GLI transcriptional repression is inert prior to Hedgehog pathway activation

The Hedgehog (HH) pathway regulates a spectrum of developmental processes through the transcriptional mediation of GLI proteins. GLI repressors control tissue patterning by preventing sub-threshold activation of HH target genes, presumably even before HH induction, while lack of GLI repression activates most targets. Despite GLI repression being central to HH regulation, it is unknown when it first becomes established in HH-responsive tissues. Here, we investigate whether GLI3 prevents precocious gene expression during limb development. Contrary to current dogma, we find that GLI3 is inert prior to HH signaling. While GLI3 binds to most targets, loss of Gli3 does not increase target gene expression, enhancer acetylation or accessibility, as it does post-HH signaling. Furthermore, GLI repression is established independently of HH signaling, but after its onset. Collectively, these surprising results challenge current GLI pre-patterning models and demonstrate that GLI repression is not a default state for the HH pathway. GLI repression has been presumed to be the default transcriptional state and important for pre-patterning tissues. Challenging current models, the authors show that GLI3 repression is inert in the limb bud before the onset of Hedgehog signaling.

T he Hedgehog (HH) signaling pathway is one of the major developmental regulators of tissue-specific development and differentiation. GLI proteins mediate transcriptional responses to the pathway in a strikingly cell-specific fashion. In the absence of HH pathway activation, GLI transcriptional repressors (GLI-R) prevent the activation of HH target genes, while upon exposure to HH, GLI proteins undergo alternative post-translational processing into transcriptional activators (GLI-A) [1][2][3][4][5] . The processing of GLI-A and GLI-R are both directly dependent on processes localized within the primary cilium and defects in ciliary components lead to characteristic misregulation of HH target genes in a large group of birth defects termed ciliopathies 6 . While GLI-A directly activates a subset of HH targets, most genes are 'de-repressed' by the loss of GLI-R alone; they do not require GLI-A-mediated activation for their expression. The importance of de-repression is exemplified in the developing limb where the phenotype of Sonic Hedgehog (Shh) null limb buds is dramatically improved in Shh;Gli3 mutants [7][8][9] . In this context, the loss of GLI3-mediated repression, even in the absence of pathway activation, is sufficient to restore expression of most GLI target genes and many aspects of limb growth and patterning.
GLI3 represses transcription, at least in part, by epigenetically regulating a subset of its own enhancers. Its properties include reduced enrichment levels of the active enhancer mark H3K27ac and reduced chromatin accessibility at a subset of HH-responsive GLI3 binding regions (GBRs), termed HH-responsive GBRs, that likely mediate the majority of HH-specific transcription 10 . As a regulator of tissue patterning, GLI3-R spatially and temporally restricts expression of HH targets, preventing sub-threshold activation of the pathway in HH-responsive tissues 1,11,12 . Consequently, GLI transcriptional repression has primarily been studied in tissues with ongoing HH signaling. Although not experimentally addressed, it is widely assumed that GLI3-R also regulates the expression of its target genes prior to the initiation of HH signaling 8,9,[13][14][15][16] . Before pathway activation, GLI3 has been proposed to have a pre-patterning role in the very early limb by repressing Hand2, thus limiting its expression to the posterior limb bud, where it is required to activate Shh expression, thereby establishing anterior-posterior polarity 8,9,[13][14][15][16][17] . Curiously, Hand2 and Gli3 are co-expressed in early limb buds, prior to their segregation into distinct posterior and anterior domains 14 , raising the question of whether GLI3-R is capable of repressing Hand2 at this time.
We initially hypothesized that GLI3-mediated repression would be important prior to HH expression for preventing premature activation of target genes by reducing H3K27ac enrichment at enhancers. Despite GLI3 binding to a majority of the same sites it occupies in the post-HH limb bud, we found that it does not regulate activation of its enhancers or their chromatin accessibility until after the activation of the HH pathway and does so independently of HH signaling. In addition, GLI target genes are not upregulated in limb buds lacking Gli3 as they are after the initiation of HH signaling, even though a subset of these genes appear competent to be repressed at this stage. Contrary to our initial hypothesis, we conclude that the GLI3-R isoform is transcriptionally inert in early limb buds, a finding that is incompatible with the pre-patterning model for limb polarity 8,9,[13][14][15][16] . Interestingly, most genes that are later repressed by GLI3 are expressed in pre-HH limb buds, suggesting that rather than preventing their activation, GLI3 repression later regulates their spatial expression after HH signaling initiates. Overall, this work demonstrates that GLI3-mediated repression of target genes is not a default state for the Hedgehog pathway; instead GLI repression is established during limb development at a time point after HH induction.

Results
GLI3-R is abundant and binds chromatin prior to HH signaling. To understand if GLI repression is established prior to HH signaling, we first defined when the pathway is activated during limb development. The earliest detection of the canonical HH target gene Gli1 was at 24 somites (24 S), where 58% of embryos at this stage had detectable Gli1 expression, while Shh was not detected until 25 S (86%) (Fig. 1a, b; Supplementary  Fig. 1a-c). As 24 S was the earliest detection of Gli1 expression, we defined that stage as the onset of HH signaling and the "pre-HH" window as 21-23 S, corresponding to embryonic day 9.25 (E9.25), a stage slightly earlier than previous reports (Fig. 1) [18][19][20] .
Confirming previous findings, GLI3-R was expressed at comparable levels at both pre-and post-HH stages (Fig. 1c, Supplementary Fig. 1d, Source Data) 14 . We then compared endogenous GLI3 FLAG binding using CUT&RUN 21 in pre-HH (E9.25, 21-23 S) and anterior E10.5 (32-35 S) limb buds, a post-HH time frame when HH signaling is firmly established in the posterior limb, but before there are morphological changes in Shh −/− limb buds 22 . In pre-HH limb buds, GLI3 bound to most regions that were bound in post-HH limb buds (82%; Fig. 1d, Supplementary Data 1). Previously, we identified a group of HHresponsive GLI3 binding regions that have reduced H3K27ac in Shh −/− limb buds (constitutive GLI repression) 10 (Supplementary Fig. 1g-e). Since this subset of GBRs seems to regulate most HH-responsive gene expression, these regions might be especially important to repress before HH signaling to prevent premature activation of enhancers. While the majority of HH-responsive GBRs (70%) were bound by GLI3 prior to HH induction, nearly a third were not bound, suggesting that GLI3 does not initially regulate a subset of HH-responsive GBRs (Fig. 1e).
GLI3 preferentially binds to poised, accessible enhancers. Because GLI3 bound to only a portion of regions in the pre-HH limb, we hypothesized that GLI3 may preferentially bind to poised enhancers. We defined poised enhancers as ATAC-seq accessible and enriched for either H3K4me1 or H3K4me2, where H3K4me1 is enriched at promoter-proximal and distal regions, while H3K4me2 is more commonly found promoter proximally [23][24][25] . Most HH-responsive GBRs that were accessible and enriched for H3K4me2 by E10.5 were already accessible (89%) and enriched for H3K4me2 (98%) in the early limb ( Supplementary Fig. 1h, i, Supplementary Data 2). In contrast, only 65% of the HH-responsive GBRs with H3K4me1 enrichment at E10.5 were also enriched for H3K4me1 by E9.25 (Supplementary Fig. 1i). We then asked if GLI3 preferentially bound to accessible, poised regions. Consistent with this scenario, most regions bound by GLI3 in pre-HH limbs were accessible (93%) and enriched for poised enhancer modifications (75%), including a defined limb-specific distal enhancer for Ptch1 26 (Fig. 1f-h;  Supplementary Fig. 1j). While nearly half of the regions not yet bound by GLI3 in the early limb overlapped with called ATACseq peaks ( Supplementary Fig. 1j), they were less accessible than GLI bound enhancers (Fig. 1g). Unbound HH-responsive GBRs also generally lacked enrichment of poised enhancer marks, H3K4me1 and H3K4me2 (Fig. 1g, i; Supplementary Fig. 1j). This is exemplified by the distal limb enhancer, GRE1 that helps regulate the HH target Gremlin, which is among the inaccessible regions that lack H3K4me1 enrichment and are not bound by GLI3 at E9.25 (Fig. 1i). This finding is consistent with previous reports demonstrating that GRE1 does not have enhancer activity until E10 (31-32 S) 27,28 . As many distal regions lacked H3K4me1 enrichment prior to HH induction, we observed a slight preference for GLI3 to be bound to promoter proximal regions ( Supplementary Fig. 1k). We conclude that GLI3 preferentially binds to poised enhancers, which may provide tissue-specific control of repression to prevent precocious expression.
GBRs are HDAC-bound with low H3K27ac before HH activation. If GLI3 represses enhancers in the early limb as it does at E10.5, then HH-responsive GBRs, which have reduced H3K27ac in E10.5 Shh −/− limb buds, should have reduced acetylation at E9.25 before HH signaling initiates. In agreement with this scenario, there was a significant reduction in H3K27ac enrichment at HH-responsive GBRs in pre-HH compared to post-HH limb buds ( Fig. 2a-c, Supplementary Data 2). In addition, only 39% (121/309) of HH-responsive GBRs have called H3K27ac peaks in the pre-HH limb, of which, 88% (107/121) are bound by GLI3 in the pre-HH limb, with only 14 GBRs not bound by GLI3 being enriched for H3K27ac at this time. In contrast to HH-responsive GBRs, we previously found most GBRs remain stably acetylated in both the presence and absence of HH signaling at E10.5 10 . Consistent with this, 86% (6385/7382; Supplementary Fig. 1e) of all E10.5 GBRs that are acetylated in E10.5 limb buds, have called H3K27ac peaks prior to HH induction. The overall reduction in H3K27ac enrichment specifically at HH-responsive GBRs in the early limb bud, supports the possibility that GLI3 could be actively repressing enhancers at this time to prevent premature activation.
Previously, we found that GLI-mediated repression is facilitated by HDACs that regulate H3K27ac enrichment at GBRs 10 . Pre-HH limb buds had similar levels of HDAC1 and HDAC2, with many GBRs enriched for both HDACs (Fig. 2d, e, Supplementary Data 1). In post-HH limb buds, most of these regions continue to be bound by HDAC1 while HDAC2 binding is greatly reduced (Fig. 2f) together with an HDAC-containing repression complex, could be repressing enhancers to prevent premature activation of target genes (Fig. 2g).
Loss of Gli3 does not prematurely activate GLI3 enhancers. We next tested the hypothesis that the reduction in acetylation at GBRs prior to HH signaling was due to GLI3-mediated reduction in the levels of H3K27ac that prevented premature activation of enhancers (Fig. 2g). We examined H3K27ac enrichment in WT and Gli3 −/− pre-HH limb buds ( Fig. 3a; Supplementary Data 2), predicting that loss of GLI3-R at this time would result in increased acetylation at GBRs, as it does in E10.5 post-HH limbs 10 (Fig. 3b). Surprisingly, and contrary to our hypothesis, there was no significant increase in H3K27ac enrichment in Gli3 −/− limb buds prior to HH induction ( Fig. 3b-f). Since GLI3mediated repression of Hand2 in the early limb is a key component of the pre-patterning model of anterior-posterior polarity 8,14,16,29,30 , we examined two GBRs located 10 kb and 85 kb downstream of Hand2 that mediate GLI repression 15 , but did not observe any increases in H3K27ac (Fig. 3e, Supplementary Fig. 2a, Supplementary Data 2). Additional GBRs around established GLI3 target genes also did not show increased H3K27ac enrichment with loss of Gli3. This was evident at genes such as Hoxd12 which lacked WT E9.25 H3K27ac at GLI-bound regions and remained unchanged in Gli3 −/− limbs, or at genes like Cdk6 31 , which had E9.25 acetylation, but did not gain further enrichment with loss of Gli3 (Fig. 3e). We next investigated other genes like Cdk6, as we considered the possibility that many enhancers, which have low levels of H3K27ac, may not be competent for activation at this stage and would therefore not have increased enrichment of H3K27ac upon loss of Gli3. To address this, we examined three outlier HH-responsive GBRs that contained significantly higher levels of H3K27ac in pre-HH than  25 WT limb buds compared to E10.5 WT limbs 10 (note that the levels of enrichment are comparable to E10.5 Shh −/− limbs). b Scatter plot of H3K27ac enrichment at HH-responsive GBRs in pre-and post-HH 10 WT limbs. HH-responsive GBRs with significant reductions (FDR < 0.05) in H3K27ac at E9.25 compared to E10.5, are denoted in dark blue, while the 3 GBRs with significantly higher H3K27ac in E9.25 limbs are in orange (FDR < 0.05 n = 2). c Heatmap of H3K27ac ChlP-seq indicating relative H3K27ac enrichment at HH responsive GBRs in pre-and post-HH 10 limbs (n = 2). d GLl3, HDAC1 and HDAC2 CUT&RUN tracks at E9.25 GLI3-bound regions, preand post-HH. e HDAC localization at E10.5 bound GBRs. Prior to HH signaling, most regions are enriched for both HDAC1 and HDAC2, while after, most GBRs are enriched for only HDAC1 (E9.25 HDAC1/2 n = 2; E10.5 HDAC1/2 n = 3). f Initially most HDAC1-bound regions are also enriched for HDAC2, while after HH-signaling, fewer HDAC1-bound regions are also enriched for HDAC2. g Prior to HH signaling, GLl3-bound regions are enriched for poised enhancer marks with lower levels of H3K27ac and HDAC enrichment. These are comparable to post-HH GLl3 enhancers where GLl3 actively represses targets. Orange shading in tracks indicates HH-responsive GBRs defined in Supplementary Fig. 1e. Scale bars = 1 kb.
post-HH limb buds (Fig. 2b) to see if the acetylation levels might further increase in the absence of GLI3 repression. Despite GLI3 binding at these regions, there was no increase in H3K27ac enrichment with loss of Gli3 before HH signaling (Supplementary Fig. 2b-d). Overall, this surprising result indicates that despite the presence of GLI3 and HDACs, GLI repression does not regulate H3K27ac enrichment at enhancers in pre-HH limb buds (Fig. 3f).
Loss of Gli3 does not prematurely activate GLI3 target genes.
To determine whether the lack of enhancer activation in E9.25, pre-HH Gli3 −/− limbs corresponded with a lack of de-repression of GLI target genes, we performed RNA-seq on pre-HH WT and Gli3 −/− limbs, as well as the anterior halves (no HH; GLIrepressed domain) of post-HH (32-35 S) WT and Gli3 −/− limbs ( Fig. 4a-d, Supplementary Data 3). In post-HH Gli3 −/− limbs, there were 159 significantly upregulated genes (FDR < 0.05), including well known signature HH target genes such as Ptch1, Hand2, Hoxd13 and Grem1 (Fig. 4c, d) 7 . However, in agreement with the lack of GLI3-regulated enhancer activity, very few genes (9) were significantly upregulated with loss of Gli3 in the early limb (Fig. 4b). These genes included Foxf1 and Osr1, which are regulated by HH signaling in several mesodermal tissues but, with the exception of Gpx6, are not upregulated in anterior E10.5 Gli3 −/− limbs [32][33][34][35] . This suggests that rather than being a product of ongoing GLI3 repression, these upregulated genes may represent outliers from contaminating flank tissue from the dissection. Alternatively, these could represent residual mRNAs from the lateral plate mesoderm prior to their epithelial to mesenchymal transition at the onset of limb initiation within the last~8 h 36 . Consistent with the former possibility, Foxf1 is present in the lateral plate mesoderm but is not detected in either WT or Gli3 −/− pre-HH limb buds (Fig. 4d). Similarly, Ptch1, a signature HH target gene was detected in the neural tube, but was also not detected in the early limb buds in WT or Gli3 −/− embryos (Fig. 4d). In support of the latter possibility, pre-HH upregulated mRNAs are reported to have an average half-life of 11.2 h in differentiating ES cells ( Supplementary Fig. 3a) 37 , and also have lower levels of retained introns than post-HH upregulated Gli3 −/− genes, suggesting that some of these transcripts may represent older mRNAs, potentially regulated by GLI3 in the lateral plate mesoderm, that are not currently being transcribed in the limb ( Supplementary Fig. 3b, c). This would be consistent with recent findings that the early limb transcriptome is a continuation of that observed in trunk progenitors 38 . We conclude that GLI3 is unlikely to be engaged in ongoing transcriptional repression in the early limb bud.
As Hand2 is the best-characterized target of GLI3 repression, we examined Hand2 expression in pre-HH and early post-HH  at that time (Fig. 4e, Supplementary Fig. 3e). Although not quantitative, these findings are consistent with previous reports that noted early co-expression of Gli3 and Hand2 at these stages 14 . We conclude that GLI3 repression of Hand2 is absent prior to the activation of HH signaling and suggests that GLI3 repression is first activated around 26 S, shortly after the initiation of HH signaling (Supplementary Fig. 1-c).
GLI3 repressor is inert in early limb buds. Our results thus far indicated that GLI repression is not active in the early limb bud. This could either be caused by inert GLI repression complexes or by target genes that were incapable of being activated.  Fig. 4a). This H3K27me3 enrichment appears to resolve over limb development, as H3K27me3 enrichment is greatly reduced or lost at most genes in post-HH limb buds, with only 21% remaining trimethylated in anterior E10.5 limb buds ( Fig. 4h; Supplementary  Fig. 4a-d). Most HH-responsive GBRs are located distally and lacked H3K27me3 enrichment, while the few GBRs enriched for H3K27me3, were generally located proximal to promoters, suggesting PRC2 repression is likely independent of GLI regulation ( Fig. 4i; Supplementary Fig. 4b).
Interestingly, there were high levels of H3K27me3 at the Hand2 locus in the early limb ( Supplementary Fig. 4f), which as noted above is widely expressed throughout the limb at this time.
Hand2 also was enriched for H3K27ac, which is mutually exclusive with H3K27me3 41 , suggesting that PRC2 is associated with Hand2 in a subset of cells that presumably do not express Hand2 in pre-HH limb buds ( Supplementary Fig. 4f). Similar to Hand2, nearly half of predicted HH target genes were enriched for both H3K27ac and H3K27me3, which was also true of the few E10.5 genes enriched for H3K27me3 (Supplementary Figs. 4d, 7c; Supplementary Data 4). At E9.25, most genes have high levels of H3K27me3 and low levels of H3K27ac, while at E10.5, when trimethylation is resolving, most genes have increased levels of H3K27ac with low levels of H3K27me3, consistent with increases in their expression levels at E10.5 ( Supplementary Fig. 4d-g). For other genes enriched for both marks, like Ptch1, the distribution of these modifications is offset, supporting the potential for these mutually exclusive marks to be present at the same locus (Fig. 4i, black brackets). The dual enrichment of H3K27ac and H3K27me3 has been proposed to enable fast, precise induction of these genes upon presence of relevant stimuli 42 . A smaller population of genes (18%), are highly enriched for H3K27me3 and completely lack acetylation at E9.25, suggesting they may not be competent for de-repression upon loss of Gli3 at this stage ( Fig. 4i; Supplementary Fig. 7c). Importantly, a third of genes are bound by GLI3 at E9.25, but have no H3K27me3 enrichment and are likely competent for de-repression upon the loss of Gli3 (Fig. 4j; Supplementary Fig. 8a). However, as we did not observe increased acetylation in E9.25 Gli3 −/− limbs at these regions as we do in E10.5 Gli3 −/− limb buds, we conclude that GLI3 repressor is inert in the early limb bud (Fig. 4j).
GLI-dependent chromatin compaction occurs after HH induction. We previously noted that there was reduced chromatin accessibility in posterior E10.5 Shh −/− (constitutive GLI repression) limb buds at HH-responsive GBRs 10 , compared to posterior WT limbs which have active HH signaling and lack GLI repression. To determine if this was caused by GLI repression, as opposed to a lack of GLI activator, we performed ATAC-seq in posterior E10.5 Shh −/− ;Gli3 −/− limb buds. Loss of Gli3 in the absence of Shh resulted in accessible chromatin at HH-responsive GBRs similar to posterior WT controls, confirming that chromatin compaction is GLI3 repressor-dependent (Fig. 5a). To determine the onset of GLI3-mediated chromatin compaction, we examined chromatin accessibility using ATAC-seq at HHresponsive GBRs in E9.25 pre-HH limb buds (21-23 S), posterior WT and posterior Shh −/− limbs at E10.5 (35 S) when HH signaling is firmly established and an intermediate timepoint, E10 (28-30 S), 8-12 h after the initiation of HH signaling. We first examined Hand2, whose expression becomes posteriorly restricted in the limb around 26 S (Supplementary Fig. 3e). We anticipated that GLI3 repression should be established by E10 (28-30 S), and predicted that chromatin at HH-responsive GBRs in Shh −/− limbs at this stage would be more compact compared to E10 (28-30 S) WT controls, in addition to pre-HH WT limbs, which contain accessible chromatin and lack active GLI repression. Alternatively, if GLI-repression was still in the process of being established, the accessibility of HH-responsive GBRs would be similar to that of pre-HH and E10 (28-30 S) WT limbs. Consistent with an absence of GLI repression, the chromatin at HH-responsive GBRs in posterior E10 (28-30 S) Shh −/− limb buds was not yet compacted. In fact, chromatin was significantly more accessible in E10 (28-30 S) Shh −/− limb buds compared to both E9.25 (21-23 S) WT limbs and posterior E10 (28-30 S) WT limbs, which have active HH signaling and little to no nuclear GLI3-R 10 ( Fig. 5a-d). This surprising result suggests that GLI repression has not been completely established in this population at this stage in development. We next examined individual GBRs to determine if there were any regions with less accessible chromatin in E10 (28-30 S) Shh −/− limb buds, indicative of initial targets of GLI3 repression. While most regions were not significantly reduced, several regions trended toward having reduced accessibility at E10 (28-30 S) in Shh −/− compared to WT, such as HH-responsive GBRs near HH targets Ptch1 and Hhip (Fig. 5e-g, blue datapoints). However, this reduction did not become significant until E10.5 ( Fig. 5e-g, red datapoints). Only a few HHresponsive GBRs had significantly reduced accessibility in E10 (28-30 S) Shh −/− limb buds including GBRs around Cdh11 and Mllt3, which were also significantly reduced at E10.5 (35 S) (Fig. 5e, f, h, i; Supplementary Data 5). Since Shh −/− limbs have constitutive GLI repression, the lack of GLI3-dependent chromatin compaction in E10 (28-30 S) Shh −/− limbs supports that GLI2-R is unlikely to functionally compensate for the lack of GLI3 repression in early limbs (see discussion). Collectively these results support that GLI repression is not fully established even up to 12 h after HH signaling would have normally become active (Fig. 5j).
GLI3 ciliary changes coincide with onset of GLI repression. Prior to entering the nucleus, GLI proteins traffic through the primary cilium and are subsequently modified into their truncated repressor form. As GLI3 repression does not initiate until after the induction of HH signaling, we asked if there was a change in the ciliary trafficking of GLI proteins that could signify changes in the processing of GLI-R. In pre-HH (E9.25, 21-23 S) limb buds, endogenous GLI3 FLAG was localized to~80% of cilia ( Supplementary Fig. 5a, b, e). There were similarly high trends in GLI3, as well as GLI2, ciliary co-localization in pre-HH WT and Shh −/− limb buds, where they were primarily localized to the ciliary tip ( Supplementary Fig. 5a-e). This indicates that this enrichment is not due to undetectable levels of SHH that might still promote GLI activation and, as GLI-R is not localized to the cilia 43 , it suggests that pre-HH ciliary localization is enriched with full-length, unprocessed GLI protein ( Supplementary  Fig. 5a-e) 43 after the onset of HH signaling, but prior to GLI repression being fully established at E9.75 (26-28 S). Here, GLI3 ciliary localization remained at comparable levels to E9.25 limbs, as well as in anterior and posterior E10.5 (32-35 S) limbs, after GLI repression is established (Supplementary Fig. 5f). Interestingly, E9.75 (26-28 S) limb buds had a significant redistribution of GLI3 within the cilia. While at E9.25 (21-23 S) and E10.5 (32-25 S), most GLI3 was localized to the ciliary tip alone with~20-40% having signal along the ciliary axoneme, at E9.75 (26-28 S), 74% of cilia had GLI3 signal along the ciliary axoneme with a corresponding reduction in the percentage of cilia with tip restricted GLI3 (Fig. 5k, l; Supplementary Fig. 5g). This increase in ciliary axoneme distribution is suggestive of an increase in the rate of GLI ciliary trafficking, which occurs at a time just prior to when we observe initial GLI-dependent chromatin compaction and the onset of GLI repression. Overall, this data is consistent with the possibility that there may be a 'steady state' of low to moderate levels of GLI3 ciliary trafficking, with a spike in the rate of trafficking coinciding with the requirement for GLI3 repression to become established during limb development shortly after the induction of HH signaling.
GLI3 is unlikely to co-regulate HAND2 targets in early limbs. Hand2 is one of the best characterized GLI3 repression targets in the limb bud, however in opposition to this, HAND2 and GLI3 have recently been shown to physically interact and synergistically activate targets in developing facial structures 45 . Similar to their co-occurrence in craniofacial tissue 45 , they are co-expressed throughout the pre-HH limb, prior to HAND2 and GLI3 segregating into distinct domains post-HH signaling 14 . Instead of being responsible for repressing targets prior to HH signaling, we questioned whether GLI3 might initially co-activate targets with HAND2, in a manner similar to that in facial tissue. Interestingly, while GLI3 bound to HAND2 binding regions both pre-and post-HH signaling near many genes ( Supplementary  Fig. 6a), other genes, including posteriorly expressed Tbx2 and Tbx3, were specifically bound by GLI3 at HAND2 binding regions in pre-HH but not post-HH limb buds ( Supplementary  Fig. 6b, light blue shading, Supplementary Fig. 8b). To determine whether GLI3 and HAND2 might initially co-activate targets specifically in the early limb, we identified 310 previously characterized HAND2 limb binding sites 14 that are bound by GLI3 pre-HH at E9.25, but not at E10.5 ( Supplementary Fig. 6b, c). We analyzed published HAND2 and GLI3 motifs 10,45 in ATAC-seq footprints to identify the co-occurrence of these factors in the early limb compared to post-HH Shh −/− limbs, which have active GLI repression and would not be expected to have HAND2 and GLI3 co-regulation. Inconsistent with E9.25 specific co-regulation of HAND2 and GLI3, there was no difference in HAND2 motif enrichment in ATAC footprints between pre-HH WT and post-HH Shh −/− (Supplementary Fig. 6c-e, Fisher's extract test p-value = 0.14). It is possible that other factors could be present at some ATAC-seq footprints in Shh −/− limbs overlapping HAND2 motif regions, thus masking our ability to detect a loss of HAND2 footprints. We also note that Tbx2 and Tbx3 are not reduced in pre-HH Gli3 −/− limb buds as would be expected if GLI3 had a significant role in co-activating these HAND2 target genes (Supplementary Data 3). While we cannot rule out the possibility of GLI3-HAND2 co-regulation at specific enhancers, we do not believe this to be a broad mechanism for initial regulation of GLI target genes.

Discussion
GLI repressors have been proposed to play a significant role in 'pre-patterning' the anterior-posterior limb bud prior to the onset of HH signaling 8,9,14,16,17,29,46,47 . Unexpectedly, we find that GLI3 does not act as a transcriptional repressor in the early limb bud before the onset of HH signaling. Although the GLI3-R isoform is produced at comparable levels and binds to chromatin, it does not mediate deacetylation of H3K27ac or chromatin compaction at enhancers, as it does after HH signaling 10 . Moreover, there is little to no upregulated gene expression in early Gli3 −/− limb buds. The lack of GLI repressor activity prior to the onset of HH signaling is inconsistent with the current pre-patterning model, instead it suggests that the repressive patterning activities of GLI3 in the limb bud are initiated concurrently with, or after, the initiation of HH signaling (Fig. 6). More broadly, these findings indicate that GLI repression is not a default state prior to the onset of HH signaling and that its activation is instead likely to be context dependent (Fig. 6c). ARTICLE GLI3-R is transcriptionally inert in early limb buds. The absence of GLI repression in the early limb bud could be caused by inert GLI-R isoforms or by target genes that are not yet competent for GLI repression (Fig. 4i, j). Our results are consistent with the co-occurrence of both of these mechanisms. GLI3 does not bind to almost a third of all HH-responsive GBRs that will be bound at E10.5, and most unbound regions having not yet acquired a poised conformation ( Fig. 1; Supplementary Fig. 1g, h;  Supplementary Fig. 7a, b) Fig. 6 Model for establishing GLl3 repression. a Model for lack of GLl3 repression prior to HH signaling. In pre-HH limb buds, many target genes are already activated by HH-independent factors. While GLl3 binds to most targets, it is inert as it is unable to regulate deposition of H3K27ac at enhancers or chromatin accessibility. After HH induction when 'active' GLl3 repression has been established, GLl3 prevents addition of H3K27ac at enhancers to spatially restrict expression of its target genes and pattern the limb. b Schematic of GLl3 spatial regulation of target genes. In pre-HH limb buds, GLl3 does not repress targets such as Hand2, which is expressed in an expanded anterior domain. In contrast, GLl3 restricts expression of targets to the posterior limb bud where HH signaling is active in post-HH limbs. c Schematic for GLI repression being established in a context-dependent manner. d Possible models for initiating GLl3 repressor activity. Left: HH-independent co-repressor(s), which may not be abundant in early limb development collaborate with GLl3 to assemble a complete repression complex. Middle: Addition of post-translational modifications (PTMs), potentially added via ciliary trafficking and processing, promote assembly of, or stabilize, a GLI repression complex. Right: A spike in ciliary GLl3 processing, as reflected by increased axonemal colocalization, increases the amount of viable GLl3 repressor, potentially through addition of PTMs as suggested above. them, as has been proposed for SOX2's role in activating GLI neural enhancers 48,49 . Although GLI3 itself could act as a pioneer factor, we do not favor this possibility. First, while many regions that lack GLI3 binding also lack poised enhancer modifications, some regions remain accessible and poised, but are not yet bound by GLI3 at E9.25 (Fig. 1f-i). This suggests that GLI3 binds preferentially to enhancers that are already poised, rather than playing a role in promoting deposition of these poised modifications. Additionally, while chromatin compaction of GBRs is GLI3-dependent, initially gaining accessibility is not, as we find that in E10.5 Shh −/− ;Gli3 −/− limbs, the chromatin is accessible, similar to E10.5 posterior WT controls (Fig. 5).
The presence of the polycomb repressive mark, H3K27me3, at the promoters of many predicted HH target genes, provides an additional mechanism of GLI3-independent repression, as it is highly enriched at many promoters in E9.25 limbs, but is greatly reduced or absent from post-HH limb buds (Fig. 4f-j; Supplementary Fig. 4). Notably, GLI3 de-repression is unlikely to mediate this reduction, as many of these genes also lack H3K27me3 in E10.5 Shh −/− limb buds 10 . This suggests that removal of H3K27me3 is largely independent of HH signaling and GLI3 regulation, despite reports demonstrating that H3K27me3 enrichment at some HH targets is resolved in a HH-dependent manner ( Fig. 4f-j, Supplementary Fig. 4) 50 . The H3K27me3 clearing at these genes is most likely mediated by the loss of unknown, GLI-independent transcriptional repressors as has been proposed within the developing brain 51 .
Unlike the previous examples, the class of genes that are bound by GLI3 and not enriched for H3K27me3 are more likely to be competent for GLI repression in pre-HH limb buds (Fig. 4j). However, these genes are also not upregulated in Gli3 −/− limb buds (Fig. 4b, d), suggesting an absence of GLI3 repressor function at this stage. Several additional lines of evidence support this suggestion. First, Hand2 is broadly expressed in the anterior domain of pre-HH limb buds where there is co-localization with GLI3-R that persists through E9.5 14 (Fig. 4d, e). Second, H3K27ac levels at HH-responsive GBRs do not increase in pre-HH Gli3 −/− limb buds (Fig. 3) as they do in post-HH Gli3 −/− limb buds, even at genes that appear competent for GLI-repression 10 (Figs. 3b, e;  4j). Third, while most HH-responsive enhancers have lower H3K27ac enrichment in pre-HH limb buds than in post-HH limb buds (Fig. 2b, c), exceptional enhancers with higher levels of H3K27ac do not further increase acetylation levels in Gli3 −/− limb buds, excluding the possibility that lack of increased H3K27ac levels are due to lack of a HH-independent activator (Supplementary Fig. 2b-d). Finally, HH-responsive GBRs are broadly accessible at pre-HH stages and up until~10 h after the time of HH induction, in contrast to the loss of accessibility seen under conditions of maximal GLI repression in post-HH E10.5 limbs (Fig. 5).
We predict that HH-independent mechanisms may be required to activate them prior to GLI3 repression becoming established. Most genes that are regulated by GLI3 in the post-HH limb are already expressed in E9.25 limb buds (Supplementary Fig. 3d), suggesting that genes might have to become activated prior to GLI3 repressing them.
Mechanisms for establishing GLI-repressor activity. One mechanism for regulating repressor activity is the variable expression of required co-factors as exemplified by the Brinker repressor 52 . Consistent with this scenario, the genes encoding several co-repressors implicated with GLI repression in various contexts, including Ski, Smarcc1 and Atrophin1 53-55 are significantly downregulated prior to HH induction compared to at E10.5 (Supplementary Data 3). Conversely, a protein enriched in early limb buds could inhibit GLI3 repressor activity, as has been shown for HOXD12 in post-HH limb buds 56 . Although HDACs are enriched at GBRs in both pre-and post-HH limb buds (Fig. 2e, g), the lack of increased H3K27ac enrichment in Gli3 −/− limb buds suggests that they are either inactive or not regulated by GLI3. Interestingly, both HDAC1 and HDAC2 initially colocalize at many GBRs at E9.25, while in post-HH limb buds, most remain bound by HDAC1 but lose HDAC2 enrichment. In other contexts, HDAC2 is required to recruit, but not maintain HDAC1 at some enhancers 57 , and a similar mechanism could enable the assembly of an HDAC1-containing GLI repression complex. As HDACs require co-repressor complexes to guide them to their substrates 58 , the absence of a functional GLI3 corepressor might prevent HDACs from properly regulating GLI3 enhancers in the early limb. Alternatively, rather than missing a co-repressor, it is possible that GLI3-R lacks additional, uncharacterized post-translational modifications required for its activity, or to functionally interact with its repression complex (Fig. 6d).
Temporal onset of GLI transcriptional repression. GLIdependent restriction of chromatin accessibility can first be detected by ATAC-seq at 28-30 S when a few HH-responsive GBRs have significantly reduced accessibility in Shh −/− limb buds. However, most regions are actually more accessible than in E10.5 Shh −/− , suggesting that GLI-mediated compaction is not widespread at this time (Fig. 5). Focusing on Hand2 as a likely direct target of GLI3 repression 9,14,15 , there is no reduction in chromatin accessibility at this locus in E10 Shh −/− limbs as is seen at E10.5 (Fig. 5a). However, GLI3 transcriptional repression of Hand2 is likely to commence around 26 S (Fig. 4e) 14 , a timepoint coinciding with the activation of Shh expression in the limb bud (Fig. 1a, b; Supplementary Fig. 1a, b) 18,20 . The lack of detectable reductions in chromatin accessibility at this time indicate that GLI-dependent compaction likely occurs as a later step in GLI3-mediated repression.
The onset of GLI transcriptional repression is accompanied by a re-distribution of ciliary GLI3 into the axoneme, suggestive of a possible increase in GLI3 trafficking that coincides with the time when GLI3 starts to functionally repress target genes. We propose that increased GLI3 trafficking is facilitated by developmental changes to unknown ciliary processing components that regulate GLI3-R activity (Fig. 6d). Since there are comparable levels of GLI3-R protein present within the pre-and post-HH limb buds (Fig. 1c), this regulation seems most likely to affect an unknown post-translational modification of GLI3-R rather than the processing of GLI3-FL into truncated GLI3-R (Fig. 6d). If this is the case, then the activation of GLI repression is dependent on the status of the cilia and GLI transcriptional repression may be dynamically regulated in different developmental contexts. Consistent with this notion, GPR161 has recently been shown to repress HH signaling by regulation of GLI-R processing, through both ciliary and extraciliary mechanisms in a tissuespecific manner 59 .
GLI2 is unlikely to serve as a repressor in the absence of GLI3. One possible explanation for the lack of GLI3 repressor activity is functional redundancy with other GLI proteins. While Gli1 is not expressed in pre-HH limb buds (Fig. 1b, Supplementary Fig. 1a,  b), GLI2 is present ( Supplementary Fig. 5c, e). GLI2 acts as a transcriptional activator in most embryonic tissues 4,60-63 but in certain contexts has weak repressor properties 4,60,[64][65][66] . Loss of both GLI2/3 in Gli2 −/− ;Gli3 −/− limbs do not further alter digit patterning in Gli3 −/− limbs 67 , suggesting that GLI2 repression does not regulate limb patterning. Similarly, there is no evidence for GLI2-R compensating for GLI3-R in Shh;Gli3 double mutants, either on the gene or enhancer level post-HH signaling 7-10 . Overall, this indicates that GLI2-R does not regulate post-HH limb patterning though it does not rule out a scenario where transient GLI2-R might initially be redundant with GLI3-R in pre-HH limb buds. However, we think this is unlikely for two reasons. First, in pre-HH WT limb buds, the GLI3-R target Hand2 is initially co-expressed with Gli3 throughout the entire limb bud before becoming posteriorly restricted by GLI3 in later limb development 14 (Fig. 4e; Supplementary Fig. 3e). In these WT pre-HH limbs, GLI2-R does not function to prematurely repress Hand2, one of the signature repression targets of GLI3. Second, GLI3-dependent reduction in chromatin accessibility at HHresponsive GBRs is a hallmark of GLI repression 10 (Fig. 5a). However, there is no chromatin compaction in pre-HH WT limb buds or in Shh −/− limb buds (with constitutive GLI2 and GLI3 repression), until nearly 12 h after HH signaling would have normally been activated (E10.0, 28-30 S) (Fig. 5a-j).
A model for establishing GLI3-mediated repression. Our findings provide genetic and genomic evidence that GLI3-R is inert prior to HH induction in the limb bud. As GLI3-dependent chromatin compaction is established in E10.5 (35 S), but not E10 (28-30 S) Shh −/− limb buds, it also suggests that GLI repression is likely established temporally during limb development through HH-independent mechanisms. Rather than preventing initial expression of HH targets, most genes repressed by GLI3 in later limb development are initially expressed in pre-HH limb buds (Supplementary Fig. 3d; Supplementary Data 3). Once repression is later established through unknown mechanisms, GLI3 functions to restrict target gene expression to the posterior limb (Fig. 6b, d). More broadly, this work indicates that GLI repression is not a default state and instead must be acquired in a contextdependent fashion. Since we have demonstrated that GLI repression is established independently of HH, tissue-specific factors may play a role in initiating GLI repression. Thus, repression may be established at different times, in relation to HH activation, depending on the tissue type (Fig. 6c). This work has major implications for our understanding of the pathway itself, specifically how GLI repression might be regulated in contextdependent mechanisms and whether genes and enhancers must gain competency to be regulated by HH signaling. In situ hybridization. Conventional in situ hybridization done as previously described 7 . Briefly embryos were fixed for 16 h at 4°C in 4%PFA/PBS and rehydrated through methanol/NaCl washes. Embryos were treated with proteinase K (10 μg/ml) for~15 min, post-fixed, and prehybridized at 70°C for 1 h before being hybridized with 500 ng/ml digoxigenin-labeled riboprobes at 70°C overnight. Samples were then incubated with RNase A, washed, blocked and incubated with a 1:5000 dilution of anti-digoxigenin antibody conjugated with alkaline phosphatase (AP) (Roche) overnight at 4°C. Extensive washing in 1XMBST was followed by incubation in NTMT prior to signal visualized with BM purple. Fluorescent in situ hybridization was performed using HCR v3.0 Molecular Instruments reagents and was performed according to manufacturer's protocol 71 except that probes were incubated overnight with a concentration of 16 nM. Embryos were embedded in Ultralow gelling temperature agarose (Sigma A5030), cleared in Ce3D + + 72 and visualized on a Zeiss LSM 710 confocal microscope and are shown as maximal intensity z-stack projections.
Chromatin Immunoprecipitation. ChIP-Seq was performed on whole E9.25 (21-23 S) forelimbs pooled from 30-40 embryos as previously described 15 with the following modifications. Cells were dissociated with 100μg/ml Liberase (Roche 05401119001) and fixed for 15 min in 1% formaldehyde. After cell lysis, chromatin was sheared in buffer containing 0.25% SDS with a Covaris S2 focused ultrasonicator. Antibodies used for ChIP experiments include H3K27ac (3μg/reaction; Abcam #ab4729) and H3K4me2 (3μg/reaction; Millipore #07-030). Libraries were generated using the NEBNext Ultra II library preparation kit with 15 cycles of PCR amplification (NEB E7645) and sequenced to a depth of >40 million reads per sample, using two biological replicates. Peaks were called using CisGenome version 2.1.0 73 . The read numbers were adjusted by library size and log2 transformed after adding a pseudo-count of 1. The differential analyses were performed using Limma version 3.14 74 ; peaks with an FDR < 0.05 were considered to have differential enrichment.
ATAC-Seq. WT E9.25 (21-23 S) forelimbs or posterior forelimbs from E10 (28-30 S) or E10.5 (35 S) forelimbs were dissected and pooled from WT embryos (n = 3, n = 3). Posterior forelimbs from E10 (28-30 S) or E10.5 (35 S) were pooled from Shh −/− embryos (n = 3, n = 3). ATAC used components from the Nextera DNA Library Preparation Kit (Illumina) as described previously 75 with the following variations. Limb buds were dissociated in 100μg/ml Liberase (Roche 05401119001) and lysed for 10 min. at 4°C, rotating, in a nuclear permeabilization buffer (5% BSA, 0.2% NP-40, 1 mM DTT in Ca + /Mg+ free PBS + protease inhibitor) and then centrifuged for 10 min. at 300 g, 4 C. Nuclei were resuspended, counted using Trypan Blue, and 30,000 nuclei from each WT and Shh −/− sample were put into each transposase reaction. The transposase reaction (30 µL volume) was carried out for 1 h at 37°C with gentle agitation. Libraries were generated using 11 cycles of PCR amplification with NEB high fidelity 2x master mix (New England Biolabs), cleaned up with SparQ PureMag beads (QuantaBio) and sequenced on an Illumina NovaSeq using PE150 or SR100 to a depth of at least 40 million reads. Peaks were called using MACS2 with a fixed window size of 200 bp and a q-value cutoff of 0.05. Read counts from the peak regions were obtained and differential analysis was performed using DESeq2 76 . Peaks with FDR < 0.05 in the differential test were considered to have significant chromatin accessibility difference.
Footprinting analysis was performed using HINT 77 based on the ATAC-seq data. After obtaining the DNA footprints, GLI3 and HAND2 motifs were mapped to the footprints which overlap with regions that have pre-HH only GLI3 binding, E10.5 HAND2 binding, and E9.25 H3K27ac signals ( Supplementary Fig. 6C) using CisGenome version 2.1.0 73 . Motif enrichment (Supplementary Fig. 6E) is calculated as the ratio between the percentage of footprints containing the motif in the E9. 25 WT samples and that in the E10.5 Shh −/− samples. Fisher's extract test is applied to test if the motif enrichment is different between E9.25 WT and E10.5 Shh −/− (GLI3 p-value = 0.87, HAND2 p-value = 0.14).
CUT&Tag. Experiments were performed as described previously 39 with the following modifications. 100,000 cells from E9.25 (21-23 S) forelimbs were dissociated, bound to Concanavalin beads and incubated overnight at 4°C on a nutator with primary antibodies. Antibodies were used at the following concentrations: 1 μg/mL of H3K4me1 (Abcam #ab8895) and 5 μg/mL of H3K27me3 (abcam #ab195477). The following day, samples were incubated at room temperature for 30 min. with secondary antibody, 1:100 Donkey anti-rabbit (Jackson ImmunoResearch #711-005-152). CUT&Tag transposases used were gifted from the Henikoff lab and EpiCyphr (now commercially available). Libraries were generated using NEB high fidelity 2x master mix with 14 PCR cycles and cleaned up to remove adapters using SparQ PureMag beads (QuantaBio). Samples were sequenced on an Illumina NextSeq 500 instrument using PEx75 to a depth of depth of 3-5 million reads. Peaks were called using SEACR 78 .
RNA-seq. Single limb pairs were collected, saved in TRIzol and later pooled (~4 limb bud pairs/replicate) for each developmental stage, E9.25 (21-23 s) and E10.5 (33-35 S). After RNA extraction, libraries were prepped with NEBNext Ultra II RNA Library Prep Kit. Samples were sequenced on a NovaSeq S1, SR100, with a minimum of 40 million reads/sample. Reads were aligned with HISAT2 and DESeq2 was used for DEG analysis.
Intron retention rates. The formula (1-N)/M was used to calculate the intron retention rate for each gene, where N = the number of reads overlapping with any of the gene's exons and M = the number of reads overlapping with the body of the gene. Limma version 3.14 74 was used to identify differential intron retention rate between E9.25 and E10.5 Gli3 −/− limbs, FDR cutoff< 0.05.
Identification of GLI target genes. To identify GLI target genes, we developed a weighted system which incorporated HH-responsive GBRs, poised chromatin modifications, active promoter marks and genes identified as being significantly downregulated in E10.5 Shh −/− RNA-seq compared to WT limb buds 7 (FDR = 0.05; WT/Shh −/− FC < 0) (GSE58222 [https://www.ncbi.nlm.nih.gov/geo/query/ acc.cgi?acc=GSE58645]). HH-responsive GBRs were intersected with poised enhancer modifications from H3K4me1 ChIP-seq (GSE86690), H3K4me2 ChIPseq and E10.5 ATAC-seq datasets 10 (GSE108880) from E10.5 WT limb buds. HHresponsive GBRs enriched for poised enhancers modifications, and thus increasing the probably of being active enhancers, were given a higher weight than ones without modifications. Genes identified as downregulated in Shh mutant limbs were intersected with H3K4me3 (GSE86698) E10.5 WT limb datasets. Genes enriched for these active promoter marks were given more weight. The weighted HH-responsive GBRs and weighted genes were then intersected based on proximity, up to 500 kb, and within the same topologically associated domain (TAD) (GSE96107). Genes were identified as a putative HH targets based on their proximity to HH-responsive GBRs and the number of HH-responsive GBRs, such that a gene with multiple HH-responsive GBRs, enriched for poised enhancer marks, would rank higher than a gene with HH-responsive GBRs found more distally.