Research Article

Conservation and Divergence of Regulatory Strategies at Hox Loci and the Origin of Tetrapod Digits

  • Joost M. Woltering,

    Affiliation: Department of Genetics and Evolution, University of Geneva, Geneva, Switzerland

  • Daan Noordermeer,

    Affiliation: School of Life Sciences, Federal Institute of Technology, Lausanne, Switzerland

  • Marion Leleu,

    Affiliation: School of Life Sciences, Federal Institute of Technology, Lausanne, Switzerland

  • Denis Duboule mail

    Affiliations: Department of Genetics and Evolution, University of Geneva, Geneva, Switzerland, School of Life Sciences, Federal Institute of Technology, Lausanne, Switzerland

  • Published: January 21, 2014
  • DOI: 10.1371/journal.pbio.1001773


The evolution of tetrapod limbs from fish fins enabled the conquest of land by vertebrates and thus represents a key step in evolution. Despite the use of comparative gene expression analyses, critical aspects of this transformation remain controversial, in particular the origin of digits. Hoxa and Hoxd genes are essential for the specification of the different limb segments and their functional abrogation leads to large truncations of the appendages. Here we show that the selective transcription of mouse Hoxa genes in proximal and distal limbs is related to a bimodal higher order chromatin structure, similar to that reported for Hoxd genes, thus revealing a generic regulatory strategy implemented by both gene clusters during limb development. We found the same bimodal chromatin architecture in fish embryos, indicating that the regulatory mechanism used to pattern tetrapod limbs may predate the divergence between fish and tetrapods. However, when assessed in mice, both fish regulatory landscapes triggered transcription in proximal rather than distal limb territories, supporting an evolutionary scenario whereby digits arose as tetrapod novelties through genetic retrofitting of preexisting regulatory landscapes. We discuss the possibility to consider regulatory circuitries, rather than expression patterns, as essential parameters to define evolutionary synapomorphies.

Author Summary

Our upper limbs differ from fish fins, notably by their subdivision into arm and hand regions, which are separated by a complex articulation, the wrist. The development of this anatomy is associated with two distinct waves of expression of the Hoxa and Hoxd genes during development. Would such a shared expression pattern be sufficient to infer homology between fish fins and mouse limbs? We investigated this question here, looking at whether the two phases of Hox gene transcription that are observed during tetrapod limb development also occur during zebrafish fin development. We find the answer is “not quite.” For although the mechanisms that regulate the expression of Hoxa and Hoxd are comparable between zebrafish fins and mouse limbs, when the genomic regions that regulate Hox gene expression in fish fins are introduced into transgenic mice, they trigger Hox gene expression in only the proximal limb segment (the segment nearest the body) and not in the presumptive digits. We conclude that although fish have the Hox regulatory toolkit to produce digits, this potential is not utilized as it is in tetrapods, and as a result we propose that fin radials—the bony elements of fins—are not homologous to tetrapod digits.


The tetrapod limb is made out of a proximal-to-distal series of long bones, the stylopod, zeugopod in the arm, and the digits in the hand, the latter of which are separated from the former two by the mesopodium, an articulation based on an array of small roundish bones [1][4]. This skeletal organisation evolved during the Devonian as an adaptation to the buoyancy-lacking environment of the land [5],[6]. The fossil record indicates that limbs evolved from fins via successive steps of distal elaboration, eventually resulting in the formation of the autopod as a tetrapod-specific evolutionary novelty, with fin radials or distal fin radials as putative evolutionary precursors of digits [7][9].

During mammalian limb development, the activity of both HoxA and HoxD gene clusters is essential and the absence of these two loci leads to rudimentary and truncated appendages [10]. All long bones of the limb require the activation of Hox genes in different though partially overlapping combinations. Initially, Hoxd9 to Hoxd11 and Hoxa11 are expressed in the developing proximal limb (the presumptive arm). Subsequently, in a second phase of transcriptional activation, Hoxd9 to Hoxd13 as well as Hoxa13 are expressed in presumptive digits [11][14]. The existence of distinct regulatory modules for long bones on either side of the mesopodial articulation (the wrist and ankle), together with the separated evolutionary trajectories of these elements, has supported the view that tetrapod limbs are genetically organized following a specific bimodal pattern of proximal (arm and forearm) and distal (digits) long bones, which as such is not present in fish fins (refs. in [1]).

The characterization of Hoxa and Hoxd expression patterns during fish fin bud development has re-enforced the view that changes in Hox genes' regulation were instrumental in the transformation of fins into limbs [1],[2],[15][19]. The exact nature of these changes, however, has remained controversial. The analysis of fin buds from various fish species lead to different conclusions regarding the existence in fishes of the late and distal phase of Hoxd gene expression, associated with the development of tetrapod digits (Figure 1A). This phase was indeed either considered as a tetrapod novelty [18],[20], implying an origin close to the stem of tetrapods, or alternatively as more ancestral and already present in fish [21][24]. This latter scenario suggests a deeper homology between fin radials and distal limb structures [16],[21],[25], which could potentially qualify distal fin radials as digit homologs [16]. Regarding Hoxa genes, the expression of Hoxa11 and Hoxa13 is largely overlapping in fins [20],[21],[24],[26], instead of the mutually exclusive patterns observed in limbs, and is thus of little help in providing a proximo-distal (P-D) reference point.


Figure 1. Regulatory mechanisms and the homology conundrum between fins and limbs.

(A) The evolutionary changes that occurred during the transition from fins to limbs are mostly unresolved, in particular concerning the most distal segment of tetrapod limbs: the digits. Mammalian proximal and distal limb regions develop along with independent phases of Hoxd expression [indicated in red (arm) and blue (digits)] and fish fin buds have been probed for the existence of similar Hoxd expression patterns. A single expression domain of 5′ Hoxd genes along the distal fin margin (in grey) was interpreted either as corresponding to the distal phase in tetrapods or, alternatively, as homologous to the proximal phase [1],[15],[16],[20],[25]. Accordingly, radials (in grey) could be homologous with digits or this homology may not exist, in which case digits are tetrapod novelties. (B) Proximal (red) and distal (blue) Hoxd gene expression domains in the developing mouse limb are derived from enhancers located within distinct 3′- (red) and 5′- (blue) regulatory landscapes. The enhancer–promoter interaction profiles within these two landscapes were shown to precisely match two topological domains [31] as determined by Hi-C using ES cells [32]. This bimodal regulatory organization in tetrapods suggests distinct evolutionary trajectories for proximal and distal limbs. The presence or absence of such a modular regulatory strategy in fish would help clarify the origin of this mechanism and the homology between fins and limbs. The DNA domain shown is approximately 3 mb large.


These analyses primarily rely on the comparison between gene expression domains whose interpretation is complicated when highly divergent structures are considered, such as fins and limbs [1], and hence whether or not any homology can be inferred from such expression analyses is unclear. The assessment of fish DNA sequences orthologous to tetrapod digit enhancers in transgenic mice indeed indicated their potential for regulating gene transcription in developing appendages [27],[28]. However, these fish enhancers, related to tetrapod digit control sequences, appeared to drive transgene expression primarily in more proximal mouse limb territories rather than in digits. As an alternative to using gene expression to infer homologies, we looked at whether a comparison between the regulatory mechanisms underlying Hox gene transcription in both tetrapod and fish appendages could be more informative.

The transcriptional regulation of the HoxD gene cluster during limb development becomes rather well understood. The successive proximal and distal waves of expression are controlled by distinct enhancer-containing regulatory landscapes, located in gene deserts on opposite sides of the gene cluster. A proximal landscape is located on the 3′ side of the gene cluster, whereas the distal (digit) landscape extends on the 5′ side [29][31]. These regulatory landscapes are regions of active enhancer-promoter interactions, as defined by chromosome conformation capture (4C), and their genomic extents and properties strongly suggest that they correspond to recently defined topological domains (Figure 1B) [31]—that is, ca. 100 kb to megabase large chromatin domains, which provide a permissive environment for long-range enhancer—promoter interactions [32],[33]. Hoxd9 to Hoxd11, which are located in the central part of the gene cluster, successively interact with either one of these 3′ and 5′-located regulatory landscapes depending upon which series of enhancers are active, thus switching their contacts from one landscape to the other at the time of the transition between cells with a proximal fate to cells forming the presumptive digits [31].

In contrast, genes that are situated at either extremity of the cluster always interact within their neighbor landscape and will not switch their contacts. Hoxd13 for instance will only interact with 5′-located enhancers and, as a consequence, will be transcribed only in the distal limb territory. Therefore, a bimodal chromatin organization of the HoxD locus prefigures the bimodal expression of Hoxd genes with their proximal and/or distal specificities, leading to the tetrapod P-D limb axis [31]. The expression patterns of Hoxa genes suggested that this collinear property of Hoxd genes may also apply to their Hoxa paralogs and each gene cluster on its own is capable of specifying a complete limb P-D axis, as demonstrated by the full deletion of either HoxA or HoxD [10],[34],[35]. However, whether or not these functional similarities reflect a conservation of regulatory strategies remained elusive. Here, we report that this mode of regulation is globally conserved between both gene clusters, suggesting that its emergence predated the origin of tetrapods. In addition, we looked at the situation in fishes and analyzed the regulatory potential of their Hox clusters in the context of transgenic mice.


HoxA and HoxD Clusters Implement Similar Global Regulatory Strategies in Limbs

We first evaluated whether the bimodal regulatory strategy observed at the HoxD cluster was particular to this locus or, in contrast, was shared with HoxA during limb development, in which case such regulatory modalities would likely predate the duplication of Hox clusters and hence the emergence of tetrapods. We looked at the expression patterns of Hoxa genes to see how well they adhere to the proximal-to-distal restrictions observed for Hoxd genes in budding limbs [12],[13],[36]. Although a weak expression of Hoxa4 was scored proximal to the digits in E12.5 limbs, strong expression of Hoxa9 and Hoxa10 was detected both in developing digits and in a more proximal domain, corresponding to the presumptive forearm. In contrast, both Hoxa13 and the Hoxa11 antisense transcript (Hoxa11as) [37] accumulate only in the distal, presumptive digit domain and in the future wrist (Figure 2A). A noticeable difference with Hoxd genes was observed, however, as Hoxa11 transcripts are absent from this distal domain, while present in the proximal territory, suggesting that Hoxa11 may escape the distal regulation imposed on the Hoxa9 to Hoxa13 genes, unlike in the case of HoxD [38], where Hoxd9 to Hoxd13 genes are coregulated in digits.


Figure 2. Hoxa gene expression in limb buds.

(A) Expression of Hoxa4, Hoxa9, Hoxa10, Hoxa11, Hoxa11 antisense (Hoxa11as), and Hoxa13 in E12.5 limb buds. The Hoxa11as transcript [37] originates from a promoter within the intron of Hoxa11 (upper panel) and is expressed like Hoxa13. (B) In control (WT) embryos, Hoxa11 and Hoxa13 are expressed in mutually exclusive domains, with Hoxa13 in the autopod and Hoxa11 in the distal zeugopod. In Hoxa13 homozygous mutants embryos, the Hoxa11 expression domain shifts into the proximal autopod, partly overlapping with Hoxa13. In Hoxa13−/−/Hoxd13+/ double mutant animals, Hoxa11-expressing cells spread further distally. The Hoxa13 probe is within the 3′ UTR and thus detects Hoxa13 transcripts in mice carrying a loss of function for this gene. Although HOX13 proteins repress Hoxa11 transcription, this latter gene has the capacity to respond to global distal enhancers, much like its Hoxa9, Hoxa10, Hoxa11, and Hoxa13 neighbors (fl, forelimb; hl, hindlimb). The anterior-to-posterior polarity of the limb buds is indicated with arrows.


It has, however, recently been shown that Hox13 group genes repress Hoxa11 in the distal limb [39]. In the absence of Hoxa13 function, the expression of Hoxa11 shifts into the distal limb bud to partially overlap with the expression of the inactivated Hoxa13 transcript, as detected by using a 3′UTR probe (Figure 2B, compare left and central panels). When doses of Hoxa13 and Hoxd13 functions were progressively removed, the distal extension was strengthened and Hoxa11 transcripts were found almost throughout the entire developing autopod (Figure 2B, right panels), much like its antisense Hoxa11as transcript in the wild-type condition. Although this result indicates that both Hoxa13 and Hoxd13 products repress Hoxa11 expression, it also demonstrates that the Hoxa11 promoter can readily respond to the distal regulation, much like Hoxa13. Therefore, it appears that the HoxA cluster, like HoxD, is the target of a distal limb (digit) global enhancer, which can regulate at least two distinct promoters. On the other hand, Hoxa4, Hoxa9, Hoxa10, and Hoxa11 are transcribed in a wider territory, including the proximal limb region, suggesting that as for the HoxD cluster, Hoxa genes are regulated by two distinct regulatory modules during limb budding and patterning.

This regulatory dichotomy may correlate with another similarity between the HoxA and the HoxD cluster—that is, the fact that HoxA too is located at the junction between two topological domains [32]. These data were, however, obtained in ES cells and thus we further characterized the three-dimensional chromatin dynamics of the HoxA cluster during limb development, in comparison with forebrain cells where all Hox genes are inactive, at least at this stage. We implemented circular chromosome conformation capture and deep sequencing (4C-seq) [40],[41], using Hoxa4, Hoxa9, Hoxa11, and Hoxa13 as baits in E12.5 dissected presumptive digits, proximal limb, and forebrain cells. The distribution of contacts over an 8 Mb DNA interval (Figure S1A), as judged by the number of sequence reads, shows that ca. 90% of the interactions are concentrated within the regions corresponding to four topological domains as determined in ES cells and located on either side of the cluster (Figure S1B, shaded area from −2 to +2), with particularly strong contacts with the regions corresponding to the two flanking domains (Figure S1B, from −1 to +1). This observation was strongly reminiscent of the situation described for the HoxD cluster (Figure S1B, bottom and [31]).

Within the domains of high interactions (i.e., the shaded areas in Figure S1B), the occurrence of contacts was quantified and no difference was observed in the distribution of interactions for either Hoxa4 or Hoxa13, when either proximal or distal limb bud samples were used. Hoxa4 establishes interactions primarily with the 3′ neighborhood of the gene cluster in both distal (Figure 3; 72%) and proximal (71%) samples, whereas Hoxa13 mostly contacts the opposite, 5′-located landscape in the same two samples (Figure 3; 66% and 63%, respectively) (the 3′ and 5′ orientation of the cluster is given following the direction of transcription of Hox genes). In contrast, both Hoxa9 and Hoxa11 increase their interactions with the 5′-located landscape in digit cells, when compared to proximal limb cells (Figure 3A,B; from 26% to 39% and from 39% to 53%, respectively). This shift in contacts observed with more centrally located Hoxa genes (Hoxa9, Hoxa11) is comparable to the situation described for the HoxD cluster (Figure 3D,E, Figure S2, and [29],[31]). The increase in interactions between these genes and the 5′ landscape in distal cells suggests that several Hoxa genes located at the 5′ extremity of the cluster are coordinately regulated in the presumptive digit domain. These results and the analogy with the HoxD cluster are in line with the presence of functional enhancer sequences in the 5′ landscape, capable of activating transcription with distal limb specificity (Figure S3) [42],[43].


Figure 3. Interaction profiles of mouse Hoxa genes.

Circular chromosome conformation capture (4C) analysis of either distal (A) or proximal (B) E12.5 dissected limb bud (schematized in the left) or forebrain (C). The proximal or distal fates of these cells are illustrated by adult skeletons (left) with the same colors as in Figure 1. Dark grey squares indicate regions of local interactions excluded from the analysis. Four interaction profiles are shown after using Hoxa4, Hoxa9, Hoxa11, and Hoxa13 as viewpoints. The genomic orientation of HoxA is inverted with respect to HoxD. The percentage of contacts for each viewpoint is given, either in 5′ or in 3′ of the gene cluster. In both samples, Hoxa4 mostly interacts with the 3′ landscape, whereas Hoxa13 is biased toward the 5′ landscape. Both Hoxa9 and Hoxa11 change their bias from increased contacts in 3′, in the proximal limb bud sample (B), to contacts in 5′ in the distal sample (A), thus resembling Hoxd genes (Figure S2) [31]. Note that the interaction profiles obtained when using either autopod, proximal limb, or brain (C) tissues are quite similar to one another, indicating a constitutive chromatin organization at the HoxA locus. The size of the displayed DNA interval is of ca. 3 Mb. (D and E) Summaries of the directional 4C signals using bar diagrams in the 3′ and 5′ flanking regions of both HoxA and HoxD clusters. The colored bars represent 100% of the signal for each of the three tissues (color code at the bottom) and for three genes in either the HoxD (D) or the HoxA (E) clusters. The position of each bar with respect to the central black line (0) represents the balance between the contacts scored either in 5′ (left in D; right in E) or in the 3′ (right in D; left in E) landscapes. The HoxA and HoxD clusters are shown in opposite orientation regarding 3′ and 5′ directions to reflect their inverse locations on chromosomes 2 and 6. The four displayed topological domains were extracted from the Hi-C ES cell dataset of Dixon et al. [32].


Because the HoxA cluster appears to respond to a bimodal regulation that shares several features with that reported for Hoxd genes [31], we propose that this operational mode is a core component of both Hoxa and Hoxd gene regulation during limb proximal to distal patterning and thus likely predates the emergence of tetrapods. The data obtained from the forebrain samples show that this partitioning between 5′ and 3′ regulatory landscapes is only partially tissue-specific and largely independent from the transcriptional activity, as noted for a large proportion of topological domains (Figure 3C,D,E and Figure S2) [32],[33].

Bimodal Partitioning of the Fish HoxA and HoxD Clusters

Because this chromatin partitioning at and around the tetrapod HoxA and HoxD clusters is associated with distal and proximal regulatory capacities, we looked at its presence in fish Hox clusters as a potential indication that distinct regulations may also be at work during fin development. Data obtained from limb tissues, brain, and ES cells all show this biased distribution in interactions corresponding to the existence of two flanking topological domains, indicating that such a structural organization exists in both expressing and nonexpressing tissues. Consequently, we used whole zebrafish embryos at day 5 postfertilization (dpf) to visualize the interaction profiles of related fish Hox clusters, instead of dissected fin bud tissues, which would have met our technical limitations due to their small size and the amount of tissue required for 4C analysis. Teleosts underwent an additional genome duplication and have up to eight Hox cluster loci [44][46], of which HoxAa, HoxAb, and HoxDa are the most relevant for fin development [24],[47]. Accordingly, we used viewpoints in Hoxa4a, Hoxa9a, Hoxa11a, Hoxa13a, Hoxa2b, Hoxa11b, Hoxa13b, Hoxd4a, Hoxd10a, Hoxd11a, and Hoxd13a for 4C experiments (Figure 4 and Figure S4).


Figure 4. Zebrafish Hox clusters are partitioned into 3′ and 5′ interaction domains.

4C analysis of zebrafish whole embryos (5 dpf, including well-developed fin buds) using as viewpoints (left) several genes within the HoxAa (A and B), HoxAb (C and D), and HoxDa (E and F) gene clusters (for HoxAa, see also Figure S4). The HoxDa cluster has a reversed chromosomal orientation when compared to both HoxA clusters. The percentages of interactions between the viewpoints and either the 5′ or the 3′ landscapes are indicated above each profile. Bar diagrams in (B, D, and F) give a summary of the signal directionality per viewpoint in the 3′ and 5′ flanking regions (compare Figure 3C,D). The blue bars are as in Figure 3. Genes located at either extremity of their clusters display a strong bias toward the flanking landscape, such as Hoxa4a (B), Hoxd4a (F), Hoxa13a (B), or Hoxd13a (F). Genes located at more central positions in the clusters [e.g., Hoxa11a (B) or Hoxd11a, (F)] show more balanced interaction profiles, like for the mouse HoxA and HoxD clusters. Dark grey squares are regions of local interactions excluded from the analysis.


We observed that fish Hox13 genes also display a clear bias in their interactions toward their 5′ flanking neighborhood (Figure 4). For example, Hoxa13a, Hoxa13b and Hoxd13a show 67%, 79% and 66%, respectively, of their total contacts with their 5′ landscapes. In contrast, only 27% of the contacts established by Hoxa4a and 26% of the contacts established by Hoxd4a were scored over their 5′ landscapes, these latter genes interacting mostly with the 3′ neighborhood of the gene clusters (73% and 74% respectively). Therefore, as in the mouse, genes located at either end of the clusters establish preferential contacts with either their 3′ or 5′ neighboring landscape. In contrast, interactions involving Hoxa9a, Hoxa11a, Hoxa11b, Hoxd10a, or Hoxd11a—that is, genes located at more central positions—are rather equally distributed between the two landscapes on either side of the cluster (Figure 4). We thus concluded that the chromatin partitioning observed in tetrapods at the HoxA and HoxD loci is also present in fishes. These results suggest that the structural component of the mechanism underlying the distinct proximal and distal phases of Hox gene expression predates the evolution of tetrapod limbs. Therefore, a resemblance greater than anticipated may exist between distal fins and limb structures, as recently proposed [16],[21],[22].

Fish Regulatory Landscapes Drive Expression in the Proximal Mouse Limb

The bias of fish Hox13 genes to contact their immediate 5′ environment suggested that, similar to their tetrapod counterparts, they might be used as the distal contribution of a bimodal regulatory strategy. We investigated the potential presence in these fish landscapes of enhancers driving limb-specific expression. Accordingly, we generated transgenic mice with fish Hox clusters including their entire 5′ flanking regions. We selected Pufferfish (Tetraodon nigroviridis) BACs because, due to the compressed genome of this species [48], they permit the transgenic analysis of entire syntenic regions.

Mice transgenic for the fish HoxAa 5′ landscape showed expression of Hoxa11a, Hoxa13a, and Evx1 in hindlimb buds, but with a proximal-only specificity, whereas no distal expression was observed despite the presence of the 5′ flanking genomic region (Figure 5A).


Figure 5. Regulatory potential of the fish HoxA and HoxD landscapes in mouse transgenic limbs.

(A) Scheme of the HoxAa BAC used for transgenesis in the mouse with the expression of the fish Hoxa11a, Hoxa13a, and Evx1 genes in mouse embryonic limbs. All genes assayed showed expression in a proximal domain, yet not in the presumptive digit domain. Note that Hoxa11a expression was not observed in forelimb buds. (B) Scheme of the HoxAb BAC (bottom) with the expression of several genes. The fish Hoxa10b, Hoxa11b, and Hoxa13b genes are expressed in a proximal domain, and transcripts are absent from the presumptive digit domain. Likewise, the 5′ flanking genes HIBADHb, TAX1BP1b, and JAZF1b respond to the same proximal regulation. A comparison with the endogenous Hoxd11 expression (mmuHoxd11) shows that limb expression of the transgenes is confined to the distal zeugopod and mesopod. (C) Two BAC clones containing either the entire 5′ (top) or 3′ (bottom) landscape flanking the HoxDa cluster with their corresponding expression patterns. Here again, expression is observed in a proximal domain but is absent from developing digits. In the various schemes, genes analyzed are shown in black. All samples are right hind limbs, dorsal views with anterior to the left, except for the endogenous mouse gene “mmuHoxd11” (B), which is a mirror image of the left hind limb of the limb bud stained for Hoxa11b to its right, in order to facilitate the comparison of transcription domains. The anterior-to-posterior polarity is indicated with arrows. (D, digits; F and T, distal parts of the femur and tibia, respectively).


Likewise, the fish Hoxa10b, Hoxa11b, and Hoxa13b genes were strongly expressed in mouse limb buds transgenic for the fish HoxAb 5′ landscape. Here again, however, the expression domain matched a proximal zone and transcription was not scored in developing digits (Figure 5B). Because the pufferfish HoxAb BAC contains both the 5′ and 3′ neighborhoods, we implemented 4C-seq on transgenic mouse limbs and could confirm that strong interactions occurred between both Hoxa13b and Hoxa11b with the 5′ flanking region, despite the transcriptional outcome, which was restricted to a proximal domain (Figure S5). Also, the HIBADHb, TAX1BP1b, and JAZF1b genes, located next to Hoxa13b, were co-expressed with HoxAb genes, further illustrating that a global regulation is associated with this 5′ landscape (Figure 5B), as is the case for the mouse locus where these genes are co-expressed along with Hoxa13 [49].

Of note, the onset of the fish Hoxa13b expression in transgenic limb occurs in distal limb bud cells located underneath the apical ectodermal ridge (AER; Figure 6A, arrowhead), similar to the “late” expression pattern of this gene during fin bud development. In mice, however, the expression territory of this fish transgene does not follow the distal extension of the bud and thus remains at a more proximal position (Figure 6A, arrows). This result illustrates the difficulty of using relative parameters such as “proximal” or “distal” when assigning homology between fins and limbs (Figure 6B). Altogether, the fish regions syntenic to the mouse HoxA cluster failed to elicit expression in presumptive digit cells during limb budding. Instead, when introduced into mice, fish HoxA genes were all transcribed in proximal limb domains.


Figure 6. The Tetraodon Hoxa13b expression domain in mice: from “distal” to “proximal.”

(A) In situ hybridization of a Tetraodon Hoxa13b probe using E9.5 to E12.5 fetuses transgenic for the Tetraodon HoxAb cluster. Top panels are dorsal views of forelimbs (anterior to the left), and bottom panels are whole mount pictures. Hoxa13b is expressed in limb buds and posterior trunk, whereas the staining in the head vesicles at E10.5 and E11.5 is a routinely observed artifact. At day E10.5, before the appearance of digits, expression initiates in the distal limb bud (arrowhead). In subsequent stages, however, this domain becomes increasingly “proximal” due to the distal expansion of the digit domain (arrows in E11.5 and E12.5 specimen). The distal expression of Hoxa13b at E10.5 is strikingly similar to the distal expression of Hoxa13b in the fish fin [24]. The anterior-to-posterior polarity is indicated with arrows. (B) Scheme illustrating the difficulty in using relative parameters such as “proximal” or “distal” to assign homologies. Due to the developmental expansion of the autopod, the zeugopod domain becomes relatively more proximally positioned within the limb bud, along with time. During digit evolution, a similar process may have occurred and structures that are distal in the fin apparently shifted to a more proximal position in the limb, due to the distal growth of the autopod. The fish Hoxa13b expression in mouse limb buds (purple color in the left scheme) in fact illustrates that distal fish fin tissues correspond to proximal limb structures after the evolution into limbs (right scheme). The fin bud scheme only depicts the endoskeletal part of the fin and not the exoskeleton, which derives from a distinct developmental lineage.


We also analyzed the HoxDa cluster by using two BACs extending either 5′ or 3′ from the fish cluster. The 5′ BAC covers a region of the fish genome syntenic with the digit regulatory landscape, upstream of the mouse HoxD cluster [29], whereas the 3′ BAC is syntenic with the proximal limb regulatory landscape [31]. In both cases, when introduced into transgenic mice, the fish Hoxda genes were expressed in a restricted domain, always located in the proximal limb bud, whereas no transcript was detected distally (Figure 5C). In this context, the fish genes were expressed according to their relative proximity to the flanking landscapes; Hoxd4a, Hoxd9a, and Hoxd10a were indeed preferentially transcribed whenever their closely located 3′ landscape was included, whereas Hoxd11a, Hoxd12a, and Evx2 were preferentially expressed when the opposite 5′ landscape was present. However, the same proximal specificity was observed in both cases, suggesting that regulatory domains exist on both sides of the fish HoxDa cluster, which are able to control fish Hoxda gene transcription in a proximal domain of the mouse limb bud, rather than in the digits (Figure 5C). These results are in agreement with the capacity of zebrafish, skate [27], and coelacanth [28] sequences orthologous to mouse HoxD “digit enhancers” to drive expression essentially in proximal, rather than distal, domains of murine transgenic limb buds.


An Ancient Bimodal Regulatory Strategy

Our results show that, similar to the HoxD gene cluster, the HoxA cluster establishes preferential contacts with the two flanking genomic neighborhoods, such that Hoxa13 strongly interacts with the telomeric DNA (i.e., with its 5′ side), whereas Hoxa4 contacts its centromeric (i.e., on the 3′ side) landscape. The existence of such a structural bias in both gene clusters suggests that the ancestral gene cluster, before its duplication at the root of the vertebrate taxon, already displayed such a general bimodal chromatin structure. This may indicate the presence of a generic regulatory constraint imposed to these gene clusters, such as the necessity to functionally separate, in space and time, the most posterior genes from their anterior neighbors, the former proteins being generally dominant over the latter [50].

This split of both HoxA and HoxD clusters into two chromatin domains precisely matches the results obtained by using Hi-C on ES cell material [32]. Interestingly, however, the same dataset reveals that neither HoxB nor HoxC seem to display this feature, suggesting it may have been lost subsequently. This might relate to the fact that these latter two gene clusters are truncated either for their anterior (HoxC) or posterior (HoxB) genes.

Constitutive Contacts

Our 4C experiments also confirmed that many interactions were present in all the tissues assayed, regardless of their transcriptional activity, as previously observed [29],[31]. In addition, the general extent of our interaction domains precisely coincided with the topological domains as defined by using the Hi-C dataset of Dixon et al. [32], further suggesting that many of those interactions associated with such topological domains are constitutive in nature. For example, the strong 3′ HoxA interacting peak observed at the border between topological domains −2 and −1 (Figure 3, Chr6: 51,120,000) was present in all tissues investigated. This peak colocalizes with both proximal limb enhancers (elements 406 and 407 and human element 1465) and branchial arch enhancers (elements 402 to 406), as reported in a genome-wide enhancer screen [43].

It may be that such a constitutive contact anchors the Hox cluster at the vicinity of tissue-specific enhancers, thus working as a priming mechanism for enhancer–promoter interactions. In this context, the presence of constitutive contacts with anchoring points rather than with the actual enhancers themselves might reflect the fact that Hox genes are regulated by multiple tissue-specific enhancers in time and space. The presence of a constitutive, poised regulatory architecture may have evolved at these loci to facilitate the successive implementation of multiple regulations, by providing a stable framework to be complemented by tissue-specific factors.

The Evo-Devo-Regulo of the Fin to Limb Transition

The problem raised by the fin-to-limb transition shows that developmental expression patterns cannot always be used to infer homologies between distinct species. Because both Hoxd13 and Hoxa13—that is, the two tetrapod genes essential for digit development—display this strong regulatory tropism towards their upstream genomic neighborhoods, we looked at whether the related fish Hox genes would display the same behavior and found that fish Hox gene clusters have the same conformational organization. This observation suggested a level of conservation between the regulation of these genes in both tetrapods and fishes higher than anticipated. However, the existence of such separated chromatin domains including the fish Hox13 genes and their flanking genomic sequences does not lead to a clear partitioning of regulatory activities, at least when introduced into transgenic mice. All fish regulatory landscapes assayed, taken from either sides of the clusters, indeed elicited comparable proximal expression in mouse limbs and were thus unable to respond to those signals, triggering the emergence of the digital plate in mouse.

The existence in fishes, of regulatory landscapes showing proximal specificities in transgenic murine limbs, may be related to the fact that fins can display elaborate proximal-to-distal patterns, as illustrated by combinations of radials and distal radials. Both zebrafish and paddlefish, as well as shark Hox, genes appear to be activated in a partially heterotopic manner consistent with the presence of these distinct fin segments [21],[22],[24], and hence such P-D patterns may result from biphasic regulations emanating from opposite regulatory landscapes. In this view, it is conceivable that the regulatory balance between these two landscapes in fishes contributes to the wide variety of P-D patterns observed in the fins of various species (Figure 7) [51]. Given the inability of fish regulatory sequences to activate transcription in the mouse digital plate, however, this P-D division would not be homologous to the regulatory partitioning observed between the arm and the hand in tetrapods.


Figure 7. Regulatory evolution and the fin-to-limb transition.

Fish and tetrapod HoxA and HoxD clusters are regulated by 3′ and 5′ regulatory landscapes, represented here as triangles due to their correspondence to topological domains [31],[32]. Enhancer (indicated with colored shapes) interactions within these domains (indicated by arrows) occur with the neighboring parts of the Hox clusters, resulting in a regulatory partition between 3′ and 5′ parts of the clusters. In fishes, this mechanism may be used for patterning the fin proximal (red) to distal (orange) (P-D) polarity, through the potential function of these two landscapes in slightly different fin domains. Variation in the regulatory balance between these 3′ and 5′ landscapes through the acquisition of novel enhancers potentially explains interspecies differences in P-D fin morphology, as for instance between zebrafish and species such as coelacanth, which possesses a more elaborate fin skeleton. Although these regulatory landscapes may underlie the P-D patterning of fin skeletons, they both elicit a proximal response when assessed in transgenic mice, and hence the fish 5′ landscape is indicated as “proximal” (orange). In tetrapods, the 5′ domain (blue) has acquired new enhancers or modified existing ones, thereby evolving a novel, more distal autopodial identity, perhaps as a response to preexisting signals emanating from the apical ectoderm.


Hijacking a Regulatory Landscape

The bimodal limb bauplan is characterized by a clear morphological separation between the long bones of the arm and the forearm, on the one side, and of the hand, on the other side. This separation is controlled by opposite regulatory landscapes and gives rise to the presence in between of a nodular articulation critical for the function of the limbs and not observed in any fish fins: the mesopodium [1][3]. Because the development of long bones requires high doses of HOX products whereas a lower dose is associated with carpal-like small bones [52], it was proposed that the mesopodium results from the offset between the proximal and distal Hox expression territories, made possible by the existence of distinct regulatory landscapes [31]. In this context, a potential scenario emerges for the fin-to-limb transition whereby two partially overlapping expression domains in fins progressively segregated to generate entirely distinct expression territories.

This model hypothesizes the evolution of a bimodal “proximal–proximal” patterning system in fins (without mesopodium) into a bimodal “proximal–distal” system in limbs, including an articulation and thus postulates the transformation of a regulatory landscape from a “proximal” to a “distal” specificity. The mechanisms underlying this “regulatory homeosis” are elusive, but modifications in the structure and function of the AER, a source of growth factors in the tip of growing limb buds, may have been instrumental. During fin development, the cessation of endoskeletal expansion coincides with the transformation of the AER into the apical ectodermal fold (AEF) from which the exoskeletal fin rays will develop [53]. Various models have predicted that the folding of the ectodermal layer plays a key role in the termination of fin distal growth, possibly due to the interruption of AER-derived proliferative signals by its dense extracellular matrix [17],[18],[53][56]. In this view, the abrogation of ectodermal folding in tetrapods may have lead to a prolonged exposure to AER signals, leading to increased Hox expression and extended distal growth, thus resulting in the formation of the autopod.

However, we show here that this model cannot fully account for our transgenic results, as expression of fish Hox genes is not observed distal to the mesopodium, even in the absence of ectodermal folding. We conclude that there is an intrinsic inability of fish enhancers to respond to the distal limb regulatory program in the mouse. Consequently, the absence of a clear distal expression territory in fin buds separated from (but concomitant with) the proximal expression domains is likely not caused by a mere physical obstacle induced by the folding of the ectodermal layer. Rather, an ancestral fish 5′ regulatory landscape may have evolved to better respond to distal ectodermal signals, and the reinforced transcription of Hox genes distally might have promoted supplementary growth by delaying ectodermal folding. This situation is illustrated by the effect of overexpressing Hoxd13a in zebrafish fins, which leads to increased distal growth at the expense of the AEF [17]. Alternatively, the capacity for ancestral fish enhancers to respond to the appropriate distal signals may have existed from early on but be repressed, in which case tetrapod loci may have simply overcome this repression, for instance, through the loss of repressor binding sequences.

It has been pointed out that chondrichthyans or actinopterygians could be more informative regarding the fin-to-limb transition than extant teleosts [17],[21], which may have lost primitive characters. Yet fins of all these species consist of both radials and distal radials and exhibit similar HoxA and HoxD expression patterns [15],[16],[20][22],[24],[47]. Furthermore, our results with transgenic pufferfish sequences are consistent with the patterns found when enhancers isolated from more primitive fishes were used. In all cases indeed [27],[28], these enhancers did not elicit expression patterns as distal as one would have expected for bona fide digit enhancers [29],[57].

In conclusion, although digits are likely formed through the action of tetrapod-specific Hox enhancers, the underlying regulatory circuitry relies upon an ancestral framework already implemented in fish, illustrating the retrofitting of preexisting genomic infrastructure. In this context, the question regarding the homology between fin radials and digits may receive different answers depending on which level is considered within the regulatory hierarchy. Fish have the necessary genes and higher order regulatory architecture to form digits and likely implement the 5′ regulatory landscape to pattern distal fin radials [16]. Accordingly, digits could be considered as a specialized type of distal radials as both structures rely on a unique ancestral regulatory strategy. However, the fish 5′ regulatory landscapes are unable to specify a distinct digit territory and, as such, this regulatory feature defines a clear tetrapod synapomorphy. Therefore, a qualification of distal radials as digits (senso “classical” homology, that is, with a common ancestral structure) is not supported by our results.

Materials and Methods

Animal Experimentation and Ethics Information

All animal experiments were performed according to Swiss regulations under license no. 1008/3482/0 (to D.D.).

Chromosome Conformation Capture

4C libraries were constructed as described before [41]. Mouse libraries consisted of 52 dissected E12.5 proximal forelimb buds, distal forelimb buds, or forebrains. Zebrafish libraries consisted of approximately 300 5 dpf embryos from the TU strain. Transgenic mouse libraries containing the Tetraodon HoxAb (C0AA043AG01) BAC contained 48 E12 proximal and distal hindlimb buds. The sequencing data for these samples were combined and processed as a whole limb sample. For the mouse baits used for 4C, the primary restriction enzyme used was NlaIII (New England Biolabs, R0125L), and the secondary restriction enzyme was DpnII (New England Biolabs, R0543M). For the zebrafish baits, the primary restriction enzyme was DpnII (New England Biolabs, R0543M), and the secondary enzyme was TaqαI (New England Biolabs, R0149M). In the latter case, DNA was cut for 8 h at 65°C. For the Tetraodon baits, assessed in transgenic mice, the primary restriction enzyme used was DpnII (New England Biolabs, R0543M), and the secondary enzyme was NlaIII (New England Biolabs, R0125L). For each viewpoint, between 1.3 and 2.6 µg of the 4C library was amplified using 16 individual PCR reactions with inverse primers containing Illumina Solexa adapter sequences (Table S2). Multiplexed samples were sequenced on the Illumina HiSeq system using 100 bp single-end reads according to the manufacturer's specifications. 4C-seq reads were sorted, aligned, and translated to restriction fragments using the 4C-seq pipeline of the BBCF HTSstation (available at [41]). Mouse samples were mapped to the ENSEMBL Mouse assembly NCBIM37 (mm9) and zebrafish samples were mapped to the ENSEMBL Zebrafish assembly Zv9. Transgenic Tetraodon samples in mouse were mapped to a custom genome containing the Tetraodon BAC (C0AA043AG01) and ENSEMBL Mouse assembly NCBIM37 (mm9), thus minimizing the chance of mapping nonspecific reads. The directionality of signal was calculated on 4C-seq patterns over the regions mentioned in Table S1A. Data are summarized in Table S1B. In the figures, smoothed 4C-seq patterns (running mean, window size 11) are visualized except in Figure S4B, which shows unprocessed data. Topological domains shown to complement the 4C-seq experiments are ES cell HiC data take from (​tabase.php) [32]. Mouse domains selected for 4C-seq analyses correspond to two topological domains located centromeric and telomeric of the clusters (i.e., four domains in total) and are described in Figure S1. The zebrafish regions were selected on basis of synteny with the mouse domains analyzed. Similar experiments involving the HoxD cluster as shown in Figure S2A,B were previously reported [31]. The experiments shown here were, however, repeated together with the analysis of Hoxa genes in order to compare datasets produced under the exact same conditions.

Construction and Genotyping of Transgenic Lines

BAC constructs were identified using the Genoscope Tetraodon genome browser (​anew/). BAC numbers and genomic positions (TETRAODON8) correspond to HoxAa, C0AB048AA04 (Chr21:2,888,799–3,037,908); HoxAb, C0AA043AG01 (Chr8:6,699,347–6,844,622); HoxDa 3′, C0AB015CD05 (Chr2:13,313,310–13,458,212); HoxDa 5′, C0AB043BH04 (Chr2:13,417,090–13,578,446). BAC clones were obtained from Genoscope, France. A PISceI meganuclease site was introduced into the vector backbones using standard EL250 cell-based recombineering technology. BAC DNA was prepared using a Nucleobond Midiprep Kit, linearized with PISceI (New England Biolabs, R0696L), incubated with SDS according to the manufacturer's instructions, 2× chloroform-phenol purified, ethanol precipitated, and dialyzed against microinjection buffer containing protamines [70 µM spermidine (Sigma, S2626), 30 µM spermine (Sigma, S3256)]. Constructs were microinjected using standard protocols for pronuclear injection. BAC lines were genotyped using primer pairs every 5 to 10 kb in combination with deep sequencing using 4C-seq and mapping of the reads on the BAC sequence to confirm its integrity. A HoxAb transgenic line was mapped using embryonic hindlimb samples (Figure S5), and adult mouse ear samples of HoxDa 3′ and 5′ BAC lines were processed using 4C-seq specifically for the purpose of integrity mapping using a viewpoint located in Hoxd11a (unpublished data). BAC diagrams in Figure 5 represent the regions that were found to be integrated using PCR (all lines) or 4C-seq data (HoxAb, HoxDa 3′, and HoxDa 5′) in the transgenic lines presented.

Mutant Mouse Lines and in Situ Hybridization

The Hoxa13 and Hoxd13 mouse knockout lines were previously described [58],[59]. In situ hybridization was performed as described [60] with a temperature of prehybridization, hybridization, and posthybridization steps increased to 68.5°C. For Tetraodon probes, the SSC concentration in the hybridization mix was lowered to 1.3× to increase specificity (for the Evx2 probe, 0.5×SSC was used), and the posthybridization SSC washes were done using 4×30 minutes 2×SSC-T. In all experiments using transgenic Tetraodon probes, wild-type embryos were coprocessed for each probe and stage to monitor specificity of the probes (unpublished data). Except in brain vesicles, susceptible to probe trapping, nonspecific signal was never observed using the conditions described above.

Probe Construction

Probes were amplified using PCR from BAC DNA or limb cDNA and cloned into pGEMTE easy vector systems (Promega A1360). Primer sequences are given in Table S3. DIG-labeled RNA probes were synthesized using Sp6 or T7 polymerase (Promega). Probes for Hoxd11 and Hoxa13 were described previously [61],[62]. The probe used to detect the Hoxa11 sense transcript was kindly provided by Dr. C. Fromental-Ramain and corresponds to a ScaI-HpaI fragment in the 3′ UTR of Hoxa11 (mm10: Chr6:52,242,847–52,243,385).

Image Acquisition and Editing

In situ hybridization images were acquired using Leica Application Suite software v3.3.1 in combination with a Leica DFC300FX camera and Leica MZFLIII microscope. Images were edited in Adobe Photoshop software using the brightness/contrast function. Limb buds shown in Figure 1B stained for Hoxa13 are right side limb buds coming from the same embryos as the left side limb buds stained for Hoxa11 and are mirrored for purpose of comparison. Similarly, the Hoxa4 sample in Figure 1A is a mirror image right side forelimb bud. In Figure 1A the same wild-type forelimb specimen is used as in Figure 1B to illustrate the wild-type expression of Hoxa11.

Accession Numbers

Unprocessed 4C-seq data for mouse and zebrafish samples are available from the Gene Expression Omnibus repository under accession number GSE47644. Ensembl IDs for genes used in this study are as follows: Mouse (Mus musculus), Hoxa4; ENSMUSG00000000942| Hoxa9; ENSMUSG00000038227| Hoxa10; ENSMUSG00000000938| Hoxa11; ENSMUSG00000038210| Hoxa11as; ENSMUSG00000086427| Hoxa13; ENSMUSG00000038203| Hoxd4; ENSMUST00000111980| Hoxd11; ENSMUSG00000042499| Hoxd13; ENSMUSG00000001819|; Zebrafish (Danio rerio), Hoxa4a; ENSDARG00000057724| Hoxa9a; ENSDARG00000096510| Hoxa11a; ENSDARG00000009045| Hoxa13a; ENSDARG00000007609| Hoxa2b; ENSDARG00000023031| Hoxa9b; ENSDARG00000007009| Hoxa13b; ENSDARG00000036254| Hoxd4a; ENSDARG00000059276| Hoxd10a; ENSDARG00000057859| Hoxd11a; ENSDARG00000059267| Hoxd13a; ENSDARG00000059256|; Pufferfish (Tetraodon nigroviridis), Hoxa11a; ENSTNIG00000001767| Hoxa13a; ENSTNIG00000009207| Evx1; ENSTNIG00000000875| Hoxa10b; ENSTNIG00000001780| Hoxa11b; ENSTNIG00000000494| Hoxa13b; ENSTNIG00000001781| HIBADHb; ENSTNIG00000018428| TAX1BP1b; ENSTNIG00000018429| JAZF1b; ENSTNIG00000018430| Hoxd4a; ENSTNIG00000001765| Hoxd9a; ENSTNIG00000016957| Hoxd10a; ENSTNIG00000001775| Hoxd11a; ENSTNIG00000001776| Hoxd12a; ENSTNIG00000001777| Evx2; ENSTNIG00000001817|.

Supporting Information

Figure S1.

Interaction profiles around the mouse HoxA and HoxD genomic loci. (A) Distribution of 4C signals (in percent of total interactions) over large genomic intervals flanking the HoxA and HoxD clusters, including 10 topological domains (−5 to +5 from centromeric to telomeric, respectively, with domains −1 and +1 neighboring the clusters) as determined by using the Hi-C dataset of [32]. The genomic coordinates for these regions are as follows: HoxA, Chr6:47,160,000–55,960,000 and HoxD, Chr2:70,960,000–80,160,000. The 4C signals are combined for all viewpoints after normalization, either in the HoxA or in the HoxD cluster (i.e., Hoxa4, Hoxa9, Hoxa11, Hoxa13 Hoxd4, Hoxd11, and Hoxd13). For each cluster, the diagram on the left shows the percentage of reads localized within each topological domain (indicated in percent), whereas the diagram on the right shows the percentage of reads per megabase (indicated as %/Mb) within each topological domains—that is, after correction for the size of the domains. The legend to the color code referring to the topological domains (numbering after [32]) is on the right. A vast majority of contacts are established within the DNA interval covered by the first two topological domains flanking the Hox clusters on either side (i.e., −1, −2, +1, +2), demonstrating the correspondence between the 4C data and the organization into topological domains [32]. DNA regions of strong local interactions [41] directly surrounding the viewpoints were excluded from the analysis. For HoxA, these are from Chr6:52,098,978–52,227,163, and for HoxD, from Chr2:74,484,971–74,607,492. (B) Interaction profiles for the HoxA and HoxD cluster in dissected digit (autopod) samples using viewpoints located in Hoxa4 and Hoxa13 (top) or Hoxd4 and Hoxd13 (bottom). A DNA interval containing six topological domains is shown (Hi-C heatmap data from [32]). For Hoxa4 (top), robust centromeric (3′) interactions are scored up to the boundary between topological domains −2 and −3 (in red), whereas on the telomeric (5′) side, Hoxa13 mostly interacts with domain +1 and weaker contacts are detected in domain +2 (blue). The same interaction patterns are seen for Hoxd genes, with clear thresholds in contacts occurring either at the boundary between topological domain −2 and −3 in the 5′ contacts (blue) or in 3′ between topological domains +2a and +2b (below and [31]). Consequently, these four topological domains (two on each side, shadowed DNA intervals) were selected for further analyses as presented in Figure 3.



Figure S2.

Interaction profiles of murine HoxD cluster genes. (A–C) 4C analysis using Hoxd4, Hoxd11, and Hoxd13 as viewpoints, in either E12.5 proximal limbs, autopod, or forebrain tissues. (A and B) In the limbs, Hoxd4 and Hoxd13 show strong interaction preferences for the topological domains located 3′ and 5′ of the cluster, respectively, whereas Hoxd11 switches from 3′ to 5′ enriched contacts between the proximal limb and the autopod samples (see [31]). (C) The 3′ and 5′ bias in contact distribution for Hoxd4 and Hoxd13 is also present in inactive forebrain cells.



Figure S3.

Limb enhancers in regions flanking the HoxA cluster. Limb enhancers with proximal and distal specificities have been reported in regions flanking the HoxA cluster. The drawings illustrate published datasets (see below). Vista mm407 (A) and Vista hs1465 (B) are located 3′ from the cluster and drive expression in proximal areas of the limb bud. In the 5′ region, enhancer activity recapitulated part of the Hoxa13 expression pattern in the autopod. Vista mm48 (C) drives expression in the distal hind limb bud, whereas Vista hs1430 (E) shows both proximal and distal specificities. The Vista enhancer sequences are after [43] (​l). The regulatory activity of BAC RPCI-23-347D13 (D), which recapitulates digit expression in a Hoxa13-like pattern, is after [42].



Figure S4.

4C analysis of HoxAa in Zebrafish embryos. (A and B) For all the viewpoints located in the fish HoxAa cluster, the distribution of 4C signals displayed a surprising gap in 3′ located regions, with an abrupt drop around the start of ube2e1 and a sudden recovery ca. 750 kb more 3′, at the start of KANSL1 (unsmoothed raw signal shown in B). These abrupt transitions are unusual in 4C interaction patterns, which typically show a smooth gradient between sequences located in each other's proximity as well as a higher “background” signal between regions in cis showing strong interactions, and hence we considered the possibility of an error in the current assembly of the zebrafish genome. The zebrafish genome is assembled based on whole genome shotgun sequencing (WGS, indicated in red) and a clone path of overlapping BAC clones (“Ctg,” in green), of which the latter is of higher quality (​/faqs.shtml). The region surrounding the HoxAa cluster consists of three contigs (Ctg1861, Ctg1872, Ctg1864). Clone contigs are in general well assembled within themselves, but their position relative to each other is not always certain. The observed signal drop corresponds to Ctg1872 plus its 5′ flanking region of whole genome shotgun assembly, up to the start of ube2e1. This region contains genes that do not belong to the synteny 3′ of the HoxA cluster, a region otherwise well conserved in other vertebrates. Ctg1872 was mapped to its current position using a genetic map, which however is rather uninformative at this close proximity to the centromere, and this contig could equally be well positioned somewhere on the 5′ side of Ctg1861—that is, much further away from the HoxAa cluster (James Torrance, Sanger Institute, Zebrafish genome project, personal communication). Ctg1864 appears reliably placed, with a synteny in line with the vertebrate conservation profile (npvf, cycsa, osbpl3a). In the absence of Ctg1872, the region of WGS assembly 3′ of Ctg1864 (containing ColQ, Snx10a, Cbx3a, and Nfe2l3) would form an uninterrupted gene synteny between Ctg1861 and Ctg1864, and hence we conclude that the placement of Ctg1872 is most likely incorrect and we thus excluded it from the data presented (dotted line Figure 4A). In any case, the analysis including or excluding Ctg1872 plus 5′ flanking WGS only leads to a marginal difference (increased 3′ bias in interaction of <4%) and does not change our interpretation or conclusions.



Figure S5.

4C interaction profiles from the transgenic Tetraodon HoxAb cluster. 4C-seq signals obtained from E12.5 mouse limb buds transgenic for the Tetraodon HoxAb cluster, using Hoxa11b and Hoxa13b as viewpoints (positions indicated with red asterisks). The 4C-seq profiles show strong interactions between the fish genes (viewpoints) and the 5′ BAC region. In addition, the presence of a smooth signal curve over the entire length of the BAC demonstrates its integrity at the integration site of the transgenic mouse line.



Table S1.

Directionality of 4C interactions on both 5′ and 3′ sides of mouse and fish Hox cluster.



Table S2.

List of primers used for 4C-seq analyses.



Table S3.

List of primers used to clone DNA probe for whole mount in situ hybridizations.




We thank F. Chabaud, R. Finazzi, and A. Solaro for their help with animals and M. Docquier and J. Rougemont for their assistance in generating and analyzing high throughput data. We also thank G. Gyapay and O. Jaillon from the Genoscope for providing BAC clones, J. Torrance from the Sanger Institute for discussion, and J. Belahbib and other members of the Duboule laboratories for sharing reagents and discussion. Computations were performed at the Vital-IT ( center for high-performance computing of the SIB Swiss Institute of Bioinformatics using tools developed by the BBCF (

Author Contributions

The author(s) have made the following declarations about their contributions: Conceived and designed the experiments: JMW DN DD. Performed the experiments: JMW DN ML. Analyzed the data: JMW DN ML DD. Wrote the paper: JMW DD.


  1. 1. Woltering JM, Duboule D (2010) The origin of digits: expression patterns versus regulatory mechanisms. Dev Cell 18: 526–532. doi: 10.1016/j.devcel.2010.04.002
  2. 2. Wagner GP, Chiu CH (2001) The tetrapod limb: a hypothesis on its origin. J Exp Zool 291: 226–240. doi: 10.1002/jez.1100
  3. 3. Gaffney ES (1979) Tetrapod monophyly: a phylogenetic analysis. Bull Carnegie Mus Nat Hist 13: 92–105.
  4. 4. Holmgren N (1952) An embryological analysis of the mammalian carpus and its bearing upon the question of the origin of the tetrapod limb. Acta Zoologica 33: 1–115. doi: 10.1111/j.1463-6395.1952.tb00360.x
  5. 5. Coates MI, Jeffery JE, Rut M (2002) Fins to limbs: what the fossils say. Evol Dev 4: 390–401. doi: 10.1046/j.1525-142x.2002.02026.x
  6. 6. Shubin N, Tabin C, Carroll S (1997) Fossils, genes and the evolution of animal limbs. Nature 388: 639–648. doi: 10.1038/41710
  7. 7. Holmgren N (1933) On the origin of the tetrapod limb. Acta Zoologica 14: 185–295. doi: 10.1111/j.1463-6395.1933.tb00009.x
  8. 8. Holmgren N (1939) Contribution on the question of the origin of the tetrapod limb. Acta Zoologica 20: 89–124. doi: 10.1111/j.1463-6395.1939.tb00494.x
  9. 9. Boisvert CA, Mark-Kurik E, Ahlberg PE (2008) The pectoral fin of Panderichthys and the origin of digits. Nature 456: 636–638. doi: 10.1038/nature07339
  10. 10. Kmita M, Tarchini B, Zakany J, Logan M, Tabin CJ, et al. (2005) Early developmental arrest of mammalian limbs lacking HoxA/HoxD gene function. Nature 435: 1113–1116. doi: 10.1038/nature03648
  11. 11. Zakany J, Duboule D (2007) The role of Hox genes during vertebrate limb development. Current Opinion in Genetics & Development 17: 359–366. doi: 10.1016/j.gde.2007.05.011
  12. 12. Nelson CE, Morgan BA, Burke AC, Laufer E, DiMambro E, et al. (1996) Analysis of Hox gene expression in the chick limb bud. Development 122: 1449–1466.
  13. 13. Haack H, Gruss P (1993) The establishment of murine Hox-1 expression domains during patterning of the limb. Dev Biol 157: 410–422. doi: 10.1006/dbio.1993.1145
  14. 14. Dolle P, Izpisua-Belmonte JC, Falkenstein H, Renucci A, Duboule D (1989) Coordinate expression of the murine Hox-5 complex homoeobox-containing genes during limb pattern formation. Nature 342: 767–772. doi: 10.1038/342767a0
  15. 15. Schneider I, Shubin NH (2013) The origin of the tetrapod limb: from expeditions to enhancers. Trends Genet 29: 419–426. doi: 10.1016/j.tig.2013.01.012
  16. 16. Davis MC (2013) The deep homology of the autopod: insights from hox gene regulation. Integr Comp Biol 53: 224–232. doi: 10.1093/icb/ict029
  17. 17. Freitas R, Gomez-Marin C, Wilson JM, Casares F, Gomez-Skarmeta JL (2012) Hoxd13 contribution to the evolution of vertebrate appendages. Dev Cell 23: 1219–1229. doi: 10.1016/j.devcel.2012.10.015
  18. 18. Sordino P, Duboule D (1996) A molecular approach to the evolution of vertebrate paired appendages. Trends Ecol Evol 11: 114–119. doi: 10.1016/0169-5347(96)81089-5
  19. 19. Yano T, Tamura K (2013) The making of differences between fins and limbs. J Anat 222: 100–113. doi: 10.1111/j.1469-7580.2012.01491.x
  20. 20. Sordino P, van der Hoeven F, Duboule D (1995) Hox gene expression in teleost fins and the origin of vertebrate digits. Nature 375: 678–681. doi: 10.1038/375678a0
  21. 21. Davis MC, Dahn RD, Shubin NH (2007) An autopodial-like pattern of Hox expression in the fins of a basal actinopterygian fish. Nature 447: 473–476. doi: 10.1038/nature05838
  22. 22. Freitas R, Zhang G, Cohn MJ (2007) Biphasic Hoxd gene expression in shark paired fins reveals an ancient origin of the distal limb domain. PLoS ONE 2: e754 doi: 10.1371/journal.pone.0000754.
  23. 23. Johanson Z, Joss J, Boisvert CA, Ericsson R, Sutija M, et al. (2007) Fish fingers: digit homologues in sarcopterygian fish fins. J Exp Zool B Mol Dev Evol 308: 757–768. doi: 10.1002/jez.b.21197
  24. 24. Ahn D, Ho RK (2008) Tri-phasic expression of posterior Hox genes during development of pectoral fins in zebrafish: implications for the evolution of vertebrate paired appendages. Dev Biol 322: 220–233. doi: 10.1016/j.ydbio.2008.06.032
  25. 25. Shubin N, Tabin C, Carroll S (2009) Deep homology and the origins of evolutionary novelty. Nature 457: 818–823. doi: 10.1038/nature07891
  26. 26. Metscher BD, Takahashi K, Crow K, Amemiya C, Nonaka DF, et al. (2005) Expression of Hoxa-11 and Hoxa-13 in the pectoral fin of a basal ray-finned fish, Polyodon spathula: implications for the origin of tetrapod limbs. Evol Dev 7: 186–195. doi: 10.1111/j.1525-142x.2005.05021.x
  27. 27. Schneider I, Aneas I, Gehrke AR, Dahn RD, Nobrega MA, et al. (2011) Appendage expression driven by the Hoxd Global Control Region is an ancient gnathostome feature. Proc Natl Acad Sci U S A 108: 12782–12786. doi: 10.1073/pnas.1109993108
  28. 28. Amemiya CT, Alfoldi J, Lee AP, Fan S, Philippe H, et al. (2013) The African coelacanth genome provides insights into tetrapod evolution. Nature 496: 311–316. doi: 10.1038/nature12027
  29. 29. Montavon T, Soshnikova N, Mascrez B, Joye E, Thevenet L, et al. (2011) A regulatory archipelago controls Hox genes transcription in digits. Cell 147: 1132–1145. doi: 10.1016/j.cell.2011.10.023
  30. 30. Spitz F, Herkenne C, Morris MA, Duboule D (2005) Inversion-induced disruption of the Hoxd cluster leads to the partition of regulatory landscapes. Nat Genet 37: 889–893. doi: 10.1038/ng1597
  31. 31. Andrey G, Montavon T, Mascrez B, Gonzalez F, Noordermeer D, et al. (2013) A switch between topological domains underlies HoxD genes collinearity in mouse limbs. Science 340: 1234167. doi: 10.1126/science.1234167
  32. 32. Dixon JR, Selvaraj S, Yue F, Kim A, Li Y, et al. (2012) Topological domains in mammalian genomes identified by analysis of chromatin interactions. Nature 485: 376–380. doi: 10.1038/nature11082
  33. 33. Nora EP, Lajoie BR, Schulz EG, Giorgetti L, Okamoto I, et al. (2012) Spatial partitioning of the regulatory landscape of the X-inactivation centre. Nature 485: 381–385. doi: 10.1038/nature11049
  34. 34. Zakany J, Duboule D (1996) Synpolydactyly in mice with a targeted deficiency in the HoxD complex. Nature 384: 69–71. doi: 10.1038/384069a0
  35. 35. Zakany J, Zacchetti G, Duboule D (2007) Interactions between HOXD and Gli3 genes control the limb apical ectodermal ridge via Fgf10. Dev Biol 306: 883–893. doi: 10.1016/j.ydbio.2007.03.517
  36. 36. Yokouchi Y, Sasaki H, Kuroiwa A (1991) Homeobox gene expression correlated with the bifurcation process of limb cartilage development. Nature 353: 443–445. doi: 10.1038/353443a0
  37. 37. Hsieh-Li HM, Witte DP, Weinstein M, Branford W, Li H, et al. (1995) Hoxa 11 structure, extensive antisense transcription, and function in male and female fertility. Development 121: 1373–1385.
  38. 38. Montavon T, Le Garrec JF, Kerszberg M, Duboule D (2008) Modeling Hox gene regulation in digits: reverse collinearity and the molecular origin of thumbness. Genes Dev 22: 346–359. doi: 10.1101/gad.1631708
  39. 39. Sheth R, Bastida MF, Kmita M, Ros M (2013) “Self-regulation,” a new facet of Hox genes' function. Dev Dyn doi: 10.1002/dvdy.24019.
  40. 40. van de Werken HJ, Landan G, Holwerda SJ, Hoichman M, Klous P, et al. (2012) Robust 4C-seq data analysis to screen for regulatory DNA interactions. Nat Methods 9: 969–972. doi: 10.1038/nmeth.2173
  41. 41. Noordermeer D, Leleu M, Splinter E, Rougemont J, De Laat W, et al. (2011) The dynamic architecture of Hox gene clusters. Science 334: 222–225. doi: 10.1126/science.1207194
  42. 42. Lehoczky JA, Innis JW (2008) BAC transgenic analysis reveals enhancers sufficient for Hoxa13 and neighborhood gene expression in mouse embryonic distal limbs and genital bud. Evol Dev 10: 421–432. doi: 10.1111/j.1525-142x.2008.00253.x
  43. 43. Visel A, Minovitsky S, Dubchak I, Pennacchio LA (2007) VISTA Enhancer Browser–a database of tissue-specific human enhancers. Nucleic Acids Res 35: D88–92. doi: 10.1093/nar/gkl822
  44. 44. Amores A, Force A, Yan YL, Joly L, Amemiya C, et al. (1998) Zebrafish hox clusters and vertebrate genome evolution. Science 282: 1711–1714. doi: 10.1126/science.282.5394.1711
  45. 45. Woltering JM, Durston AJ (2006) The zebrafish hoxDb cluster has been reduced to a single microRNA. Nat Genet 38: 601–602. doi: 10.1038/ng0606-601
  46. 46. Kuraku S, Meyer A (2009) The evolution and maintenance of Hox gene clusters in vertebrates and the teleost-specific genome duplication. Int J Dev Biol 53: 765–773. doi: 10.1387/ijdb.072533km
  47. 47. Takamatsu N, Kurosawa G, Takahashi M, Inokuma R, Tanaka M, et al. (2007) Duplicated Abd-B class genes in medaka hoxAa and hoxAb clusters exhibit differential expression patterns in pectoral fin buds. Dev Genes Evol 217: 263–273. doi: 10.1007/s00427-007-0137-4
  48. 48. Jaillon O, Aury JM, Brunet F, Petit JL, Stange-Thomann N, et al. (2004) Genome duplication in the teleost fish Tetraodon nigroviridis reveals the early vertebrate proto-karyotype. Nature 431: 946–957. doi: 10.1038/nature03025
  49. 49. Lehoczky JA, Williams ME, Innis JW (2004) Conserved expression domains for genes upstream and within the HoxA and HoxD clusters suggests a long-range enhancer existed before cluster duplication. Evol Dev 6: 423–430. doi: 10.1111/j.1525-142x.2004.04050.x
  50. 50. Duboule D, Morata G (1994) Colinearity and functional hierarchy among genes of the homeotic complexes. Trends Genet 10: 358–364. doi: 10.1016/0168-9525(94)90132-5
  51. 51. Coates MI (1995) Limb evolution. Fish fins or tetrapod limbs–a simple twist of fate? Curr Biol 5: 844–848. doi: 10.1016/s0960-9822(95)00169-2
  52. 52. Villavicencio-Lorini P, Kuss P, Friedrich J, Haupt J, Farooq M, et al. (2010) Homeobox genes d11-d13 and a13 control mouse autopod cortical bone and joint formation. J Clin Invest 120: 1994–2004. doi: 10.1172/jci41554
  53. 53. Thorogood P (1991) The development of the teleost fin and implications for our understanding of tetrapod evolution. In Hinchliffe JH, Summerbell D, editor. Developmental patterning of the vertebrate limb. New York: Plenum Press. pp. 347–354.
  54. 54. Zhang J, Wagh P, Guay D, Sanchez-Pulido L, Padhi BK, et al. (2010) Loss of fish actinotrichia proteins and the fin-to-limb transition. Nature 466: 234–237. doi: 10.1038/nature09137
  55. 55. Yano T, Abe G, Yokoyama H, Kawakami K, Tamura K (2012) Mechanism of pectoral fin outgrowth in zebrafish development. Development 139: 2916–2925. doi: 10.1242/dev.075572
  56. 56. Hodgkinson VS, Ericsson R, Johanson Z, Joss JMP (2009) The apical ectodermal ridge in the pectoral fin of the Australian Lungfish (Neoceratodus forsteri): keeping the fin to limb transition in the fold. Acta Zoologica 90: 253–263. doi: 10.1111/j.1463-6395.2008.00349.x
  57. 57. Spitz F, Gonzalez F, Duboule D (2003) A global control region defines a chromosomal regulatory landscape containing the HoxD cluster. Cell 113: 405–417. doi: 10.1016/s0092-8674(03)00310-6
  58. 58. Fromental-Ramain C, Warot X, Messadecq N, LeMeur M, Dolle P, et al. (1996) Hoxa-13 and Hoxd-13 play a crucial role in the patterning of the limb autopod. Development 122: 2997–3011.
  59. 59. Kmita M, Kondo T, Duboule D (2000) Targeted inversion of a polar silencer within the HoxD complex re-allocates domains of enhancer sharing. Nat Genet 26: 451–454.
  60. 60. Woltering JM, Vonk FJ, Muller H, Bardine N, Tuduce IL, et al. (2009) Axial patterning in snakes and caecilians: evidence for an alternative interpretation of the Hox code. Dev Biol 332: 82–89. doi: 10.1016/j.ydbio.2009.04.031
  61. 61. Gerard M, Chen JY, Gronemeyer H, Chambon P, Duboule D, et al. (1996) In vivo targeted mutagenesis of a regulatory element required for positioning the Hoxd-11 and Hoxd-10 expression boundaries. Genes Dev 10: 2326–2334. doi: 10.1101/gad.10.18.2326
  62. 62. Warot X, Fromental-Ramain C, Fraulob V, Chambon P, Dolle P (1997) Gene dosage-dependent effects of the Hoxa-13 and Hoxd-13 mutations on morphogenesis of the terminal parts of the digestive and urogenital tracts. Development 124: 4781–4791.