Polycomb safeguards imaginal disc specification through control of the Vestigial–Scalloped complex Free

A fundamental task of metazoans is to organize populations of cells into specialized tissues or organs. Cells first start out totipotent, transition through a pluripotency phase, and then enter a lineage-committed state (Ferrell, 2012; Riveiro and Brickman, 2020; Smith, 2017). This concept is enshrined in Waddington's epigenetic landscape model of embryonic development, which today forms the basis of modern developmental biology (Waddington, 1940, 1957). Numerous studies have shown that cells of one lineage can be converted into cells of another lineage through the simple expression of one or more transcription factors (Konstantinides and Desplan, 2020; Masserdotti et al., 2016; Takahashi and Yamanaka, 2015). For instance, the forced expression of Pax6 induces the reprogramming of non-ocular tissues into retinal structures (Chow et al., 1999; Halder et al., 1995). Several transcription factors that induce cellular reprogramming interact directly with histone-modifying proteins, thereby indicating that altering the chromatin landscape is important for the transdetermination of tissue fate (Ang et al., 2011; Apostolou and Hochedlinger, 2013; Li et al., 2021; Mansour et al., 2012).

In this study, we investigate the role that one epigenetic factor, Polycomb (Pc), plays in establishing the fate of the eye-antennal imaginal disc of the fruit fly, Drosophila melanogaster. Pc is a member of the Polycomb group (PcG) of epigenetic repressors, and it is responsible for recognizing or ‘reading’ the H3K27me3 mark (Cao et al., 2002; Czermin et al., 2002; Müller et al., 2002). Pc plays an important role in the specification of Drosophila tissues as all thoracic and the first seven abdominal segments of null mutant embryos are transformed into the eighth abdominal segment (Lewis, 1978). Similarly, the adult T2 and T3 legs of a dominant Pc mutant are partially transformed into T1 legs (Lewis, 1947). Lastly, the frequency of leg-to-wing transdetermination events increases when PcG member levels are reduced (Lee et al., 2005).

Imaginal discs are excellent model systems for elucidating the mechanisms underlying cell and tissue specification, in part because their fates can be dramatically altered by manipulations of key molecular determinants (Weasner and Kumar, 2022). For example, the fate of the eye is controlled by the retinal determination (RD) network of 14 transcription factors (Kumar, 2010). In their absence, the developing eye field is transformed into head epidermal tissue (Hoge, 1915; Hunt, 1970; Mardon et al., 1994; Milani, 1941; Sved, 1986; Weasner and Kumar, 2013). Similarly, in the wing imaginal disc, the loss of several wing selector genes results in complete loss of the wing blade (Ng et al., 1995; Sharma and Chopra, 1976; Stevens and Bryant, 1985; Williams et al., 1991). In contrast, the forced expression of eye and wing selector genes is sufficient to alter the fate of the targeted disc and provides additional insight into how tissue fate is controlled. For instance, forced expression of RD network members induces the formation of ectopic eyes in all imaginal discs (reviewed by Kumar, 2010) and ectopic activation of the wing selector gene vestigial (vg) transforms halteres into wings and induces wing outgrowths within the eye (Kim et al., 1996; Prasad et al., 2003; Simmonds et al., 1998).

Here, we report on the mechanisms by which the fate of the eye-antennal disc is specified. When this disc is fragmented, regenerating cells along the wounded edge are often reprogrammed and give rise to wing tissue (Hadorn, 1968, 1978). Reprogramming of the eye into a wing also occurs in response to several genetic perturbations (Edwards and Gardner, 1966; Goldschmidt and Lederman-Klein, 1958; Katsuyama et al., 2005; Kobel, 1968; Kurata et al., 2000; Masuko et al., 2018; Ouweneel, 1970; Postlethwait, 1974; Simmonds et al., 1998). Interestingly, there are no recorded instances of the eye being transformed into any other imaginal disc. This contrasts with the adjacent antenna, which can be reprogrammed to develop into wing, leg, labial and genital discs (Hadorn, 1978; Weasner and Kumar, 2022). We set out to understand why the eye field is specifically reprogrammed into a wing disc rather than one of the other imaginal discs.

In a previous study, we discovered that a reduction in Pc expression reprograms the eye into a wing (Zhu et al., 2018). Here, we show that reducing Pc expression results in ectopic activation of the wing selector gene vestigial (vg) within the eye field. The repression of vg expression by Pc during normal eye development is likely to be direct as functional Polycomb response elements (PREs) are present within the vg locus (Ahmad and Spens, 2019; Herzog et al., 2014; Okulski et al., 2011). In the developing wing, Vg interacts with the DNA-binding protein Scalloped (Sd) to promote its fate (Delanoue et al., 2004; Halder and Carroll, 2001; Halder et al., 1998; Kim et al., 1996; Klein and Arias, 1998; Pimmett et al., 2017; Simmonds et al., 1998). The adult wing is severely reduced in size or eliminated if the Vg–Sd complex is disrupted (Bownes et al., 1981; Gruneberg, 1929; Lindsley and Grell, 1968; Williams and Bell, 1988). In contrast, because vg is not normally expressed in the developing eye, Sd interacts with the transcriptional activator Yorkie (Yki) instead. The Yki–Sd complex goes on to regulate the final size of the compound eye (Koontz et al., 2013; Meserve and Duronio, 2015). Our results suggest that the eye field is prevented from developing into a wing by Pc-mediated repression of the vg locus. In the absence of this repression, ectopic activation of Vg within the eye field allows for the inappropriate formation of the Vg–Sd complex and the reprogramming of the eye into a wing. Thus, our findings suggest that selective regulation of Sd-binding partners by PcG proteins guides the eye primordium along the path towards its final fate.

We made a second surprising discovery when we reduced Pc expression in vg and sd loss-of-function mutant backgrounds. Instead of the disc simply reverting to its normal state, it undergoes massive hyperplastic growth instead. This is reminiscent of the immense growth expansion seen when certain PcG factors are misregulated (Beira et al., 2018; Bunker et al., 2015; Classen et al., 2009; Loubiere et al., 2016; Martinez et al., 2009) or when epithelial polarity is lost (Bunker et al., 2015; Enomoto et al., 2021; Pagliarini and Xu, 2003; Wu et al., 2010) in the eye-antennal disc. The explosive growth that we observe is consistent with the role of Sd as a default repressor of tissue growth (Koontz et al., 2013). Unexpectedly, histological analysis of the mutant tissue indicates that multiple domains of the overgrown discs take on the fate of other imaginal discs. RNA sequencing (RNA-seq) confirms the presence of several exogenous imaginal discs. Our results bring to mind the diverse collection of cell types that are found within mammalian tumors (Dagogo-Jack and Shaw, 2018; Genovese et al., 2019; Ju, 2021; Li et al., 2022; McGranahan and Swanton, 2017; Tammela and Sage, 2020). As such, information gleaned from hyperplastic imaginal discs, may also provide insight into mechanisms that give rise to intratumor heterogeneity. In total, our results highlight the complex developmental landscape that tissues must navigate on the way to adopting final fates.

Upon reduction of Pc, the dorsal eye field is transformed into a wing imaginal disc. These discs show ectopic activation of the Hox gene Antennapedia (Antp) and the wing selector gene vg as well as loss of the RD network genes ey and eya (Zhu et al., 2018). Removal of this epigenetic repression results in the inappropriate activation of wing selector genes and the transformation of the eye into a wing. Because simultaneous reduction of Pc and Antp levels results in dramatically fewer instances of eye-to-wing transformations, it was proposed that Antp plays a central role in this fate decision. However, the eye-to-wing transformation is not observed in most Antp gain-of-function mutants (Scott et al., 1983) or when Antp is forcibly expressed in the developing eye (Jorgensen and Garber, 1987; Kurata et al., 2000; Schneuwly et al., 1987). As such, the mechanism by which Pc maintains the fate of the eye remains unknown.

We used RNA interference (RNAi) to individually knock down 18 out of 20 PcG members in the eye-antennal disc (Fig. 1, Table S1). PcG factors are ubiquitously expressed across different tissues (Buchenau et al., 1998; DeCamillis and Brock, 1994; Paro and Zink, 1993) and, consistent with these studies, our RNA-seq data (Table S2) show that transcripts of all PcG members, and their known binding partners, are expressed within wild-type (WT) eye-antennal and wing discs as well as Pc knockdown eye-antennal discs (Fig. S1A-D). Therefore, knocking down PcG members in the eye-antennal disc should enable us to elucidate the function of these factors in eye development. Because we focused on the effects of knocking down Pc, we validated the Pc RNAi line by showing that Pc transcript levels are reduced in knockdown discs compared with WT eye-antennal discs (Fig. S1E). Using a Pc-specific antibody we have previously shown the RNAi line to be effective in knocking down Pc expression in the eye (Zhu et al., 2018).

Because Antp is robustly expressed in Pc knockdown eye-antennal discs and required for wing formation (Fig. 1A-B″) (Fang et al., 2022; Paul et al., 2021; Zhu et al., 2018), we first screened for ectopic activation of Antp and transformation of the eye into a wing. Surprisingly, although we detected Antp expression when knocking down five other PcG members [pleiohomeotic (pho), Scm-related gene containing four MBT domains (Sfmbt), Sex combs extra (Sce), Sex comb on midleg (Scm) and calypso], the expression levels and pattern of Antp in these discs differed from that of the Pc knockdowns and the eye did not transform into a wing. In pho, Sfmbt and calypso knockdown discs, ectopic Antp expression was largely restricted to the peripodial epithelium (Fig. 1C-E′). The corresponding adults had bristle, head capsule, and antennal defects but compound eyes that were similar to those of WT (Fig. 1C″-E″). Reductions of Sce resulted in the expression of ectopic Antp just within the antenna (Fig. 1F). This differed drastically from that of Pc knockdown discs, in which Antp was expressed throughout the antennal field and ectopic wing tissue (Fig. 1B). This may explain why the reduction of Sce induced an antenna-to-leg transformation (Fig. 1F″) but not the eye-to-wing transformation (Fig. 1B″). Interestingly, in contrast to dominant gain-of-function Antp mutants, in which the antenna is transformed into the second thoracic T2 leg (le Calvez, 1948a,b,c; Yu, 1949), reductions in Sce transformed the antenna into a sex comb-bearing T1 leg (Fig. 1F″, arrow). Lastly, whereas knocking down Scm resulted in ectopic Antp expression within the eye and antenna, the fate of the antenna was primarily altered (Fig. 1G-G″). These five ectopic Antp expression patterns were consistently observed when PcG factors were knocked down with three GAL4 drivers: DE-GAL4 (dorsal eye field), ey-GAL4 (throughout the eye field) and c311-GAL4 (peripodial epithelium) (Fig. 1H). Because ectopic Antp expression does not always result in an eye-to-wing transformation, these results suggest that the eye-to-wing transformation is caused by the potential de-repression of other wing selector genes. This is more in line with studies suggesting that Antp plays only a minor role in forewing development (Carroll et al., 1995; Tomoyasu et al., 2005). In a prior study, we also noted that the expression of several retinal determination genes is dramatically reduced within the region of the disc that is being transformed into a wing (Zhu et al., 2018). The loss of network expression is associated with an erasure of eye fate (reviewed by Kumar, 2010). We propose that eye fate is normally driven by Pax6-mediated activation of the RD gene regulatory network (Gehring and Ikeo, 1999; Kumar, 2010) and maintained by Pc-dependent repression of other wing selector genes.

Given that Pc functions to establish the expression pattern of Hox genes during embryonic development (Janody et al., 2004; Lewis, 1978; Ringrose and Paro, 2007; Schwartz et al., 2006; Struhl and Akam, 1985; Tolhuis et al., 2006) and to maintain imaginal disc fate (Katsuyama and Paro, 2011; Lee et al., 2005), perturbation of PcG repression might be expected to result in the ectopic expression of these genes within the eye field. Transcript counts from RNA-seq analysis of WT and Pc knockdown eye-antennal discs indicated that the Hox genes can be divided into two groups when Pc is knocked down (Fig. 2A″-H″). The first group includes Antp, Sex combs reduced (Scr), Ultrabithorax (Ubx), and Abdominal-B (Abd-B); these genes are normally silenced within the eye-antennal disc, but their expression is elevated within the disc upon Pc knockdown (Fig. 2A″-D″). The second group comprises labial (lab), proboscipedia (pb), Deformed (Dfd) and abdominal-A (abd-A). The expression of these Hox genes did not change significantly in Pc knockdown discs (Fig. 2E″-H″). Of these, pb and Dfd are normally expressed within the developing head epidermis and peripodial epithelium, but not the eye (Aplin and Kaufman, 1997; Diederich et al., 1991).

Because the ectopic expression of Hox genes is known to induce changes in tissue fate, such as the transformation of the antenna into a leg (Scott et al., 1983), we forcibly expressed each one within the developing eye field using ey-GAL4 (Fig. 2, Table S1) to determine whether one or more Hox genes are involved in the Pc-dependent eye-to-wing transformation. Consistent with prior reports, the overexpression of Antp alone resulted in either eyeless (83%) or headless (17%) flies, but no eye-to-wing transformations (Fig. 2A,A′). So, although Antp may contribute to the establishment of this transdetermination event, it alone is not sufficient to do so. Like the forced expression of Antp, the overexpression of several remaining Hox genes severely reduced the size of the eye field and thus resulted in predominantly eyeless adults (Fig. 2A-D,G,H). The effect of ectopic Scr on the developing eye was slightly milder with 52% of adults having small eyes (Fig. 2B,B′) whereas the rest were eyeless (Fig. 2C′,D′,G′,H′). Overexpression of lab and pb eliminated the eye-antennal disc completely (Fig. 2E,F, asterisks) and resulted in headless pharate adults (Fig. 2E′,F′) Our results show that Hox gene overexpression severely impairs eye, ocellar and head fate. Antp and Pb proteins, when ectopically expressed within the eye-antennal disc, inhibit eye development by physically binding to Eyeless (Ey) and preventing it from binding to its target enhancers (Benassayag et al., 2003; Plaza et al., 2008, 2001). Eye development may be disrupted by a similar mechanism when the other Hox genes are upregulated. These results indicate that inappropriate levels of Hox genes, which result from the loss of Pc, may destroy the pre-programmed eye fate, but do not induce the eye-to-wing transformation.

To identify putative selector genes underlying the eye-to-wing transdetermination event, we used RNA-seq to compare the transcriptome profiles of Pc knockdown eye-antennal discs (ey>Pc RNAi) with WT eye-antennal and wing discs throughout third larval instar development [96, 108, 120 and 144 h after egg lay (AEL)] (Fig. S2, Table S2). When Pc was knocked down, eye discs were developmentally delayed by approximately 24 h – the larvae pupate at 144 h instead of 120 h AEL. This delay was consistent throughout the developmental window described above, as hierarchical clustering analysis showed that the transcriptome of each Pc knockdown disc roughly clustered with that of the 24 h earlier WT eye-antennal disc (Fig. S2B). We therefore compared the transcriptome of Pc knockdown discs with WT eye-antennal and wing discs throughout ‘early’ (108 h versus 96 h AEL), ‘mid’ (120 h versus 108 h AEL) and ‘late’ (144 h versus 120 h AEL) third instar development (Fig. S3A-C). Differential expression analysis (log2 fold change, adjusted P<0.05) of these developmental windows identified 516 early genes, 230 mid genes and 233 late genes with elevated expression in both Pc knockdown eye-antennal discs and WT wing discs compared with WT eye-antennal discs (Fig. 3A-D). We then compared these candidate lists with each other to determine which genes are potentially underlying the Pc-dependent eye-to-wing transformation. Although numerous genes were specific to early, mid and late development, there were also several candidates that were elevated during multiple stages (Fig. 3E,F, Table S3). Enriched Gene Ontology (GO) analysis showed that early genes are important for lipid metabolism and biosynthetic processes (Fig. S3D). Although we found no significant enrichment clusters in the mid or late candidates, the category shared by early and mid-stage genes showed enrichment for fatty acid biosynthesis, metabolism, and translation elongation (Fig. S3E).

Our RNA-seq data showed that Antp is only enriched in Pc knockdown discs during early and late third instar development (Fig. 3F). This provides additional support for our model that Antp may be required, but is not sufficient, for the eye-to-wing transformation. We identified other wing selector genes, such as nubbin (nub), as being enriched only at late developmental stages. These genes may be pivotal for continuing the transdetermination of the eye into a wing, but are unlikely to be responsible for initiating the switch in fate. We ectopically expressed genes with available UAS lines (Fig. 3F, bold) with ey-GAL4 and found that the wing genes nub and wingless (wg) are insufficient to transform the eye into a wing (Table S1). As genes that are upregulated at just one or two stages cannot induce the eye-to-wing transformation, we hypothesized that to initiate transdetermination, and maintain the push towards a novel fate, selector genes should be consistently enriched in Pc knockdown discs at all developmental stages. Thirteen candidates comprise this category, including the Hox gene Ubx and two wing selector genes – vg and apterous (ap). We further hypothesized that reductions in Pc should lower the levels of the PcG-dependent repressive mark H3K27me3 at these loci and allow for these genes to be ectopically expressed. These factors would then be able to inappropriately activate the downstream wing gene regulatory network and force a transformation of the eye into a wing.

To test this hypothesis, we performed cleavage under targets & release using nuclease (CUT&RUN) on whole imaginal discs and determined the profile of PcG-specific marks (H3K27me3 and H2AK119Ub) in Pc knockdown eye-antennal discs as well as both WT eye-antennal and wing discs (Fig. 4, Fig. S4). We found that genome-wide H3K27me3 peaks in Pc knockdown eye-antennal discs resemble those of WT wing discs more closely than those of WT eye-antennal discs (Fig. 4A). This is consistent with the fact that the eye had transformed into a wing. We note that once the knockdown of Pc had begun to induce the eye-to-wing transformation, the ey-GAL4 driver was turned off on the transformed tissue. This is consistent with the normal wing where the ey-GAL4 driver was not expressed. As such, the increase in H2K27me3 marks seen in Pc knockdown discs is a result of the change in fate. Additionally, as we used whole imaginal discs for CUT&RUN, the H3K27me3 profile represents a mixture of cell types.

The H2AK119Ub profiles of Pc knockdown discs appeared more similar to WT eye-antennal discs (Fig. 4B). This is not surprising as the role that ubiquitylation plays in PcG repression is not fully understood (Blackledge et al., 2014; Cohen et al., 2020; Cooper et al., 2014; Kahn et al., 2016; Kalb et al., 2014; McGinty et al., 2014; Pengelly et al., 2015; Tamburri et al., 2020). An examination of peak profiles within candidate genes expressed throughout third instar development (Fig. 4C) showed that the 13 genes fall into three categories: (1) elevated transcript levels and diminished H3K27me3, (2) elevated transcript levels but minimally changed H3K27me3, and (3) candidates with generally low transcript counts (>5). As eight of the 13 genes showed elevated transcript levels and diminished levels of H3K27me3 in Pc knockdown discs compared with WT eye-antennal discs (category 1), this supports our hypothesis that reduction of Pc expression prevents either the maintenance or propagation of H3K27me3 and this ultimately allows candidates to be ectopically expressed and alter the fate of the eye-antennal disc.

The most promising candidate is vg, as forced expression of this gene alone in the eye field induces an eye-to-wing transformation similar to what we observe in Pc knockdown discs (Kim et al., 1996; Simmonds et al., 1998) (Fig. 5A,D). When vg was forcibly expressed, wing outgrowths were seen along the edges of all four quadrants (anterior, posterior, dorsal and ventral) of the eye at varying frequencies (Fig. S5). This contrasts to Pc knockdown discs, in which ectopic wing tissue was only seen in the dorsal portion of the eye. This ability remains unique to vg as overexpression of other wing selector genes (ap, nub and wg) did not induce any wing-like outgrowths (Table S1). During normal development, vg is expressed in a broad stripe within the wing pouch but is entirely absent from the eye-antennal disc (Fig. 5B,C) (Kim et al., 1996; Williams et al., 1991; Zhu et al., 2018). In transformed Pc knockdown discs, vg transcripts were differentially expressed throughout third instar development and Vg protein was localized exclusively to the pouch region of the ectopic wing (Fig. 5E,F). The presence of two functional PREs within the vg locus (Ahmad and Spens, 2019; Herzog et al., 2014; Okulski et al., 2011; Ringrose et al., 2003; Srinivasan and Mishra, 2020) suggests that it is repressed by PcG proteins during normal development of the eye-antennal disc (Fig. 5G, asterisks).

Our CUT&RUN analysis (Fig. 5G) further supports this hypothesis as we found high levels of H3K27me3 at the vg locus in the eye disc (where vg is normally repressed). Conversely, in the wing disc (where vg is expressed across the wing pouch) the vg locus was decorated with significantly lower levels of H3K27me3. In Pc knockdown eye-antennal discs, H3K27me3 levels at the vg locus were drastically reduced and appeared similar to wing discs. Although the chromatin profile H2AK119Ub was also altered in Pc knockdown discs, it did not truly resemble the profile of either eye or wing discs. These results demonstrate that when Pc levels are reduced, the repressive chromatin landscape of vg is diminished, and this results in it being ectopically expressed within the transdetermining eye-antennal disc.

Although it is known that ectopic Vg in the eye (Kim et al., 1996; Simmonds et al., 1998) can yield wing outgrowths, the question of how Vg is able to convert this tissue remains. In the wing disc, its binding partner, Sd, is enriched within a similar stripe across the pouch. The Vg–Sd complex promotes wing fate by activating downstream wing fate genes (Garg and Bell, 2010; Halder and Carroll, 2001; Halder et al., 1998). Sd also promotes growth of the wing pouch by forming a complex with the Yorkie (Yki) transcriptional activator (Koontz et al., 2013; Wu et al., 2008). In the developing eye (where Vg is absent), Sd is present at low levels throughout the disc and promotes its growth via the Yki–Sd complex (Campbell et al., 1992; Guss et al., 2013; Koontz et al., 2013; Wu et al., 2008). A possible mechanism underlying the eye-to-wing transformation is that Vg, which is now present in the eye because of knocking down Pc, competes with Yki for binding to Sd. This could recreate the genetic conditions that are present within the wing disc and, as a result, the eye field is reprogrammed into a wing. We knocked down sd in animals that were overexpressing vg (Fig. 6A-B′) and, as expected, the adult eye no longer transformed into wing tissue and instead had a malformed eye that appeared like that of sd knockdown animals (Fig. 6C,C′). We conclude that the eye-to-wing transformation seen in Pc knockdown animals is due to ectopic Vg taking advantage of Sd, which is already present in the eye-antennal disc. The newly formed Vg–Sd complex then forces the eye field to transdetermine into a wing by initiating activation of the downstream wing gene regulatory network.

To test further whether formation of the Vg–Sd complex drives the Pc-dependent eye-to-wing transformation, we genetically disrupted the complex by knocking down Pc in vg¹ or sd¹ loss-of-function mutant backgrounds. We expected that loss of either gene would prevent the eye from being transformed into a wing and possibly restore eye development. When Pc was knocked down in either mutant background, the eye-to-wing transformation was suppressed and levels of ectopic Antp were further elevated (Fig. 6D,E). Not only does this result support a cardinal role for Vg–Sd in eye-to-wing transformation, but it also further demonstrates that Antp is not driving this reprogramming event. Reduction of both vg and Pc (vg¹; DE>Pc RNAi) resulted in early pupal lethality (Fig. 6D′), whereas the combined reduction of sd and Pc (sd¹; ey>Pc RNAi) resulted in pharate lethal adults with ‘tumorous’ head structures (Fig. 6E′).

Surprisingly, vg¹; DE>Pc RNAi and sd¹; ey>Pc RNAi discs undergo dramatic hyperplastic growth. This is reminiscent of eye-antennal discs in which other PcG factors have been depleted (Bunker et al., 2015; Classen et al., 2009; Loubiere et al., 2016; Martinez et al., 2009). Both vg¹; DE>Pc RNAi and sd¹; ey>Pc RNAi discs showed patterns of tissue organization that indicate it is being specified into several types of imaginal discs. We used immunohistochemistry and RNA-seq to determine whether both discs have adopted fates outside of the eye-antennal disc. We first treated the two types of hyperplastic discs with Wingless (Wg), Ubx, Distal antenna (Dan) and Dachshund (Dac) antibodies to distinguish between eye, antennal, wing, leg and haltere imaginal discs. The spatial distribution of Wg protein is an excellent indicator of wing versus haltere fate. In both wing and haltere discs, wg is expressed in a stripe across and around the pouch; however, the shape of the pouch is oval in the wing disc and circular in the haltere disc (Prasad et al., 2003). Differences in the shape of the wg expression pattern makes it relatively easy to distinguish these two tissues from each other (Fig. 7A). We also used Ubx antibodies to distinguish between wing and haltere discs as Ubx is distributed within a few peripodial cells of the wing pouch but is expressed in a large swathe of cells within the haltere disc (Fig. S6G-I) (Weatherbee et al., 1998). The spatial distribution of Wg protein can normally be used to distinguish the antenna from the leg as it is differentially expressed within these discs (Fig. 7A). However, because the eye-antennal disc in vg1; DE>Pc RNAi and sd1; ey>Pc RNAi has undergone hyperplastic growth, it is grossly deformed, and wg expression alone cannot be reliably used to identify regions of the disc that might have acquired these fates.

To better distinguish between putative antennal and leg tissue, we instead treated hyperplastic discs with antibodies that recognize Dachshund (Dac). Dac protein is distributed in several concentric rings within the leg disc whereas its spatial expression pattern within the antennal field resembles a signet ring (Fig. 7D) (Mardon et al., 1994). Dan antibodies can also distinguish between antenna and leg discs as Dan protein is expressed throughout the antennal disc but completely absent from the leg disc (as well as the wing and haltere disc, Fig. S6A-C) (Emerald et al., 2003). Thus, the dual presence of both Dac and Dan indicates an antennal disc whereas the sole presence of Dac would indicate a leg imaginal disc (Fig. S6B,C,E,F). The spatial distribution of Wg, Ubx, Dac and Dan proteins within both vg¹; DE>Pc RNAi and sd¹; ey>Pc RNAi eye-antennal discs indicated that heterogeneity exists within the hyperplastic tissue (Fig. 7B,C,E,F). Based on these patterns, we determined that an individual disc can simultaneously contain eye, antenna, leg, wing and haltere tissue (Fig. 7G,H). We calculated the frequency of each type of transformation by dividing the number of each specific transformation by the total number of transformations that we could identify in 30 discs. The percentages of transformations were consistent across the different antibodies. Interestingly, whereas the eye and antennal fields have the capacity to be transformed into all possible imaginal disc fates, the eye-to-haltere and antenna-to-haltere transformations were only seen in sd¹; ey>Pc RNAi eye-antennal discs.

Using RNA-seq, we determined the transcriptomes of vg¹; ey>Pc RNAi (vgPcKD) and sd¹; ey>Pc RNAi (sdPcKD) eye-antennal discs and compared these profiles back to those of WT and ey>Pc RNAi (PcKD) eye-antennal discs (Fig. 7I, Table S4). Comparing the lists of differentially expressed genes with each other allowed us to identify potential molecular signatures underlying unique changes in tissue specification and proliferation. vgPcKD and sdPcKD discs shared the differential expression of 648 genes (Fig. 7I, Tables S4-S6). This is expected as both types of discs showed hyperplastic growth and undergo transdetermination events (Fig. 7B-H). Interestingly, the percentage of unique fate transformations within the eye and antennal fields differed between vgPcKD and sdPcKD discs and the emergence of haltere tissue appeared only within sdPcKD discs (Fig. 7B-H). As such, both types of mutant discs also contained unique sets of differentially expressed genes. For example, vgPcKD discs were characterized by the differential expression of a unique set of 438 genes, whereas the expression of a completely different set of 439 genes was specifically altered in sdPcKD discs (Fig. 7I, Tables S4-S6). These phenotypic and molecular differences were surprising as disrupting the Vg-Sd complex with either vg or sd loss-of-function mutants was expected to result in similar mutant phenotypes and gene expression changes. The differences could be explained by the fact Sd and Vg are not obligate partners (Deng et al., 2009; Vissers et al., 2018; Zhang et al., 2017); therefore, mutating each one likely disrupts different sets of biochemical complexes. Despite the phenotypic differences between vgPcKD, sdPcKD and PcKD discs, all three mutant tissues shared the differential expression of 277 genes (Fig. 7I, Tables S4-S6). GO analysis of the elevated genes identified enrichment clusters for imaginal disc development, morphogenesis, and cell fate specification (Fig. 7J).

To infer the trends in gene expression behind the varying disc phenotypes, we categorized each upregulated and downregulated gene based on its primary annotated function (Tables S5 and S6). We observed reductions in the expression of selector genes that are normally tasked with promoting the fate of the eye, antenna and head epidermis. We also identified the complementary upregulation of selector genes that function outside of the eye-antennal disc. Additionally, complex changes in the expression of several Hox genes (both up- and downregulation) were observed. Finally, we observed changes in the expression of genes that encode known members of several cell proliferation pathways. Although the roles that these genes play in transformations within the eye-antennal disc were not directly tested here, we predict that the dramatic increase in tissue size and changes in tissue fate seen in vgPcKD and sdPcKD discs are caused, in part, by the combined changes in the expression of these factors.

Our data suggest a possible mechanism that explains why the eye predominantly transforms into a wing and offer the surprising observation that, under the right circumstances, the eye will adopt a multitude of other imaginal disc fates. We compared the transcriptome and epigenome profiles of WT eye-antennal discs, WT wing discs and Pc knockdown eye-antennal discs with each other and identified several genes that have higher transcript counts and lower H3K27me3 levels throughout the third larval instar (Figs 3 and 4). Of these, the vg gene merited special consideration as it is a known wing selector gene (Kim et al., 1996; Williams et al., 1991), is transcriptionally silent in the WT eye disc (Zhu et al., 2018), contains two functional PRE elements (Ahmad and Spens, 2019; Okulski et al., 2011; Ringrose et al., 2003), and induces an eye-to-wing transformation when ectopically expressed within the eye field (Fig. 5) (Kim et al., 1996; Simmonds et al., 1998). PcG-mediated regulation of additional genes is important for maintaining eye fate because the expression of H3K27M histone proteins (which cannot be methylated) interferes with and blocks Vg-mediated eye-to-wing transformation (Ahmad and Henikoff, 2021).

We note that reducing the expression of several PcG factors, such as Pc, Sce and Scm, has effects outside of the developing eye and induces antenna-to-leg transformations. This is in part due to the fact that the ey-GAL4 driver is expressed throughout the entire eye-antennal disc during late embryogenesis and the first larval instar (Kumar and Moses, 2001) and thus drives the knockdown of PcG factors throughout the entire eye-antennal disc for a portion of its development. Previously, we have shown that the requirement for Pc in the suppression of Antp expression within the eye-antennal disc takes place during these two developmental stages (Zhu et al., 2018). This is the likely developmental window for the antenna-to-leg transformations that are associated with Pc knockdown. Although not tested here, the requirement for Sce and Scm to suppress Antp expression may be during the same developmental window. It is unclear, however, why Antp expression is only strongly seen within the antenna of Sce knockdowns.

We investigated the mechanism by which ectopic Vg alters the fate of the eye. Vg is a transcriptional activator (MacKay et al., 2003) that binds to the DNA-binding protein Sd and promotes wing formation (Halder et al., 1998; Simmonds et al., 1998). The binding of Vg to Sd appears to alter the enhancer specificity of the Sd protein and forces it to bind only to wing-specific target genes (Garg and Bell, 2010; Halder and Carroll, 2001). This implies that in tissues such as the eye, Sd will bind to a distinct set of genes that are not involved in wing formation. Within the eye, the transcriptional activator Yki binds to Sd (Yki–Sd) to relieve the default repression of growth-promoting genes and to promote re-entry of quiescent cells into the cell cycle during tissue regeneration (Koontz et al., 2013; Meserve and Duronio, 2015; Wu et al., 2008). Our model is that ectopic activation of Vg within the eye (as a result of reductions in Pc levels) allows for Vg to bind Sd and redirect it away from cell proliferation genes and onto wing fate-promoting genes instead (Fig. 8A,B). A similar model has been proposed for the wing itself where the formation of Vg–Sd complexes within the pouch indirectly reduces tissue growth by titrating Sd away from Yki (Koontz et al., 2013). Our model is supported by the fact that reducing sd expression prevents the Vg-mediated eye-to-wing transformation (Fig. 6). Our findings suggest that Pc mediates the eye versus wing decision by selectively repressing vg within the developing eye primordium while allowing its expression within the developing wing pouch.

How might vg expression be activated in the absence of Pc repression within the eye field? Several studies have shown that, in the developing wing, vg expression is activated by the Notch and Wingless signaling pathways (Bernard et al., 2009; Couso et al., 1995; Djiane et al., 2014; Koelzer and Klein, 2006; Neumann and Cohen, 1996; Zecca and Struhl, 2007a,b). Similarly, vg expression is activated within the eye when Antp and the Notch pathway are simultaneously activated, but not when Antp is ectopically expressed alone (Katsuyama et al., 2005; Kurata et al., 2000). Based on these prior studies, we propose that the Notch and Wingless pathways, which play important roles in the developing eye (Baker, 2007; Silver and Rebay, 2005; Tio et al., 1996; Voas and Rebay, 2004), could activate vg expression in the absence of Pc-mediated repression. Under these conditions, the formation of the Vg–Sd complex would shift the balance away from eye development (Yki–Sd) and towards wing formation (Vg–Sd).

To test our model further, we used vg and sd loss-of-function mutants to disrupt the Vg–Sd complex while simultaneously reducing Pc expression levels. Our expectation was that the eye-to-wing transformation would be simply blocked, and eye development might even be restored. Although the percentage of eye-to-wing transformations was reduced, we surprisingly observed that the mutant eye-antennal discs underwent dramatic levels of hyperplastic growth (Fig. 6). This tumor-like growth is reminiscent of discs in which other PcG members have been depleted (Classen et al., 2009; Medina et al., 2021) or when epithelial polarity is lost within the eye-antennal disc (Bunker et al., 2015; Enomoto et al., 2021; Pagliarini and Xu, 2003; Wu et al., 2010). Furthermore, we made the unexpected discovery that the vgPcKD and sdPcKD hyperplastic discs do not just consist of amorphous tissue, but instead are regionally specified as other imaginal discs (Figs 7 and 8C,D). Our bulk RNA-seq analysis of these mutant eye-antennal discs suggests that once the primary push towards a wing fate is eliminated (through disruption of the Vg–Sd complex), cell proliferation genes likely drive hyperplastic growth while changes in Hox and selector genes disrupt the pre-programmed fate and force the tissue to transdetermine into several new imaginal disc fates, respectively (Figs 7 and 8C,D, Tables S5 and S6).

These dramatic changes in proliferative state and fate are the result of removing multiple genetic elements; in this case, we are combining Pc knockdown with vg or sd loss-of-function mutants. How can this inform normal development and potential disease states? In terms of the former, our results suggest that the developing eye-antennal disc is more multipotent than previously thought and undergoes a hierarchy of developmental decisions. In particular, we show that the eye and antennal portions of the disc have the capacity to adopt several alternate disc fates, including those of the wing, haltere, leg and antenna. This is consistent with other imaginal discs (Hadorn, 1968, 1978). However, several of these developmental choices have remained hidden and have only been revealed through the use of simultaneous gene knockdowns and loss-of-function mutants. Support for this approach comes from other studies in which the simultaneous knockdown of multiple transcription factors and/or epigenetic regulators also revealed novel changes in the fate of the eye-antennal disc (Ordway et al., 2021; Palliyil et al., 2018). When considering the onset of disease, it is well documented that mutations in multiple genes or the combination of multiple risk factors are the underlying cause of many disease states, such as cancer, schizophrenia and Parkinson's disease (Al Hajri et al., 2020; Cabezudo et al., 2020; Maynard et al., 2001). The simultaneous knockdown of multiple genes can, in certain instances, also serve as models for human disease. For example, individual knockdowns of the Drosophila genes prune (pn) and Killer of prune (K-pn; abnormal wing discs, awd) have effects on the color of the eye, but their combined loss results in a dramatically different phenotype that mimics human neurofibromatosis (Biggs et al., 1988; Hackstein, 1992). The unexpected hyperproliferation and diversity in tissue fates that we observe in the vgPcKD and sdPcKD discs is surprisingly reminiscent of the intratumor heterogeneity that is seen in a wide range of human tumors (Ju, 2021; Li et al., 2022; McGranahan and Swanton, 2017). As such, the induction of hyperplastic imaginal discs could serve as tractable model systems for elucidating the genetic and molecular mechanisms that underly tumor heterogeneity.

The following fly stocks for experiments shown in the main text were obtained from the Bloomington Drosophila Stock Center (BDSC), Indiana University, Bloomington, USA: ey-GAL4 8221, UAS-Pc RNAi 33964, UAS-Scr 7302, UAS-Antp 7301, UAS-Ubx 911, UAS-Abd-B 913, UAS-lab 7300, UAS-pb 7298, UAS-abd-A 912, UAS-Dfd 7299, UAS-pho RNAi 42926, UAS-Sfmbt RNAi 28677, UAS-calypso RNAi 56888, UAS-Sce RNAi 67924, UAS-Scm RNAi 55278, UAS-vg 37296, vg¹ 432, sd¹ 1027, UAS-sd RNAi 29352, c311-GAL4 5937 and DE-GAL4 (Georg Halder, Katholic University, Leuven, Belgium). Other screening stocks are noted in Table S1. All stocks listed in Table S1 are from the BDSC. All crosses were carried out at 25°C. For scoring of all eye-antennal discs and adult phenotypes, n=30.

The following antibodies were used: rat anti-Elav [Developmental Studies Hybridoma Bank (DSHB), 7E8A10, 1:100], mouse anti-Antp (DSHB, 8C11, 1:100), mouse anti-Wg (DSHB, 4D4, 1:800), mouse anti-Dac (DSHB, 2-3, 1:5), mouse anti-Ubx (DSHB, FP3.38, 1:100), rabbit anti-Vg (K. Guss, Dickinson College, Carlisle, PA, USA; 1:100), FITC-conjugated donkey anti-rabbit (Jackson ImmunoResearch, 711-095-152, 1:100), FITC-conjugated donkey anti-rat (Jackson ImmunoResearch, 712-095-153, 1:100), rat anti-Dan (J. Curtiss, New Mexico State University, Las Cruces, NM, USA; 1:500), Cy3-conjugated donkey anti-mouse (Jackson ImmunoResearch, 715-165-151), Alexa Fluor 647-conjugated phalloidin (Thermo Fisher Scientific, A22287, 1:20), and Rhodamine phalloidin (Thermo Fisher Scientific, R415, 1:100). Drosophila eye-antennal imaginal discs were prepared as previously described (Spratford and Kumar, 2014). Fluorescent images were taken using a Zeiss Axioplan II compound microscope and processed using Fiji/ImageJ (Schindelin et al., 2012) and Adobe Photoshop software. Adult flies were imaged with either a Zeiss Discovery V12 or a Leica M205FA stereo microscope.

Samples were collected from eye-antennal or wing discs throughout third instar development (84-144 h AEL). Fifty eye-antennal imaginal discs were dissected from ey>Pc RNAi or w¹¹¹⁸ larvae, 20 wing discs from w¹¹¹⁸ larvae, or 25 eye-antennal discs from vg¹; DE>Pc RNAi or sd¹; ey>Pc RNAi larvae were dissected as previously described (Spratford and Kumar, 2014). Total RNA was extracted with the RNeasy Plus Mini Kit (QIAGEN, 74134) following the kit protocol. A polyA-strand specific library was generated using 0.3 μg RNA with the Illumina TruSeq Stranded mRNA library prep kit. Libraries were prepared and sequenced by the Center for Genomics and Bioinformatics (CGB, Indiana University, Bloomington, IN, USA).

Chromatin was extracted and purified from ten third instar eye-antennal or wing imaginal discs as previously described (Weasner et al., 2023). The following antibodies were used: rabbit anti-trimethyl-histone H3 (Lys27) (C36B11) (Cell Signaling Technology, 9733, 1:51), rabbit anti-Ubiquityl-Histone H2A (Lys119) (D27C4) (Cell Signaling Technology, 8240, 1:51), and normal rabbit IgG (Cell Signaling Technology, 2729, 1:51). Cleaved chromatin fragments purified with the MinElute PCR Purification Kit (QIAGEN, 28004) following the kit protocol.

Read quality was assessed with fastqc (http://www.bioinformatics.babraham.ac.uk/projects/fastqc;v0.11.9). Genome indices were generated using STAR (Dobin et al., 2013) and were aligned to the dm6 Drosophila genome (with the additional setting ‘--genomeSAindexNbases 13’ for Drosophila). Aligned reads were counted using the Subread (Liao et al., 2019) function ‘featureCounts’ with the settings ‘-F GTF -t exon -g gene_id --minOverlap 10 --largestOverlap --primary -s 2 -T’. bigWig files of RNA-seq counts were generated with the deepTools (Ramírez et al., 2016) (v3.5.1) function ‘bamCoverage’ with the settings ‘--normalizeUsing CPM -bs 1 --smoothLength 25 --filterRNAstrand [forward or reverse] -p 16’. Downstream analysis was performed in Rstudio (v4.2.1). Differential expression analysis was performed with DESeq2 LO and heatmaps were generated with ComplexHeatmap (Gu et al., 2016). Differentially expressed transcripts were filtered with the dplyr package (https://github.com/tidyverse/dplyr) to select for genes ‘enriched in WD′ (in ead_vs_wd_change) and ‘enriched in absence of Pc’ (in pckd_vs_ead_change). Candidate overlap between each time point was then compared using VennDiagram (Chen and Boutros, 2011). GO analysis was performed with the clusterProfiler (Yu et al., 2012) ‘eGO’ function (settings: OrgDb=‘org.Dm.eg.db’; ont=‘BP’; pAdjustMethod=‘fdr’; keyType=‘SYMBOL’). Additional scripts for downstream analysis can be found in the supplementary information.

Read quality was assessed with fastqc (http://www.bioinformatics.babraham.ac.uk/projects/fastqc; v0.11.9). Genome indices were generated using Bowtie2 (Langmead et al., 2009) and were aligned to the Drosophila dm6 genome and Escherichia coli EB1 genomes (with the settings ‘min_fragment=10 max_fragment=700’). The quality and consistency of raw aligned reads were assessed with the deeptools (v3.5.1) functions ‘multiBamSummary’ and ‘plotCorrelation’ (with the settings ‘--whatToPlot heatmap --corMethod spearman –plotNumbers’). The number of E. coli aligned reads were accessed with the samtools (Danecek et al., 2021) (v1.15.1) command ‘view -c -F 4’ and used to generate an appropriate scale factor with the calculation: 10,000/[# E. coli aligned reads]. Normalized bigWig and bedgraph files of Drosophila CUT&RUN reads were generated with the deepTools (Ramírez et al., 2016) (v3.9/3.5.1) function ‘bamCoverage’ with the corresponding ‘--scaleFactor and --outFileFormat [bigwig or bedgraph]’ setting. Initial visualization of bigWig reads was carried out in Integrative Genomics Viewer (v2.8.0); figures of chromatin tracks were generated using Gviz (Hahne and Ivanek, 2016) in Rstudio (v4.2.1). Peaks of normalized bedgraph tracks were called with the SEACR (Meers et al., 2019) web interface (‘norm’ settings). deepTools functions were used to generate all downstream analysis: Peak matrices were generated with ‘computeMatrix reference-point –referencePoint center –missingDataAsZero -a 2000 -b 2000’. The heatmap of CUT&RUN peaks was generated from matrix files with ‘plotHeatmap --matrixFile --whatToShow ‘plot, heatmap, and colorbar’. Additional scripts for downstream analysis have been deposited in GitHub (https://github.com/brownhe717/Brown_etal_2023).

We are grateful to Kirsten Guss for the anti-Vg antibody, to the Developmental Studies Hybridoma Bank (DSHB) for other primary antibodies, to Georg Halder for the DE-GAL4 fly strain, to the Bloomington Drosophila Stock Center for all remaining fly strains, to both Gabe Zentner and Robert Policastro for technical support with bioinformatics, to the Light Microscopy Imaging Center (LMIC) for use of the Leica M205FA stereo microscope, and to the Center for Genomics and Bioinformatics (CGB) for the generation of sequencing libraries and next-generation sequencing.