A Gli silencer is required for robust repression of gremlin in the vertebrate limb bud

The Hedgehog (Hh) pathway is one of the primary signaling mechanisms underlying developmental patterning in organisms ranging from Drosophila to mammals (Ingham et al., 2011). Although progress has been made in understanding the processes underlying signal transduction and the genetic consequences of perturbing Hh signaling, the transcriptional mechanism by which Hh signaling regulates expression of its target genes remains poorly understood.

In vertebrates, Hh ligands ultimately regulate transcription by controlling the activity of Gli transcription factors (Gli1-3), homologs of the Drosophila Cubitus Interruptus (Ci) protein. In the presence of Hh signaling, full-length, activated Gli proteins translocate into the nucleus where they activate transcription (Bai et al., 2004; Bowers et al., 2012; Dai et al., 1999; Matise et al., 1998; Sasaki et al., 1999; Wang et al., 2000, 2007). In the absence of Hh signaling, Gli2 and Gli3 are processed by proteolytic cleavage into truncated proteins that act as transcriptional repressors (Litingtung et al., 2002; Pan et al., 2006; Wang et al., 2000; Wen et al., 2010). Evidence from a number of studies indicates that all forms of Gli transcription factors can bind the same 9 bp motif sequence (Hallikas et al., 2006; Muller and Basler, 2000; Peterson et al., 2012). The transcriptional output of Gli target genes is influenced both by the quality of the Gli motif sequence and the presence of tissue-specific co-factors. In addition, the genomic regions surrounding Gli target genes often contain multiple Gli-binding regions, suggesting the possibility that multiple cis-regulatory modules (CRMs) could interact together to regulate gene expression (Biehs et al., 2010; Oosterveen et al., 2012; Peterson et al., 2012; Vokes et al., 2008). If these interactions exist, they could potentially result in redundant, synergistic or dose-dependent regulation of target genes.

Gli transcriptional targets fall into two distinct groups: genes that require Gli activation for transcription (Gli activator genes), and genes that are transcribed in the absence of Gli repression (Gli derepression genes). Gli activators could potentially play quantitative roles in regulating the expression levels of a subset of this latter class. The behavior of target genes in response to a gradient of Hh signaling suggests that competition between Gli activators and repressors could drive threshold responses that restrict the boundary of Gli-activator target gene expression (Jacob and Briscoe, 2003; Ruiz i Altaba, 1997; Wang et al., 2000). Studies that have manipulated Gli expression levels in the chick neural tube support this competition model (Oosterveen et al., 2012). The mechanism by which Gli repression prevents expression of its target genes is poorly understood, but in some cases relies on interactions between Gli repressors and specific transcription factors (Oosterveen et al., 2012). Mouse neural tubes lacking the major Gli transcriptional repressor Gli3 have a relatively modest change in target gene expression boundaries with no change in ventral neural fates and more subtle changes to intermediate identities, an effect that could be due to the robustness of the neural-specific downstream regulatory network (Balaskas et al., 2012; Persson et al., 2002).

Sonic hedgehog (Shh), the Hh ligand expressed in the limb bud, has graded activity emanating from the most posterior region of the limb bud. The dose and duration of Shh signaling are crucial for specifying digits and regulating growth (Ahn and Joyner, 2004; Harfe et al., 2004; Towers et al., 2008; Yang et al., 1997; Zhu et al., 2008). Compared with Shh^−/− embryos, Shh^−/−;Gli3^−/− embryos have a substantial rescue in limb growth and digit formation. The expression of many genes that are lost in Shh^−/− limb buds are restored in Shh^−/−;Gli3^−/− embryos but with symmetrical gene expression patterns along the anterior-posterior axis. This contrasts with their asymmetric expression in wild-type embryos and is exemplified by Grem1, an important Shh target gene that encodes a protein playing key roles in regulating differentiation (reviewed by Rabinowitz and Vokes, 2012). Expression of Grem1 in the limb expands anteriorly in Gli3^−/− embryos, is severely downregulated in Shh^−/− embryos and is rescued in Shh^−/−;Gli3^−/− embryos (Aoto et al., 2002; Litingtung et al., 2002; Panman et al., 2006; te Welscher et al., 2002; Zúñiga et al., 1999). Collectively, these studies illustrate the profound importance of Shh in counteracting Gli3-mediated repression of target genes.

Despite the central role of Gli proteins in regulating Hh signaling responses, the mechanism by which Gli activator and repressor proteins collaboratively regulate target genes remains poorly understood and is an impediment to defining target genes and gene regulatory networks. We used a CRM that is embedded within a global control region for the mouse gremlin locus to perform the first genetic characterization (loss of function) of a Gli-responsive CRM. We find that the GRE1 (Gli responsive element 1) acts as both an enhancer and a silencer. GRE1 enhancer activity requires sustained Hh signaling to drive activity. In the anterior limb bud, GRE1 acts as a Gli-dependent silencer. The silencer activity is necessary for providing robust repression of Grem1 in the distal-anterior limb.

In a genome-wide chromatin immunoprecipitation study, we previously identified a 438-bp Gli3-binding region located >100 kb downstream of Grem1 that exhibited enhancer activity in transient transgenic limb buds in a region partially overlapping with Grem1 gene expression (Fig. 1A) (Vokes et al., 2008). Enhancer activity is dependent on the presence of at least one Gli motif as mutations of the motif resulted in a complete lack of enhancer activity in G0 transgenic embryos (Vokes et al., 2008). We sought to characterize Gli enhancer regulation in the context of this CRM, which is henceforth referred to as GRE1 (Gli responsive element 1). Embryos derived from three founder lines of stable transgenics had β-galactosidase activity in posterior limb bud mesenchyme in an identical domain to that previously reported for transient transgenics (Vokes et al., 2008). We selected one line, Tg(Rr26-lacZ)438Svok, henceforth referred to as GRE1LacZ, for further analysis. β-Galactosidase activity was first detected in embryos at embryonic day (E) 10.0 (31-32 somites; Fig. 1B), well after the reported onset of Grem1 expression at ∼E9 (Benazet et al., 2009; Zúñiga and Zeller, 1999). The enhancer had activity in the posterior limb within a subregion of the Shh-responsive domain. Shh expression initiates in the limb bud at ∼28 somites (E9.75) (Charité et al., 2000), and the lag in reporter expression is consistent with the reported kinetics of Shh-mediated induction of Grem1 (Benazet et al., 2009). By E10.5, β-galactosidase activity was strongly upregulated and persisted until late E11.5. By E11.75, expression was reduced and had retreated from the distal limb mesenchyme. Expression was nearly absent by E12.0 except for faint staining in the proximal middle of the condensing digit mesenchyme (Fig. 1C-F). No expression was detected after E12.0, correlating with the termination of Shh activity in the limb (Echelard et al., 1993; Harfe et al., 2004). Although the enhancer analyses focused on forelimb expression, we observed similar domains in the hindlimbs (Fig. 1G).

To determine if GRE1 is responsive to Shh signaling, we examined enhancer activity at E10.5 in Shh gain- and loss-of-function backgrounds. In contrast to wild-type or heterozygous littermates, Shh^−/−;GRE1LacZ^+/− embryos had no detectable β-galactosidase activity (6/6 embryos) and, consistent with previous studies, Grem1 gene expression was highly downregulated (Fig. 2B,B′) (Zúñiga and Zeller, 1999). We also examined expression by activating high levels of Hh signaling throughout the limb bud using a Cre-inducible, dominant-active allele, Rosa^SmoM2 (Jeong et al., 2004). Prx1Cre;Rosa^SmoM2c/+;GRE1LacZ^+/− embryos expressed both the Grem1 transcript and β-galactosidase activity throughout the entire distal limb bud (11/11 embryos; Fig. 2C,C′), indicating that high levels of Hh pathway activity were sufficient to activate GRE1LacZ along the anterior-posterior axis. Grem1 gene expression appeared patchy (Fig. 2C′), and although the reason for this expression in unclear, it is consistent with observations from another study that also activated the Hh pathway throughout the limb bud (Butterfield et al., 2009). Because PrxCre is active throughout the limb mesenchyme (Logan et al., 2002), the distal restriction of GRE1 enhancer activity suggested that additional, distal factors are also required for Grem1 expression. We concluded that Shh is both necessary and sufficient for enhancer activation.

We next examined enhancer activity in Gli3^−/−;GRE1LacZ^+/− embryos at E10.5. Consistent with previous reports, Grem1 gene expression expands anteriorly in Gli3^−/− embryos (Fig. 2D′). By contrast, the enhancer activity domain, marked by β-galactosidase staining, does not expand anteriorly (Fig. 2D). Instead, the domain is significantly reduced in all Gli3^−/− embryos (5/5) compared with heterozygous littermates (P=0.0007). The reduction in enhancer activity suggests a role for Gli3 as an activator in the posterior limb. Consistent with this, Gli3^−/− limbs at this stage had significantly reduced levels of the Gli activator target gene Gli1 (supplementary material Fig. S1A). There was also a trend towards a 25% reduction in Shh levels that did not reach statistically significant levels (supplementary material Fig. S1B). These results are consistent with previous studies that have shown that Gli3^−/− limb buds have reduced Gli activator levels as a combination of the direct reduction in Gli3 activator and reduced levels of Shh (Bai et al., 2004; Galli et al., 2010; Wang et al., 2007).

In Shh^−/−;Gli3^−/−;GRE1LacZ^+/− embryos at E10.5, Grem1 expression persists in limb buds in a depolarized fashion, as shown previously (Fig 2E′) (Aoto et al., 2002; Litingtung et al., 2002; te Welscher et al., 2002). However, the limb buds had an absence of β-galactosidase staining (3/3 embryos; Fig. 2E), indicating that Gli activation is required for GRE1 enhancer activity. This is consistent with our previous work that identified a Gli motif that was essential for driving enhancer activity (Vokes et al., 2008). Although we focused on enhancer activity at E10.5, we noticed a single Shh^−/−;Gli3^−/−;GRE1LacZ^+/− embryo at E11 that had a thin stripe of β-galactosidase activity extending in a symmetrical arc across the distal limb bud (supplementary material Fig. S2E). A similar arc was seen in the anterior of E11 Gli3^−/−;GRE1LacZ embryos whereas no activity was present in Shh^−/− embryos (supplementary material Fig. S2). It is presently unclear whether this expression reflects a weak, late role for Gli3 repression in restricting the enhancer domain or some type of indirect activation (see Discussion).

To determine the time period during which GRE1 requires Gli activator for enhancer activity in the posterior limb, we used an established ex vivo limb bud culture assay, treating GRE1LacZ^+/− limb buds with the Hh pathway inhibitor cyclopamine (Panman et al., 2006). We cultured one forelimb in media containing cyclopamine while the contralateral side was cultured in control media, providing an internal control for staging and embryo variability (Fig. 3B). As expected from the lack of activity in Shh^−/− embryos (Fig. 2B), limb buds cultured in cyclopamine at stages before enhancer activity is detected (29-30 somites) resulted in a complete loss of β-galactosidase (Fig. 3A,A′). In limb buds cultured at 31-32 somites, there is a strong reduction (61%) in the size of the enhancer activity domain compared with the control side (Fig. 3C-C″; P=0.0004). Limbs cultured at 33-34 somites have more modest reductions (39%) in the size of the enhancer domain (Fig. 3D-D″; P=0.0065). The domain size no longer depends on Shh signaling from 35-36 somites onwards (Fig. 3E-E″; P=0.3333). These results indicated that Shh signaling is required for expanding the domain of enhancer activity until 35-36 somites.

Because residual β-galactosidase protein could persist after the cessation of transcriptional activity from the reporter, it was not possible to determine if Shh is required to maintain enhancer activity with this approach. To circumvent this problem, we performed additional limb bud cultures on 32- and 38-somite embryos and measured lacZ expression by qRT-PCR. As a control to ensure that the experimental conditions resulted in robust inhibition of Shh signaling, we measured the expression of the obligate Shh target gene Gli1 (Panman et al., 2006). When forelimbs from 32-somite embryos were cultured, they had an 84% reduction in Gli1 gene expression and a 75% reduction in lacZ expression. Similarly, forelimbs cultured from 38-somite embryos had a 70% reduction in Gli1 and also had a 64% reduction in lacZ (Fig. 3F). The change in gene expression at later stages contrasts with the stable expression domains indicated by β-galactosidase staining (Fig. 3E-E″). We concluded that establishing the enhancer domain requires Shh signaling transiently until 35 somites, whereas enhancer activity within the domain continues to require sustained Shh signaling.

In addition to requiring Hh signaling, GRE1 could potentially be negatively regulated by fibroblast growth factor (FGF) signaling, which has previously been shown to negatively regulate Grem1 during later limb specification (Verheyden and Sun, 2008). To test this, we cultured additional GRE1LacZ limb buds in the presence and absence of the FGF inhibitor SU5402 (Mohammadi et al., 1997). Consistent with previous results, Grem1 was expanded in the distal-anterior limbs in SU5402-treated cultures, but GRE1LacZ activity was not inhibited by FGF signaling (supplementary material Fig. S3).

To determine the effect of GRE1 on Grem1 expression, we examined G0 transgenics at E11-11.5 (41-48 somites) containing a previously generated bacterial artificial chromosome (BAC) in which lacZ was inserted into the Grem1 coding region (Zuniga et al., 2004). Consistent with previous studies, β-galactosidase activity was restricted to the posterior limb in β-galactosidase-expressing embryos (0/6 have anterior expression; Fig. 4A) (Zuniga et al., 2012, 2004). The expression domain was similar although not identical to GRE1LacZ expression at the same stage (Fig. 1D), and we hypothesized that deletion of GRE1 would reduce or eliminate reporter gene expression. Unexpectedly, the deletion resulted in ectopic anterior limb bud expression in most embryos (7/10; Fig. 4B). The ectopic anterior expression suggested the presence of additional CRM(s) that either individually or collectively have pan-limb enhancer activity. It also suggested that GRE1 is acting as a silencer in the anterior limb. We generated a third set of constructs in which GRE1 was re-inserted into the BAC with mutations in the two Gli DNA-binding motifs present in the CRM (see Materials and Methods). These forelimbs also contained ectopic anterior expression (3/3 embryos; Fig. 4C). We concluded that Gli-binding regions within GRE1 mediate silencer activity, preventing ectopic Grem1 transcription. This is consistent with the anterior expansion of Grem1 observed in Gli3^−/− embryos (Fig. 2D′).

The results described so far suggested that GRE1 likely mediates the transcriptional repression of Grem1. We then generated mice containing a deletion of GRE1 (supplementary material Fig. S4). Gremlin^{ΔGRE1/ΔGRE1} forelimbs expressed gremlin and formin 1 at levels that are indistinguishable from wild-type control forelimbs (Fig. 5A-C). Gremlin^{ΔGRE1/ΔGRE1} mice were viable and fertile with normal skeletal patterning (Fig. 5D-G). Embryos containing one null allele of Grem1 (Khokha et al., 2003) and a second allele harboring the deletion of GRE1 also had normal skeletal patterning (supplementary material Fig. S5). These results indicate that GRE1 is not necessary for normal skeletal development. The Grem1 gene expression domain was nearly normal in Gremlin^{ΔGRE1/ΔGRE1} embryos at E11 (Fig. 5A,B), but at earlier stages, the distal anterior boundaries of expression were more diffuse (Fig. 6A,B). In light of these results, we hypothesized that redundant Gli-dependent CRMs might regulate Grem1. Two additional Gli-binding regions are present within the gremlin locus (Vokes et al., 2008). One of these regions was recently shown to have Shh-responsive enhancer activity and to be crucial for mediating BAC reporter activity in transgenic embryos (Zuniga et al., 2012). We hypothesized that our Gli CRM might be redundant with other GREs under normal conditions but still required for robust regulation of Grem1.

Studies in Drosophila have tested the robustness of transcriptional responses to shadow enhancers by examining CRM deletion phenotypes at the outer ranges of permissive temperatures or by removing one copy of an upstream regulator (Frankel et al., 2010; Perry et al., 2010). We used the latter strategy to examine Gremlin expression in Gremlin^{ΔGRE1/ΔGRE1} embryos containing a single copy of Gli3, which is sufficient to prevent the distal-anterior expression of Grem1 seen in Gli3^−/− embryos (te Welscher et al., 2002) (Fig. 2D′). At E10.5, both wild-type and Gli3^+/− littermates have a sharp boundary of Grem1 expression that is restricted from the most distal-anterior mesoderm in the forelimbs (n=7; Fig. 6A,C, brackets). Gremlin^{ΔGRE1/ΔGRE1} littermates have forelimbs with less pronounced distal-anterior borders of Grem1 and with weak ectopic expression in the anterior limb mesoderm directly adjacent to the apical ectodermal ridge (n=10; Fig. 6B, dashed arrow). By contrast, Gremlin^{ΔGRE1/ΔGRE1};Gli3^+/− littermates have forelimbs with ectopic distal-anterior Grem1 expression that is broader and stronger than in Gremlin^{ΔGRE1/ΔGRE1} forelimbs (n=8; Fig. 6D). This expression is significantly different from Gli3^+/− (P=0.0002) or Gremlin^{ΔGRE1/ΔGRE1} forelimbs (P<0.0001), indicating a genetic interaction between Gli3 and the Gremlin^ΔGRE1 allele.

An expansion of GREM1 into the anterior distal mesenchyme would inhibit bone morphogenetic proteins (BMPs), causing an expansion in anterior growth (Lopez-Rios et al., 2012; Pizette and Niswander, 1999). This growth would probably result in anterior polydactyly, which is also seen in mice with reduced BMP activity (Dunn et al., 1997; Selever et al., 2004). In the mixed genetic background present in our colony, the presence of the Gli3^+/− ‘extra toes’ allele only rarely results in mice or embryos with fully polydactylous digits. In this study, all of the Gli3^+/− embryos had a single nub (a fleshy outgrowth that sometimes contains a single speck of cartilage) but none of them had distinct polydactylous digits (18/18 hindlimbs; supplementary material Table S1; Fig. 6O). Gremlin^{ΔGRE1/ΔGRE1} littermates have normal digit patterning (14/14 hindlimbs; supplementary material Table S1; Fig. 6N). By contrast, Gremlin^{ΔGRE1/ΔGRE1};Gli3^+/− littermates have a distinct, polydactylous digit in three out of eight hindlimbs (Fig. 6P), a significant difference from Gli3^+/− embryos (P=0.0215). Gli3^+/− forelimbs displayed a spectrum of phenotypes ranging from completely normal digits (7/17) to polysyndactyly (4/17) (supplementary material Table S1). Gremlin^{ΔGRE1/ΔGRE1};Gli3^+/− forelimbs uniformly contained a polysyndactylous thumb (8/8), a significant increase in frequency compared with Gli3^+/− embryos (P=0.0005; Fig. 6I-L). Gremlin^ΔGRE1/+;Gli3^+/− embryos also contained a high proportion of polysyndactylous forelimbs (23/28; supplementary material Table S1). These results suggest that GRE1 has silencer activity that is required for robust anterior repression of Grem1. Our result is consistent with previous studies showing a genetic interaction between Gli3 and BMP4 (Dunn et al., 1997; Lopez-Rios et al., 2012). To determine if GRE1 might also be required to provide a Gli activator input for robust Gli enhancer activity, we performed a parallel analysis of compound crosses with Shh^+/− mice. Gremlin^{ΔGRE1/ΔGRE1};Shh^+/− embryos have no genetic interaction (supplementary material Fig. S6), suggesting that the enhancer properties of GRE1 are either completely redundant or biologically irrelevant. We concluded that silencer activity through GRE1 is required for robust, Gli-dependent repression of Grem1 in the anterior limb (schematized in Fig. 6Q-T).

In this study, we have performed the first genetic characterization of a vertebrate Gli CRM. Within the limb bud, most putative Gli target genes are associated with multiple Gli-binding regions (Vokes et al., 2008). Our results, summarized in supplementary material Table S2, suggest that one role for multiple, distinct Gli-binding regions around Gli target genes is to provide a robust silencing response that buffers against genetic perturbations. This is in contrast to the Fgf8 and HoxD loci, for which multiple enhancers with similar activity domains have been proposed to additively or synergistically amplify transcription (Marinić et al., 2013; Montavon et al., 2011). Our results further suggest that Gli silencers prevent transcriptional activity driven by additional, Gli-independent CRMs. We also show that GRE1 can act as a Gli-activator-dependent enhancer in the posterior limb, although the biological role for this activity is unclear (see section on Gli enhancer activity).

We propose a model in which Gli repressors bind to multiple Gli-dependent CRMs in the anterior limb, providing a robust silencing activity that prevents ectopic activation of Grem1 that would otherwise be driven by at least one additional Gli-independent CRM that is active throughout the distal limb (a pan-limb enhancer). Gli repressor-mediated silencing results in the anterior repression of Grem1 in the absence of threshold levels of Gli activator complexes. In the posterior limb, where Gli activator activity is high and Gli repressor activity is low, GRE1 silencing activity is lost and Gli-activator complexes provide enhancer activity. We have synthesized these results in a model for how Grem1 is regulated by Gli proteins within the limb (Fig. 7).

GRE1 enhancer activity is detected in the posterior limb in a spatial and temporal fashion that correlates with Shh signaling (Fig. 1). GRE1 requires Gli activation for initiating and sustaining activity at E10.5 and ectopic Gli activator signaling is sufficient to drive GRE1 expression throughout the anterior-posterior axis (Fig. 2). These results suggest that GRE1 enhancer activity is primarily regulated by Shh signaling. GRE1 enhancer activity is transiently reduced in E10.5 Gli3^−/− limb buds (Fig. 2D). Our results (supplementary material Fig. S1) are consistent with several studies showing that Gli3^−/− limbs have reduced levels of Gli activation caused by a combination of reduced levels of Gli proteins and a reduction in Shh (Bai et al., 2004; Bowers et al., 2012; Galli et al., 2010; Wang et al., 2007).

In marked contrast to Grem1 gene expression, the GRE1 enhancer domain does not expand in E10.5 Gli3^−/− limb buds (Fig. 2D,D″). We rule out the possibility that Gli repressors do not work through GRE1 because our subsequent experiments indicate that it does indeed mediate Gli repressor-mediated silencing of Grem1 (Figs 4, 6) and it is bound by Gli3 repressor in chromatin immunoprecipitation assays (Vokes et al., 2008). The behavior of GRE1 contrasts with the behavior of a dpp wing imaginal disc CRM in Drosophila, in which both repressor and activator functions of Ci can be detected in the same enhancer element (Muller and Basler, 2000). Within the mammalian neural tube, studies have reported conflicting conclusions regarding the role for Gli3 in restricting the boundaries of Gli activator enhancers or genes (Balaskas et al., 2012; Oosterveen et al., 2012; Peterson et al., 2012).

In contrast to the absence of anterior expression at E10.5, there is a thin, anterior domain of enhancer activity at E11 in Gli3^−/− and Shh^−/−;Gli3^−/− embryos (supplementary material Fig. S2C-E). Previous experiments with G0 transgenics indicated that a Gli motif within GRE1 is absolutely required for enhancer activity at E11 (Vokes et al., 2008). It is presently unclear whether this expression reflects a weak, late role for Gli3 repression in restricting the enhancer domain, an artifact of the enhancer construct or transgenic line, or some type of indirect activation. If this represents biological derepression, a possible model would be the presence of an unknown anterior activator in the anterior limb that is repressed by Gli3 but activates late GRE1 activity. This would be consistent with our previous G0 transgenic enhancer results because they were in wild-type limbs and so the hypothetical anterior activator would still be repressed (Vokes et al., 2008).

There are several possible explanations for the lack of anterior expansion of GRE1LacZ in E10.5 Gli3^−/− limb buds. The first is that Gli repressors might not compete with Gli activators to limit the anterior domain of enhancer activity. In this scenario, enhancer activity is driven solely by threshold-dependent Gli activation. The lack of baseline anterior activity would prevent visualization of the silencer activity in an enhancer reporter assay. A second possibility is that the GRE1LacZ transgenic construct is incapable of responding normally to Gli repressors because it is removed from its normal chromosomal environment. Indeed, our experiments suggest that Gli repressors do regulate the activity of additional CRMs in the gremlin locus (Figs 4, 6). Taken out of context, GRE1 could also have altered affinities for Gli activator and repressor complexes that prevent its anterior expansion in Gli3^−/− embryos. A third possibility is that residual Gli repressor activity is sufficient to prevent anterior expansion of GRE1LacZ in Gli3^−/− limb buds. Consistent with this, recent work has indicated that there is a genetic role for Gli2 repressor in skeletal patterning in the absence of Gli3 (Bowers et al., 2012). However, Grem1 expression appears largely symmetrical along the anterior-posterior axis in Gli3^−/− limb buds, suggesting that the remaining Gli repressor activity mediated by Gli2 might not be sufficient to repress Grem1 (Fig. 2D′) (Aoto et al., 2002; Litingtung et al., 2002; te Welscher et al., 2002). Additional studies examining GRE1LacZ enhancer activity in Gli2^−/−;Gli3^−/− limb buds would be necessary to determine if GRE1 itself is more sensitive to Gli2 repression than the overall Grem1 gene expression pattern would suggest.

Two models for Gli repression have been proposed (Wang et al., 2010). In one, Gli3 repressor acts as an inert decoy competing with Gli activator to regulate the transcription of target genes (Oosterveen et al., 2012; Wang et al., 2010). In the second, Gli repressor behaves like a conventional transcriptional repressor, recruiting transcriptional co-repressors that actively shut down transcription (Wang et al., 2010). Whereas the first model would apply specifically to Gli activator target genes, the second model could, in principle, apply both to Gli activator target genes and to genes that only require Gli derepression. GRE1 displays properties that are associated with both classes of Gli target genes. Grem1 is a Gli derepression gene and Gli3 works through GRE1 as a silencer, preventing transcription directed by additional CRMs that would otherwise lead to ectopic distal-anterior expression. This mechanism of repression is distinct from that of conventional CRMs, for which repressor activity is integrated at the CRM level with each CRM then acting as an autonomous module to regulate gene expression. In future studies, it will be interesting to determine the mechanism of repression, which could function as a basal regulator of transcriptional activity. Alternatively, Gli3 might specifically inactivate one or more CRMs.

In the posterior limb bud, it is unclear whether Gli activators are simply indicative of a derepressed environment that permits additional CRMs to drive expression or if they also provide a quantitative contribution as enhancers to increase Grem1 transcription. The only evidence suggesting GRE1 is an enhancer is the enhancer activity of the isolated element in transgenic limb buds (Figs 2, 3). Although this fits the generally accepted criteria for an enhancer, there is no genetic evidence for reduced Gli activator responses in either Gremlin^ΔGRE1/− or Gremlin^{ΔGRE1/ΔGRE1};Shh^+/− embryos (supplementary material Figs S5, S6). There is also no observable reduction of posterior β-galactosidase activity in the BAC transgenics harboring a deletion in GRE1 (Fig. 4B). The lack of any detectable phenotype suggests that in the context of the native genomic locus, the enhancer activity is absent, trivial or completely redundant with additional Gli-dependent CRMs. The ambiguity over the contribution of enhancer activity is represented in Fig. 7, suggesting that the major purpose of GRE1 enhancer activity lies in counteracting Gli repression rather than providing quantitative levels of activation. In this way, GRE1 could act as a binary switch, causing transcription to be on or off in different domains (Fig. 7). This model provides a mechanism for how Shh signaling imposes asymmetric expression of ‘pre-patterned’ genes that would, in the absence of any Gli regulation, be symmetrically expressed throughout the limb bud. It also suggests that the inclusion of Gli-driven CRMs into the locus of pre-patterned limb might have provided an evolutionary mechanism for regulating asymmetric gene expression in a pre-existing pattern.

Within the context of this study, there appear to be at least three distinct CRMs regulating Grem1. This is consistent with previous studies that describe a complex regulatory locus for Grem1 (Vokes et al., 2008; Wang et al., 1997; Zuniga et al., 2012,, 2004). Several proteins have also been shown to regulate Grem1 at various developmental time points. In particular, BMPs and HoxA/D transcription factors both regulate Grem1 along the anterior-posterior axis. Their activity and expression domains make them excellent candidate regulators for the Gli-independent pan-limb enhancer (Fig. 7) (Benazet et al., 2009; Capdevila et al., 1999; Nissim et al., 2006; Sheth et al., 2013). Intriguingly, HoxA/D conditional mutants lack most Grem1 expression with the exception of a posterior domain that appears to be nearly identical to the Gli CRM enhancer domain (Fig. 1D) (Sheth et al., 2013). Although our model depicts pan-limb enhancer activity with one CRM as the simplest possibility (Fig. 7), it is certainly possible that this activity integrates multiple Gli-independent enhancers active in distinct or overlapping domains.

Recently, a second GRE that lies closer to the transcriptional start site has been characterized. Although it does not contain a high affinity Gli motif within the core region, it is nonetheless bound by Gli3 in chromatin immunoprecipitation assays and requires Shh expression for enhancer activity in mutant embryos (Zuniga et al., 2012). Unlike GRE1, the more proximal GRE is essential for Grem1 transcriptional activity in the same BAC reporter used in this study (Zuniga et al., 2012). Notably, GRE1 is not sufficient to activate transcriptional activity in its absence. This more proximal GRE could integrate Gli signaling with additional, Shh-independent facets of Grem1 or there could be additional, uncharacterized Gli-dependent element(s). Our study was limited to the contribution of a single CRM, and future studies will be required to determine if there are higher-order chromatin interactions among the individual CRMs regulating Grem1, as has been suggested for the Fgf8 and HoxD loci (Marinić et al., 2013; Montavon et al., 2011). In Drosophila, Ci (Gli) repressors have been proposed to work cooperatively by binding to several distinct sites within a CRM regulating dpp (Parker et al., 2011). The presence of an additional Gli CRM in the gremlin locus raises the intriguing possibility that Gli proteins binding to distinct CRMs might nonetheless be able to cooperatively repress Grem1 in the context of a higher order chromatin structure.

Given the crucial role for Shh in regulating Grem1 and the significant derepression observed when GRE1 was deleted in transgenic BAC reporters (Fig. 4B), we were initially surprised at the subtle phenotypes seen upon deleting the Gli CRM. Embryos and mice lacking GRE1 have no detectable skeletal phenotype. Nonetheless, embryos do have subtle shifts in Grem1 expression (Fig. 6A,B), and when one copy of Gli3 is removed GRE1 is required for the repression of Grem1. It is possible that the enhanced phenotype seen in ΔGRE1;Gli3 compound heterozygous embryos (Fig. 6) is due to the presence of another allele co-segregating with GRE1. The primary support that this interaction occurs between GRE1 and Gli3 is that it is consistent with interactions observed between Gli3 and BMPs (which should have reduced anterior activity with ectopic Grem1 expression) (Dunn et al., 1997; Lopez-Rios et al., 2012).

Both the subtle changes in expression pattern and the requirement of the CRM as a mechanism for buffering genetic variation are analogous to the shadow enhancers described in Drosophila (Barolo, 2012; Frankel et al., 2010). Shadow elements are defined by the genetic interactions of two genetically defined CRMs (Frankel, 2012) and further genetic studies involving multiple Gli-bound elements would be required to determine if the Gli CRM is functioning as a shadow repressor of Grem1. Our study, focused exclusively on a single Gli CRM, is the first to address the potential genetic role that multiple Gli-bound CRMs play in regulating transcription. Multiple Gli-binding sites are associated with many predicted Gli target genes (Peterson et al., 2012; Vokes et al., 2008) and we propose that they may act as a general mechanism for mediating robust transcriptional responses to Hh signaling.

Experiments involving mice were approved by the Institutional Animal Care and Use Committee at the University of Texas at Austin (protocol AUP-2010-00166).

The genomic coordinates in this study are reported in the Genome Reference Consortium mm10 build 38 genomic assembly. The GRE1LacZ transgenic line, officially named Tg(Rr26-lacZ)438Svok (MGI:5052053), was generated by pronuclear injection using the previously described enhancer reporter construct containing the 438-bp Gli binding region (chr2:113640843-113641280) (Vokes et al., 2008). The BAC transgenic constructs were generated using the Quick&Easy BAC Modification Kit (Gene Bridges) to modify a previously generated BAC containing lacZ within the Grem1 transcript (Zuniga et al., 2004). The homology arms for both targeting vectors were chr2:113,640,295-113,640,842 and chr2:113,641,281-113,641,757. After targeting, the FRT-flanked neomycin-resistance cassette was removed with a heat shock-inducible FlpE construct (Gene Bridges), leaving a 69-bp FRT site and linker sequence precisely in place of the Gli CRM (chr2113640843-113641280). The Gli binding sites within the CRM were mutated from TAGGTGGTC (chr2:113641085-113641088) to TACCACGTC and from CACCTCCCA (chr2:113641174-113641177) to CACGTGGCA; the mutated CRM was flanked by a 5′ EcoR1 site and a 3′ 70-bp linker sequence that included the FRT site. The official name for the Gremlin^ΔGRE1 allele is Rr26<tm1Svok> (MGI:5486166). This allele results in the replacement of the 438-bp CRM sequence with an 89-bp sequence containing a single LoxP scar. Further details on the generation of this allele are provided in the supplementary material Fig. S4.

When applicable, limbs were cultured for 15 h in 10 μM cyclopamine dissolved in absolute ethanol (Toronto Research) or 0.125% absolute ethanol for controls. After incubation, limb buds were separated from adjacent tissues and processed for qRT-PCR, β-galactosidase staining or in situ hybridization. All limb buds were assayed for β-galactosidase activity by staining overnight using established methods (Whiting et al., 1991). Skeletal preparations were performed as described previously (Allen et al., 2011).

RNA was extracted from a single forelimb bud (or paired forelimbs in Fig. 5 and supplementary material Fig. S1) using the RNA-Aqueous 4-PCR Kit (Ambion) and subsequently treated with DNAse1. cDNA was synthesized from 300 ng of total RNA with random hexamers using SuperScript II (Invitrogen). qRT-PCR experiments were performed using Power SYBR Green (Applied Biosystems) on a Viia7 system (Applied Biosystems). Target gene expression was determined by amplifying with the following primer pairs: F-Fmn1: GACGCCGCACCAACTTTATG; R-Fmn1: GGCCTCTGACA-GGGGTTTTT; F-Gapdh: GGTGAAGGTCGGTGTGAACG; R-Gapdh: CTCGCTCCTGGAAGATGGTG; F-Gli1: CCCAGCTCGCTCCGCAAACA; R-Gli1: CTGCTGCGGCATGGCACTCT; F-Grem1: ACTCGTCCACAGCG-AAGAAC; R-Grem1: TCATTGTGCTGAGCCTTGTC; F-Jag1: GTGCTA-CAATCGTGCCAGTG; R-Jag1: GGGGACCACAGACGTTAGAA; F-LacZ: GGGCCGCAAGAAAACTATCC; R-LacZ: TCTGACAATGGCAGATCCCA; F-Shh: TCTCGAGACCCAACTCCGAT; R-Shh: GACTTGTCTCCGATCCC-CAC. The Gapdh primers are widely used and the lacZ primers were previously described (Jeong et al., 2002). Unless specified otherwise (Fig. 5), gene expression levels were normalized to Gapdh.

Unless indicated otherwise, statistical significance was measured using Fisher's Exact Test with a two-tailed P-value. The compound crosses of mice used to determine genetic interactions were compared with littermate controls.

We thank Dr Aimée Zuniga for providing a BAC construct; Dr Susan Mackem for providing the HoxB6CreER line; Dr Richard Harland for providing the Gremlin null line; and the Mouse Genetic Engineering Facility at the University of Texas at Austin for generating transgenic mice. We are grateful to Dr Simone Probst and the other members of Dr Rolf Zeller's laboratory for teaching us the limb bud culture technique. We thank Dr Seema Agarwala, Dr Paul Krieg and Margo Hennet for comments on the manuscript.