Variant PRC1 competes with retinoic acid-related signals to repress Meis2 in the mouse distal forelimb bud

Polycomb group (PcG) proteins form two major multimeric protein complexes known as PRC1 and 2 (polycomb repressive complexes 1 and 2) (Simon and Kingston, 2009). Both complexes are evolutionarily conserved from metazoans to mammals, and play synergistic roles to repress developmental genes, such as the Hox cluster. PRC1 mediates histone H2A mono-ubiquitylation at lysine 119 (H2AK119ub1) via E3 ubiquitin ligase activity by RING1A and 1B (RING finger protein 1A and 1B; also known as RING1 and RNF2, respectively), whereas PRC2 mediates histone H3 lysine 27 methylation (H3K27me1/2/3) by the function of EZH1 and 2 (enhancer of zeste homologs 1 and 2). The H3K27me3 marks are further recognized by the PRC1 components CBX2, 4, 6, 7 or 8 (chromobox protein 2, 4, 6, 7 or 8), in turn leading to robust gene repression.

PRC1, furthermore, constitutes six subcomplexes, each incorporating different combinations of PCGF (polycomb group ring finger) proteins (PCGF1 to 6) (Farcas et al., 2012; Gao et al., 2012; He et al., 2013; Wu et al., 2013). The canonical PRC1 (cPRC1) complexes contain PCGF2 (MEL18) or PCGF4 (BMI1) and bind H3K27me3 via the CBX proteins (Aranda et al., 2015; Blackledge et al., 2015; Isono et al., 2013; Kondo et al., 2016). A variant (or non-canonical) PRC1 (vPRC1 or ncPRC1), containing PCGF1, localizes to CpG islands (CGIs) and mediates H2Aub1 (Blackledge et al., 2014). A different vPRC1 complex, containing PCGF3 and 5 (PCGF3/5-PRC1), mediates H2Aub1 and H3K27me3 at the inactive X chromosome (Almeida et al., 2017). Finally, another variant PRC1 complex, containing PCGF6 (PCGF6-PRC1), mediates H2Aub1 and H3K27me3 at meiosis- and/or germ cell-related gene promoters (Endoh et al., 2017).

The role for PcG factors in patterning and cellular differentiation, by repressing developmental genes, is well established. Similarly, it is also well known that morphogenetic signals, such as retinoic acid (RA) or RA-related molecules synthesized by fetal tissues, play a role in the activation of genes during development (Cunningham and Duester, 2015). Based on these arguments, it could be hypothesized that PcG factors and developmental signals oppose each other to regulate gene repression, or activation. During limb development, strong RA-related signals facilitate proximal specification of forelimb bud by activating the Meis (Meis homeobox) genes, such as Meis2 and Meis1, whereas weak RA signals are insufficient to activate Meis, resulting in the progression of distal specification, although this model remains controversial (Cunningham et al., 2013; Mercader et al., 2000; Probst et al., 2011; Rosello-Diez et al., 2014; Yashiro et al., 2004). Our recent study, however, showed that PcG factors repress Meis2 to mediate distal specification of the forelimb bud where RA-related signals are weak (Yakushiji-Kaminatsui et al., 2016). Therefore, the Meis2 gene provides an excellent system to probe the competitive relationship between PcG factors and developmental signals.

Here, we examined whether RA-related signals and PcG factors antagonize each other with respect to Meis2 regulation during proximal-distal (PD) specification of forelimb bud. Indeed, we show that RING1 proteins competitively regulate RA-related signaling to polarize Meis2 expression along the PD axis. Based on mathematical modeling and simulation, we propose that RING1 adjusts the threshold of RA-related signals for Meis gene regulation. We further sought to clarify the specific PRC1 subtype(s) that are involved in mediation of Meis2 downregulation in the distal forelimb bud. By performing genetic analyses, we reveal that the variant PCGF3/5-PRC1 complex is required for Meis2 repression. Taken together, we have discovered a role for PcG factors (i.e. PCGF3/5-PRC1) in repression of Meis2 expression by creating de novo PcG-repressive domain(s) through antagonizing RA-related developmental signals.

We investigated whether PcG-mediated repression and RA-related signaling are mutually linked at the Meis2 locus. Previous reports show that the Meis2 promoter is bound by retinoic acid receptors (RARs) in F9 embryonal carcinoma cells, inducing transcriptional activation (Chatagnon et al., 2015; Leid et al., 1992; Mangelsdorf and Evans, 1995) (Fig. 1A, tracks 1, 2). Consistent with this notion, we found that RARs bound the Meis2 promoter region in embryonic day (E) 11.5 embryos in the proximal forelimb bud, but much less so in the distal forelimb bud, reflecting the active status of Meis2 expression specifically in the proximal forelimb bud (Fig. 1A, tracks 3, 4). We further investigated local enrichment of H3K27 acetylation (H3K27ac), which is known to be mediated by co-activators such as p300 (Ep300) and CBP (CREBBP) that form complexes with RARs in an RA-dependent manner (Tie et al., 2009). Consistent with these findings, a study by McKeown et al. (2017) using human cell lines revealed that local enrichments of H3K27ac and RARs were concurrently induced by an RARα agonist at their common targets, including MEIS2 and MEIS1 (McKeown et al., 2017) (Fig. S1A,B). Of note, RAR-mediated signals may mediate local H3K27ac by stimulating co-activators. We did indeed observe considerable enrichment of H3K27ac at the promoter of Meis2 in the proximal forelimb bud, but depletion of the same mark in the distal forelimb bud (Fig. 1A, tracks 5, 6). These findings suggest that RA-related signals upregulate Meis2 expression via activation of an RAR/co-activator pathway in the proximal forelimb bud (Bajpe et al., 2013; Chen et al., 2013; Jin et al., 2011).

In contrast, our previous report (Yakushiji-Kaminatsui et al., 2016) showed RING1B binding at the Meis2 locus, causing transcriptional repression in the distal forelimb bud (Fig. 1A, tracks 7, 8). We therefore investigated whether this inverse correlation between RA-related signals and PcG activity at the Meis2 locus in developing forelimb bud is functionally interlinked. To this end, we orally administered all-trans RA (ATRA) at a dose of 100 mg/kg to pregnant females at E10.25 (Fig. 1B, left). ATRA administration induced ectopic Meis2 expression in the distal forelimb buds (Fig. 1B, right), in combination with decreases in both RING1B (PRC1 component) binding and H3K27me3 (indicator of PRC2 activity) deposition at the Meis2 promoter in the distal forelimb bud (Fig. 1C, top). Similarly, RING1B binding and H3K27me3 levels were also downregulated in the RING1B binding site at the 3′ region of the Meis2 (RBS). The RBS associates with the repressed promoter and facilitates RING1B binding (Fig. 1C, top) (T.K., Y.K. and H.K., unpublished observations). We further confirmed whether ATRA treatment activates the RAR/co-activator pathway at Meis2. We investigated local H3K27ac levels, which correlate with RA signal intensity (Fig. S1A,B), and found a considerable increase in H3K27ac in the distal forelimb bud (Fig. 1C, bottom). We also found an increase in another activating mark, H3K4 trimethylation (H3K4me3) (Fig. 1C, bottom). Collectively, these data indicate that activation of the RAR/co-activator pathway contributes to the eviction of PcG factors from the Meis2 locus.

We went on to investigate whether ATRA-mediated PcG eviction affects the contact between promoter and RBS in the Meis2 locus, which is mediated by RING1 proteins (Kondo et al., 2014; Yakushiji-Kaminatsui et al., 2016) (Fig. 1D). We used the DNA-FISH technique to measure the distance between promoter and RBS and found that the promoter-RBS contact was dissolved in the distal forelimb bud by ATRA treatment (Fig. 1D). These findings support the model that Meis2 activation in the distal forelimb bud by strong RA signals involves eviction of PcG factors.

To validate the role of endogenous RA-related signals in eviction of PcG factors, we took advantage of a loss of function of CYP26B1 mutant (Cyp26b1 KO), which shows sustained RA signaling throughout the limb bud due to a block in degradation of endogenous RA or its related molecules (Yashiro et al., 2004). Consistent with our hypothesis that ATRA mediates eviction of PcG, we observed depletion of RING1B binding and H3K27me3 at both promoter and RBS regions in distal forelimb buds in the Cyp26b1 KO (Fig. 1E), which was reminiscent of the alterations in PcG enrichment seen in the ATRA-treated distal forelimb buds (Fig. 1C, top). Taken together, these data further support the notion that endogenous RA-related signals mediate polarization of Meis2 expression (i.e. create Meis2-positive or -negative domains) in forelimb bud by inhibiting PcG activity.

We investigated whether ATRA-mediated signals are linked with PcG activity in general or in a locus-specific manner at PcG target genes during PD specification. We selected Meis1, Pitx2, Hoxa11 and Hoxa13 genes, which exhibit proximally or distally skewed expression, and are targeted by both PcG factors and RARs in the forelimb bud at E11.5 (Fig. S1C). Meis1 is structurally related to Meis2 and shows strong RAR binding and H3K27ac enrichment at the proximal forelimb bud. In contrast, RING1B is enriched at the distal forelimb bud, similar to Meis2 (Fig. S1C, left). Furthermore, we have previously shown that Meis1 is repressed in the distal forelimb bud in a PcG-dependent manner, but is activated by ATRA administration (GSM1716758, published microarray data; Yakushiji-Kaminatsui et al., 2016). The Pitx2 gene is also bound by RARs and H3K27ac in the proximal forelimb bud, but repressed by RING1 in the distal forelimb bud (Fig. S1C, middle). Unlike Meis2, however, Pitx2 is not activated by ATRA (GSM1716758, published microarray data; Yakushiji-Kaminatsui et al., 2016). The Hoxa11/13 genes, located at the 3′ region of the HoxA gene cluster, are bound by H3K27ac in the distal forelimb bud and by RING1B in the proximal forelimb bud, but do not show enrichment of RARs (Fig. 1C, right). In addition, Hoxa11/13 genes are expressed in the distal forelimb bud, and are downregulated by ATRA, likely via RAR binding to the 5′ region of HoxA cluster (GSM1716758, published microarray data; Yakushiji-Kaminatsui et al., 2016).

We compared RING1B, H3K27me3, H3K4me3 and H3K27ac levels at the above genes, and observed reduction of RING1B and H3K27me3 at both promoter and RBS regions of Meis1, similar to observations with respect to Meis2 (Fig. S1D). However, binding of RING1B or H3K27me3 did not change at Pitx2, Hoxa11 and Hoxa13 promoters upon ATRA treatment, in either proximal or distal forelimb bud (Fig. S1E). Intriguingly, H3K4me3 and H3K27ac levels were differentially regulated at these genes, indicating that the effect of ATRA on these genes is not necessarily the same (Fig. S1C,D). Taken together, our data show that target binding of PcG factors is sensitive to RA-related signals at specific gene loci, such as Meis2 and Meis1, in the forelimb bud.

The above evidence clearly indicates that RA-related signals negatively regulate PcG function at Meis genes. Reciprocally, PcG factors regulate RA sensitivity of Hox genes and Stra8 (Bel-Vialar et al., 2000; Yokobayashi et al., 2013). Our previous report, showing RING1-dependent modulation of RA sensitivity during PD specification of forelimb bud (Yakushiji-Kaminatsui et al., 2016), is consistent with these notions.

By using the Meis2 gene as a model, we sought to clarify whether RA sensitivity in forelimb bud could be inversely regulated by PcG factors. We challenged Ring1B conditional knockout mutants (Prx1-Cre;Ring1B^fl/+ or Prx1-Cre;Ring1B^fl/fl) or control (Ring1B^fl/+ or Ring1B^fl/fl) embryos (E10.5) with ATRA by maternal oral gavage (5 mg/kg). At this dose, we observed, by in situ hybridization, ectopic activation of the Meis2 gene in a part of the control forelimb bud (Fig. 2A). We obtained control and mutant embryos by crossing Ring1B^fl/fl females with Prx1-Cre;Ring1B^fl/+ males. Six hours after ATRA administration, four out of 11 control embryos exhibited subtle upregulation of Meis2 in the distal limb bud (Fig. 2A, bottom left). Prx1-Cre;Ring1B^fl/+ embryos showed more pronounced expansion of the Meis2 expression domain (n=9/14) (Fig. 2A, bottom middle). In contrast, Prx1-Cre;Ring1B^fl/fl (Ring1B-KO) embryos showed robust expansion of the Meis2 expression domain reaching as far as the distal tip of forelimb buds (n=10/11) (Fig. 2A, bottom right). Therefore, sensitivity of Meis2 expression to RA signals is enhanced in the Ring1B mutants, indicating a reciprocal role for PcG factors in regulation of RA function (Fig. 2A).

To rule out the possibility that the phenotype described above arises from alteration of RA-related signals along the PD axis, we took advantage of a RARE-lacZ reporter allele (Rossant et al., 1991), which is a standard indicator for RA-related signals. Indeed, we found no obvious changes in the lacZ expression pattern in the Ring1A/B-dKO forelimb bud (Fig. 2B). Based on these results, we concluded that PcG factors limit RA-related signal transduction at Meis2. Taken together, the above observations indicate that RA-related signals and PcG activity compete with each other to regulate Meis2 expression in the forelimb bud and polarize the transcription of Meis2 along the PD axis.

Here, [X] indicates the concentration of chemical X, and K and h are reaction constants (Lehninger, 1970). The expression level of Meis2 in each cell is defined by an input [RA], and a threshold (K′)^h. The input [RA] depends on the position of the cell along the PD axis. (K′)^h represents the sum of [PcG]^h and K^h, and acts as a threshold for regulation of Meis2 expression (Fig. 2C). RING1B is uniformly expressed along the PD axis (Yakushiji-Kaminatsui et al., 2016), indicating that the threshold (K′)^h is constant, independent of the position of cells within the forelimb bud. Thus, for a given RA gradient, the expression level of RING1 should adjust the threshold of RA-related signaling and determine the Meis2 expression region along the PD axis. Conversely, in the Ring1 mutant, the affinity of RA-related signals for the regulatory regions of Meis2 will increase in each cell throughout the limb bud (shown by a green arrow in Fig. 2C). This model is also consistent with our experimental observations (Fig. 2A).

We asked whether this model could explain the respective gene expression patterns in wild-type or control (Ring1A-KO), ATRA-treated wild-type, Ring1A/B-dKO and Prx1-Cre;Ring1A^−/−;Ring1B^fl/fl;Meis2^fl/fl (Ring1A/B;Meis2-tKO) forelimb buds. To this end, we took advantage of a previously proposed network involving RA-related signals, CYP26B1 and FGF activities during early PD specification (Probst et al., 2011). In this network, we included the RING1 proteins as threshold adjusters for Meis1/2 expression. We considered both Meis2 and Meis1 genes, because both are regulated by RA-related signals and inhibit distal specification (Capdevila et al., 1999; Mercader et al., 2009). We further developed our model by combining the following interactions (i-vii, shown by solid arrows in Fig. 2D) (also see Materials and Methods): Meis2 is activated by RA-related signals (Mercader et al., 2000) (i); FGFs released from apical ectodermal ridge (AER) activate CYP26B1 expression (ii); the activated CYP26B1 degrades RA (Probst et al., 2011; Yashiro et al., 2004) (iii); Meis2 suppresses the expression of Cyp26b1 (Rosello-Diez et al., 2014) (iv) and Fgf8 at AER (figure 4B in Yakushiji-Kaminatsui et al., 2016) (v); Meis2 also suppresses Hoxa13 (Rosello-Diez et al., 2014; figure 4B in Yakushiji-Kaminatsui et al., 2016) (vi); and PcG factors inhibit Meis1/2 (figure 1E in Yakushiji-Kaminatsui et al., 2016) (vii). Two more potential interactions (indicated by dashed arrows in Fig. 2D), induction of Hoxa13 by AER-derived FGFs (Vargesson et al., 2001) (viii) and binding of PcG factors to the Hoxa13 promoter in the proximal forelimb bud (ix), may exist (Fig. S1E). However, we did not include these interactions (viii and ix) in the mathematical modeling because Hoxa13 induction by AER (viii) was reported to be mediated by RA/Meis axis (Rosello-Diez et al., 2014) and PcG binding to the Hoxa13 promoter (ix) was permissive to repression mediation in our previous study (Yakushiji-Kaminatsui et al., 2016).

We examined whether our model could recapitulate the expression patterns of Fgf8, Meis1/2 and Hoxa13 in ATRA-treated wild type, Ring1A/B-dKO or Ring1A/B;Meis2-tKO mutant forelimb buds (Fig. 2E, also see Table 1). Indeed, computer simulations revealed that ATRA treatment upregulated Meis1/2 expression in the distal forelimb bud and inhibited distal forelimb-specific genes such as Fgf8 and Hoxa13 (Fig. 2E). In the Ring1A/B-dKO, lowering the threshold for RA-related signals upregulated Meis1/2 expression, and downregulated Fgf8 and Hoxa13 in the distal forelimb bud (Fig. 2E). Furthermore, in the Ring1A/B;Meis2-tKO forelimb bud, Meis2 depletion, despite sustained expression of Meis1, partially restored Fgf8 and Hoxa13 levels (Fig. 2E). Collectively, these results support a role of RING1 as a threshold adjuster for RA-related signals to regulate Meis2 expression in forelimb bud.

Having shown that PcG factors play a role in modulation of the threshold of RA-related signaling at Meis2, we sought to clarify the specific PRC1 subcomplex(es) that play a role in downregulation of Meis2 expression. Because different PCGF (1 to 6) proteins give rise to different PRC1 subcomplexes, we surveyed forelimb phenotypes in embryos with each of the six Pcgf genes (Pcgf1-6) individually knocked out. Our previous reports indicated that PCGF2, PCGF4 and PCGF6 are dispensable for skeletal patterning of the forelimb, along the PD axis (Akasaka et al., 2001; Endoh et al., 2017; Isono et al., 2005). Therefore, in the present study we explored the role of PCGF1, PCGF3 and PCGF5 in either Meis2 expression or skeletal patterning of the forelimb by using respective mutant alleles (Fig. S2A,B,D). Deletion of Pcgf1 at E8.5 did not alter Meis2 expression in the forelimb buds at E10.5 (Fig. S2A; T.K. and H.K., unpublished observations). Deletion of Pcgf3 caused perinatal lethality and skeletal defects, in both zeugopod and scapula, similar to the Ring1B-KO, and posterior homeotic transformations of the axis (Fig. 3A, Fig. S2B,C). Finally, the Pcgf5 mutant embryos developed normally to adulthood and showed normal skeletal patterns (Fig. S2D,E; Y.K. and H.K., unpublished observations). Thus, our results reveal that the PCGF3-PRC1 complex mediates PD specification in the forelimb bud.

As PCGF3 and PCGF5 participate in similar protein complexes and possess redundant functions (Almeida et al., 2017; Gao et al., 2012), we deleted both Pcgf3 and Pcgf5 to investigate the role of PCGF3/5-PRC1. We first examined Pcgf3/5-dKO embryos, in which both genes are constitutively depleted, and found them to be embryonic lethal around E9.5 and to exhibit general growth defects at E7.5 (Almeida et al., 2017). We further observed similar phenotypes in ERT2Cre;Pcgf3^fl/fl;Pcgf5^fl/fl embryos treated with 4-hydroxytamoxifen at the pre-implantation stage (Fig. 3B). We thus used ERT2Cre;Pcgf3^fl/fl;Pcgf5^fl/fl embryos treated with tamoxifen at E8.5 (Pcgf3/5-dKO) to examine the roles of Pcgf3 and Pcgf5 in the developing forelimb bud and observed considerable expansion of the Meis2 expression domain in the distal forelimb bud at E10.5 (Fig. 3C). We further noted depletion of H3K27me3 deposition at the Meis2 promoter in the Pcgf3/5-dKO (Fig. 3D). We checked the level of contact between the promoter and RBS at Meis2, and found that the promoter and RBS were dissociated in the distal forelimb bud (Fig. 3E). Therefore, the forelimb bud phenotype observed in the Pcgf3/5-dKO resembles that of the Ring1A/B-dKO. Taken together, PCGF3/5-PRC1 mediates Meis2 downregulation in distal forelimb bud by inducing a PcG-repressive domain at the Meis2 locus.

We designed the present study to elucidate the interactions between PRC1 and RA-related signals for polarizing Meis2 gene repression along the PD axis of the forelimb bud, by combining genetic studies on various PRC1 mutants and mathematical modeling. We did indeed find a novel role for the variant PCGF3/5-PRC1 in repression of Meis2 expression in the distal forelimb bud, likely by adjusting the threshold for RA-related signals (Fig. 3F).

We therefore propose that PCGF3/5-PRC1 possesses two distinct functions: initiation of PcG-mediated silencing of Meis2, and sensing of RA-related signals. The first function of PCGF3/5-PRC1 is consistent with our previous observation that showed a role for PCGF3/5-PRC1 in X chromosome inactivation, via recruitment of the PRC2 complex (Almeida et al., 2017; Pintacuda et al., 2017). PCGF3/5-PRC1-mediated Meis2 repression similarly involves PRC2 recruitment, likely via RING1-dependent H2A mono-ubiquitylation (Blackledge et al., 2014). Another mechanism by which PRC1 could adjust the threshold for RA-related signals was revealed by treating Ring1B mutants with ATRA, as well as by a mathematical model in which the threshold-adjuster function is integrated into gene regulatory networks for PD patterning. We find that the computer simulations recapitulate the expression patterns of Meis1/2, Fgf8 and Hoxa13 at E10.5 under various experimental conditions, and thus support our genetic model. As already mentioned, our model does not consider the degree of tissue outgrowth of the forelimb bud, as the time scale of these biochemical reactions induced by developmental signals (see Fig. 2C) should be much faster than tissue growth. Previous studies, however, indicate that tissue outgrowth and epigenetic changes at target genes could be contributing factors for differentiation of the distal forelimb bud into prospective zeugopod and autopod regions (Rosello-Diez et al., 2014; Uzkudun et al., 2015). Of note, Meis1/2 genes are stably repressed in the distal forelimb bud (Mercader et al., 2000, 2009). Thus, our study indicates that PcG factors play two distinct roles in the control forelimb specification. First, at the E10.5 forelimb region, PcG factors regulate Meis1/2 expression as a threshold adjuster of RA-related signals. Second, in the later stages of limb development (post E10.5), PcG factors mediate robust gene repression to lock-down tissue specification stably. Full elucidation of the roles of PcG factors will therefore extend our understanding of PD specification of the forelimb bud.

It is noteworthy that the molecular mechanisms by which PCGF3/5-PRC1 and RA-related signals are linked are not fully understood yet. Our study reveals that ATRA-mediated inhibition of PcG binding at Meis2 accompanies accumulation of H3K27ac, which is mediated by p300 and CBP co-activators (Jin et al., 2011; Tie et al., 2009) (Fig. 1 and Fig. 3G, left). It is known that RARs interact with both co-activators or co-repressors, in an RA-dependent manner, to facilitate active or repressive chromatin domains, respectively (Chen et al., 1997; Nagy et al., 1997). We thus speculate that in the absence of RA-related signals, or when such signals are less abundant, co-repressors mediate the interactions between PcG factors and RARs. Indeed, co-repressors such as CTBP and histone deacetylases can associate with both RAR complexes and PcG factors, in the absence of exogenous RA (Sánchez et al., 2007; Sewalt et al., 1999; van der Vlag and Otte, 1999). We therefore hypothesize that PCGF3/5-PRC1 cooperates with RARs to form complexes with co-repressors, in the absence of abundant RA-related signals (Fig. 3G, right). Intriguingly, it has been reported that PCGF3/5-PRC1 could also be functionally linked with p300 to mediate gene activation (Zhao et al., 2017). Selective interactions between PCGF3/5-PRC1 and either co-activators or co-repressors that depend on the amount of RA may contribute to polarization of Meis2 expression in the forelimb bud (Fig. 3G). PCGF3/5-PRC1 may partly mediate the RA-dependent competition between co-activators and co-repressors to adjust the threshold for RA-related signals.

Previous studies have revealed functional interactions between PcG factors and not only RA-related signals but also developmental signal pathways such as Hedgehog, WNT and TGFβ (Boukarabila et al., 2009; Gaarenstroom and Hill, 2014; Marino, 2005). Importantly, these pathways also associate with co-activators in a ligand-dependent manner (Akimaru et al., 1997; Itoh et al., 2000; Roose et al., 1998). Molecular mechanisms that potentially link PcG factors and RARs, in a ligand-dependent manner, to downregulate Meis2 expression could also be employed in other morphogenetic signal pathways to alter target gene expression during development. Our study, by showing a link between PCGF3/5-PRC1 and developmental regulation of the Meis2 gene, will therefore contribute to a better understanding of the regulatory mechanisms for developmental signal-dependent recruitment of PcG factors at target genes. Intriguingly, we also show that activities of PcG factors were not affected by excess RA signals at Pitx2, Hoxa11 or Hoxa13 promotors, suggesting another aspect of PcG activity in protecting CGI promoters from unnecessary developmental signals.

All animal experiments were carried out according to the in-house guidelines for the care and use of laboratory animals of the RIKEN, Yokohama Institute, Japan. Cyp26b1^+/−, RARE-hsplacZ, Prx1-Cre;Ring1B^fl/fl, Prx1-Cre;Ring1A^−/−;Ring1B^fl/fl and Prx1-Cre;Ring1A^−/−;Ring1B^fl/fl;Meis2^fl/fl mouse lines were described previously (Rossant et al., 1991; Yashiro et al., 2004; Yakushiji-Kaminatsui et al., 2016). Generation of Pcgf3^fl/fl and Pcgf5^fl/fl mice is described in Fig. S2B,D. Pcgf3-KO mice were obtained by crossing Pcgf3^+/− mice. Pcgf3/5 double mutant embryos were obtained by mating male ERT2Cre;Pcgf3^fl/fl;Pcgf5^fl/fl and female Pcgf3^fl/fl;Pcgf5^fl/fl mice and subsequent intraperitoneal tamoxifen injection. Generation and analysis of Pcgf1^fl/fl mice will be described elsewhere. To induce deletion of Pcgf1, Pcgf3 and/or Pcgf5, we injected 0.1 ml of 15 mg/ml tamoxifen (Sigma-Aldrich) into the peritoneal cavity at E8.5 and sampled embryos at E10.5. Pcgf3/5 double mutant embryos shown in Fig. 3B were obtained by the following procedure: embryos were generated by in vitro fertilization and cultured until the blastocyst stage. To induce deletion of Pcgf3/5, embryos at the morula stage were further cultured in KSOM medium with 4-hydroxytamoxifen (7.7 µM, Sigma-Aldrich) for 24 h, followed by transplantation to the foster mothers (E2.5) and sampling at E7.5. Primers used for genotyping of respective alleles are listed in Table S1.

Skeletons of fetal and newborn mice were stained with Alizarin Red and Alcian Blue as described previously (Parr and McMahon, 1995). Whole-mount in situ hybridization was performed according to the methods described by Wilkinson and Nieto (1993). Whole-mount detection of β-galactosidase activity by X-gal staining was carried out as described previously (Loughna and Henderson, 2007).

Retinoic acid (RA) treatment was performed as previously described (Yashiro et al., 2004). Wild-type embryos (BDF2) that were transplanted into ICR female recipients received all-trans RA (ATRA) (Sigma-Aldrich) at a dose of 100 mg/kg body weight in sesame oil (Sigma-Aldrich) at E10.25 by maternal oral administration and were sampled 6 h (E10.5) or 18 h (E11.0) after ATRA treatment. Embryos obtained by crossing Ring1B^fl/fl females with Prx1-Cre;Ring1B^fl/+ males received ATRA at a dose of 5 mg/kg body weight at around E10.25 and were sampled 6 h (around E10.5) after oral administration. Experimental control pregnant females were administered sesame oil (vehicle) only. These embryos were subjected to in situ hybridization, DNA-FISH and ChIP-qPCR analyses.

DNA-FISH analysis was performed as previously described (Kondo et al., 2014; Yakushiji-Kaminatsui et al., 2016). Confocal images were captured with 65 nm pixels in xy and 300 nm steps in the z-plane by microscopy (Olympus UPlanSApo 1003 NA 1.40 and PlanApo N 603 NA1.42). We deconvoluted the images and measured the distances of foci using the Volocity application (Improvision). P-values were determined by χ² test.

Micro-dissected proximal and distal forelimb buds at E10.5 from wild-type (BDF2) embryos treated with vehicle or ATRA (+RA), Cyp26b1^+/+ (wild type) and Cyp26b1^−/− and distal forelimb buds at E10.5 from Pcgf3^fl/fl;Pcgf5^fl/fl (control) or ERT2-Cre; Pcgf3^fl/fl;Pcgf5^fl/fl embryos were used for ChIP-qPCR. The most proximal (Meis2-expressing domain) and distal tips of forelimb buds at E11.5 from wild-type (CD1) embryos were used for ChIP-seq. ChIP was performed as described previously (Yakushiji-Kaminatsui et al., 2016). For each immunoprecipitation, anti-RING1B (Atsuta et al., 2001), anti-H3K27me3 (07-449, Millipore), anti-H3K4me3 (07-473, Millipore), anti-H3K27ac (ab4729, Abcam) and anti-RAR (sc-773X, Santa Cruz) were used. ChIP-qPCR was performed on an Mx3005P system using the Brilliant SYBR Green QPCR Master Mix (Agilent Technologies). ChIP-qPCR analysis using RING1B/H3K27me3 and H3K4me3/H3K27ac (except for Pcgf3/5 mutant embryos) was carried out with three or two biological replicates, respectively. ChIP-qPCR analysis using Pcgf3/5 mutant was performed with six biological replicates. Primer sequences are listed in Table S1. For ChIP-Seq, at least 5 ng of purified DNA was used to make libraries and the material was sequenced with 100-bp single-end reads on an Illumina Hiseq2500 according to the manufacturer's protocol (Illumina).

Demultiplexed ChIP-seq reads were mapped onto the mm10 or hg19 using Bowtie (Version 0.12.7) (Langmead et al., 2009), with parameters ‘-m1 –strata –best’ according to conditions described previously (Riising et al., 2014), and PCR duplicates were removed from mapped reads using SAMtools (Version 0.1.18) (Li et al., 2009). All ChIP-seq data shown in this research were normalized to get a 1× depth of coverage by using bamCoverage (Version 2.5.0), and the difference (subtraction of vehicle from treated samples of the normalized number of reads in Fig. S1A) was analyzed using bamCompare (Version 2.5.0) (Ramírez et al., 2014, 2016). All analysis was performed with the Denis Duboule lab Galaxy server (the Bioteam Appliance Galaxy Edition, bioteam.net, bioteam.net/products/galaxy-appliance) (Afgan et al., 2016). For Fig. S1, H3K27ac and RAR ChIP-seq data before and after tamibarotene treatment of human AML cells were obtained from the NCBI SRA database (SRP103029, see also Table S2). For Fig. S1B, short reads were aligned on the human genome (NCBI version 38) using bowtie2 (version 2.3.4) with default options and significantly enriched regions were detected using MACS2 (version 2.1.0) with an option to process broad peaks. Overlaps between RAR and H3K27ac were evaluated using inclusion of RAR and/or H3K27ac peaks in merged broad peaks that had significantly enriched and at least more than twice the number of reads than whole cell extract. RPKM values in all merged regions were calculated for all experiments and changes after tamibarotene treatment were evaluated.

We tested the network model (shown in Fig. 2D) to see if it could account for the expression patterns of Meis1/2, Fgf8 and Hoxa13 in ATRA-treated wild-type (+RA), Ring1A/B-dKO or Ring1A/B;Meis2-tKO forelimb buds at E10.5. We did not consider the impact of tissue growth, as the time scale of tissue growth is considerably slower than chemical reactions or diffusions. We sought to recapitulate the gene expression pattern at E10.5 (see Probst et al., 2011) by focusing on the RA-FGF gradient in the forelimb bud. We assumed that enzymatic reactions, such as association/dissociation between RA and its receptors and between transcription factors and their binding regions, are at equilibrium as time scales of these reactions are much quicker than those of gene expression.

The expression of Ring1A or Ring1B did not change between proximal and distal regions of the forelimb bud (Yakushiji-Kaminatsui et al., 2016). Thus, in our model the expression level of PcG proteins was constant (P). RING1A/B regulated proximal gene expression by competing with RA signaling. RA is inactivated by CYP26B1 and all variables decay linearly at rate γ_*. The expression of proximal genes, CYP26B1, and distal genes were modeled by a Hill-type function with maximum production rates α_*. Expression of the proximal genes (here, we considered both Meis2 and Meis1 as proximal genes because of their potential redundancy) is competitively regulated by RA and PcG proteins. RA and FGF8 are produced in the flank and in the AER, respectively, and both diffuse into the forelimb bud mesenchyme with a diffusion constant D_*. In this simulation, we used a fixed value of RA (R₀) at the flank. We assumed FGF8 influx from the AER was negatively regulated by Meis1/2 based on our previous results. The Fgf8 expression at the AER was downregulated in the Ring1A/B-dKO but was considerably restored in the Ring1A/B;Meis2-tKO mutants (Yakushiji-Kaminatsui et al., 2016). Zero-flux conditions were applied to other boundaries. We also assumed that Meis1/2 suppresses Hoxa13. Meis2 and Hoxa13 show complementary expression patterns in different conditions [e.g. in wild type or control (Ring1A-KO), ATRA-treated wild type (+RA) and Ring1 mutants] and downregulation of Hoxa13 in the Ring1A/B-dKO was considerably restored in the Ring1A/B;Meis2-tKO mutants (Yakushiji-Kaminatsui et al., 2016). This is further supported by our unpublished observations showing Meis2 binding to the Hoxa13 or Cyp26b1 promoters (T.K. and H.K., unpublished observations) and another study showing a PcG-mediated interaction of the Meis2 gene locus with the HoxA cluster in mouse embryonic stem cells (Schoenfelder et al., 2015).

Computer simulations were performed by applying the following relative values to respective components of equations. For RA and FGF8, normalized concentrations relative to their source levels were used (for FGF8, normalization by the source value was carried out for M=0). The values of K_* in Hill functions were chosen so that the expression patterns of CYP26B1, proximal and distal genes at the equilibrium, were similar to those observed in the wild type. Hill coefficients (h) were set to 2, and simulation results remained mostly unchanged for values other than 2. For Ring1A/B-dKO and Ring1A/B;Meis2-tKO, a smaller value for K_MR(P) compared with the wild type was adopted (see Table 1 for parameters). For Ring1A/B;Meis2-tKO, the value of α_M was also set to be smaller than the wild type, but was not zero as Meis1 is still expressed. For ATRA treatment, a larger value for the source level of RA was used.

We thank Dr Janet Rossant (The Hospital for Sick Children, Canada) and Dr Hiroshi Hamada (RIKEN, Japan) for providing RARE-lacZ and Cyp26b1^+/− mice, respectively, and our animal facility group staffs for animal care and two-cell transplantations to foster mothers. We deeply appreciate Dr Denis Duboule for his generous support.