Since most HH targets can be activated by loss of GLI repression, we hypothesized that enhancers may be activated by HH signaling when GLI repression is relieved. To test this, we first identified active GLI enhancers in the developing limb at embryonic day 10.5 (E10.5), when high levels of HH target gene expression are observed. We used an endogenously FLAG tagged Gli3 allele (Lopez-Rios et al., 2014; Lorberbaum et al., 2016) to identify GLI3 binding regions by ChIP-seq and then identified regions enriched for H3K27ac, a marker associated with active enhancers (Heintzman et al., 2007; Heintzman et al., 2009; Creyghton et al., 2010; Rada-Iglesias et al., 2011; Cotney et al., 2012). Altogether we identified 7,282 endogenous GLI3 binding regions (GBRs), with the majority of regions enriched for H3K27ac (83%; 6,064/7,282 GBRs) in wild-type (WT) limb buds which have active HH signaling (Figure 1—figure supplement 1A; Figure 1—source data 1). Nearly all nuclear GLI3 is present in the anterior half of the limb bud in the repressor form with little or no nuclear GLI3 present in the posterior half, consistent with previous findings (Wang et al., 2000) (Figure 1—source data 1B). Therefore, the GBRs identified in this study are likely to exclusively represent GLI3-repressor binding regions.

Next, we asked if HH signaling was required for the activation of GLI enhancers by performing ChIP-seq for H3K27ac in Sonic hedgehog (Shh) null E10.5 forelimbs, prior to overt phenotypes in Shh nulls, and comparing H3K27ac enrichment to that in WT limbs (Chiang et al., 2001) (Figure 1A; Figure 1—source data 2). Since Shh^-/- forelimbs have constitutive GLI repression, we hypothesized that in the absence of HH signaling, GLI repressors may prevent activation of their enhancers. We found that most H3K27ac enriched regions were present in both WT and Shh^-/- embryos (98.3%; 58,720/59,729 H3K27ac peaks); however a subset of 2,113 WT H3K27ac enriched regions had acetylation that was significantly reduced or completely lost in the absence of HH signaling (Figure 1—source data 2). We then asked whether those regions with reduced acetylation in the absence of HH signaling include GLI-bound enhancers by intersecting H3K27ac enrichment with the endogenous GBRs identified. We found that 94% of GBRs (5,715/6064 GBRs) with acetylation in WT limbs also retain H3K27ac in Shh^-/- limb buds, which we have termed Stable GBRs (Figure 1C,D). GBRs that remain stably acetylated regardless of HH signaling likely function as active enhancers whose activity is not predominantly regulated by HH signaling. However, H3K27ac enrichment was reduced or lost in the absence of HH signaling in a smaller subset of GBRs, suggesting that GLI repressor may regulate the activity of this group of enhancers. Within this GBR class with HH-responsive acetylation, we identified populations of GBRs that had either significant reductions (termed HH-sensitive; n = 148) or a complete absence of H3K27ac enrichment (termed HH-dependent; n = 201) in Shh^-/- limb buds (Figure 1B,C). The latter two categories are henceforth collectively referred to as HH-responsive GBRs. As H3K27ac is not exclusively localized to enhancers, we also examined the enrichment of histone H3K4me1, a general marker of primed and active enhancers, at these GBRs using publicly available data (ENCODE Project Consortium, 2012) (Figure 1—source data 3). In WT limb buds, 82% of HH-responsive GBRs are enriched for H3K4 mono-methylation, supporting that these regions are likely to act as enhancers (HH-sens: 123/148, 83%; HH-dep: 162/201, 81%).

To determine if HH-responsive GBRs are associated with HH target genes, we examined biologically validated GLI enhancers in the Gremlin and Ptch1 loci that mediate limb-specific transcription of these HH targets and found that they are among the HH-responsive class of GBRs (Figure 1E,F) (Vokes et al., 2008; Zuniga et al., 2012; Li et al., 2014; Lopez-Rios et al., 2014). This suggests that HH-responsive enhancers may regulate limb-specific gene expression in response to HH signaling. Consistent with this possibility, we found that HH-responsive GBRs are highly enriched around the TSS (2 kb upstream to 1 kb downstream) of genes that have reduced expression in Shh^-/- limb buds (Lewandowski et al., 2015). In contrast, Stable GBRs have minimal, albeit still significant enrichment around HH target genes (p=0, permutation test; Figure 1G).

We observed many HH-responsive H3K27ac regions that change acetylation status in response to HH signaling but are not bound by GLI3. This prompted us to ask asked if these regions cluster near GBRs. HH-responsive non-GLI binding regions cluster together and are significantly enriched around HH-responsive GBRs, and to a lesser extent, near Stable GBRs (Figure 1—figure supplement 1C,D). We conclude that HH-responsive GBRs cluster with other HH-responsive regulatory regions, and are strongly associated with HH target genes, supporting their role in driving gene expression in response to HH signaling during limb development.

Although Stable GBRs are not highly enriched at HH target genes, 62% of them (3,544/5,715) are located in close proximity to the promoters of genes (2 kb upstream to 1 kb downstream of TSS), compared to 26% (91/349) of HH-responsive GBRs (Figure 1H). Most promoter-associated Stable GBRs (90%; 3,190/3,544) are found at promoters associated with CpG islands (defined as a TSS with a CpG region within 5 kb upstream to 2.5 kb downstream), a quality typically associated with housekeeping genes, and genes that tend to be more broadly expressed and less tissue-specific (Zhu et al., 2008). To examine how different classes of GBRs might be differentially regulated, we examined their GLI binding motifs. A significantly higher percentage of HH-dependent and HH-sensitive GBRs contain GLI motifs compared to Stable GBRs (69.7% HH-dep., 57.4% HH-sens., 39.5% Stable). HH-dependent and HH-sensitive GBRs also contain a higher density (1.51 HH-dep., 1.47 HH-sens., 1.27 Stable) and higher quality of GLI motifs compared to Stable GBRs (5.75 HH-dep., 5.56 HH-sens., 4.88 Stable)(Figure 1I). Interestingly, we did not uncover high levels of enrichment of other motifs using de novo motif analysis (Figure 1—source data 4). Additionally, Stable GBRs are slightly more conserved than HH-dependent, but not HH-sensitive GBRs (see Discussion) (Figure 1—figure supplement 1E).

GLI activators have been shown to recruit demethylases that remove H3K27me3, a hallmark of the Polycomb repressor complex (PRC2) to promote transcriptional activation of several HH target genes, most notably Gli1 and Ptch1 (Margueron and Reinberg, 2011; Shi et al., 2014; Lorberbaum et al., 2016). If PRC2 is recruited by GLI repressors, there should be enrichment of H3K27me3 at HH-responsive enhancers in Shh^-/-, where maximal levels of GLI repression would lead to recruitment of PRC2 and thus methylation at these enhancers. Contrary to this prediction, we identified a minimal number of HH-responsive GBRs enriched for H3K27me3 in E10.5 Shh^-/- limb buds (31/349 GBRs; Figure 2A–C, Figure 2—source data 1). As reported for MEFs (Shi et al., 2014), these methylated GBRs include the pathway target Gli1 in addition to other pathway target genes such as Ptch1 and Ptch2 (Figure 2B). In contrast, most HH-responsive GBRs (318/349) and signature target gene promoters (14/22) lack enrichment of H3K27me3 in the absence of HH signaling (Figure 2C; Figure 2—figure supplement 1; Figure 2—source datas 1 and 2 ). We conclude that while the PRC2 complex has the potential to regulate a small number of HH pathway target genes, it is not the primary mechanism by which GLI repressors prevent target gene expression.

We considered two possible mechanisms by which GLI repression could regulate H3K27ac enrichment in response to HH signaling: first, GLI repression could cause large-scale modifications to chromatin at enhancers resulting in an overall loss of their identity as enhancers. Alternatively, GLI repressors could regulate H3K27ac specifically. To address the first mechanism, we asked if HH regulates H3K4me2, another histone modification enriched at active enhancers and most promoters (Ernst et al., 2011; Pekowska et al., 2011; Wang et al., 2014). Consistent with H3K4me2 being enriched at promoters and our finding that Stable GBRs are enriched around promoters, we find H3K4 di-methylation at 73% of Stable GBRs (4,172/5,715), while only 26% of HH-responsive GBRs (91/349) which are less enriched around promoters. None of the GLI-bound H3K4me2 enriched regions had significant reductions in H3K4me2 in Shh^-/- limbs compared to WT controls (Figure 2D,E). Furthermore, essentially all peaks remained unchanged between the two genotypes, where only 12 peaks were reduced in Shh^-/- limbs, none overlapping with GLI binding regions or non-GBR HH-responsive peaks (Figure 2—source data 3).

H3K4me2 marked most Stable GBRs, but only a subset of HH-responsive GBRs which are primarily located within 2 kb upstream to 1 kb downstream of TSS (79% (72/91) of H3K4me2+ HH-responsive GBRs are near promoters). Since we found that most HH-responsive GBRs in wildtype limb buds are enriched for H3K4me1 (see results above), we asked if this mark was altered at HH-responsive GBRs in response to HH signaling. We performed ChIP on WT and Shh^-/- limb buds and assessed enrichment of H3K4me1 at several HH-responsive GBRs by quantitative PCR, selecting intergenic regions that would not overlap with promoters (regions are at least 7 kb from the nearest TSS). All tested regions retained H3K4me1 enrichment in Shh^-/- limb buds (Figure 2F). We conclude that HH-responsive regions retain enrichment of other active or poised enhancer marks, suggesting that HH signaling and GLI repression specifically regulate H3K27ac enrichment at these regions.

The dynamic acetylation of HH-responsive GBRs, yet unaltered methylation of H3K4 in Shh^-/- limb buds are properties consistent with ‘poised’ enhancers, which retain H3K4me1 and accessible chromatin in the absence of H3K27ac (Heintzman et al., 2009; Creyghton et al., 2010; Rada-Iglesias et al., 2011). Therefore, if HH-responsive enhancers are not active but ‘poised’ in the absence of HH, we predicted that chromatin accessibility would be unchanged in response to HH signaling. Using ATAC-seq to measure regions of open chromatin, we compared the accessibility of GBRs between WT and Shh^-/- posterior limb buds, a fraction providing a more homogenous WT population of cells exposed to HH signaling (Figure 3A; Figure 3—source data 1) (Buenrostro et al., 2013; Buenrostro et al., 2015). Overall, in HH stimulated WT limbs, 87% of Stable GBRs (4,978/5,715) are accessible, while only 66% of HH-responsive GBRs (232/349) are accessible, suggesting a more restricted accessibility of HH-responsive GBRs even in WT conditions (Figure 3B–C). To determine if these regions are likely to be enhancers, we analyzed the co-enrichment of the enhancer markers H3K4me1 and H3K4me2 at ATAC accessible (ATAC+) and inaccessible (ATAC-) HH-responsive GBRs. 93.5% (217/232) of ATAC+ regions are co-enriched with H3K4me1/2 while 72% (84/117) of ATAC- regions are co-enriched with H3K4Me1/2. These results suggest that most of the ATAC- regions are likely to correspond to real enhancers though at a somewhat reduced frequency compared to ATAC+ regions. Contrary to expectations for a poised enhancer, both HH-sensitive and HH-dependent GBRs have significantly reduced accessibility compared to Stable GBRs in the absence of HH signaling, with the majority of HH-responsive GBRs being more compact in Shh^-/- compared to wild-type limbs (Figure 3D–F). Overall, we conclude that HH-responsive GBRs are less accessible than Stable GBRs, with access being further restricted in Shh^-/- limb buds, which have constitutive GLI repression.

The presence of multiple GLI proteins and their bifunctional roles as both transcriptional activators and repressors has made it challenging to determine how HH genes are primarily regulated. To test the roles of activator and repressor on enhancers, we performed H3K27ac ChIP on Shh^-/-;Gli3^-/- limb buds (devoid of GLI activators and most GLI repressors). We hypothesized that loss of H3K27ac at most HH-responsive enhancers in the absence of HH signaling is due to constitutive GLI repression preventing acetylation of GLI enhancers. Thus, in Shh^-/-;Gli3^-/- limbs, we predicted H3K27ac should be maintained at HH-responsive enhancers. Alternatively, if GLI activator is required, H3K27ac would remain absent or reduced as it does in Shh^-/- limbs (Figure 1A).

To overcome the reduced tissue available for ChIP samples, we optimized a ‘MicroChIP’ approach to allow ChIP-seq on single pairs of limb buds and assessed H3K27ac enrichment at GLI enhancers in E10.5 Shh^-/-;Gli3^-/- limb buds (Figure 4A; Figure 4—source data 1). As anticipated, there was reduced signal compared to our standard protocol, however we were still able to detect many of the HH-responsive GBRs (59%; 207/349) and most Stable GBRs (91%; 5,211/5,715). Consistent with expectations, HH-responsive GBRs associated with Gli1 and Ptch1, which require GLI activation (Litingtung et al., 2002; te Welscher et al., 2002), had greatly reduced H3K27ac enrichment in the double mutants along with a small number of additional GBRs (24 total; Figure 4B,D,E). However, consistent with a GLI repression-driven model, most HH-responsive GBRs retained or increased H3K27ac enrichment in the absence of both GLI activator and repressor (88%; 183/207; Figure 4C–E). Despite being unchanged in Shh^-/- limbs, Stable GBRs had slight but significant increases in H3K27ac enrichment (Figure 4F), indicating that on a population level, some of these regions respond to GLI repression (see Discussion).

In a parallel series of experiments, we noted that HH-responsive GBRs have higher levels of H3K27ac enrichment in posterior limb halves, where HH is active, compared to anterior limb halves, which have little exposure to HH and are dominated by GLI repression (Figure 4G). This contrasts with Gli3^-/- limb buds, where H3K27ac levels in anterior halves are comparable to those in posterior halves in many GBRs, as both domains lack GLI repression (Figure 4H). Together these results strongly support a GLI repressor centric mode of regulation where GLI de-repression is responsible for activation of most GLI limb enhancers. We conclude that GLI activator does not mediate acetylation levels at most HH-responsive GBRs.

The simplest interpretation of the above results is that GLI repressor regulates the activity of histone deacetylases (HDACs) at HH-responsive GBRs, in which loss of an HDAC-GLI repressor complex leads to acetylation. To test this, we cultured limb buds in the presence of the HDAC inhibitors FK228 (Furumai et al., 2002) or SAHA for 2 hr. As expected, there were greatly upregulated levels of H3K27ac within two hours of treatment (Figure 5—figure supplement 1A,B). We then dissected the anterior halves of limb buds cultured in control or HDAC inhibitor-containing media and compared the levels of H3K27ac enrichment at HH-responsive GBRs previously shown to have enriched H3K27ac levels in posterior limb halves (+HH, no GLI repression) (Figure 4G). Inhibition of HDACs with both FK228 and SAHA resulted in increased acetylation at HH-responsive enhancers compared to untreated control anterior limb buds (Figure 5A). The increased enrichment of H3K27ac acetylation in HDAC-inhibited anterior limb buds was comparable to that seen in posterior limb buds (Figure 4G). HDACs could regulate H3K27ac activity in a GLI-responsive fashion through a variety of different mechanisms including direct interactions with responsive GBRs, potentially mediated by a repression complex including GLI3 proteins and HDACs. We asked if GBRs were bound by HDACs, focusing on HDAC1, which along with HDAC2 is preferentially inhibited by FK228 (Furumai et al., 2002). We identified HDAC1 binding regions in E11.5 limb buds by CHIP-seq and intersected them with GBRs. 78% (4,109/5,282) of stable GBRs and 41% (144/349) of HH-responsive GBRs overlapped with HDAC1 peaks (Figure 5B–D), consistent with a possible role for HDAC1 in regulating H3K27ac levels. We conclude that GLI repressors regulate H3K27ac levels at HH-responsive GBRs through HDACs (see discussion).

Having identified distinct classes of GBRs that respond differently to HH signaling, we next addressed the biological significance of these properties. To this end, we used the VISTA enhancer database to identify a total of 305 Stable and 23 HH-responsive GBRs that had been tested for enhancer activity in transgenic embryos (Visel et al., 2007). While nearly half of each class have enhancer activity in the limb, HH-responsive GBRs tend to drive activity specific to the HH-responsive posterior limb bud, while Stable GBRs tend to have activity throughout the limb or regions that are not responsive to HH (Figure 6A,B) (Ahn and Joyner, 2004; Probst et al., 2011; Lewandowski et al., 2015). Additionally, HH-responsive enhancers are active more specifically within the limb (drive expression in an average of 1.9 tissues) while Stable GBRs are more broadly active throughout the embryo (drive expression in an average of 2.9 tissues; p<0.01; Figure 6C; Figure 6—source data 1). While all GBRs examined in the VISTA database with limb activity are by definition enriched for H3K27ac, 91% of HH-responsive GBRs and 95% of Stable GBRs are also enriched for H3K4me1. Additionally, all GBRS are enriched for at least two markers of enhancers (H3K27ac, H3K4me1, H3K4me2, ATAC) while most are enriched for 3–4 of these markers (67% HH-responsive GBRs; 93% Stable GBRs) (Figure 6D,E).

These results suggest that Stable GBRs act as general enhancers that drive expression in multiple tissues, while HH-responsive GBRs mediate tissue-specific expression. To test this in another biological context, we treated HH-responsive NIH3T3 cells with and without the HH agonist purmorphamine, identified H3K27ac enriched regions by ChIP-Seq, and assessed the H3K27 acetylation status of different classes of limb GBRs. Strikingly, only 12% (42/349 GBRs) of HH-responsive limb GBRs are acetylated in response to HH signaling in NIH3T3 cells. An additional 18% (63/349 GBRs) of HH-responsive limb enhancers have stable acetylation in NIH3T3 cells, while most lack any activity. In contrast, 70% (4,001/5715) of Stable GBRs in the limb are still active in NIH3T3 cells in both untreated and HH stimulated cells (Figure 6F,G; Figure 6—source data 2). We conclude HH-responsive GBRs are tissue specific enhancers that mediate HH signaling, while Stable GBRs have broadly expressed enhancer activity.

Experiments involving mice were approved by the Institutional Animal Care and Use Committee at the University of Texas at Austin (protocol AUP-2016–00255). The Gli3^Xt-J and Shh^tm1amc null alleles have been described previously (Hui and Joyner, 1993; Dassule et al., 2000) and were maintained on a Swiss Webster background. The Gli3^3XFLAG allele, with an N-terminal 3XFLAG-epitope, (Lopez-Rios et al., 2014; Lorberbaum et al., 2016) was maintained on a mixed background. For ChIP and ChIP-seq experiments, fresh E10.5 (32–35 somite) forelimb buds were pooled from multiple litters to obtain sufficient Gli3^-/- and Shh^-/- mutant embryos along with somite matched controls (Swiss Webster embryos for Gli3^-/- experiments and a mixture of WT and heterozygous littermates for Shh^-/-) embryos. For ATAC-seq, fresh pairs E10.5 (35 somite) posterior forelimb buds were dissected from individual embryos.

To inhibit HDAC1/2, E10.5 embryos (32–35S) were dissected in warm limb bud culture media (Panman et al., 2006) and explants still attached to the body wall were cultured in 250 nM of HDAC inhibitor FK228 (Selleckchem S3020), 20 μM of the HDAC inhibitor SAHA (Selleckchem MK0683), or DMSO vehicle control, for two hours at 37C. For each condition, 20–25 embryos were used (n = 4). After incubation, the explants were changed into fresh media (without inhibitor) to dissect anterior limb buds. Cells from anterior limbs were then dissociated and processed for ChIP.

NIH3T3 were authenticated by and purchased from ATCC (NIH3T3 CRL-1658). They have tested negative for Mycoplasma. Cells were seeded on 6 cm plates with 5 × 10^5 cells and grown for three days until completely confluent. They were then switched to low serum (0.5%) and treated with 400 nM purmorphamine (Stemgent 04–0009) or 0.01% DMSO (vehicle control) for 2 days. Under these conditions, a representative purmorphamine-treated sample had substantial elevation of the canonical HH target genes Ptch1 and Gli1 compared to controls (47-fold and 697-fold enrichment, respectively). NIH3T3 cells (ATCC CRL-1658) were authenticated and purchased directly by vendor, and tested negative for Mycoplasma.

Whole limb buds from a single litter were lysed for 1 hr at 4C. For fractionation, 500,000 cells from limb buds were then dissociated with 100 ug/mL Liberase (Roche 05401119001), resuspended in CSKT buffer (10 mM PIPES pH6.8, 100 mM NaCl, 300 mM sucrose, 3 mM MgCl₂, 1 mM EDTA, 1 mM DTT, 0.5% TritonX-100, incubated on ice for 10 min, and centrifuged for 5 min @ 5000 g. The cytoplasmic fraction (supernatant) and nuclear pellet were each resuspended in loading dye and boiled. Western blots were incubated with the following primary antibodies for 1 hr at room temperature in 3% milk: 1:4000 M2 Flag (Sigma 3165),1:4000 H3 (Cell Signaling 4499), 1:1000 GAPDH (Cell Signaling 5174), 1:1000 H3K27ac (Abcam Ab4729), 1:2000 B-actin (Cell Signaling 8457). Secondary antibodies were incubated for 1 hr at room temperature in 3% milk: 1:5000 Donkey anti-mouse (Jackson 715-035-150), Donkey anti-rabbit (Jackson 711-005-0152).

ChIP experiments were performed as previously described (Vokes et al., 2008) with the following modifications. Histone ChIPs were performed on whole E10.5 (32S-35S) forelimbs pooled from 6 to 8 embryos. The GLI3-FLAG ChIP and the H3K27ac ChIP on cultured and treated limbs were performed on E10.5 (32–35S) forelimbs from 20 to 25 pooled embryos. The HDAC1 ChIP was performed on pooled forelimbs and hindlimbs from 30 E11.5 (40–44S) embryos. Cells were dissociated with 100 ug/ml Liberase (Roche 05401119001) and fixed: 15 min for H3K27ac. 30 min for GLI3-FLAG and 7 min for HDAC1 at room temperature in 1% formaldehyde. After cell lysis, chromatin was sheared. H3K27ac ChIP samples were sheared in buffer containing 0.25% SDS with a Covaris S2 focused ultrasonicator using the following settings: Duty Cycle: 2%, Intensity: 3, Cycles/burst: 200, Cycle time: 60 s, Power mode: frequency sweeping. GLI3-FLAG ChIP samples were sheared using a Branson Sonifier for 10 cycles, 30 s on/60 s off, intensity 3.5. HDAC1 samples were sheared using a Diagenode Bioruptor for 5, 10 min cycles: 30 s on/60 s off, on high power. Sheared chromatin was then split into 3 ChIP reactions and incubated with antibody-dynabead preparations overnight. The H3K27ac antibodies for conventional ChIP were from Diagenode (C15200184) and Abcam (ab4729), while the H3K27Ac antibody for MicroChIPs was from Diagenode (C15410196). Additional antibodies recognized H3K4me1 (Millipore ABE1353) H3K4me2 (Millipore 07–030) and H3K27me3 (Abcam ab195477), FLAG (Sigma F3165) and HDAC1 (Abcam ab7028). Beads were washed 5 times with RIPA buffer (1% NP40, 0.7% Sodium Deoxycholate, 1 mM EDTA pH8, 50 mM Hepes-KOH pH7.5, 2% w/v Lithium Chloride) and 1 time with 100 mM Tris pH8, 10 mM EDTA, 8.0, 50 mM NaCl and then eluted at 70°C for 15 min. For HDAC1 ChIPs beads were washed twice with low salt buffer (0.1% Deoxycholate, 1% Trition X-100, 1 mM EDTA, 50 mM Hepes-KOH pH 7.5, 150 mM NaCl), once in high salt buffer (0.1% Deoxycholate, 1% Trition X-100, 1 mM EDTA, 50 mM Hepes-KOH pH 7.5, 500 mM NaCl), once in LiCl buffer (250 mM LiCl, 0.5% NP-40, 0.5% Deoxycholate, 1 mM EDTA, 10 mM Tris-HCl pH 8), and 2x washes in TE buffer (10 mM Tris-HCl pH 8, 1 mM EDTA). Crosslinking was reversed overnight at 70°C. Chromatin was purified and concentrated, then subjected to quantitative PCR and/or library preparation and sequencing. Quantitative PCR-based analysis was performed using SensiFAST SYBR-LoROX (Bioline BIO-94020) on a Viia7 system (Applied Biosystems). ChIP regions subsequently tested by qPCR are referred to in the figures by the unique peak ID number (Figure 1—source data 2). For each biological replicate, 2–3 technical replicates were performed for each qPCR reaction and the Ct values were averaged. Chromatin enrichment was determined by calculating delta delta Ct method (Livak and Schmittgen, 2001) against a control region (C1).

Primers are described below. Primers are identified by their H3K27ac Peak ID. Primers labeled #1–5 are HH-dependent GBRs.

H3K27ac ID	Primers	GBR coordinate	GBR type	Comments
32467 (#1)	F: ACGCAGGCAGTTCCAATACA R: AGGGACTTCACCCAGTTCCA	Chr2:113640572–113641614	HH-dep.	GRE1 (near Grem1)
15198 (#2)	F: CCCTCCATTCTCCCTCCTTA R: GGACCTTTCCGTTGAAGTGA	Chr13:63950822–63952750	HH-dep.	randomly selected GBR
2666 (#3)	F: CTGGCTCCCAGAATCTCTCA R: TGTGCCCCATCTCTTTCAG	Chr1:155211962–155213426	HH-dep.	randomly selected GBR
45094 (#4)	F: GGGAGGGGTGAACTTGTCTT R: TGCAAATGAACACACGCATA	Chr5:134073187–134074116	HH-dep.	randomly selected GBR
20941 (#5)	F: TTCCCAGCTCAAGGTCATGT R: AGGAGGCAATGAAGACACTGG	Chr15:86429678–86430690	HH-dep.	randomly selected GBR
41492	F: AGAAGGACTCCTATGTGGGTGA R: ACTGACCTGGGTCATCTTTTCA	NONE	NONE	Beta actin- normalizing target
41492	F: AGCTAACAGCCTGCCCTCTG R: TTTTCCGGTGGTACCCTACG	NONE	NONE	Beta actin-normalizing target for H3K4me1
NONE (C1)	F: GCCAGAATTCCATCCCACTA R: CCAATAACCTGCCCTGACAT	NONE	NONE	negative normalizing

Samples were processed for ‘MicroChIP’ using the Diagenode True MicroChIP kit (Cat #C01010130) with the following modifications. Briefly, individual limb pairs (~100 k cells) of wildtype, Shh^-/- and Shh^-/-;Gli3^-/- E10.5 embryos (33–34S) were processed separately by dissociating limb buds with 100 ug/mL Liberase (Roche 05401119001), crosslinked for 10 min, lysed and then sheared. Samples were sheared on a Diagenode BioRuptor for six cycles on high, 30 s on/off and processed through shearing while genotyping in parallel for Shh^-/-;Gli3^-/- and wildtype littermates (Shh^+/+;Gli3^+/+). Sheared chromatin was then incubated with H3K27ac antibody (Diagenode C15410196) overnight and Protein A magnetic beads (Diagenode C03010020) the following day for 2 hr. Chromatin-bound beads were washed, eluted and de-crosslinked and purified using MicroChIP DiaPure columns (Diagenode C03040001).

The ChIP-seq raw datasets from this study have been deposited in GEO (GSE108880) (see Source Data for Figures 1–5 for processed ChIP-seq and ATAC-seq data). The H3K4me1 data used in this study (GSE86690) were processed and analyzed as all other ChIP experiments were done, described below. All chromosomal coordinates refer to the mm10 version of the mouse genome.

After ChIP was performed as described above, libraries were generated using the NEBNext Ultra II library preparation kit with 15 cycles of PCR amplification (NEB E7645) or generated using the MicroPlex library prep kit (Diagenode C05010012) and sequenced to a depth of >40 million reads per sample for both ChIP and ‘MicroChIP’ experiments, using two biological replicates. Peaks were called using CisGenome version 2.1.0 (Ji et al., 2008). To identify differentially enriched peaks in the WT and Shh^-/- limb buds (or control and purmorphamine-treated NIH3T3 cells), the peaks were merged to determine how many WT, WT input, Shh^-/- and Shh^-/- input reads overlapped with the peak region. The read numbers were adjusted by library size and log2 transformed after adding a pseudo-count of 1. The differential analysis between WT and WT input used limma (Ritchie et al., 2015). The FDR of the differential test was obtained and peaks with FDR < 0.05 are determined as having differential signal between WT and WT input. The same differential analysis procedure was repeated to compare between Shh^-/- and Shh^-/- input, and between WT and Shh^-/-. To determine GLI motif quality, de novo motif discovery was performed on the 1000 GBRs with the highest quality using the flexmodule_motif function in CisGenome to identify the GLI motif. The GLI motif was mapped to the mouse genome using the motifmap_matrixscan_genome function in CisGenome software with default parameters.

Individual pairs of posterior forelimb fractions were dissected from 35 somite wildtype (n = 2) or Shh^-/- embryos (n = 2). ATAC used components from the Nextera DNA Library Preparation Kit (Illumina) as described previously (Buenrostro et al., 2015) with the following variations. 5,000 cells from each sample were added into each reaction and cells were lysed on ice for 8 min. prior to centrifugation. Libraries were generated using 18 cycles of PCR amplification with NEB high fidelity 2x master mix (New England Biolabs), cleaned up with AMPure XP beads (Beckman Coulter) and sequenced on an Illumina NextSeq 500 using PEx75 to a depth of 30 million reads. Peaks were called using MACS2 with a fixed window size of 200 bp and a q-value cutoff of 0.05. Differential analysis of wildtype versus Shh^-/- peak signals was performed essentially as described for ChIP above using limma (Ritchie et al., 2015).