one-way ANOVA followed by Tukey's multiple comparison tests (Table S4).

The authors thank Drs Ken Cadigan, Xiang-Dong Liu and Yu-Feng Pan for critical reading of the paper. We also thank Dr Ken Cadigan and the Bloomington Drosophila Stock Center, the Vienna Drosophila RNAi Center and the Developmental Studies Hybridoma Bank for flies, cell lines and antibodies.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200153.