Collaborative repressive action of the antagonistic ETS transcription factors Pointed and Yan fine-tunes gene expression to confer robustness in Drosophila

Genetic and epigenetic mechanisms together produce the spatiotemporal gene expression dynamics that drive accurate and robust developmental transitions. At the genetic level, combinatorial codes of competing and collaborating transcriptional activators and repressors are recruited to individual cis-regulatory enhancers to determine precise gene expression outputs (Ma, 2005; Bauer et al., 2010). Analogously, at the epigenetic level, activating and repressive marks facilitate open or closed chromatin states that respectively promote or preclude expression, and more nuanced regulation can be achieved by the simultaneous presence of activating and repressive marks (Reynolds et al., 2013; Lagha et al., 2012). For example, at many developmentally important genes, specific combinations of inherently conflicting histone modifications permit RNA polymerase (pol) II to initiate transcription but then stall, keeping gene expression off yet poised for rapid activation (Schwartz et al., 2010; Gaertner et al., 2012). Although it is known that chromatin looping can physically coordinate the transcriptional complexes assembled at enhancers across a locus with the promoter-proximal complexes that orchestrate RNA pol II pause and release, the mechanisms by which these two layers of regulation are actually integrated to fine-tune gene expression dynamics during development are just beginning to be elucidated (reviewed by Gaertner and Zeitlinger, 2014; Liu et al., 2015; Meng and Bartholomew, 2017).

The Drosophila ETS transcriptional repressor Yan [also known as Anterior open (Aop); Nüsslein-Volhard et al., 1984; Rogge et al., 1995] and activator Pointed (Pnt) provide a useful model system for exploring how activator and repressor inputs are balanced to control developmental gene expression. Genetic and biochemical analysis of several enhancer elements, including the muscle heart enhancer (MHE) that drives the segmental pattern of even skipped (eve) expression in the cardiogenic mesoderm, has showcased competition between Yan and Pnt for access to consensus ETS motifs as a mechanism for directing rapid off-on gene expression transitions in response to upstream signals. Thus, prior to signaling, Yan outcompetes Pnt to repress target gene expression, thereby stabilizing the uncommitted precursor state. Following pathway activation, Yan is targeted for rapid degradation, allowing Pnt access to sites previously occupied by Yan. This turns on formerly repressed gene expression programs to promote a differentiative transition (Brunner et al., 1994; Klaes et al., 1994; O'Neill et al., 1994; Rebay and Rubin, 1995; Hsu and Schulz, 2000).

The results of several recent studies motivated us to reconsider the universality of this regulatory mechanism with respect to all Yan target genes and to ask whether more complicated Yan-Pnt interactions might also contribute to regulation of well-studied targets such as eve. First, chromatin immunoprecipitation-sequencing (ChIP-seq) studies have shown that Yan occupies chromatin in broad stretches of clustered peaks, binding preferentially to enhancers associated with developmentally important genes and signaling pathway effectors (Webber et al., 2013a). Simple binary off-on regulation of all of these putative targets seems unlikely. Second, a comparative survey of Yan and Pnt protein expression throughout development revealed extensive co-expression, particularly in tissues in which receptor tyrosine kinase (RTK) signaling levels are presumed low (Boisclair Lachance et al., 2014). This raises the possibility of more complicated interactions than might be needed if their expression were always mutually exclusive as it is in the embryonic midline. Indeed, two recent studies focused on eve highlight the importance of properly balanced Yan and Pnt repressive and activating inputs at the MHE before, during and after a cell fate transition, and emphasize the use of long-range interactions between the MHE and other Yan-bound elements as a mechanism for ensuring robust regulation (Webber et al., 2013b; Boisclair Lachance et al., 2018). How Yan/Pnt-mediated regulatory mechanisms might be coordinated with epigenetic mechanisms that influence gene expression is not known.

In this study, we report the discovery of a completely unexpected role for Pnt in recruiting or stabilizing Yan occupancy and repression at regulatory elements across the genome. In wild-type embryos, we find that Yan and Pnt have virtually identical genome-wide occupancy patterns and that the two actually co-occupy individual enhancers. Whereas the majority of Pnt occupancy is Yan independent, the majority of Yan occupancy is Pnt dependent, a finding that positions Pnt as an anchor with respect to establishing Yan occupancy and repression. Further challenging the model of exclusive Yan-Pnt regulatory antagonism, gene expression analyses predict that in addition to the classic opposing Pnt and Yan inputs at select targets, Yan and Pnt together negatively regulate many target genes. Pnt also facilitates chromatin binding of the TLE co-repressor protein Groucho, raising the possibility of context-specific roles for Pnt as a repressor. Focusing on the target gene eve, synergistic interactions between Pnt, Yan, Groucho and factors associated with pausing of RNA pol II fine-tune Eve expression to ensure robust cell fate specification. We propose that the collaborative action of an opposing activator-repressor pair establishes repressive complexes that collaborate with the pol II pausing machinery to create a locus-wide poised state that both prevents spurious gene activation and ensures timely induction of expression following signaling cues.

To investigate how regulatory inputs from Yan and Pnt are integrated across their target gene loci, we used ChIP-seq to generate a genome-wide map of Pnt-bound regions in stage 11 Drosophila embryos and then compared it with that of Yan. The two occupancy profiles were strikingly similar, including at loci of the known Yan/Pointed targets argos (aos), even skipped (eve) and mae (also known as edl) (Fig. 1A, Fig. S1A,B). Using high-confidence bound regions identified with the model-based analysis of ChIP-seq (MACS) peak-calling tool (Zhang et al., 2008), we calculated that 82% of Yan-bound peaks overlapped with Pnt-bound peaks. In the instances when a peak was called only in the Pnt dataset, visual inspection of the tag density pileups often revealed a subthreshold accumulation of reads in the Yan sample (Fig. S1A,B). Consistent with the similar binding landscapes, central motif enrichment analysis showed that the consensus sequences recognized by Mothers against dpp (Mad) and ETS transcription factors were the two most enriched motifs in Pnt-bound peaks (Fig. S1C), exactly as has been shown for Yan-bound peaks (Webber et al., 2013a). Assigning Pnt-bound regions to the nearest gene produced a list of genes with significant overlap with a similarly generated Yan target list, and thus near identical enrichment of gene ontology (GO) terms (Table S1, Fig. S1D).

Although Yan and Pnt are co-expressed extensively in stage 11 embryos (Boisclair Lachance et al., 2014), we expected the overlapping occupancy profiles to reflect mutually exclusive Yan or Pnt binding to specific enhancers, consistent with current understanding of their antagonistic relationship. To assess this, we selected a subset of bound regions for which an ability to respond appropriately to Pnt and Yan activating and repressive inputs had been previously demonstrated in S2 cell transcriptional reporter assays (Webber et al., 2013a). To our surprise, sequential ChIP (ChIP-reChIP) followed by qPCR revealed Yan-Pnt co-occupancy at six of the seven regions tested (Fig. 1B). This suggested that the mechanisms that organize Yan and Pnt chromatin occupancy and contributions to gene expression regulation are more complicated than previously assumed.

Yan-Pnt co-occupancy of an enhancer could result from either interdependent or independent recruitment. Based on the accepted model of Yan and Pnt function in which Yan-mediated repression maintains cells in an uncommitted progenitor-like state, we predicted that Yan is more likely to be the initiator, perhaps recruiting Pnt to poise bound target regions for subsequent activation in response to signaling. We therefore first asked whether binding of Pnt to its target regions depends upon Yan by examining Pnt chromatin occupancy in yan null mutant embryos. In contrast to our predictions, the binding landscape of Pnt was broadly conserved in the absence of Yan (Fig. 1C,D, Table S2), suggesting that Pnt recruitment occurs primarily independently of Yan.

To determine whether Yan recruitment was similarly independent of Pnt, we profiled Yan chromatin occupancy in pnt null mutant embryos. In contrast to expectations, comparison of ChIP-seq signal profiles revealed a global reduction in Yan occupancy (Fig. 1E,F). This finding was validated independently by ChIP-qPCR at all targets tested (Fig. 1G). Indirect immunofluorescence analysis confirmed comparable Yan protein levels in wild-type and pnt mutant embryos (Fig. S2), ruling out the most trivial explanation for globally reduced occupancy. In further support for a direct role for Pnt in facilitating Yan recruitment to chromatin, Yan occupancy was preferentially reduced at regions identified as bound by both Yan and Pnt versus regions identified as bound by Yan alone (Fig. 1H). We conclude that Pnt plays a crucial and unexpected role in the recruitment and/or stabilization of Yan binding across the genome.

Yan's unanticipated dependency upon Pnt for proper occupancy motivated us to consider a non-canonical role for Pnt as a repressor, and a collaborative rather than antagonistic relationship with Yan in this capacity. The central prediction was that the expression of genes subject to such Pnt-Yan cooperative repression should increase upon loss of either Pnt or Yan. To test this, we utilized an unpublished analysis of mRNA expression changes in pnt or yan mutant embryos that we had performed with a custom Agilent microarray made with probes from Yan-bound genes identified by ChIP (for ChIP targets, see Webber et al., 2013a). We first identified probes for which expression was significantly changed (P<0.05) in pnt mutants versus wild type (Fig. 2A) and then selected a handful of up- or downregulated targets for qPCR validation. Comparison of array and qPCR results revealed broad agreement between the two datasets, confirming the overall quality of the array data (Fig. 2B). As a second point of validation, we determined whether mRNA levels of the known Yan/Pnt targets aos, mae and eve exhibited the expected opposite response to loss of Pnt or Yan. Consistent with expectation, expression of aos and mae was reduced in pnt mutants and increased in yan mutants. In contrast, although eve levels were elevated in the absence of Yan, they were not significantly changed in pnt mutants, a finding perhaps in keeping with the stochastic and rather modest loss of Eve expression that has been described in pnt mutant embryos (Halfon et al., 2000).

Although a handful of studies have uncovered roles for Pnt in negative regulation of gene expression, including hid in the embryo, yan in the eye disc and asense in the larval brain (Kurada and White, 1998; Rohrbaugh et al., 2002; Zhu et al., 2011), Pnt has been characterized exclusively as a transcriptional activator (Klämbt, 1993; Scholz et al., 1993; Brunner et al., 1994; O'Neill et al., 1994; Gabay et al., 1996; Schwartz et al., 2010). We were therefore intrigued by the set of genes upregulated in the pnt mutant. Although some of these expression increases could reflect indirect regulation, because the genes used in the analysis were selected based on chromatin occupancy, the approach should enrich for changes resulting from loss of direct Pnt-mediated regulation. Focusing on genes with upregulated expression, which would reflect loss of Pnt repressive inputs, we assessed their response to loss of Yan. If the ability of Pnt to repress transcription depends on its ability to recruit Yan, then a similar set of genes should be upregulated in both mutants; indeed, a strong positive correlation was observed (R²=0.7, P<0.0001; Fig. 2C).

To gain insight into the developmental processes that might be regulated by coordinated Pnt-Yan repression, we identified the ontologies of the upregulated genes using the PANTHER classification system (Mi et al., 2013). Upregulated genes were enriched for categories associated with muscle cell fate commitment and cardioblast differentiation. These GO terms were absent from ontology analyses performed with downregulated genes (Table S3). Considering these differences in light of the Yan and Pnt expression patterns in the stage 11 embryo suggests that in the mesoderm, where Yan and Pnt are co-expressed and RTK signaling levels are low (Gabay et al., 1997; Boisclair Lachance et al., 2014), the two collaborate as repressors to stabilize the unspecified state.

A second prediction of a model in which Pnt contributes repressive function to gene regulation is that it should recruit co-repressor proteins, either directly or through its interaction with Yan. To identify likely candidates, we examined the modENCODE database (Contrino et al., 2012; www.modencode.org) to compare available co-repressor genome-wide occupancy patterns with those of Yan and Pnt. The binding landscape of the co-repressor Groucho (Gro) immediately stood out. Because the published datasets were not appropriately stage-matched to our work, we performed ChIP-seq analysis of Gro in stage 11 wild-type embryos. The results confirmed the similarity of the Gro binding landscape to that of Yan and Pnt. Intersecting high confidence peaks of Gro with the Yan and Pnt datasets revealed a 54% and 37% overlap, respectively, and heat-map analysis of Yan/Pnt co-bound regions suggested even greater overlap (Fig. 3A).

To determine whether proper Gro occupancy requires Pnt, we performed ChIP-seq analysis of Gro in a pnt mutant background. Western blot analysis revealed no significant change in Gro protein levels in pnt mutant versus wild-type embryos (Fig. S3A,B) and Gro occupancy was only moderately affected at regions of the genome without nearby Pnt binding (Fig. 3B,C). In contrast, analogous to our finding of reduced Yan occupancy in pnt null animals, Gro binding was reduced in regions of the genome where Gro and Pnt profiles normally overlap (Fig. 3C). Comparison of the ChIP-seq peaks suggested that loss of Gro occurred at regions that also displayed reduced Yan occupancy in pnt null embryos. For example, Gro occupancy was lost across the neuralized locus (Fig. 3D), in patterns similar to those observed for Yan loss, but was barely reduced at the turtle locus, which does not bind Yan (Fig. 3E). Plotting the ratio of Gro occupancy at bound regions in pnt mutants relative to the wild-type control confirmed that the reduction of Gro in the absence of pnt is more severe at Yan and Gro co-occupied sites, than at sites that are not bound by Yan (Fig. 3F). Taken together, these data indicate that Pnt recruits both Gro and Yan to common regulatory elements, raising the possibility of coordinated Yan-Pnt-Gro occupancy and repression of the associated target gene.

We tested this prediction by correlating gene expression changes in pnt mutant embryos with the changes in Yan and Gro occupancy described above. Of the 320 genes associated with Yan and Gro occupancy loss in pnt mutants, 129 were represented in the custom microarray. Of these, 107 were differentially expressed in the absence in of pnt, with 72% displaying upregulated expression (Fig. 3G). Using the converse approach, upregulated genes identified in the microarray had reduced Yan and Gro signal intensity in pnt mutants relative to wild type; this list included the validated Gro target E(spl)mbeta-HLH (Fig. S3C, Fig. S4). We conclude that the loss of Yan and Gro occupancy that occurs in pnt mutant embryos reflects a novel mechanism by which Pnt recruits and collaborates with these two repressive factors to negatively regulate expression at a significant subset of target genes.

Having defined a novel role for Pnt in recruitment of Yan and Gro, we next investigated how these interactions influence expression at a specific locus. The heart identity gene eve provided an ideal vantage point to do this because of the already deep mechanistic understanding of how Yan repressive and Pnt activating inputs are organized at specific enhancers (Halfon et al., 2000; Webber et al., 2013b; Boisclair Lachance et al., 2018). In stage 11 embryos, Eve is expressed in segmentally arrayed clusters of cells in the developing cardiogenic mesoderm. Yan and Pnt exert antagonistic inputs at the level of a pattern-driving MHE, such that in yan mutant embryos extra Eve⁺ cells are specified, whereas in pnt mutant embryos, the number of Eve⁺ cells specified is reduced (Halfon et al., 2000). Additional repressive input is provided by D1, a Yan-responsive element, deletion of which results in elevated and more variable Eve expression (Webber et al., 2013b).

Matching the pattern of Yan occupancy (Webber et al., 2013a), tag density profiles of Pnt at the eve locus revealed enrichment at both the D1 and MHE regulatory regions (Fig. 4A); Pnt occupancy at the D1 region, which genetically appears to be dedicated to dampening Eve expression (Webber et al., 2013b), further supports the suggestion of a role for Pnt in repressive regulation. Analysis of the Yan genome-wide ChIP dataset in pnt mutant embryos suggested that Pnt is required for proper Yan occupancy at both the MHE and D1; ChIP-qPCR confirmed this dependency (Fig. 4B). To test whether Yan and Pnt are co-bound, we performed sequential ChIP. Although we were unable to detect co-occupancy at the eve MHE, simultaneous occupancy was detected at the eve D1 region (Fig. 4C). One explanation for the negative results at the MHE is that we are simply below the detection threshold. Indeed, both the genome-wide ChIP data sets and the ChIP-qPCR confirmation experiments always show low enrichment at the MHE, perhaps indicating that Yan/Pnt occupancy/co-occupancy of this pattern-driving enhancer occurs in only the small subset of mesodermal cells from which Eve⁺ pericardial cells are specified. Alternatively, Yan and Pnt might co-occupy the D1 but not the MHE, with 3D interactions enabling D1-bound Pnt to recruit/stabilize Yan occupancy at both the D1 and the MHE. Although further testing will be required to distinguish between these possibilities, in support of the latter, long-range interactions between the D1 and MHE stabilize Yan occupancy at the two elements (Webber et al., 2013b; Boiclair Lachance et al., 2018).

Because complete loss of pnt results in reduced Eve expression, we devised an alternative genetic strategy to assess repressive function. We reasoned that embryos heterozygous for Yan might provide a suitably sensitized background to reveal a role for Pnt-mediated repressive regulation. We first assessed Eve expression levels in animals heterozygous for either yan or pnt and compared these with Eve levels in double heterozygotes. The yan and pnt loss-of-function alleles were fully recessive, with no significant change in Eve expression detected relative to wild-type control (Fig. 4D). In contrast, in yan/+;pnt/+ embryos, Eve levels were significantly elevated and extra Eve⁺ cells were specified (Fig. 4D,E). We repeated the experiment using a functional Eve-YFP BAC transgene (Webber et al., 2013b) and again measured elevated Eve levels in doubly heterozygous animals compared with single heterozygotes (Fig. S5). Together, these data suggest a cooperative function for Yan and Pnt in negative regulation of eve.

We extended the dose-sensitive genetic interaction analysis to assess involvement of the co-repressor Gro, occupancy of which at both the eve MHE and D1 is reduced in the absence of pnt (Fig. 4A). Consistent with previous studies of Gro’s repressive input at eve (Helman et al., 2011), we observed increased Eve levels and extra Eve⁺ cells in gro/+ embryos; both phenotypes were enhanced in pnt/gro doubly heterozygous animals (Fig. 4D,E). We conclude that Yan, Pnt and Gro work collaboratively to negatively regulate Eve levels and Eve⁺ cell fate specification in the cardiogenic mesoderm.

The inherent conflict of recruiting both active and repressive histone marks to a given bound region is characteristic of a balanced chromatin state, whereby genes are held silent, but poised for transcriptional activation (Schwartz et al., 2010; Gaertner et al., 2012). Analogously, at the transcription factor level co-occupancy by the activator-repressor pair Pnt/Yan could both prevent inappropriate activation of eve under sub-threshold signaling conditions and prime the locus for rapid transcriptional activation following the onset of upstream signaling. Meta-analysis of publicly available ChIP-seq data for three different chromatin modifications from 4-8 h embryos, which includes the stage 11 time point used in our ChIP experiments, provided circumstantial evidence that eve may be poised. Specifically, the combination of negligible H3K27ac, which exclusively marks active enhancers, prominent H3K4me1, a mark of both poised and active enhancers, and prominent H3K27me3, a mark that in the absence of H3K27ac indicates a poised state, suggests that the eve locus may be poised (Rada-Iglesias et al., 2011; Bonn et al., 2012; Koenecke et al., 2017; Fig. 5A).

Poised genes commonly employ an RNA pol pausing strategy whereby RNA pol II is recruited and initiates transcription, but then pauses downstream of the transcription start site (TSS) until it receives the appropriate signaling cues for pause release (reviewed by Gaertner and Zeitlinger, 2014). Recent work in Drosophila cell lines implicates Gro in RNA pol II pausing (Kaul et al., 2014), an intriguing association given our discovery of a role for Pnt in recruiting Gro, including to the eve MHE and D1 enhancers (Fig. 3D,E, Fig. 4A). Furthermore, meta-analysis of genome-wide ChIP datasets revealed frequent overlap between Yan and Pnt occupancy patterns with those of components of the pausing machinery, including Trithorax-like (Trl, also known as GAGA factor) and Bric à brac 1 (Bab1) (Fig. 5B; Contrino et al., 2012; Tsai et al., 2016). Focusing on eve, the locus bears hallmarks of pausing with high pol II occupancy near the TSS, together with low incidence of H3K4me3 and overlap with the pausing factor Trl (Fig. 5C; Contrino et al., 2012; Lee et al., 2008; Fuda et al., 2015; Gaertner et al., 2012; Tsai et al., 2016).

Using eve as our model, we assessed genetic interactions between Pnt and members of the pausing machinery. We first tested whether heterozygosity for either Trl or bab1 could influence Eve expression and observed no significant change in Eve levels (Fig. 6A). In contrast, embryos doubly heterozygous for either pnt and Trl or pnt and bab1 displayed significantly increased Eve levels with a corresponding increase in the number of Eve⁺ cells specified (Fig. 6A,B). Similar changes in Eve expression and number of Eve⁺ cells were observed in embryos doubly heterozygous for either Trl and bab1 or yan/gro and Trl (Fig. 6A,B, Fig. S6). A trend towards increased Eve expression was also observed in pnt/Nelf-E double heterozygotes, although the relative increase was not statistically different from control, perhaps because of the maternal contribution of Nelf-E and/or the multi-subunit nature of the NELF complex (Wang et al., 2010; Wu et al., 2005). Survival to adulthood of animals doubly heterozygous for either pnt and Trl or pnt and bab1 was half that of any of the three single heterozygotes (Fig. 6C), suggesting that interactions between Pnt and the pausing machinery may play a broader role in development beyond eve.

The precision with which multipotent cells commit to specialized fates relies on regulated de-repression of gene transcription. Thus, in stem cells and in early embryos, concomitant with the initial opening of chromatin domains by pioneer factors, conflicting epigenetic marks deposited at the promoters of many developmentally important genes recruit yet stall RNA pol II, thereby maintaining repression and multipotency. How such epigenetic-based repressive poising is coordinated with the transcription factors that respond to inductive cues to direct specific cell fate transitions as development proceeds is not well understood. Our study positions the ETS1 homolog and transcriptional activator Pointed (Pnt) as a key integration point between the transcriptional repressive complexes that assemble at regulatory elements across a locus and the molecular complexes that establish, maintain and release RNA pol II pausing. The results not only re-define the classic Yan-Pnt cell fate switch paradigm in Drosophila, but more broadly uncover a novel strategy by which genetic and epigenetic regulation is coordinated to confer robustness to developmental cell fate transitions.

The accepted model for Yan and Pnt function predicts mutually exclusive occupancy at enhancers, with RTK signaling triggering the transition from an initial Yan-bound repressed state in uncommitted progenitors to a subsequent Pnt-bound activated state that drives cell fate acquisition. Our study paints a different picture in which Pnt plays a role in establishing and stabilizing that initial Yan-bound repressed state, and in fact co-occupies many regulatory elements with Yan. It is important to note that because the sequential ChIP analysis was not a genome-wide study and because whole embryos rather than single cells were profiled, it is possible that Yan-Pnt co-occupancy only occurs at a small subset of targets and that the broad similarity in overall occupancy patterns primarily reflects a mix of exclusively Yan or exclusively Pnt binding at the same enhancers in different cells. Arguing against this, the genome-wide loss of Yan occupancy in pnt mutants and normal Pointed occupancy in yan mutants positions Pointed as the critical determinant and/or stabilizer of Yan binding and repression, with co-occupancy likely relevant to the mechanism.

We therefore speculate that Pnt is required first to set up Yan occupancy and repression and second to respond to RTK signaling by activating target gene expression; its activating function is thus epistatic to its repressive role, explaining why predominantly loss-of-function phenotypes have been described for pnt mutants. Use of the same transcription factor to dictate both the repressive regulation that maintains the initial multipotent state and the subsequent activation that changes it, enables a level of temporal coordination of gene expression dynamics that may be crucial to the robustness of differentiative transitions. We also note that although Yan/Pnt function has been studied primarily in the context of RTK signaling, the two are co-expressed in many tissues across development, including those presumed to have low RTK signaling input (Gabay et al., 1997; Boisclair Lachance et al., 2014), and co-occupy regulatory elements across a broad swath of signaling pathway genes and crucial developmental regulators that are unlikely all to be regulated downstream of RTK signaling. Thus, Pnt-Yan-Gro enhancer co-occupancy may provide a modular repressive mechanism that can be adapted to a variety of regulatory situations.

How Pnt-Yan co-occupancy is organized/facilitated by the DNA sequence of each enhancer will be interesting to explore. One possibility is that Pnt initially interacts with all ETS binding sites to open up a regulatory element, but then gets displaced at a subset of sites upon recruitment of Yan, perhaps remaining bound only at sites required for subsequent activation. Alternatively, distinct sequence preferences rather than affinity differences might result in Pnt occupancy of only a specific subset of ETS binding sites, leaving others free for Yan to bind. This model also supports a variation in mechanism by which Pnt and Yan are recruited jointly, rather than sequentially, to establish the initial repressed state, with co-occupancy essential to stabilize Yan binding. Our recent work exploring how the cis-regulatory organization of the eve MHE organizes Yan and Pnt inputs supports the idea that distinct sequence preferences enable simultaneous occupancy and hence complex integration of repressive and activating inputs (Boisclair Lachance et al., 2018).

The mechanism by which Yan occupancy depends on Pnt also remains to be elucidated. One possibility is that Pnt recruits Yan directly. However, to date our efforts to detect Yan-Pnt protein-protein interactions, either in vitro, in two-hybrid screens, or in standard co-immunoprecipitation experiments, have yielded negative results. Indirect protein-level interaction mechanisms, such as bridging the complex with Gro or with another transcription factor, may thus be more likely. The strong overlap between Mad and ETS binding sites noted in Yan/Pnt-bound regions genome-wide (Webber et al., 2013a) makes the Dpp effector Mad an intriguing candidate. Alternatively, rather than nucleating specific protein complexes, Pnt may establish or interpret a local chromatin state that permits Yan binding. Analogous pioneer-like activity has been described for a few other ETS factors, including PU.1 (also known as Spi1) and ETV2 (reviewed by Iwafuchi-Doi and Zaret, 2014; Kanki et al., 2017). As pnt encodes two alternatively spliced products, Pnt-P1 and Pnt-P2, that contain the same DNA-binding domain but different amino-terminal activation domains and exhibit different patterns of expression and signal responsiveness (Klämbt, 1993; Scholz et al., 1993; O'Neill et al., 1994; Brunner et al., 1994; Gabay et al., 1996; Shwartz et al., 2013), it will be important to re-evaluate the role of each isoform during cell fate transitions with respect to the establishment of Yan/Gro binding, target gene repression and the subsequent switch to activation.

Regardless of the precise mechanism, Yan's reliance on Pnt for its stable recruitment provides a plausible explanation for a previous unexpected finding that doubling Yan dose does not lead to increased or ectopic DNA occupancy (Webber et al., 2013a). More broadly, the strategy of the activator recruiting the repressor could provide an occupancy feedback circuit that buffers against fluctuations in activator or repressor concentrations. For example, the standard competition model predicts that a multipotent cell with lower than normal Pnt will over-recruit Yan; we speculate that such Yan-dominated repression would be sluggish in response to inductive cues. In contrast, a system in which Yan occupancy depends on Pnt might be buffered against such variation, because the consequence of lower Pnt levels would be less efficient Yan recruitment, which should maintain the appropriate Yan-Pnt balance.

We speculate that this precisely poised Pnt-Yan-mediated repressive state is achieved through close coordination with the RNA pol II pausing machinery (summarized in Fig. 7). Several pieces of evidence support this idea. For example, not only is Pnt essential for proper Yan occupancy, but it also recruits the co-repressor Gro to the same set of regulatory elements. Prior genome-wide analyses of Gro occupancy and function in embryos and cultured cells have shown that Gro-regulated genes are enriched for epigenetic marks and promoter proximal transcripts commonly associated with paused RNA pol II (Kaul et al., 2014; Chambers et al., 2017). Our demonstration of eve de-repression in embryos doubly heterozygous for gro and Trl provides the first genetic evidence of a possible direct mechanistic link between Gro repressive complexes and the RNA pol II pausing machinery. Our study also emphasizes the likely importance of Pnt to the Gro-paused RNA pol II connection. For example, included among the set of genes showing coordinately disrupted Yan-Gro occupancy and de-repression in pnt mutant embryos is E(spl)mbeta-HLH, a target previously shown to be regulated by Gro-dependent RNA pol II pausing in cultured cells (Kaul et al., 2014 and Fig. S4). The web of synergistic genetic interactions between pnt, yan, gro and mutants in RNA pol II pausing factors, such as Trl, Bab1 and NELF, further supports such a model.

RNA pol II pausing establishment and release is also closely linked to Polycomb (PcG) repressive complexes in both Drosophila and mammals (Schwartz et al., 2010; Gaertner et al., 2012; Bernstein et al., 2006; Ferrai et al., 2017) and PcG repression has in turn been connected to Gro (Abraham et al., 2015). For example, a recent study describes how recruitment of the Hox family transcriptional activators AbdA and Ubx reduces PcG binding at RNA pol II paused genes to promote release and transcriptional activation (Zouaz et al., 2017). Similarly, ETS1, the mammalian ortholog of Pnt, promotes release of paused RNA pol II to activate angiogenic gene expression, although connections to PcG complexes were not investigated (Chen et al., 2017). Trl, which our study connects genetically to Pnt-Gro-Yan repressive mechanisms, helps direct PcG proteins to Polycomb response elements, or PREs, and thus contributes to PcG repressive activity (Mahmoudi et al., 2003; Mishra et al., 2003; Mulholland et al., 2003). Given that eve is a PcG target gene, with a validated PRE (Dura and Ingham, 1988; Fujioka et al., 2008; Kim et al., 2011), it may provide an ideal context for elucidating the molecular levels of integration between Yan-Pnt-Gro and PcG repressive complexes in relation to RNA pol II pausing.

In conclusion, we propose that, analogous to the use of conflicting epigenetic marks to poise RNA pol II, the inherent conflict of co-occupancy by an activator-repressor pair such as Pnt-Yan establishes an exquisitely sensitive and dynamic repressive mechanism that confers robustness to developmental gene expression regulation. Because loss or misexpression of ETS transcription factors contributes to many cancers and because oncogenic transformation relies on dysregulated use of normal developmental pathways, exploration of these ideas in mammalian systems may provide new insight into human disease.

All chromatin immunoprecipitation (ChIP) was from stage 11 embryos (approximately 5-7 h after egg lay) processed as previously described (Webber et al., 2013a) and summarized in supplementary Materials and Methods. Sequential ChIP of Yan and Pnt was performed by first immunoprecipitating Yan (guinea pig anti-Yan; Webber et al., 2013a). The protein-DNA complex was eluted in reduced volume and diluted 1:10 in ChIP lysis buffer before performing the second round of ChIP with a rabbit GFP antibody (rabbit anti-GFP, A6455, Invitrogen, Lot# 1603336). A no antibody control (mock-treated) processed identically to experimental samples was included. For ChIP-seq, two biological replicates and an input sample were sequenced on an Illumina Hi-seq instrument according to the Illumina protocols. The raw sequence data were aligned to the April 2006 D. melanogaster genome using BWA (Li and Durbin, 2009). Following standard practice in the field (Robertson et al., 2007; Rozowsky et al., 2009; Zhong et al., 2010), having first confirmed consistency between replicates, the two IP reads were combined and peak detection performed using MACS software with an mfold of 3.40 and otherwise default parameters. MACS-defined bound regions were assigned to the nearest gene using the UCSC genome browser and gene ontology enrichment analyses were performed (see supplementary Material and Methods for further details). Top-scoring bound regions were also subjected to Centrimo analysis to identify centrally enriched binding motifs in the Pnt ChIP-seq dataset (see supplementary Materials and Methods for more details). SPP was used to calculate genome-wide tag density profiles. Default parameters were used, with the exception that the scale.by.dataset.size=T option was used to normalize tag density by the total dataset size to make it comparable across samples (Kharchenko et al., 2008). ChIP-qPCR was performed as previously described (Webber et al., 2013a) and is summarized in supplementary Materials and Methods.

A custom expression array was designed on the Agilent GE 8×15 K platform. The microarray included 7080 probes for putative Yan target genes identified by Webber et al. (2013a) (designed using the Earray software by Agilent), 1894 probes for random genes and 536 control probes. Total RNA was extracted from stage 11 wild-type, yan null or pnt null embryos with TRIzol Reagent (Invitrogen) following the manufacturer's protocol, and purified using the RNeasy Mini Kit (Qiagen). Total RNA was labeled and hybridized to the microarray using the Quick Amp One-Color Labeling Kit (Agilent) as described by the manufacturer. Triplicate experiments for each genotype were performed. Expression array data were first analyzed by the Feature Extraction Software (Agilent) using default parameter settings, and the processed signals for the probes were then used for downstream analysis. Linear models were generated for each array using only the probes in the random gene set, and then signals were normalized across all arrays. After normalization, the average signal for each probe across triplicate experiments of each genotype was calculated, and signal fold changes between wild type and the pnt or yan mutants were computed. t-tests were performed at the probe level, and probes with a P-value less than 0.05 were selected as significant (Table S4). For microarray validation, RNA was isolated, reverse transcribed and the resultant cDNA subjected to Real-Time PCR as described in the supplementary Materials and Methods.

Sorted bed files were produced using the BEDOPS wig2bed script (Neph et al., 2012). To manage the large number of calculations, a Visual Basic Application (VBA) for Microsoft Excel was used to generate matrices of read density data for groups of given bound regions ±5 kb from the midpoint of each individual region. The matrices were used for (1) calculating ratios of transcription factor (TF) occupancy in mutant versus wild type, (2) producing aggregate read density profiles by averaging read density across all peaks and (3) generating TF binding heat maps. For further details, see supplementary Materials and Methods.

Bed files of MACS-defined bound regions and sequence aligned reads for wild-type and mutant datasets for each factor were generated. The differential binding analysis software, MAnorm (Shao et al., 2012), was used to generate a merged set of bound regions for each factor with quantitative values of differential binding and associated P-values for each peak. These datasets were used to describe patterns of peak gain or peak loss, where peak loss corresponds to normalized M values of >0.5 and peak gain of <−0.5 (Fig. 1D,F, Tables S2 and S5).

For quantification of Eve levels and numbers of Eve⁺ cells, embryos were stained as described by Webber et al. (2013a) with either 1:10 mouse anti-Eve [3C10, Developmental Studies Hybridoma Bank (DSHB)] or 1:1000 rabbit anti-GFP (A6455, Invitrogen, 1603336). Using a Zeiss 880 confocal microscope, serial 0.8 µm z-sections were taken through the Eve-positive mesodermal cells and maximum projections generated. Expression intensity was calculated as the mean pixel intensity for each Eve⁺ cluster minus the mean background pixel intensity, normalized to the average cluster intensity of the control imaged in the same session. For cell counts, Eve⁺ cells were counted by going through z-stack projections of the relevant slices. For rescue experiments, stage 11 embryos of each genotype were hand-selected, transferred to vials and incubated at 25°C until adults emerged.

For Yan expression analysis, wild-type or pntΔ88 embryos were stained as described by Boisclair-Lachance et al. (2014) with 1:10,000 guinea pig anti-Yan (Webber et al., 2013a). For western blot analysis described in Fig. S3, stage 11 wild-type or pntΔ88 embryos were dechorionated in 50% bleach and homogenized in 50 µl of SDS sample buffer [250 mM Tris-Cl (pH 8), 10% SDS, 50% glycerol, 50% β-mercaptoethanol, 0.04% bromophenol blue]. Samples were passed through a 27G needle 10 times and boiled for 10 min prior to running on an 8% SDS-PAGE gel. After transfer to PVDF, blots were probed with 1:100 mouse anti-Gro (DSHB) antibody and 1:2000 mouse anti-tubulin (T9026, Sigma) antibody, which served as a loading control.

The following stocks were obtained from the Bloomington Drosophila Stock Center: w¹¹¹⁸, pnt^Δ88/TM3,Sb1, w1;Trl^13C/TM6B,Sb¹,Tb¹, Df(3L)bab^AR07,bab1^AR07,bab2^AR07/TM6B,Tb¹ and y¹,w^67c23; P{w^+mC]y^+mDint2=EPgy2}EY07065/TM3, Sb¹,Ser1. Additional stocks used include: gro^MB36 (Jennings et al., 2008), pnt^AF397 (Rebay et al., 2000), yan^ER443 and yan^E833 (Karim et al., 1996), Pnt-GFP (Boisclair Lachance et al., 2014), and Eve-YFP (Webber et al., 2013b). To allow genotyping of stage 11 embryos, stocks were rebalanced over twist-Gal4>UAS-GFP marked second and third chromosome balancers.

Data are presented as mean±s.e.m. except where otherwise described and minimum sample sizes are reported in each figure. Data were plotted and analyzed for statistical significance using Graphpad Prism software. Statistical significance was determined either by a two-tailed t-test where appropriate, or alternatively with a one-way ANOVA in combination with Tukey's multiple comparison tests to compare two or more groups. P-values less than 0.05 were considered to be statistically significant.

We thank Pieter Faber, Mikayka Marchuk and Abhilasha Cheruku in the University of Chicago Genomics Facility for help with ChIP-seq and microarray. Jean-Francois Boisclair Lachance, Kohta Ikegami, Rebecca Spokony and Matthew Slattery provided many helpful discussions and comments on the manuscript. We acknowledge the Bloomington Drosophila Stock Center (NIH P40OD018537) and the Developmental Studies Hybridoma Bank (created by the NICHD of the NIH) for reagents.