De novo recruitment of Polycomb-group proteins in Drosophila embryos

Polycomb group proteins (PcG) are conserved epigenetic regulators that were first identified as repressors of Drosophila Hox genes (Lewis, 1978). Their known targets have since been expanded to include numerous developmental and cell cycle regulators in widely divergent organisms. Mammalian PcG proteins play central roles in maintaining stem cell pluripotency and regulating lineage-specific differentiation (Comet et al., 2016; Oktaba et al., 2008). Their misexpression in mammals contributes to a wide range of cancers (Laugesen et al., 2016; Richly et al., 2011).

Drosophila PcG proteins function within three major multimeric complexes: Polycomb Repressive Complex 1 (PRC1), PRC2 and Pleiohomeotic (Pho) Repressive Complex (PhoRC) (Simon and Kingston, 2013). PhoRC comprises the DNA-binding PcG protein Pho and Sexcomb-on-midleg (Scm)-related gene containing four mbt domains (Sfmbt) (Klymenko et al., 2006). Pho binds directly to sites within Polycomb Response Elements (PREs), which serve as docking sites for other PcG proteins (Kassis and Brown, 2013). Multiple mechanisms have been proposed by which PhoRC may help recruit or stabilize target binding by PRC1 and PRC2 (Simon and Kingston, 2013). Pho has been shown to interact directly with subunits of both PRC1 and PRC2 (Mohd-Sarip et al., 2002, 2005; Wang et al., 2004). Sfmbt directly interacts with Scm, which can interact with both PRC1 and PRC2 (Frey et al., 2016; Kang et al., 2015). The catalytic subunit of PRC2, Enhancer of zeste [E(z)] trimethylates histone H3 at lysine 27 (H3K27me3) (Cao et al., 2002; Czermin et al., 2002; Müller et al., 2002). A core component of canonical PRC1, Polycomb (Pc), contains a chromodomain that specifically binds to H3K27me3, facilitating recruitment and/or stable binding of canonical PRC1 (Cao et al., 2002; Fischle et al., 2003; Min et al., 2003; Wang et al., 2004). Alternative models have been proposed in which PRC1 either binds to targets completely independently of PRC2 or recruits, or stabilizes binding by, PRC2, possibly by a mechanism involving PRC2 affinity for noncanonical PRC1-dependent monoubiquitylated histone H2A (reviewed by Dorafshan et al., 2017).

PcG proteins do not initiate transcriptional repression, but rather take over repression from gene-specific transcription factors (Jones and Gelbart, 1990; Riising et al., 2014; Simon et al., 1992; Struhl and Akam, 1985). The mechanisms by which PcG proteins recognize the repressed state are not understood. Initial recruitment of PcG proteins may be governed by multiple factors, including the presence of gene-specific transcription factors, the local chromatin landscape and the transcriptional state (Guertin and Lis, 2010; Paro, 1990; Yuan et al., 2012). Once PcG-mediated repression is established, it may be maintained through an indefinite number of cell cycles in the absence of the initial repressors. However, as demonstrated in stem cells, PcG-mediated repression may be reversed to allow selective activation of target genes (Jepsen et al., 2007).

The Drosophila gap gene giant (gt) has been previously identified as a PcG target (Pelegri and Lehmann, 1994). Its expression, like that of other gap genes, starts at the syncytial blastoderm stage and is initially regulated by a combination of maternally and zygotically expressed transcription factors (Eldon and Pirrotta, 1991; Lott et al., 2011). In wild-type embryos, PcG repression of gt is largely redundant with gene-specific transcription factors during early stages of embryogenesis, but gt ectopic expression is observed in Pc mutants by mid-embryogenesis (Nègre et al., 2006). Given what is known about the maintenance of repression by PcG at Hox genes, it is likely that initial recruitment of PcG complexes at gt temporally overlaps binding by these transcription factors in order to ensure repression upon the eventual decay of the local transcription factors.

Following fertilization, Drosophila nuclei undergo a rapid series of 13 synchronous mitotic cycles within a syncytium. The first extended interphase occurs after nuclear cycle 13 (nc13) mitosis. This coincides with the midblastula transition (MBT), which is marked by the onset of widespread zygotic transcription. Through nc13, chromatin accessibility is limited by the rapid pace of nuclear divisions that is overcome at a small number of pre-MBT genes with the help of Zelda (Zld), a pioneer factor (Foo et al., 2014; Moshe and Kaplan, 2017; Schulz et al., 2015; Sun et al., 2015). A checkpoint during midblastula transition coincides with RNA polymerase II binding and increased chromatin accessibility that allows for zygotic programs to set in (Blythe and Wieschaus, 2015, 2016). In concordance with this, Li et al. (2014) proposed a transition from a naive chromatin state in which histones are largely unmodified, followed by histone acetylation at limited genomic sites, and finally the establishment of domains of histone methylation, including H3K27me3, during nc14. Thus, the incorporation of epigenetic mechanisms into transcriptional regulation appears to only occur after cycles of DNA replication and mitosis have sufficiently slowed.

The molecular mechanisms by which PcG proteins initially distinguish between active and repressed target genes and the molecular steps that lead to establishment of PcG-mediated repression remain elusive. This is due to the technical difficulty of obtaining homogenous populations of cells in which a PcG target gene is uniformly transitioning from repression by gene-specific transcription factors to PcG-mediated repression. As a result, the focus of most research on the in vivo activities of PcG proteins has been on the maintenance phase of PcG silencing. These studies have used imaginal discs from Drosophila larvae or cultured cells to obtain populations of cells in which a PcG target gene is uniformly repressed. More recently, experimental approaches have been developed to overcome problems of cell heterogeneity, including single cell assays (Rotem et al., 2015) and FACs sorting, and isolating cells from particular embryonic parasegments or specific tissues (Bowman et al., 2014; Bonn et al., 2012; Gisselbrecht et al., 2013). Although such approaches have been helpful in addressing the problem of cell heterogeneity, one still faces the problem of the inability to capture temporally synced cell populations that are transitioning from repression by gene-specific transcription factors to PcG proteins.

In this study, to circumvent the problem of cell heterogeneity, we have used mutant alleles of three maternal effect loci to produce Drosophila embryos in which gt is ubiquitously repressed by maternally expressed Hunchback (Hb). To study the transition of gt repression from gene-specific transcription factors to PcG proteins, we use temporally synced embryos in chromatin immunoprecipitation (ChIP) assays to examine the spatial and temporal recruitment of PcG proteins, as well as Hb and a gt activator, Caudal (Cad). Our findings are consistent with a model in which PRC2-dependent H3K27 trimethylation precedes and is required for de novo recruitment of PRC1.

After the ninth nuclear cycle (nc9), Drosophila nuclei migrate to the periphery of the embryo forming the syncytial blastoderm. Synchronous mitotic cycles continue until cell membranes separate nuclei during cycle 14, forming the cellular blastoderm. After nc13, a prolonged interphase coincides with the MBT and zygotic genome activation (ZGA). Although a small number of genes are transcribed pre-MBT, prior to this point maternally loaded mRNA and proteins are predominantly responsible for early developmental events (Lasko, 2012).

Transcription of gt begins at nc12 (Eldon and Pirrotta, 1991; Lott et al., 2011). It is initially controlled by two maternally expressed activators: Bicoid (Bcd) (Driever et al., 1989) and Cad (Rivera-Pomar et al., 1996). Bcd forms an anterior to posterior gradient within the embryo. Cad forms a posterior to anterior gradient owing to translational repression of uniformly distributed maternal cad mRNA by Bcd (Dubnau and Struhl, 1996; Rivera-Pomar et al., 1996). gt is negatively regulated by maternally expressed Hunchback (Hb_mat), which is restricted to the anterior half of the embryo due to Nanos (Nos)-dependent translational repression of hb mRNA (Tautz, 1988; Wharton and Struhl, 1991). Proper posterior localization of nos mRNA in the oocyte requires the activities of several proteins, including Oskar (Osk) (Ephrussi et al., 1991). As a result of the combined activities of maternally expressed Bcd, Cad and Hb, gt is initially expressed in two broad stripes. By cellular blastoderm, this pattern of gt expression is refined due to the combined activities of these maternal regulators and zygotically expressed transcription factors, including Hb (Hb_zyg) (Niessing et al., 1997).

Based on this understanding of the maternal regulation of gt and the observation that gt is a PcG target gene (Pelegri and Lehmann, 1994), we developed a genetic system in which gt is ubiquitously repressed and in which the PcG is required to maintain its repression following degradation of Hb_mat. bcd osk torsolike (tsl) females produce embryos that lack the anterior gt activator Bcd and ubiquitously express the gt repressor Hb_mat due to lack of localized Nos (Fig. 1A) resulting in ubiquitous repression of gt (Fig. 1B). Zygotic expression of Hb requires its anterior activator Bcd and the terminal system, which is disabled in embryos from tsl females (Tautz, 1988; Driever et al., 1989). Therefore, following degradation of ubiquitously distributed Hb_mat, Hb_zyg fails to replace it, yet gt is not expressed (Fig. 1C,D). At syncytial blastoderm, Cad_mat is also ubiquitously translated in embryos from bcd osk tsl mothers due to lack of Bcd activity, but apparently is unable to activate gt expression due to the repressive effects of Hb_mat. Zygotic expression of Cad also is dependent on the terminal system (Mlodzik and Gehring, 1987). Therefore, in embryos derived from bcd osk tsl females, both Hb_mat and Cad_mat are ubiquitously expressed, but neither Hb nor Cad is zygotically expressed (Fig. 1A,C, Fig. S1).

In order to confirm that the maintenance of gt repression in the absence of Hb_zyg is PcG dependent, embryos were collected from E(z)²⁸ bcd osk tsl/E(z)⁶³ bcd osk tsl females that were crossed to E(z)⁶¹ males. E(z)²⁸ and E(z)⁶¹ are temperature-sensitive alleles (Jones and Gelbart, 1990; Phillips and Shearn, 1990). E(z)⁶³ is a null allele (Jones and Gelbart, 1990). At permissive temperature (18°C), PcG activity is essentially wild type and gt repression was maintained. However, at restrictive temperature (29°C) lack of E(z) activity resulted in gt derepression (Fig. 1E).

We performed a time-course series of ChIP experiments using embryos produced by bcd osk tsl females in order to examine the sequence of events that occur as control of gt repression transitions from Hb to the PcG. Embryos were collected for 30 min and then allowed to continue developing until they reached the ages described in Table 1. Following formaldehyde fixation, embryos were carefully hand sorted in order to remove older embryos (Fig. S2). The first three 30 min time points approximate nuclear cycles 10-12, 13 and the first half of nc14 (14a). The fourth time point, nc14b, includes cellular blastoderm and early gastrulating embryos.

The wild-type pattern of gt expression in blastoderm stage embryos is controlled by four enhancers located upstream of the gt promoter (Fig. 2A) (Berman et al., 2002; Schroeder et al., 2004). gt_(-1) [−0.05 to −1.3 kb upstream of the transcription start site (TSS)] produces the posterior stripe and the major anterior stripe, gt_(-3) (−1.3 to −2.5 kb) produces the posterior gt stripe, gt_(-6) (−4.3 to −6.5 kb) produces the anterior tip of expression that is apparent by late cellular blastoderm and gt_(-10) (−8.8 to −10.5 kb, not shown here) produces the anterior stripe. We have previously mapped two gt PREs, PRE1 and PRE2, to regions that overlap gt enhancers gt_(-1) and gt_(-6), respectively (Abed et al., 2013) (Fig. 2A). Early gt expression is dictated by gradients of maternally expressed transcription factors that bind to sites within the gt_(-3) and gt_(-10) enhancers (Hoermann et al., 2016). Cad_mat is the only gt_(-3) activator, whereas Hb_mat is responsible for repression of gt at this regulatory element and controls the anterior boundary of the posterior stripe. Both Hb_mat and Cad_mat are ubiquitously distributed in bcd osk tsl syncytial blastoderm embryos (Fig. 1, Fig. S1; Niessing et al., 1997). As expected, ChIP for Hb and Cad, showed they were primarily localized at PCR region 6, which is within the gt_(-3) enhancer, in nc10-12 embryos. In nc13 embryos, the signals for both Hb and Cad increased further at region 6 and were weakly detected near the promoter (region 4) before sinking to near-background levels at nc14a. These observations are consistent with the known regulation by Hb and Cad of gt_(-3), and the degradation of these maternally expressed proteins by nc14.

gt PRE1 encompasses the promoter and includes PCR region 4. PRE2 is located ∼6 kb upstream of the promoter and includes PCR region 9 (Fig. 2; Abed et al., 2013). Pho and Sfmbt were detected at PRE1 and PRE2 as early as nc10-12 (Fig. 3). Pho signals gradually increased at PRE1 from nc10-12 to nc14b. Sfmbt signals at PRE1 fluctuate, but remained well above background during this time. At PRE2, Pho signals were more variable across the time course, especially as illustrated by the nc14a time point. However, by nc14b, which coincides with separation of nuclei by cell membranes and the start of cellular blastoderm stage (Foe, 1989), Pho signals at both PRE1 and PRE2 increased to approximately eightfold enrichment compared with approximately fourfold enrichment at nc10-12. Sfmbt remained low at PRE2 until nc14b (Fig. 3). Thus, it appears that recruitment of PhoRC to PRE1 precedes its stable recruitment to PRE2.

Anti-E(z) and anti-Pc antibodies were used to detect recruitment of PRC2 and PRC1, respectively (Fig. 4). Although the abundance of both E(z) and Pc in these embryos is approximately comparable with their levels in wild-type embryos (Fig. S3), these proteins were differentially temporally recruited to gt. E(z) was observed at most gt regions at levels that are slightly above background in nc10-12 embryos with an approximately twofold enrichment at regions 3 and 4 (PRE1), 9 (PRE2) and 10. E(z) signals remained essentially unchanged at nc13. Consistent with these weak E(z) signals, H3K27me3 was detected near background levels at most gt regions in nc10-12 and nc13 embryos. At nc14a, E(z) signals increased to approximately threefold enrichment at region 4 (PRE1) and increased to approximately two- to threefold enrichment at other gt regions, 7, 9 and 10. H3K27me3 signals roughly mimicked the increase in E(z) signals at nc14a and showed a peak of approximately threefold enrichment at region 4. By nc14b, E(z) signals remained constant or slightly increased (regions 3 and 6), whereas H3K27me3 signals showed clear increases across all gt regions. Similar to E(z), Pc signals were low, but slightly above background, at most gt regions in nc10-12 and nc13 embryos. However, unlike E(z), Pc signals remained low in nc14a embryos. It is striking that Pc signals increased sharply in nc14b embryos, 30 min after the initial increase in E(z) and H3K27me3 signals, and coinciding with the further increase in H3K27me3 across the gt region. Thus, stable Pc binding appears to immediately follow PRC2-dependent trimethylation of H3K27 at gt.

In order to directly test whether PRC2 is needed for PRC1 recruitment, ChIP assays were performed using homozygous E(z)⁶¹ embryos that were produced by homozygous females. E(z)⁶¹ is a temperature-sensitive allele that displays loss-of-function phenotypes at 29°C, but behaves as a nearly wild-type allele at 18°C (Jones and Gelbart, 1990). Embryos were collected at 18°C or 29°C and allowed to develop until they reached cellular blastoderm/early gastrulation (nc14b). Disruption of PRC2 activity (at 29°C) resulted in loss of both E(z) and Pc signals at gt (Fig. 5). These results are similar to those of our previous studies using imaginal discs (Savla et al., 2008; Wang et al., 2004). However, whereas previous studies described the hierarchical roles of various PcG proteins in maintaining PcG proteins at a silenced locus, our current studies provide evidence for PRC1 dependence on PRC2 for initial recruitment to a target gene as PcG-mediated silencing is established. Somewhat surprisingly, we also observed reduced Pho signals at both PRE1 and PRE2 in embryos lacking PRC2 activity. This may indicate some level of dependence of PhoRC on PRC1 and/or PRC2 for stable PRE binding.

Most investigations into the activities of the PcG proteins in vivo or in cell culture have focused on the maintenance phase of PcG silencing. In this study, we provide the first description of the temporal order of recruitment of PcG proteins at a target gene while PcG-mediated silencing is established. This was made possible by the generation of embryos, derived from bcd osk tsl females, in which the PcG target gene, gt, is initially ubiquitously repressed by a maternally expressed gene-specific transcription factor, Hb. As Hb_mat is degraded, PcG proteins assume the responsibility of maintaining gt repression, as shown by gt derepression in cellular blastoderm embryos from bcd osk tsl females that also lack E(z) activity (Fig. 1).

Similar to expression of Hb_mat in embryos produced by bcd osk tsl females, the gt activator Cad_mat is also ubiquitously expressed at syncytial blastoderm due to lack of translational repression of maternally deposited cad mRNA by Bcd (Fig. S1). ChIP assays revealed the presence of both Cad and Hb at the gt_(-3) enhancer in nc10-12 embryos (Table 2; Fig. 2). This is not surprising, given that there are both overlapping and non-overlapping Hb and Cad sites within the gt_(-3) enhancer (Fig. S4; Schroeder et al., 2004). Given uniform lack of gt expression in these embryos (Fig. 1B), it is unlikely that Hb and Cad are independently bound to the gt_(-3) enhancer in different nuclei. Rather, we suggest that Hb may be preventing activation by Cad, despite its presence. Hb and Cad play antagonistic roles at the enhancers of other genes, such as even skipped (Clyde et al., 2003; Small et al., 1992; Vincent et al., 2018). Therefore, co-occupancy of the gt_(-3) enhancer by Hb and Cad is not unprecedented, but the mechanism by which Hb prevents gt activation is unclear. After nc13, the signals for both Hb and Cad drop to near background levels (Fig. 2, Table 2). This is consistent with the timing of the degradation of maternally expressed Hb and Cad (Mlodzik et al., 1985; Tautz et al., 1987).

PhoRC components Pho and Sfmbt (Table 2, Fig. 3) were detected at gt PRE1 and PRE2 as early as Hb and Cad were detected at the gt_(-3) enhancer (nc10-12). PhoRC is the only PcG complex with the ability to bind directly to PREs (Brown et al., 1998). The presence of PhoRC at gt so early in development is consistent with the genetically defined maternal effect of pho, the requirement of Sfmbt⁺ for oogenesis, and maternal deposition of pho and Sfmbt transcripts (Breen and Duncan, 1986; Tomancak et al., 2002). In addition, PhoRC is known to directly interact with and maintain PRC2 (Mohd-Sarip et al., 2002, 2005; Wang et al., 2004) and PRC1 (Frey et al., 2016; Mohd-Sarip et al., 2002, 2005) at genomic targets in imaginal discs and cultured cells. Consistent with a role in recruitment of PRC1 and PRC2, PhoRC is enriched at PRE1 and PRE2 prior to stable binding of Pc or E(z) (Table 2). Pho signals continue to increase at PRE1 during nc14a and nc14b. This may be due to Pho occupying more binding sites or more stably binding to its sites in later stages. It is also possible that the increasing duration of interphase from nc10 to nc14 plays a role (Farrell and O'Farrell, 2014). Although there are discrepancies in the literature (Black et al., 2016; Steffen et al., 2013), the association of Pho with chromosomal targets may decrease during mitosis. Therefore, the gradual increase in Pho signals, at least at PRE1, may reflect the increasing representation of interphase nuclei in later stage embryos. Pho signals at PRE2 are more variable across the time course. In particular, their temporary reduction at the nc14a time point is puzzling. Consistent with the late increase of Pho at PRE2, Sfmbt signals also increase approximately twofold from nc14a to nc14b (Fig. 3). One possible explanation for the temporal differences in PhoRC binding to the two PREs is changes in local chromatin accessibility during the transition to zygotic control of gene activation following nc13 mitosis. Data from Blythe and Wieschaus (2016) (Fig. 6) support this model. PRE1, which includes the pre-MBT gt promoter, is accessible as early as nc11, whereas the PRE2 region, which overlaps with the late-acting gt_(-6) enhancer (Abed et al., 2013; Schroeder et al., 2004) only begins to become accessible late in nc13.

E(z) and Pc signals are both weakly positive across the gt regulatory region in nc10-12 and nc13 embryos (Fig. 4, Table 2). Although the relative contributions of maternally expressed versus zygotically expressed E(z) and Pc differ (described in more detail below), both are maternally expressed and ubiquitously distributed in blastoderm stage embryos (Carrington and Jones, 1996; Denell and Frederick, 1983; Fig. S3). E(z) signals are slightly higher at regions 3 and 4. It is possible that PhoRC may contribute to PRC2 and/or PRC1 recruitment to the vicinity of PRE1. However, neither E(z) nor Pc is particularly enriched at PRE2 in nc10-12 or nc13 time points. Therefore, this apparent weak association of PRC2 and PRC1 with much of the gt regulatory region may involve PhoRC-independent mechanisms. H3K27me3 signals remain low during these time points, suggesting that the unstable association of PRC2 with chromatin prevents it from carrying out H3K27 trimethylation.

At nc14a, E(z) signals increase at multiple gt regions, particularly at region 4 (PRE1), consistent with a potential role for PhoRC in stabilizing PRC2 binding. It is striking that PRC2 does not stably bind to gt until at least 1 h after PhoRC recruitment. Stabilization of PRC2 binding in early nc14 coincides with two obvious events. First, the activator Cad is lost from the gt_(-3) enhancer. It is possible that Cad, or a Cad-recruited co-activator, may inhibit PRC2 binding or deposition of H3K27me3. Second, stable binding of PRC2 also coincides with the MBT and lengthening of interphase. Multiple mechanisms have been proposed to regulate the timing of this transition, including titration of various factors as the nuclear to cytoplasmic ratio increases, the altered activities of nuclear cycle and DNA replication checkpoints (Farrell and O'Farrell, 2014), and the activity of a major genome activator such as Zelda, which contributes to transcriptional activation of a subset of genes pre-MBT and to increased chromatin accessibility (Blythe and Wieschaus, 2016; Harrison et al., 2011; Nien et al., 2011). Any of these factors, as well as simply the short duration of interphases, could conceivably inhibit stable PRC2 binding prior to the MBT. As alternatives to, or possibly in addition to, such inhibitory effects, PRC2 stable binding may require the presence of additional PcG proteins. These could include additional PRE-binding proteins or proteins such as Polycomb-like (Pcl), which is a substoichiometric subunit of PRC2 (Nekrasov et al., 2007), or Sex comb on midleg (Scm), which interacts with both PRC2 and PRC1 (Kang et al., 2015; Peterson et al., 2004). Both Pcl and Scm have been implicated in recruitment and/or maintenance of these complexes at genomic targets (Kang et al., 2015; Oksuz et al., 2018; Perino et al., 2018; Savla et al., 2008; Wang et al., 2004).

Concurrent with apparent PRC2 stabilization at nc14a, H3K27me3 signals increase in roughly the same pattern as E(z) across gt. It is notable that Pc signals remain low at nc14a, indicating that stabilization of PRC2 binding and the onset of H3K27 methylation precede stable binding by PRC1. During the subsequent 30 min (detected at nc14b), E(z) signals become somewhat more broadly distributed, H3K27me3 signals further increase in amplitude and spread, and Pc signals sharply increase across the gt region. These observations are consistent with the requirement of PRC2 and H3K27me3 for stable recruitment of PRC1.

All Drosophila PcG genes are expressed both maternally and zygotically. However, the relative contributions of their maternally expressed versus zygotically expressed products to PcG repression in embryos vary widely. As a general rule, PRC2 components must be provided maternally, whereas PRC1 components may be provided by zygotic expression. For example, lack of maternal E(z)⁺ results in embryonic lethality and strong homeotic phenotype, even if a copy of E(z)⁺ is provided paternally (Jones and Gelbart, 1990). esc mutant alleles exhibit similar strong maternal effects (Struhl, 1983). By contrast, lack of maternal Pc⁺ is completely rescued by one paternal copy of Pc⁺ (Haynie, 1983). Other PRC1 genes exhibit similar weak maternal effects that are rescued by zygotic expression (Breen and Duncan, 1986; Dura et al., 1988). In studies investigating the ability of LexA-fusion proteins to repress a reporter gene in embryos, LexA-tethered PRC1 components Pc, Ph, Psc or Su(z)2 were able to repress the reporter when expressed zygotically. However, LexA-Esc fusion protein was able to repress the reporter only when expressed maternally (Poux et al., 2001a,b). These observations support the idea that the components of PRC2 must be expressed in sufficient quantities prior to the MBT checkpoint and the progressive ZGA in order to be available when needed at the beginning of nc14 (Harrison and Eisen, 2015). However, sufficient quantities of PRC1 may be provided by expression during or after the MBT. The relative temporal requirements for expression of PRC2 and PRC1 components correlate well with the increased signals of E(z) (PRC2) and H3K27me3 during the first half of nc14 followed by increased Pc (PRC1) signals at nc14b. It has been suggested that maternally deposited H3K27me3 is needed for PcG-mediated regulation of target genes in embryos (Zenk et al., 2017). However, our observed timing of PRC2 stabilization and H3K27me3 deposition at gt closely agrees with the previously reported appearance of this histone modification during nc14 at many genomic loci (Li et al., 2014). Our results also are consistent with a recent report of the formation of PcG-dependent chromatin loops after the MBT (Ogiyama et al., 2018).

Based on the relative temporal binding of PcG complexes to PRE1 and deposition of H3K27me3, we propose a three-tier model for de novo recruitment of PcG complexes (Fig. 7). Chromatin accessibility at enhancers and promoters in early embryos is facilitated by pioneer factors or transcriptional activators like Zld and Gaf (Blythe and Wieschaus, 2016; Moshe and Kaplan, 2017). During syncytial blastoderm, and no later than nuclear cycles 10-12, PhoRC acts as a PcG nucleation factor (Fig. 5A) (Schuettengruber et al., 2014; Zaret and Carroll, 2011). PRC1 and PRC2 sample permissive chromatin sites (Fig. 7A) (Klose et al., 2013) and accumulate at those sites lacking antagonistic chromatin modifications or activators. Once the embryo reaches the MBT, chromatin binding by PRC2 is stabilized, potentially involving direct interaction with PhoRC. The longer duration of nc14 interphase may also allow stably bound PRC2 to begin deposition of H3K27me3, which forms a positive-feedback loop in which Esc binding to H3K27me3 further stimulates the catalytic activity of PRC2, leading to additional methylation of H3K27 (Margueron et al., 2009) (Fig. 7B). The presence of PRC2 and H3K27me3 contribute to stable recruitment of PRC1, possibly owing to the binding of the Pc chromodomain to H3K27me3 (Cao et al., 2002; Wang et al., 2004), leading to the formation of repressive chromatin domains (Fig. 7C). PhoRC may also stabilize binding by PRC1, possibly involving Scm, which has been shown to interact directly with both PRC1 and PRC2 (Kang et al., 2015; Peterson et al., 1997).

The relative temporal recruitment of PcG complexes to PRE2 is less clear, which raises the possibility that a different model for recruitment may be more appropriate for this region. Unlike the Pho signals at PRE1, which steadily increase from nc10-12 to nc14b, Pho signals at PRE2 are reduced, albeit still positive, at nc14a and then increase at nc14b. Similarly, Sfmbt signals are low at PRE2 prior to nc14b. Thus, we are unable to distinguish the timing of increased PhoRC signals at PRE2 from increases in H3K27me3 and Pc signals. Two models seem most plausible. First, it is possible that increased PhoRC presence at PRE2 may in fact precede that of PRC1, but that it occurs within this 30 min window. In this case, recruitment at PRE2 may closely resemble recruitment at PRE1. Alternatively, interactions with other proteins may be needed to stabilize PhoRC at PRE2. These could be other PRE-binding proteins or other PcG complexes. For example, it has been previously proposed that interaction with PRC1, possibly mediated by Scm, may stabilize PhoRC binding to PREs (Frey et al., 2016; Kahn et al., 2014). Indeed, as shown by examination of E(z)⁶¹ embryos, the presence of Pho at both PRE1 and PRE2 is reduced in the absence of PRC1 and PRC2. Therefore, PRC1 and/or PRC2 may participate in a feedback loop to maintain or enable stable PhoRC binding to PREs.

It is also possible that other, yet to be identified, late acting factors are needed to stabilize PhoRC binding. However, the most parsimonious explanation for the relatively late increase of PhoRC signals at PRE2 may be that initial PhoRC binding is inhibited by inaccessible chromatin (Fig. 6) and that it may be unnecessary to invoke roles for additional factors. Whereas PRE1 is accessible at least by nc11, PRE2 becomes accessible post-MBT, which correlates with the relative timing of PhoRC to the two PREs. E(z) signals at PRE2 do not increase from nc14a to nc14b. It may be that the relatively low level of PhoRC at PRE2 is sufficient to recruit PRC2 at nc14a. Alternatively, one or more unidentified factors may initially help to stabilize PRC2 binding to PRE2. Whether the low PhoRC levels at PRE2 are sufficient for stable recruitment of PRC2, or whether PRC2 is initially recruited to PRE2 by a PhoRC-independent mechanism, is not clear at this time. The extent to which this model may apply to other genomic sites remains to be determined, and it is likely that it will only apply to recruitment of canonical PRC1, which includes Pc, and not to the non-canonical PRC1, RAF (Lagarou et al., 2008). However, based on the differential recruitment at PRE1 and PRE2, de novo recruitment of PRC2 and PRC1 may be affected by a variety of factors in addition to PhoRC. Indeed, definitive tests for the validity of the three-tier recruitment model we propose at gt will require reciprocal depletions of components of each of the PcG complexes in order to determine their respective contributions.

In summary, the temporal order of de novo recruitment of PRC2 followed by PRC1 at gt is consistent with the hierarchical recruitment model of PcG proteins (Wang et al., 2004). Alternative models for PcG protein recruitment have been proposed in which PRC1 either binds to sites independently of PRC2 or recruits PRC2, thus reversing the hierarchy of recruitment described in Fig. 7 (Dorafshan et al., 2017). These models are largely based on the affinity of PRC2 for PRC1-dependent monoubiquitylated H2A or genomic distributions of PcG proteins during the maintenance phase of PcG-mediated repression. It is quite possible that multiple mechanisms may contribute to stabilizing the chromatin association of PcG complexes subsequent to their initial recruitment, and that this may include reciprocal positive feedback between PRC1 and PRC2. It will be interesting to see whether these alternative models apply to de novo recruitment.

Embryos were collected at 25°C, fixed and processed as previously described (Jones and Gelbart, 1990). Anti-Hb antibody was diluted 1:1000, anti-Gt was diluted 1:500 (Kosman et al., 1998), anti-E(z) was diluted 1:200, anti-Pc was diluted 1:50, anti-Cad was diluted 1:400. Biotin-SP-conjugated goat anti-rabbit secondary antibodies (Jackson ImmunoResearch) were diluted 1:10,000. Streptavidin-horseradish peroxidase (Jackson ImmunoResearch) was diluted 1:5000. Signals were detected by incubating embryos in 1 mg/ml diaminobenzidine (Sigma #D12384) in 0.1 M Tris-HCl (pH 7), 1% NiCl₂, 0.003% H₂O₂for ∼20 min. Reactions were stopped by washing with PBST (0.01% Trition-X-100). Embryos were dehydrated and mounted in Permount (Fisher Scientific, SP15). Images were obtained using a Zeiss Axiovert 200M microscope.

Embryos were collected for 30 min periods and aged at 25°C for the time after egg lay (AEL) until they reached the appropriate stage (Table 1). Embryos were crosslinked as previously described (Blythe and Wieschaus, 2015) except that 108 µl of formalin (Fisher #BP531) was added to 2.0 ml PBS+0.5% Triton X-100 and 6 ml heptane. After 15 min, the fixation was quenched with 0.125 M glycine. Embryos were thoroughly washed with ice-cold PBST (0.5% Triton-X-100), resuspended in 1 ml of PBST with 1× protease inhibitor (Sigma #P8340), and placed on ice until they were hand-sorted using a Zeiss Primovert microscope to remove older contaminating embryos based on their morphology (Harrison et al., 2011). Representative embryos from each stage are shown in Fig. S2. Embryos were weighed, flash frozen and stored at −80°C. Aliquots were made to obtain the following masses of embryos for each embryonic stage per tube: nc14b, 10 mg; nc14a, 10 mg; nc13, 20 mg; and nc10-12, 40 mg. The embryo masses for each stage contain roughly the same amount of DNA per aliquot. Each aliquot was sufficient for a ChIP experiment with nine antibodies and input genomic DNA.

For each antibody, 50 µl of protein A magnetic beads (Pierce #88846) were washed with PBST-3% BSA, then blocked for 1 h with 1 ml PBST (0.1% Triton), 3% BSA at 4°C. Antibody dilutions were prepared in 250 µl of 1× PBST, 3% BSA as follows: Hb, 5 µl; Cad, 5 µl; Pc, 10 µl (Wang et al., 2004); Pho, 5 µl (Brown et al., 2003); E(z) (Carrington and Jones, 1996), 10 µl; H3K27me3 (Millipore, 07-449), 0.2 µl; H3 (Abcam, ab1791), 0.5 µl; mock, 0.5 µl (anti-IgG, Cell Signaling, 2729). For Ab-bead capture, the blocked antibodies were added to the magnetic beads on a rotating wheel at room temperature for 15 min, then stored at 4°C for 2 h. Embryos were thawed on ice and homogenized in 700 µl RIPA buffer supplemented with 1 mM DTT, 1× Sigma protease inhibitor (Sigma-Aldrich, P8340). Homogenized embryos were centrifuged at full speed at 4°C and resuspended in 1 ml of RIPA buffer (at 10 mg/ml for nc14a and nc14b, 20 mg/ml for nc13, 40 mg/ml for nc10-12) and sonicated on ice with a microtip probe and Misonix sonicator 3000: 30% power, 15 s pulses, 45 s pauses, total of 4.5 min. Approximately 90% of the resulting DNA fragments were <1 kb with a 30-35% peak at 1 kb (not shown). Following centrifugation, the sonicate was precleared with 80 µl protein A agarose/salmon sperm DNA (Millipore, 16-157) for 1 h at 4°C. The mixture was centrifuged and 100 µl supernatant aliquots were added to each of the antibody beads. RIPA buffer (500 µl) was added to each sample and incubated overnight at 4°C while rotating. A 20 µl supernatant aliquot was used as input DNA and treated with 0.5 µl of 10 mg/ml RNAse A for 30 min at 37°C. The volume of input DNA was made up to 100 µl using elution buffer [50 mM Tris-HCl (pH 7.5), 10 mM EDTA, 1% SDS, 300 mM NaCl], and 2.5 µl proteinase K (10 mg/ml to a final concentration of 0.25 mg/ml) and incubated overnight at 65°C. After overnight incubation, antibody bead-chromatin samples were centrifuged and washed on a magnetic stand with the following buffers for a maximum of 1 min each: 1× low-salt wash buffer [0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl (pH 8.0), 150 mM NaCl], 1× high-salt wash buffer [0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl (pH 8.0, 500 mM NaCl] and 2× with LiCl wash buffer [0.25 M LiCl, 1% NP-40, 1% SDC, 1 mM EDTA, 10 mM Tris-HCl (pH 8.0)], and then with 1× with TE [10 mM Tris-HCl (pH 8.0), 1 mM EDTA]. After the last washing step, 100 µl elution buffer was used to resuspend the beads and elute at 65°C for 15 min. Samples were centrifuged at full speed, and the supernatants transferred to fresh tubes. TE (95 µl) and 5 µl 10 mg/ml proteinase K were added (final concentration of 0.25 mg/ml) and incubated at 65°C for 4 h to reverse crosslinks. Samples and input DNA were extracted with phenol/chloroform, and chloroform and ethanol precipitated for 2 h at −20°C following addition of 40 µg glycogen. The pellets were resuspended in 40 µl water and further purified using 1.5× volume of Agencourt Ampure beads (Beckman Coulter, A63880). Samples were resuspended in 112 µl water. An additional 112 µl of water was added to the input sample to make a 10% input genomic DNA for qPCR standard.

qPCR was performed using Quanta Biosciences Perfecta SYBR green supermix in the Rotor Gene RG3000 thermocycler. PCR reactions for each ChIP experiment were performed in triplicate unless otherwise indicated. DNA (5 µl) was used for each PCR reaction, or an equivalent of the DNA immunoprecipitated from 0.045 mg of nc14b embryos, 0.0045 mg for 10% input, 0.00045 mg for 1% input and 0.00009 mg for 0.2% input genomic DNA for the same stage. Approximately equivalent amounts of DNA were added per PCR reaction from different stages due to a 1:1:2:4 mass ratio of embryos used per ChIP. PCR was performed for 45 cycles: 95°C for 2 min for initial denaturation; annealing for 30 s at various annealing temperatures for each primer set; and extension for 15 s at 72°C. PCR primers used and corresponding annealing temperatures are listed in Table S1. Rotor Gene 5 software was used to determine Ct values and automatically set the threshold. The sample concentrations were then determined using the 2^−ΔΔCT method (Rao et al., 2013), and then normalized to an internal control: Pka-C1. The Pka-C1 gene is maternally deposited and zygotically expressed (Tomancak et al., 2002; Lott et al., 2011), and is not bound by PcG proteins or the transcription factors Hb and Cad.

The bcd⁷osk⁶ (3252) and tsl⁴ (3289) fly stocks were obtained from the Bloomington Drosophila Stock Center (NIHP40OD018537). The tsl⁴ allele was crossed onto the bcd⁷osk⁶ chromosome to make one bcd osk tsl stock. A second bcd osk tsl stock, containing the bcd⁶ osk⁶ tsl^PZREV32 third chromosome, was a gift from Leslie Stevens (The University of Texas at Austin, TX, USA) (Stevens et al., 2003). Because both bcd osk tsl chromosomes contain unidentified recessive lethal mutations, these two stocks were crossed to generate bcd⁷ osk⁶ tsl⁴/bcd⁶ osk⁶ tsl^PZREV32 females for embryo collections. E(z) bcd osk tsl strains were generated by crossing E(z)²⁸ (Phillips and Shearn, 1990) onto the bcd⁶ osk⁶ tsl^PZREV32 chromosome and E(z)⁶³ (Jones and Gelbart, 1990) onto the bcd⁷ osk⁶ tsl⁴ chromosome. The presence of all alleles was confirmed by complementation tests and/or examination of embryo cuticular phenotypes. All chromosomes were balanced over TM3, Sb Ser. E(z)⁶³ bcd⁷ osk⁶ tsl⁴/TM3 and E(z)²⁸ bcd⁶ osk⁶ tsl^PZREV32 /TM3 stocks were crossed to generate E(z)⁶³ bcd⁷ osk⁶ tsl⁴/ E(z)²⁸ bcd⁶ osk⁶ tsl^PZREV32 females. These females were crossed to E(z)⁶¹ males (Jones and Gelbart, 1990) for embryo collections.

Rabbit anti-Cad, and anti-Hb antisera were raised against His₆-fusion proteins containing either Cad residues 2-240 or Hb residues 2-313. Injections and bleeds were performed at Covance Research Products. Antibodies were affinity-purified essentially as previously described (O'Connell et al., 2001). Antisera was preadsorbed against His₆-Pcl fusion protein, then antibodies were affinity purified using His₆-Cad or His₆-Hb fusion proteins, respectively.

We thank Shelby Blythe for sharing his ChIP protocol and ATAC-seq data ahead of publication, Steven Small for the anti-Gt antibody, Judy Kassis for anti-Sfmbt and anti-Pho antibodies, and Leslie Stevens and the Bloomington Drosophila Stock Center for fly stocks. We also thank Chao-ting Wu for support of J.A.A. in the latter stages of this work.