Co-option of an ancestral cloacal regulatory landscape during digit evolution

Abstract

The fin-to-limb transition in vertebrate evolution has been central to the study of how development underlies evolutionary change. In this context, the functional analysis of Hox gene regulation to infer evolutionary trajectories has been critical to explain the origin of new features. In tetrapods, the transcription of Hoxd genes in developing digits depends on a set of enhancers forming a large regulatory landscape^1,2. The presence of a syntenic counterpart in zebrafish, which lacks digits, suggests deep homology³ or shared developmental foundations underlying distal fin and limbs. However, how this regulatory program evolved has remained unresolved. We genetically evaluated the function of the zebrafish Hoxd regulatory landscapes by comparatively assessing the effects of their full deletions. We show that, unlike in mice, deletion of these regions in fish does not disrupt hoxd gene transcription during distal fin development. By contrast, we found that this deficiency leads to the loss of expression within the cloaca, a structure related by ancestry to the mammalian urogenital sinus, and that distal hox13 genes are essential for correct cloacal formation. Because Hoxd gene regulation in the mouse urogenital sinus relies on enhancers located in this same chromatin domain that controls digit development, we propose that the current regulatory landscape active in distal limbs was co-opted as a whole in tetrapods from a pre-existing cloacal regulatory machinery.

Main

The skeletal pattern of tetrapod limbs has been conserved since the Devonian Period, with a universal architecture of segments along the proximal-to-distal axis. The stylopod is a single bone (upper arm or thigh) attached at one end to the torso and at the other end to two zeugopod bones (forearm or leg), and the most distal are the mesopod (wrist and ankle) and autopod (hand and foot). The formation of this generic pattern began before the water-to-land transition, because sarcopterygian fishes display structures clearly related to proximal tetrapod limb structures⁴. However, when homologies are considered between fin structures and the most distal parts of tetrapod appendages (mesopod and autopod), the extent to which fishes possess homologous skeletal elements remains debatable. Although mesopodial elements and extensive distal segments are present in sarcopterygian fins, the presence of true digital homologues remains controversial^5,6.

Because the HoxA and HoxD gene clusters have been shown to be instrumental in making tetrapod limbs^7,8,9,10, their expression domains during fin development were used to infer the presence of an autopod-related structure in fishes. In particular, Hoxa13 and Hoxd13 have been studied because of their specific autopodial expression in tetrapod limbs¹⁰ and because their combined inactivation in mice produces autopodial agenesis⁷. An analysis of hox13 genes in the teleost fin suggested that a ‘distal program’ also exists in fishes, because their combined inactivation leads to the loss of distal structures, implying that a genetic regulatory network, or part thereof, would have preceded digit formation in tetrapods^11,12. In such a scenario, the autopods of tetrapods are proposed to form from the postaxial vestiges of an ancestral sarcopterygian fin^13,14. The partial retention of expression patterns that presage the emergence of digits in ray-finned and chondrichthyan fishes is nevertheless suggestive of a common regulatory program shared among vertebrates, the deployment of which, in different species, accompanied changes in form¹³.

During tetrapod limb bud development, a series of enhancers within a large regulatory landscape positioned 3′ of the HoxD gene cluster (3DOM) control the transcription of Hoxd genes up to Hoxd11 in a proximal expression domain. These expression domains encompass the tissue of the future stylopod (upper arm) and zeugopod (forearm) (Extended Data Fig. 1a (green and schemes on the left))¹⁵. Posterior–distal limb bud cells then switch off these enhancers and activate another large regulatory landscape (5DOM), located on the other side of (5′ to) the gene cluster. This region is enriched in conserved enhancer elements that have been found to control the formation of digits by activating Hoxd13 and its closest neighbours (Extended Data Fig. 1a (blue)). Supporting this model, the deletion of 3DOM abrogated the expression of all Hoxd genes in the proximal limb domain¹⁵, whereas the deletion of 5DOM removed all Hoxd messenger RNAs (mRNAs) from the forming autopod².

In teleosts, a further whole-genome duplication (TWGD) occurred, leading to the presence of two hoxd cluster orthologues. However, the hoxdb cluster was lost, resulting in a 1:1 orthology between the single remaining zebrafish hoxda cluster and the tetrapod HoxD cluster. The hoxda genes are also expressed during early fin bud development, with progressively nested expression domains comparable with the amniote situation^13,16. At a later stage, transcription of both hoxd9a and hoxd10a persists in the ‘preaxial’ (anterior) part of the fin bud only (Extended Data Fig. 1b (magenta)), whereas hoxd11a, hoxd12a and hoxd13a transcripts are restricted to ‘postaxial’ (posterior) cells (Extended Data Fig. 1b (orange)), as is the case in the emerging fin bud¹⁶. For the latter genes, combined inactivation has revealed their function during distal fin skeletal development^11,17. However, despite the identification of DNA elements orthologous to mouse distal enhancers¹², the functionality of the 3DOM and 5DOM regulatory landscapes in zebrafish has not yet been addressed. Owing to this gap in knowledge, the existence of comparable bimodal regulation of Hoxd gene expression has remained controversial.

To address these issues, we deleted both zebrafish hoxda regulatory landscapes to assess the extent to which appendage regulation by the orthologous 3DOM and 5DOM regions was conserved between zebrafish and mice. Embedding these results within a phylogenetic and genomic framework, we propose a new hypothesis of co-option for the origin of the distal regulation of appendages and genitalia in vertebrates.

Zebrafish hoxda locus

The zebrafish hoxda locus shares a high degree of synteny with the HoxD locus in mammals, reflecting broad conservation, given the key patterning role of this complex in the development of many axial structures. The gene cluster is flanked by two gene deserts: 3DOM (3′-located domain) and 5DOM (5′-located domain). As in mammals, the extent of both 3DOM and 5DOM corresponds to topologically associating domains (TADs), and 3DOM is split into two sub-TADs (Extended Data Fig. 2). This remarkable similarity in three-dimensional conformations, although with a 2.6-fold difference in size between the mouse versus zebrafish loci, is further supported by the conserved position and orientation of critical CTCF binding sites within the gene clusters and their enrichment at TAD and sub-TAD borders (Extended Data Fig. 2).

Interspecies genomic alignments revealed conserved sequences within 5DOM across vertebrates, whereas little conservation was scored in 3DOM (Extended Data Fig. 3a,b). Within the 5DOM comparison, we identified several previously annotated mouse enhancers in zebrafish^12,18 (see “Enhancers in the fish 5DOM”). Consistent with the apparent conservation of chromatin structure, we found the same global organization of both coding and non-coding elements as in the mouse landscape. When compared with the size of the Hox cluster, the relative sizes of both the 3′ and 5′ gene deserts were larger in mouse than in zebrafish, and the zebrafish 5DOM was found to be larger than 3DOM, in contrast to the mouse situation (Extended Data Fig. 3c). Because the overall genomic organization of both HoxD loci is well conserved between mammals and fishes, we concluded that these two flanking gene deserts and their TADs are ancestral features predating the divergence between ray-finned fishes and tetrapods, probably conserved because of important regulatory functions. Whether these domains have or retain Hox gene regulation, as initially defined at this locus in the mouse^1,19, remains unclear.

Zebrafish hoxda regulatory landscapes

To address the potential function(s) of zebrafish hoxd gene deserts, we explored the histone modification profiles for H3K27 acetylation and H3K27 trimethylation using the cleavage under targets and release using nuclease (CUT&RUN)²⁰ assay with the posterior trunk as a source of cells, that is, a domain where most hox genes are active. As a control sample, we used the corresponding dissected heads where hoxda genes were not expressed (Extended Data Fig. 4a,b). H3K27ac-positive marks were enriched over 3DOM, which was devoid of negative H3K27me3 marks (Extended Data Fig. 4c (green and magenta, respectively)), suggesting the strong involvement of 3DOM in transcriptional regulation, as seen within the hoxda cluster itself. By contrast, H3K27me3 marks were enriched over 5DOM. This analysis revealed that both zebrafish gene deserts might indeed serve as regulatory landscapes, with distributions of histone marks comparable with tetrapods in the same developmental context.

To assess the functional potential of both hoxda gene deserts, we generated zebrafish mutant lines carrying full deletions of either 5DOM (hoxda^del(5DOM), referred to as Del(5DOM)) or 3DOM (hoxda^del(3DOM) or Del(3DOM)), using CRISPR–Cas9 chromosome editing. We first examined the impact of these large deletions on hoxd13a, hoxd10a and hoxd4a expression using whole-mount in situ hybridization (WISH), spanning from 36 h post-fertilization (hpf), that is, from the onset of hoxd13a expression¹⁶, to 72 hpf. In Del(3DOM) mutant embryos, the expression of both hoxd4a and hoxd10a completely disappeared from the pectoral fin buds (Fig. 1a (right and middle panels, arrowheads)). The same effect was observed at all stages analysed (Fig. 1a). These data are consistent with those of a similar analysis in mice, in which the limb proximal expression domain was no longer visible upon deletion of the 3DOM landscape¹⁵. This demonstrates that, similar to tetrapods, enhancers controlling the transcription of hoxd3a to hoxd10a during fin bud development are located in the adjacent 3′ landscape. Thus, 3DOM has an ancestral regulatory function in the development of proximal paired appendages. Expression of hoxd13a in postaxial cells, however, remained unchanged, with a global transcript distribution indistinguishable from that of wild-type fin buds (Fig. 1a (left panels, arrowheads)). These data indicate that the control of hoxd13a expression is distinct from that impacting hoxd3a to hoxd10a, as is also the case for tetrapods² (Extended Data Fig. 1a).

Fig. 1: Regulation of hoxda genes in pectoral fins lacking the 3DOM and 5DOM regulatory landscapes.

a–d, Expression of hoxd13a, hoxd10a and hoxd4a using WISH at 36 hpf, 48 hpf, 60 hpf and 72 hpf in zebrafish embryos with either the 3DOM (a) or 5DOM (b–d) regulatory landscape deleted. Wild-type and homozygous mutant embryos derived from the same cross are shown side by side. WISH was performed on at least five embryos for each combination of probes and genotypes, with four or more technical replicates, except for the Hoxd10 probe, which had only one technical replicate. a, Expression of both hoxd10a and hoxd4a is completely lost in mutant fin buds lacking 3DOM (arrows), whereas expression of hoxd13a is identical to that of wild-type embryos (arrows). b, In fin buds lacking 5DOM, expression of hoxd13a, hoxd10a and hoxd4a is identical to matched wild-type embryos up to 48 hpf (arrows). c,d, However, at 60 hpf (c) and 72 hpf (d), a decrease in intensity was observed throughout, particularly marked in the distal aspect of the fin bud (arrows). The degree of this attenuation is variable among mutant fish (Supplementary Fig. 1). WT, wild type. Scale bars, 50 µm.

To determine whether hoxd13a transcription was controlled by enhancers present within 5DOM, we similarly analysed Del(5DOM) zebrafish embryos using WISH. Consistent with the regional control of Hox gene transcription, neither hoxd4a nor hoxd10a expression was affected in the mutant Del(5DOM) fin buds (Fig. 1b (arrowheads)). Up to 48 hpf, hoxd13a transcripts were virtually unaffected, with a pattern closely matching that of the control fin buds (Fig. 1b (arrowheads) and Supplementary Fig. 1). However, at later stages of development, the 5DOM deletion led to variably attenuated patterns of hoxd13a expression, whereas its overall mRNA distribution remained globally similar to that of the wild type. We confirmed this diminished expression in late-stage mutant fins using the hybridization chain reaction (HCR), but this trend was not significant (Extended Data Fig. 5). These results indicate that although some regulatory activity may exist within the 5DOM in fishes, the most significant regulatory inputs for hoxd13a are located outside the domain (Fig. 1c,d and Extended Data Fig. 5). This situation differs from that in mice, in which a similar deletion abrogates hoxd13 expression entirely.

These two genomic regions also control expression in other axial systems in mice^21,22. Thus, we extended our analysis to assess shared components of regulation between these regulatory landscapes. Mutant Del(3DOM) embryos did not reveal visible differences in expression in the trunk (hoxd13a, hoxd10a and hoxd4a), the pseudo-cloacal region (hoxd13a) or the branchial arches and rhombomeres (hoxd4a) (Extended Data Fig. 6a). Del(5DOM) embryos also showed comparable expression to wild-type controls, except for the complete disappearance of hoxd13a transcripts from the pseudo-cloacal region (Fig. 2). We noticed a temporary reduction in hoxd13a expression in the tailbud (Fig. 2a); yet, this deficit was no longer detectable at 36 hpf. These results revealed that in zebrafish, 5DOM-located enhancers regulate the hoxd13a genes in the cloacal area from its onset of expression until at least 72 hpf, whereas neither hoxd10a nor hoxd4a is expressed there (Extended Data Fig. 6). As previously reported, both hoxd13a and hoxa13b^23,24 transcripts were found in the region around the nascent pronephric ducts and the hindgut. These structures eventually converge towards a single pseudo-cloacal complex that exits the body with adjacent openings that never completely fuse. In 72-hpf larvae, hoxd13a mRNAs appeared in the posterior gut in both control and mutant samples. However, transcripts were still absent in the mutant cloacal region (Fig. 2c (black and red arrows, respectively)). This demonstrates that these two expression specificities are regulated separately.

Fig. 2: Effects of deleting 5DOM on hoxd13a regulation in the pseudo-cloacal region.

a–c, Expression of hoxd13a is completely lost in the cloaca of 16-hpf (a), 36-hpf (b) and 72-hpf (c) embryos lacking 5DOM (open red arrowheads), whereas in embryos lacking 3DOM it is identical to controls (filled red arrowheads), indicating that 5DOM is required for hoxd13a activity in the pseudo-cloacal region. At 16 hpf, a temporary decrease in hoxd13a expression was observed in the tailbuds lacking 5DOM (open black arrowhead), but this effect was no longer observed at later stages (filled black arrowheads). WISH was performed on at least two embryos per condition. b, Enlargement of the cloacal region showing hoxd13a transcripts mostly lining the very end of the intestinal canal, converging towards the cloacal region. c, At 72 hpf, hoxd13a expression was detected in the posterior epithelial part of the gut in both control and mutant larvae (black arrow), indicating that expression in the cloacal region (red arrowheads) responds to a separate regulatory control. Scale bars, 200 µm (a–c (whole embryos)), 50 µm (a–c (enlarged views)).

The cloaca evolved at the base of the craniate lineage as a single orifice for the digestive, urinary and reproductive tracts, as found in birds and squamates. In mammals, a cloaca initially forms early in embryonic development, but as the embryo grows, it divides into different openings for the urogenital and digestive systems. To evaluate whether the observed 5DOM regulation of hoxd13a in the zebrafish pseudo-cloacal region is a derived or ancestral condition, we examined the developing mouse urogenital sinus (UGS), a structure derived from the mammalian embryonic cloacal area.

Hoxd gene regulation in the UGS

The UGS, positioned below the urinary bladder, is derived from a cloacal rudiment originating from the hindgut and ectodermal tissue^25,26. During mid-gestation, as the nephric and Müllerian ducts grow towards the posterior end of the embryo, they meet and fuse with the invaginating cloaca. We performed WISH on dissected urogenital systems from control murine male and female embryos at embryonic day 18.5 (E18.5) (Fig. 3a,b). All genes tested, except Hoxd13, were detected in the anterior portions of the urogenital system, including the kidneys, uterus and deferens ducts^27,28 (Extended Data Fig. 7a,b). By contrast, Hoxd13 expression was restricted to UGS in both male and female embryos, along with Hoxd12, Hoxd11 and, to a weaker extent, Hoxd10 (Fig. 3 and Extended Data Fig. 7). These are the same four genes that respond to both the digit and external genital long-range regulations exerted by 5DOM²⁹, thus suggesting a transcriptional control coming from this same 5′-located domain.

Fig. 3: Hoxd gene expression in the mouse urogenital system.

a, Schematic representations of male and female urogenital systems. UGS is indicated with a red circle. b, WISH of Hoxd13 in representative female and male urogenital systems. Hoxd13 is selectively expressed in UGS. c, Schematic representation of the two HoxD inversion alleles. The locations of the inversion breakpoints are depicted with red arrows. Hox genes shown in shades of purple. d, Hoxd13 expression in urogenital systems of mice carrying the inversions (WISH, left panel; RT–qPCR, right panel). HoxD^{Inv(Itga6–attP}: wild type (n = 4 samples) and inversion (n = 3 samples); HoxD^{Inv(Itga6–nsi)d11lac)}: wild type (n = 3 samples) and inversion (n = 3 samples). Expression of Hoxd13 in UGS was abolished when the target genes were disconnected from 5DOM. The boxes represent the interquartile range (IQR), with the lower and upper hinges denoting the first and third quartiles (25th and 75th percentiles). Whiskers extended from the hinges to the furthest data points within 1.5 times the IQR. The upper whisker reached the largest value within this range, whereas the lower whisker extended to the smallest value within 1.5 times the IQR from the hinge. Each experiment was repeated independently at least twice with similar results (b,d). B, bladder; K, kidney; O, ovary; T, testis. Scale bar, 1 mm (b,d).

We verified this using an engineered inversion that keeps the HoxD cluster linked with 5DOM but takes them far away from 3DOM (Fig. 3c; HoxD^{Inv(Itga6–AttP)})³⁰. In this allele, Hoxd13 transcription in UGS was unaffected (Fig. 3d). We then tested a comparable inversion, yet with a breakpoint immediately 5′ to the HoxD cluster, thus disconnecting 5DOM from all Hoxd genes (Fig. 3c; HoxD^{Inv(Itga6–nsi)d11lac)})³¹. This inversion led to a virtually complete loss of Hoxd13 transcription (Fig. 3d), suggesting that most, if not all, UGS-specific enhancers were located within 5DOM. We confirmed this using a large BAC transgene containing only HoxD with no flanking sequences³⁰ (Extended Data Fig. 7c,d) introduced into mice lacking both copies of the HoxD locus³² (Extended Data Fig. 7c). In this mutant line, Hoxd13 transcription was not detected (Extended Data Fig. 7c (arrows)). Finally, we examined the β-galactosidase staining of a LacZ reporter integrated into the same BAC transgene. Whereas the reporter was strongly active in fetal kidneys as expected²², UGS was not stained (Extended Data Fig. 7d). By contrast, a comparable LacZ enhancer–reporter transgene integrated into 5DOM within the inversion, separating 5DOM from the HoxD cluster (Extended Data Fig. 7d; HoxD^{Inv(Itga6–nsi)d11lac)}) robustly stained E18.5 UGS, again supporting the presence of UGS enhancers within 5DOM rather than within the HoxD cluster (Extended Data Fig. 7d).

We quantified the reduction in Hoxd gene expression in the HoxD^{Inv(Itga6–nsi)d11lac)} allele using RNA sequencing (RNA-seq) on E18.5 UGS of males and females. In both cases, Hoxd13, Hoxd12 and Hoxd10 transcription levels dropped abruptly when compared with the wild-type samples, whereas the transcription level of other Hoxd genes was not affected (Extended Data Fig. 8a). Altogether, these genetic configurations demonstrated that mammalian 5DOM contains UGS enhancers, similar to zebrafish 5DOM. It also showed that the Hoxd genes responsive to this regulation (Hoxd13–Hoxd10) are the same subgroup that responds to both digit and external genital enhancers.

Identification of mouse UGS enhancers

To identify UGS enhancers within the mouse 5DOM, we used three scanning deletion alleles covering 5DOM² (Fig. 4a (red)) and measured the change in expression by reverse transcription–quantitative polymerase chain reaction (RT–qPCR) (Extended Data Fig. 8b). In the HoxD^{Del(Atf2–SB1)} allele, the most distal portion of 5DOM was removed, with no impact on Hoxd gene expression levels. However, when either the central (HoxD^{Del(SB1–Rel5)}) or the most proximal (HoxD^{Del(Rel5–Rel1)}) portions of 5DOM were removed, transcription of Hoxd13, Hoxd12 and Hoxd10 was significantly reduced, indicating that these two 5DOM intervals contain UGS enhancers (Extended Data Fig. 8b).

Fig. 4: UGS enhancers located in 5DOM.

a, Chromatin accessibility (ATAC-seq; blue track) and H3K27ac (green track) and H3K27me3 (red track) ChIP–seq profiles from micro-dissected male UGS at E18.5. The red lines on top delineate the three deletions within 5DOM with the percentage of Hoxd13 expression left in UGS after each deletion (Extended Data Fig. 8). Hoxd genes are in purple. Blue rectangles indicate previously described 5DOM enhancers. The dashed box highlights an H3K27ac-positive cluster of ATAC-seq peaks lacking H3K27me3 and containing three enhancer sequences: GT2, Island E and CsB. ATAC-seq and H3K27ac, n = 2 experiments, one representative example is shown; H3K27me3, n = 1 experiment. b, Regulatory potential of the GT2, Island E and CsB elements when cloned into a lacZ reporter cassette. GT2 induces robust lacZ expression in UGS of both male and female embryos, whereas Island E shows weaker expression. The CsB transgene induces robust expression in males (no data available for females). Each experiment was repeated independently at least twice with similar results. Scale bars, 100 kb (a), 1 mm (b).

We then measured chromatin accessibility using the assay for transposase-accessible chromatin with sequencing (ATAC-seq) and profiled H3K27ac and H3K27me3 histone marks associated with either active or inactive chromatin, respectively, using chromatin immunoprecipitation followed by sequencing (ChIP–seq) on micro-dissected male UGSs (Fig. 4a). We identified a cluster of several conspicuous ATAC-seq and H3K27ac signals located approximately 200 kb upstream of Hoxd13, in a region encompassing the Rel5 breakpoint, that is, between the Del(SB1–Rel5) and the Del(Rel5–Rel1) deletions (Fig. 4a (dashed box)). Within this 67-kb large region, the ATAC and H3K27ac signals matched three elements previously characterized as enhancer sequences, yet with distinct tissue specificities. The GT2 and Island E sequences were identified as pan-dorsal and proximal-dorsal genital tubercle-specific enhancers, respectively^33,34, whereas the CsB element was reported as a neural and distal limb and fin enhancer element^1,12,35. Within these two elements, the main ATAC peak was positioned in a small region that was relatively depleted for the H3K27ac mark (Extended Data Fig. 9), which is a hallmark of active enhancer elements³⁶. We tested these three putative enhancers in an enhancer–reporter assay, and all three sequences were able to drive robust lacZ expression in UGS, closely matching the expression of posterior Hoxd genes in this area in both male and female specimens (Fig. 4b). This indicates that in mammals, 5DOM contains a set of multiple enhancer elements that control the transcriptional activation of Hoxd genes in UGS.

Enhancers in the fish 5DOM

Although the GT2 and CsB elements showed some islands of sequence conservation across bony fish hoxda loci, Island E was present only in mammals (Extended Data Fig. 9). Because the teleosts underwent another whole-genome duplication (TWGD), such sequences may have been reshuffled to different genomic locations and hence might bias our interpretation. To assess the impact of TWGD on the structure of 5DOM, we aligned the genomes of 13 vertebrate species, including holosteans (gar and bowfin), cartilaginous fishes (skate and catshark) and the coelacanth, a lobe-finned fish that did not undergo further whole-genome duplication. First, we projected the alignment onto the zebrafish genome (Supplementary Fig. 2a). With one exception (Supplementary Fig. 2a (red arrow)), all sequences conserved between zebrafish and fugu, two distantly related teleosts, were conserved in the gar and bowfin genomes. This suggests that TWGD did not lead to widespread loss or rearrangement of conserved 5DOM elements despite the genetic redundancy introduced by duplication. This may be related to the loss of the hoxdb paralogue cluster, thus leading to the maintenance of all regulatory constraints that were initially applied to the hoxda locus before duplication.

In support of this conclusion, conservation blocks within zebrafish alignments largely covered the hox genes, which have many paralogues yet not the 5DOM regulatory landscape, demonstrating that the potential hoxdb counterpart of 5DOM was also lost after TWGD. Finally, although tetrapods and ray-finned fishes are more closely related to each other than to cartilaginous fishes, the 5DOM of ray-finned and cartilaginous fishes exhibited a higher degree of conservation, indicating sequence divergence in tetrapods following their split from the coelacanth lineage (Supplementary Fig. 2a (black arrows)).

To assess the impact of TWGD when a non-teleost species was used as a reference, we projected a 13-way alignment onto the mouse genome (Supplementary Fig. 2b). This revealed a strong positive correlation between conservation and phylogeny. Conservation blocks between the mouse and any species of fish (teleost or non-teleost) were largely the same, although with a few cases in which sequences were lost specifically in the zebrafish genome (Supplementary Fig. 2b (black arrows)). To confirm that the conservation blocks truly represent the corresponding loci across species, we repeated the alignments using the 5DOM region only in mouse, gar and zebrafish (Supplementary Fig. 2c). This confirmed that, at least at the hoxda locus, TWGD neither disrupted nor reshuffled any conserved sequence beyond what would be expected on the basis of phylogenic distance. We concluded that TWGD does not bias the analysis of 5DOM regulatory function.

Because of the low sequence conservation between the mammalian and fish 5DOM, we looked at potential regulatory sequences in the fish counterpart, either by producing ATAC-seq profiles (Supplementary Fig. 3 (three profiles on top)) or by mining single-cell ATAC-seq datasets^37,38 (Supplementary Fig. 3 (bottom four profiles)). The quality and reproducibility of these datasets can be verified by the signals recovered over the hoxda gene cluster itself. The range of accessibility peaks was scored in common with the micro-dissected head, cloaca and tailbud samples at 30 hpf, with little conservation with the main mouse sequences (such as for the limbs; bottom line). The extraction of cell clusters identified as tailbud mesoderm (24 hpf), predicted cloaca (14 hpf) and pectoral fins at both 48 hpf and 72 hpf from a previous study³⁸ (Methods) revealed common and specific peaks (Supplementary Fig. 3). Notably, although the predicted cloaca and pectoral fin cell types shared a few accessibility peaks, the profiles were clearly different, with many more peaks in the former than in the latter.

Enhancer deletion in zebrafish

Because of the transgenic activity of CsB during both fish and mouse appendage development^18,35, the ambiguity between the ATAC peak in pectoral fin cells and the possibility of a variable effect of the 5DOM deletion on hoxd13a expression during pectoral fin development, we sought to remove the CsB element from the zebrafish genome and observe the effect on hoxd13a regulation. To have an efficient readout over developmental time and track potential expression changes with increased sensitivity, we generated an endogenous knock-in reporter line that expresses tdTomato from the hoxd13a locus (Extended Data Fig. 10a,b). We used CRISPR–Cas9 to delete CsB in cis to the hoxd13a^{Tg(hsp70:tdTomato)} endogenous reporter and assessed its expression in the developing cloaca, fins and tail using confocal microscopy. Fish carrying the CsB deletion did not exhibit any decrease in fin signal compared with siblings with an intact CsB (Extended Data Fig. 10c), demonstrating that, unlike in the mouse, the zebrafish CsB sequence did not have a critical role in the regulation of hoxd13a in the developing appendage. This result agrees with previous findings that the pufferfish CsB sequence was unable to drive distal transgene expression in mouse³⁵, indicating that the teleost CsB element either does not contain distal fin regulatory functionality or that a potential functionality cannot be revealed by our deletion approach. Expression was also compared between the cloacal and tailbud regions of CsB-deleted and CsB-intact reporter animals, and no major differences in reporter signal were scored. If anything, the CsB-deleted animals exhibited slightly increased activity in the cloacal region at the 19-somite stage (Extended Data Fig. 10c–e), an observation reminiscent of the complex regulatory interactions occurring within this landscape, as reported in the mouse counterpart during the outgrowth of external genitals²¹.

Ancestral cloacal regulatory landscape

The conservation for more than 350 million years of Hoxd13 regulation in the cloacal region by 5DOM-located sequences suggests an ancestral role of Hoxd13 in cloacal morphogenesis. This critical function was demonstrated in mice, in which the combined mutation of both Hoxa13 and Hoxd13 had a drastic effect on the development of the posterior part of the digestive and urogenital systems^28,39, causing an absence of any detectable UGS²⁸. Previous studies have revealed the expression of most hox13 genes in the intestine and cloacal regions of developing zebrafish (Supplementary Fig. 4). Additionally, hoxa genes are differentially regulated during patterning of the goby fish cloacal region⁴⁰. Although these results indicate functional conservation⁴¹, the requirement of hox13 genes for cloaca development in fishes remains to be established. Therefore, we examined whether zebrafish hox13 mutants exhibited cloacal phenotypes.

Wild-type zebrafish exhibited a pseudo-cloacal configuration, in which the hindgut and pronephric duct exited the trunk through separate but adjacent openings. The outlet of the hindgut was anterior to that of the pronephric duct, and a septum resided in between (Fig. 5a,e). Homozygous single mutants of hoxa13a, hoxa13b and hoxd13a were indistinguishable from the wild-type arrangement, as were animals triply heterozygous for these genes (Fig. 5b,f and Supplementary Fig. 5). However, the combined hoxa13a;hoxa13b double homozygous mutants exhibited connection between the hindgut and pronephric duct before exiting the body through a single opening (Fig. 5c,g and Supplementary Table 1). The loss of Hoxa13 paralogues also affected the pronephric duct and hindgut length at the level of the median fin fold (Supplementary Fig. 5). A more severe phenotype was observed in hoxa13a;hoxa13b;hoxd13a triple mutants, in which the septum was dysmorphic and the hindgut and pronephric duct were fused, resulting in a large shared lumen and outlet (Fig. 5d,h and Supplementary Table 1). These results revealed a conserved requirement of Hox13 function for the normal patterning of the termini of the digestive and urogenital systems across vertebrates.

Fig. 5: Loss of hox13 paralogues in the zebrafish resulted in defects in the cloacal region.

a–h, Confocal microscopy of phalloidin-labelled cloacal regions of wild-type and hox13 mutant zebrafish at 6 days post-fertilization shown in a single channel (a–d) and with pseudo-colouring (e–h). Pseudo-colouring indicates hindgut (blue), pronephric duct (yellow) or fused ducts (green). The phenotypes reported here were consistent across individuals collected from four independent crosses of triply heterozygous parents. a,e, Wild-type fish have adjacent but distinct openings for the hindgut (blue line) and pronephric duct (yellow line) (n = 10), as do hoxd13a mutants (n = 6) (b,f). c,g, hoxa13a;hoxa13b double mutants exhibited fusion of the hindgut and pronephric duct and a single opening (green line) (n = 4). d,h, hoxa13a;hoxa13b;hoxd13a triple mutants showed connection of the hindgut and pronephric duct to form a large shared lumen (green) with a single opening (green line) (n = 4). The consistent phenotypes observed in the double and triple mutant classes were significant when each class was individually compared with wild type using Fisher’s exact test (wild type, n = 10; mutant, n = 4; P = 9.9 × 10⁻⁴). Scale bar, 30 μm.

Discussion

Hox regulation and fin-to-limb transition

The expression and function of Hoxd and Hoxa genes have been central to hypotheses attempting to explain the evolutionary change from fins to limbs^5,11,13,14. By comparing their complex transcription patterns across actinopterygian, chondrichthyan and sarcopterygian fishes, various efforts have been made to relate the two types of paired appendages. These analyses have led to the conclusion that despite being composed of different types of skeletons, the development of actinopterygian fin rays and digits has a common regulatory architecture^11,42. Deletion of the two TADs flanking the zebrafish hoxda cluster has shown that the essential digit regulatory landscape characterized in tetrapods indeed has a structural counterpart in teleosts^43,44 (Extended Data Fig. 1). However, unlike in limbs, hoxda gene expression in distal fins is for the main part maintained in the absence of this regulatory landscape. Although a contribution from 5DOM to hoxd13a regulation probably exists¹², most of the regulatory control may reside within the gene cluster itself, probably in the vicinity of the hoxd13a, hoxd12a and hoxd11a genes, that is, the three genes sharing the same expression in postaxial cells¹⁶.

This observation confirmed the results obtained when assaying fish 5DOM conserved sequences as transgenes, either in zebrafish or mice^12,18. In addition, the lack of any strong effect of our deletion of the fish CsB, either in pectoral fins or in cloacal cells, supports the distinct functionalities of this enhancer sequence between mammals and fishes. CsB may thus play a role during zebrafish development that is only tangentially overlapping with that of its murine counterpart, in particular during differentiation of V0 interneurons, as driven by the gene Evx2 (ref. ²³) located nearby in 5DOM. This also explains why the zebrafish lnpa gene, which, like Evx2, is embedded into 5DOM, is not expressed in the emerging distal pectoral fin buds¹⁶, whereas the mouse counterpart has a strong distal expression owing to enhancer hijacking¹.

The presence of a functionally truncated distal appendage 5DOM regulatory landscape in teleosts may potentially illustrate an intermediate step in the full co-option of this regulation, as achieved in tetrapods (see “Co-options of a whole regulatory landscape”). Alternatively, it may reflect a secondary loss of distal enhancers either in the actinopterygian lineage or in the aftermath of TWGD because such an event may lead to the reorganization of paralogous regulatory landscapes. However, our phylogenetic analysis of sequence conservation within 5DOM showed that the latter option is unlikely for this specific locus, which does not maintain any paralogous copy in zebrafish. In fact, the structural organization of 5DOM in zebrafish is very close to that of non-teleost fishes, such as the gar or bowfin, with very minor changes matching phylogenetic distances. Therefore, our conclusions are considered to apply to fishes in general rather than to teleosts only.

Ancestral cloacal regulation

Zebrafish hoxa13a and hoxd13a, as well as other hox13 paralogues³⁷, are strongly expressed in and around the developing cloacal region^23,24. This is an area where the extremities of both the gastro-intestinal tract and the reproductive and urinary systems come together, although their openings remain separated, unlike in some chondrichthyan fishes or other vertebrates, in which the tubes coalesce into a single opening (such as in sharks or birds). This pseudo-cloacal structure is disrupted in zebrafish carrying hox13 mutant alleles, with an abnormal fusion between the intestinal and pronephric openings, thus giving rise to a single, abnormal, cloacal opening. Likewise, the developing murine UGS expresses Hoxd13 (ref. ²⁷), and double mouse Hoxd13–Hoxa13 mutant animals have severely malformed posterior regions^28,39, with no distinguishable UGS²⁸, illustrating that the evolutionary conservation of this regulatory landscape is accompanied by shared functional effects.

We also documented that, as with zebrafish, the control of murine posterior Hoxd genes in the cloaca is achieved by enhancers located within the 5′ located regulatory landscape, that is, in the same genomic region that regulates expression in both digits and external genitals. In mice, several 5DOM enhancers are somewhat versatile, such as the GT2 sequence, which is both UGS and genitalia-specific³³, whereas CsB is UGS and digit-specific¹. Other enhancer sequences, however, seem to have kept a unique specificity, such as ‘Island 2’, the strongest Hox digit enhancer identified thus far⁴⁵, which is located in a different area of 5DOM² and absent in zebrafish¹². These observations illustrate the ‘functional adaptation’ of enhancers, which could be facilitated by spatial proximity within the same large chromatin domain, thus triggering the sharing of upstream factors⁴⁶. Finally, all the regulatory specificities encoded in this 5DOM landscape control the same subset of posterior Hoxd genes in tetrapods (from Hoxd13 to Hoxd10), suggesting that although groups of enhancers can be reused for new tissue types, there is a constraint on which genes they can target.

Co-options of a whole regulatory landscape

In vertebrates, Hox13 genes are located within a TAD that is distinct from the one containing more anterior Hox genes and their regulations^15,47. This condition prevents Hox13 from being activated too early and hence too anteriorly in the body axis, which is detrimental to the embryo owing to the potent posteriorizing function of these proteins⁴⁸. As a result, Hoxd13 was probably the main target gene that triggered and stabilized the various evolutionary co-options of 5DOM regulations because of its location within the 5DOM TAD and through its function to organize posterior or distal body parts together with its Hoxa13 paralogue^7,28,39. Our results indicate that the initial functional specificity of this regulatory landscape was to organize a cloacal region, which is the posterior part of the intestinal and urogenital systems (Fig. 6). This conclusion is supported by the documented expression of Hox13 paralogues in the cloacal regions of paddlefish^41,49, cat shark^49,50 and lampreys⁵¹, suggesting that this pattern is characteristic of the common ancestor of craniates. Altogether, our results indicate that the regulatory landscape involved in the evolution of genitalia and limbs first arose to drive the formation of the cloaca and was co-opted, supporting the development and diversification of genital and digital morphologies.

Fig. 6: Evolutionary co-option of the HoxD 5DOM regulatory landscape.

Schematic representation of posterior Hoxd gene regulation by the 5DOM regulatory landscape (top left) and (at least) three developmental contexts in which this landscape is functional (top right). On the left are shown the phylogenetic relationships between taxa where distal fins, distal limbs and external genitals emerged, whereas on the right, the corresponding 5DOM regulatory contributions to these structures are indicated; ‘0’ denotes the absence of any given structure. In this hypothetical view, the 5DOM cloacal regulation is an ancestral feature. In actinopterygian fishes, 5DOM lightly contributes to hoxda gene regulation in the postaxial and distal territories of paired fin buds. The regulatory importance of 5DOM in distal fin territories has increased in sarcopterygian fishes. In amniotes, the 5DOM contribution expands to take over the entire regulation of posterior Hoxd genes in digits, as suggested by many enhancers with mixed specificities. Similarly, a distinct yet overlapping set of 5DOM-located enhancers entirely control Hoxd gene expression in the genital tubercle. It is difficult to infer the temporal sequence of the latter two co-options of this regulatory landscape. However, because genitalia are late amniote specializations, it is conceivable that elaboration of digital character arose initially, a sequence also supported by the appearance of digits in sarcopterygian fishes. A second co-option of this multifunctional regulatory landscape might have occurred along with the evolution of external genitals, facilitated both by the developmental proximity between external genitals and the embryonic cloacal region where posterior Hox genes are initially expressed⁵², and by the tight developmental relationships between amniote limbs and genitals^27,52.

Methods

Animal husbandry and ethics

All experiments using mice were approved and performed in compliance with the Swiss Law on Animal Protection (Loi fédérale sur la Protection des Animaux) under licence numbers GE45/20 and GE81/14. All animals were kept as a continuous backcross with C57BL6 × CBA F₁ hybrids. The mice were housed at the University of Geneva Sciences III animal colony, with light cycles between 07:00 and 19:00 in the summer and 06:00 and 18:00 in winter. Temperatures were maintained between 22 °C and 23 °C, with humidity levels between 45% and 55%. The air was renewed 17 times per hour. Zebrafish (Danio rerio) were maintained according to standard conditions⁵³ under a 14 h/10 h on/off light cycle at 26 °C, with set points of 7.5 and 600 μS for pH and conductivity, respectively. All zebrafish husbandry procedures were approved and accredited either by the Federal Food Safety and Veterinary Office of the canton of Vaud, Switzerland (no. VD-H23), by the animal committees of Rutgers University under protocol no. 201702646 or under the guidance of the Institutional Animal Care and Use Committee (IACUC) of Boston Children’s Hospital. AB, Tu and TL were used as wild-type strains and were obtained from the European Zebrafish Resource Center. The hoxda^Del(3DOM) and hoxda^Del(5DOM) mutants were generated for this study. Zebrafish embryos were derived from freely mating adults. Wild-type sibling hoxda^Del(3DOM) and hoxda^Del(5DOM) homozygous embryos were obtained by crossing the corresponding heterozygous mutant. Embryos were collected within 30 min after spawning and incubated at 28.5 °C in fish water, shifted to 20 °C after reaching 80% epiboly and grown at 28.5 °C to the proper developmental stage according to a previous study⁵⁴. Pigmentation was prevented by treating the embryos with 0.002% N-phenylthiourea from 1 day post-fertilization (dpf) onwards. Sex was determined for animals used in the E18.5 UGS mouse experiments. Animals in other mouse experiments and in zebrafish experiments were not sexed. The sample size was not predetermined by a statistical test. Randomization and blinding were not conducted because the mutant and control animals were processed together in the same batch and grouped on the basis of their genotypes.

Generation of deletions in zebrafish

The hoxda^Del(3DOM) and hoxda^Del(5DOM) mutant alleles were generated using the CRISPR–Cas9 system described in a previous study⁵⁵. The sequences of the CRISPR RNAs (crRNAs) used are listed in Supplementary Table 2. Loci were identified using the GRCz11 zebrafish genome assembly available on Ensembl. The corresponding genomic regions were amplified and sequenced from fin clips. Adults carrying verified target sequences were isolated and then selected for breeding to generate eggs for genome editing experiments. The guide RNA target sites were determined using the open-source software CHOPCHOP (http://chopchop.cbu.uib.no/index.php). Chemically synthesized Alt-R crRNAs and Alt-R trans-activating CRISPR RNAs (tracrRNAs) and the Alt-R Cas9 protein were obtained from Integrated DNA Technologies (IDT). To test the efficiency of these guide RNAs in generating the expected mutant alleles, we injected boluses ranging from 100 µm to 150 µm and containing 5 μM of the duplex crRNAs, tracrRNA and Cas9 ribonucleoprotein complex into the cytoplasm of one-cell-stage embryos. Injecting the ribonucleoprotein complex solution in a 100-µm bolus gave less than 5% mortality. With this condition, 30% of the embryos carried the 5DOM deletion and 15% carried the 3DOM deletion. For each condition, we extracted the genomic DNA of 20 individual larvae at 24 hpf for genotyping⁵⁶. Identification of hoxda^Del(3DOM) and hoxda^Del(5DOM) mutants was performed using polymerase chain reaction (PCR). Amplification of evx2 was used as a control to confirm the presence or absence of 5DOM. The PCR mix was prepared using Phusion High-Fidelity DNA Polymerase (New England Biolabs), and primer sequences are listed in Supplementary Table 2. In parallel, 120 larvae per allele were raised to adulthood. To identify founders, F₀ adults were outcrossed with wild type and 25 embryos were genotyped. Three and four independent founders were obtained for the hoxda^Del(5DOM) allele and hoxda^Del(3DOM), respectively. Two founders of each deletion were verified by Sanger sequencing (Supplementary Data 1) and used for further experiments.

Generation of knock-in reporter line

The endogenous hoxd13a reporter line (hoxd13a^{Tg(hsp70:tdTomato)}) was produced using a CRISPR–Cas9-mediated Gbait vector knock-in approach^57,58. A guide targeting the coding region of exon 1 of hoxd13a (hoxd13a_KI_crRNA) was co-injected with a Gbait vector targeting guide (GFP_crRNA) and Gbait:hsp70l:tdTomato plasmid¹⁷. The injected embryos were screened for endogenous reporter RFP signal in expected hoxd13a expression domains, and positive individuals were raised to adulthood to outcross and recover F₁ germline founders. To verify vector insertion and orientation in founders, genomic primers (hoxd13a_KI_F and hoxd13a_KI_R) were each paired with primers internal to the insert (LacZ_F and hsp70_R) for PCR and Sanger sequencing. The vector was oriented in the reverse direction relative to the endogenous promoter in the hoxd13a^{Tg(hsp70:tdTomato)} line, but reporter expression matched previously published in situ hybridization data and an hoxd13a knock-in line (hoxd13a^egfp) generated independently by another research group⁵⁹. Genotyping primers are listed in Supplementary Table 2.

Removal of CsB in cis to hoxd13a ^{Tg(hsp70:tdTomato)}

To delete the CsB sequence from the chromosome carrying the hoxd13a^{Tg(hsp70:tdTomato)} endogenous reporter, each individual crRNA was duplexed with tracrRNA and injected at a final concentration of 6.25 μM with 1 μg Alt-R S.p. Cas9 Nuclease V3 (IDT). To estimate guide efficiency, DNA was extracted from four pools of three embryos each from 12 injected embryos and 12 control siblings and analysed using the T7 endonuclease 1 mismatch detection assay⁶⁰. Embryos injected with efficient guides were raised to adulthood to outcross and identify founders. Guides flanking the CsB region (CsB_g1_crRNA and CsB_g2_crRNA) were injected into the hoxd13a^{Tg(hsp70:tdTomato)} background. The injected embryos were sorted by RFP signal at 1 dpf, and 16 positive animals from each clutch were screened for CsB removal using PCR with deletion-spanning primers (CsB_g1_F and CsB_g2_R) that did not amplify the intact locus under short elongation conditions. Clutches exhibiting a high frequency of CsB removal were raised to adulthood, and individuals were outcrossed to T5D wild type to obtain embryos carrying hoxd13a^{Tg(hsp70:tdTomato)}-Del(CsB) chromosomes. To identify CsB deletions in cis to the reporter, outcrossed embryos were sorted for RFP and then genotyped for the CsB deletion. One F₀ injected parent (purple male 3) produced gametes with hoxd13a^{Tg(hsp70:tdTomato)}-Del(CsB) chromosomes at high frequency (approximately 25%), as well as gametes in which the CsB in cis to the reporter was left intact. Sanger sequencing of the deletion-spanning PCR product from 16 embryos revealed that each hoxd13a^{Tg(hsp70:tdTomato)}-Del(CsB) chromosome carried an identical deletion, suggesting clonality. Embryos resulting from outcrosses of this injected individual (purple male 3) were used in a subsequent expression analysis. The sequences of the crRNAs and genotyping primers used are listed in Supplementary Table 2. Sanger sequences of zebrafish founders are listed in Supplementary Data 1.

Quantification of hoxd13a ^{Tg(hsp70:tdTomato)} expression

Outcrossed progeny with RFP signal from the endogenous reporter were collected at the 19-somite stage and at 72 hpf. Embryos were fixed in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) (pH 7.4) for 2 h at room temperature with agitation in light-blocking containers, rinsed two times for 10 min each in PBS with 0.01% Tween 20 (PBST) and then incubated overnight at 4 °C in PBST with DAPI. The next day, the embryos were washed twice for 30 min each with PBST and processed for genotyping, as described above for the analysis of cloacal morphology of hox13 mutants, except that the head was removed for DNA extraction and the fins and trunk were retained for analysis. The embryos were imaged on a Zeiss LSM 800 confocal microscope to analyse hoxd13a^{Tg(hsp70:tdTomato)} expression. The laser and filter settings were optimized individually for each stage and tissue type to be compared, and then these settings were kept constant across CsB-intact and CsB-deleted individuals. For the 19-somite stage, the cloaca and tailbud were imaged simultaneously as a single piece of trunk, but for the 72-hpf animals, the fins, cloaca and tail were dissected and imaged separately. Maximum projection images were produced from each scan and then exported as TIF files for analysis in ImageJ⁶¹. Each image was cropped to a specific region of interest (ROI) containing the specific expression domain, and ImageJ was used to measure the mean grey value for pixels in the region. The ROI size for each tissue was as follows: 75 μm × 75 μm for cloaca (19-somite and 72 hpf), 250 μm × 250 μm for 19-somite tailbud, 200 μm × 200 μm for 72-hpf tails and 100 μm × 250 μm for 72-hpf pectoral fins. For each tissue, the average mean grey value was calculated from CsB-intact individuals and used to normalize signal intensity values so that the average CsB-intact intensity for each tissue was equal to 1. The average relative intensities for CsB-intact and CsB-deleted tissues were compared using Welch’s t-test in R (ref. ⁶²).

Zebrafish hox13 mutant lines

Frameshift loss-of-function alleles hoxa13a^ch307, hoxa13b^ch308 and hoxd13a^5bpins were previously generated¹¹. The zebrafish lines were propagated and maintained, as described in a previous study⁶³. To generate compound hox13 mutants, animals that were triple heterozygous for hoxa13a, hoxa13b and hoxd13a were intercrossed. The resulting larvae were fixed at 6 dpf in 4% PFA in PBS for 2 h at room temperature, with rocking agitation. After fixation, the larvae were rinsed twice for 5 min each in PBS with added 1% Triton X-100 (PBSX). To visualize the cloacal anatomy by labelling filamentous actin, the larvae were then incubated in PBSX with fluorophore-conjugated phalloidin (Sigma-Aldrich P1951; phalloidin-tetramethylrhodamine B isothiocyanate) added to a final concentration of 5 U ml⁻¹ overnight at 4 °C, with rocking agitation. The larvae were then rinsed twice with PBSX for 1 h each.

For genotyping, the phalloidin-labelled larvae were cut in half, separating the head, yolk and pectoral fins from the cloaca and tail. The head half was used for genotyping, and the tail half was stored at 4 °C for later analysis. DNA was extracted from the head half by digesting tissue in proteinase K diluted to 1 mg ml⁻¹ in 20 μl of 1× PCR buffer (10 mM Tris-HCl, 50 mM KCl and 1.5 mM MgCl₂) for 1 h at 55 °C, followed by heat inactivation at 80 °C for 20 min. The digested tissue was then subjected to brief vortexing, and then 1 μl was used directly as template for genotyping PCR, with primers listed in Supplementary Table 2. For thermocycling, after an initial step at 94 °C for 2 min, reactions were cycled 40 times (15 s at 94 °C, 15 s at 58 °C and 20 s at 72 °C) and finished with 5 min at 72 °C. The PCR products were then heteroduplexed on a thermocycler by heating to 95 °C for 10 min and then gradually cooled by 1 °C every 10 s until a final temperature of 4 °C was reached. Heteroduplexed PCR amplicons were then run on a high-percentage agarose gel to determine the genotype by product size.

To analyse cloacal morphology, fixed phalloidin-labelled tails were imaged using a Zeiss LSM 800 confocal microscope. After acquiring a full confocal stack through the cloacal region, a midline frame that demonstrated the hindgut and pronephric duct morphology was selected. In a separate set of quantifications, juveniles were photographed using a Leica M205 FCA stereotype microscope, PLANAPO 1.0× zoom lens and Leica MC170 HD camera. Using the pencil tool in Illustrator, we traced the internal lumen of the hindgut and pronephric duct complex from the level of the proximal end of the median fin fold to the terminal exit. A perpendicular line was then drawn to measure the width of the complex. The lengths of these lines were measured using Illustrator and were used for statistical analysis.

Mutant mouse stocks

The following mouse lines used in this study were previously reported: Inv(Itga6-nsi)d11lac³¹, Inv(Itga6–attP) and tgBAC(HoxD)³⁰, Del(HoxD)³² and Del(Atf2–SB1), Del(SB1–Rel5) and Del(Rel5–Rel1)².

Whole-mount in situ hybridization

The zebrafish and mouse antisense probes used in this study are listed in Supplementary Data 2 and 3, respectively. For zebrafish, WISH was performed, as described⁵⁶, at 58 °C for all riboprobes (hybridization temperature and saline–sodium citrate washes). Whole-mount embryos were photographed using a compound microscope (SZX10; Olympus) equipped with a Nomarski optics and a digital camera (DP22; Olympus). Genotyping of individual embryos was performed after photographic documentation using the primers listed in Supplementary Table 2. Wild-type and mutant embryos originated from the same clutch of eggs produced by heterozygote crosses and underwent WISH in the same well. Details on the number of embryos per experiment and genotype are provided in Supplementary Table 3. Murine urogenital systems were isolated from E18.5 embryos and processed following a previously reported WISH procedure⁶⁴, with some specific adjustments. For proteinase K treatment, urogenital systems were incubated for 20 min in proteinase K diluted to 20 µg ml⁻¹ in PBST. For the refixation step, a solution of 4% PFA containing 0.2% glutaraldehyde was used. The hybridization temperature was 69 °C, and the temperature of the post-hybridization washes was 65 °C. Staining was performed using BM-Purple (Roche; 11442074001) for approximately 4 h at room temperature.

Hybridization chain reaction

HCR in situ hybridization was performed, as previously described⁶⁵, with slight modifications. Embryos were fixed in 4% PFA in PBS at 4 °C overnight with rocking, washed three times for 5 min in PBS with 0.1% Tween (PBST) and then dehydrated in methanol washes (25%, 50% and 75% in PBST) for 3 min each, followed by five 10-min washes and one 50-min wash in 100% methanol. The embryos were stored at −20 °C in methanol for at least 48 h before starting the hybridization protocol. The embryos were rehydrated in methanol (75%, 50% and 25% methanol in PBST), washed twice with PBST and pre-hybridized in hybridization buffer (Molecular Instruments) at 37 °C for at least 1 h. The embryos were then incubated in 200 μl of a hybridization solution with hoxd13a probes (IDT oPools; Supplementary Table 4) at a concentration of approximately 65 nM each overnight at 37 °C. After 18–24 h in probe solution, the embryos were washed four times for 15 min each using a probe wash buffer (Molecular Instruments) at 37 °C. The embryos were then washed twice for 5 min each at room temperature with 5× SCCT on a rocker before incubation in amplification buffer (Molecular Instruments) for at least 1 h. The amplification solution with B2 546 amplifiers (Molecular Instruments) was prepared by heating 3 μl of hairpin 1 (3 μM) and 3 μl of hairpin 2 (3 μM) to 95 °C for 90 s, followed by snap-cooling. After 30 min, hairpins 1 and 2 were mixed and added to 200 μl of amplification buffer. The embryos were incubated in amplification solution overnight at room temperature on a rocker. After 18–24 h of incubation in amplification solution, the embryos were washed at least four times for 30 min each with 5× SCCT at room temperature on a rocker. The embryos were stored at 4 °C in 5× SCCT for 1 day until they were genotyped and mounted for confocal microscopy. Before genotyping, the embryos were washed in PBST with DAPI for 1 h. DNA was extracted from the dissected head of each embryo. The pectoral fins were then microdissected using tungsten needles and mounted in PBST for confocal imaging using an inverted Zeiss LSM 800. Wild-type fins were used to optimize the laser and filter settings, which were maintained across all samples during data collection. After image acquisition, post-processing was performed on the maximum-intensity projections of each sample to reduce non-specific background signals. Specifically, the black value was changed from 0 to 50 uniformly for each image using the Zeiss Zen imaging software. These scans were then exported as TIF files for analysis in ImageJ⁶¹. The images were cropped to an ROI of 180 μm × 120 μm in size containing the hoxd13a fin expression domain. ImageJ was used to measure the mean grey value of ROI from each fin, and the average mean grey value was calculated from the wild-type fins. This average was used to normalize the signal intensity values such that wild-type fins had an average value of 1. The normalized relative intensities of wild-type and 5DOM deletion mutant fins were then compared using Welch’s t-test in R⁶².

Mouse genotyping

For extemporaneous genotyping, yolk sacs were collected and placed into 1.5-ml tubes containing rapid digestion buffer (10 mM EDTA (pH 8.0) and 0.1 mM NaOH) and then placed in a thermomixer at 95 °C for 10 min with shaking at 900 rpm. While the yolk sacs were incubating, the PCR master mix was prepared using Z-Taq (Takara; R006B) and primers (Supplementary Table 2) and aliquoted into PCR tubes. The tubes containing lysed yolk sacs were then placed on ice to cool briefly and quickly centrifuged at a high speed. The lysate (1 μl) was placed in the reaction tubes and cycled 32 times (2 s at 98 °C, 2 s at 55 °C and 15 s at 72 °C). The PCR reaction (20 μl) was loaded onto a 1.5% agarose gel, and electrophoresis was run at 120 V for 10 min. When samples could be kept for some time, a conventional genotyping protocol was applied using tail digestion buffer (10 mM Tris (pH 8.0), 25 mM EDTA (pH 8.0), 100 mM NaCl and 0.5% SDS) added to each yolk sac or tail clipping at 250 μl along with 4 μl of proteinase K at 20 mg ml⁻¹ (Eurobio; GEXPRK01-15) and incubated overnight at 55 °C. The samples were incubated at 95 °C for 15 min to inactivate the proteinase K and stored at −20 °C until ready for genotyping. Genotyping primers (Supplementary Table 2) were combined with Taq polymerase (ProSpec; ENZ-308) in 25-μl reactions, cycled twice with Ta = 64 °C and then cycled 32 times with Ta = 62 °C.

Mouse RT–qPCR

UGSs were collected from E18.5 male embryos separately and placed in 1× diethyl pyrocarbonate–PBS on ice. A small portion of the remaining embryo was collected for genotyping. The UGSs were transferred into fresh 1× diethyl pyrocarbonate–PBS and then placed into RNAlater (Thermo Fisher Scientific; AM7020) for storage at −80 °C until processing. Batches of samples were processed in parallel to collect RNA using RNeasy extraction kits (QIAGEN; 74034). After isolating total RNA, first-strand complementary DNA (cDNA) was produced with SuperScript III VILO (Thermo Fisher Scientific; 11754-050) using approximately 500 ng of total RNA input. The cDNA was amplified with Promega GoTaq 2X SYBR Mix and quantified on a Bio-Rad CFX96 Real-Time System. Expression levels were determined by the difference between the cycle threshold (Ct) of the gene of interest (GOI) and the reference gene Tbp, calculated as dCt = Ct(GOI) − Ct(Tbp). They were normalized to 1 for each condition by subtracting each dCT from the mean dCT for each wild-type set. Finally, expression was evaluated by the power 2 minus this normalized dCT. Supplementary Table 2 contains the primer sequences used for quantification. RT–qPCR measurements were taken from distinct embryos. Box plots for expression changes and two-tailed unequal variance t-tests were produced in DataGraph 4.6.1. The boxes represent the IQR, with the lower and upper hinges denoting the first and third quartiles (25th and 75th percentiles). Whiskers extend from the hinges to the furthest data points within 1.5 times the IQR. The upper whisker reaches the largest value within this range, whereas the lower whisker extends to the smallest value within 1.5 times the IQR from the hinge.

Mouse RNA-seq

E18.5 male and female UGSs were collected by means of dissection separating the bladder from the UGS, including the proximal urethra in males and the vagina in females. Tissues were stored in RNAlater (Thermo Fisher Scientific; AM7020) and processed in parallel using RNeasy extraction kits (QIAGEN; 74034). RNA quality was assessed using an Agilent Bioanalyzer 2100 with RNA integrity number scores greater than 9.5. RNA sequencing libraries were prepared at the University of Geneva Genomics Platform using Illumina TruSeq Stranded Total RNA with Ribo-Zero Gold Ribo-deleted RNA kits to produce strand-specific 100-bp single-end reads on an Illumina HiSeq 2000. Raw RNA-seq reads were processed with Cutadapt v.4.1 (-a GATCGGAAGAGCACACGTCTGAACTCCAGTCAC -q 30 -m 15)⁶⁶ to remove TruSeq adapters and bad-quality bases. Filtered reads were mapped to the mouse genome mm39 using STAR v.2.7.10a⁶⁷ using ENCODE parameters with a custom gtf file⁶⁸ on the basis of Ensembl version 108. This custom GTF file was obtained by removing readthrough transcripts and all non-coding transcripts from a protein-coding gene. Fragments per kilobase of transcript per million mapped read values were evaluated using Cufflinks v.2.2.1 (refs. ^69,70) with the options --max-bundle-length 10000000 --multiread-correct --library-type ‘fr-firststrand’ -b mm10.fa --no-effective-length-correction -M MTmouse.gtf -G. Box plots depicting expression levels in distinct embryos were generated using the same methodology as that used for RT–qPCR.

ATAC-seq

Mouse and fish tissues were isolated and placed into 1× PBS containing 10% fetal calf serum on ice. Collagenase (Sigma-Aldrich; C9697) was added to 50 μg ml⁻¹ and incubated at 37 °C for 20 min with shaking at 900 rpm. Cells were washed three times in 1× PBS. The number of cells was counted, and viability was confirmed to be greater than 90%. An input of 50,000 cells was processed according to a previous description³⁶. Sequencing was performed on École Polytechnique Fédérale de Lausanne (EPFL) Gene Expression Core Facility (GECF) using an Illumina NextSeq 500. We analysed in a manner similar to a previous study⁷¹. Raw ATAC-seq paired-end reads were processed with Cutadapt v.4.1 (-a CTGTCTCTTATACACATCTCCGAGCCCACGAGAC -A CTGTCTCTTATACACATCTGACGCTGCCGACGA -q 30 -m 15)⁶⁶ to remove Nextera adapters and bad-quality bases. Filtered reads were mapped on mm39 for mouse samples and danRer11 in which alternative contigs were removed for fish samples using Bowtie 2 v.2.4.5 (ref. ⁷²) with the following parameters: --very-sensitive --no-unal --no-mixed --no-discordant --dovetail -X 1000. Only pairs mapping concordantly outside of mitochondria were kept (Samtools v.1.16.1) (ref. ⁷³). The PCR duplicates were removed using Picard v.3.0.0 (http://broadinstitute.github.io/picard/index.html). The BAM files were converted to BED using bedtools v.2.30.0 (ref. ⁷⁴). Peaks were called, and coverage was generated by MACS2 v.2.2.7.1 with --nomodel --keep-dup all --shift -100 --extsize 200 --call-summits -B. Coverages were normalized to million mapped reads.

ChIP–seq

Male UGSs were isolated and placed into 1× PBS containing 10% fetal calf serum on ice. ChIP–seq experiments were performed, as previously described⁷⁵. Briefly, they were fixed for 10 min in 1% formaldehyde at room temperature, and the crosslinking reaction was quenched with glycine. Subsequently, nuclei were extracted, and chromatin was sheared using a water-bath sonicator (Covaris E220evolution ultrasonicator). Immunoprecipitation was performed using the following anti-H3K27ac (Abcam; ab4729) or anti-H3K27me3 (Merck Millipore; 07–449). Libraries were prepared using the TruSeq protocol and sequenced on an Illumina HiSeq 4000 (100-bp single-end reads) according to the manufacturer’s instructions. CTCF was reanalysed using datasets from previous studies^43,71. The accession numbers are listed in Supplementary Table 5. Raw ChIP–seq single-end or paired-end reads were processed using Cutadapt v.4.1 (-a GATCGGAAGAGCACACGTCTGAACTCCAGTCAC for single-end reads and -a CTGTCTCTTATACACATCTCCGAGCCCACGAGAC -A CTGTCTCTTATACACATCTGACGCTGCCGACGA -q 30 -m 15)⁶⁶ to remove TruSeq or Nextera adapters and bad-quality bases. Filtered reads were mapped on mm39 for mouse samples and danRer11 in which alternative contigs were removed for reanalysis of fish samples using Bowtie 2 v.2.4.5 (ref. ⁷²) with the default parameters. Only alignments with a mapping quality above 30 were kept (Samtools v.1.16.1)⁷³. Peaks were called, and coverage was generated by MACS2 v.2.2.7.1 with --call-summits -B (and --nomodel --extsize 200 for single-end reads). Coverages were normalized to million mapped reads/pairs.

Mouse enhancer–reporter assay

Transgenic embryos were generated, as described³³. Primers were designed to amplify genomic DNA from the region around the observed ATAC and H3K27Ac peaks (Supplementary Table 5). These primers included extra restriction sites for either XhoI or SalI at the 5′ ends. The PCR fragments were cleaned using a QIAGEN Gel Extraction Kit (28704). The PCR fragment and the pSKlacZ reporter construct (GenBank X52326.1)⁷⁵ were digested with XhoI or SalI and ligated together using the Promega 2X Rapid Ligation kit (C6711). Sanger sequencing confirmed that the correct sequences were inserted upstream of the promoter. Maxipreps of the plasmid were prepared and eluted in 1× IDTE (11-05-01-13). Pro-nuclear injections were performed, and embryos were collected at approximately E18.5 and stained for lacZ. UGSs were collected from E18.5 embryos in ice-cold 1× PBS in a 12-well plate. All steps were performed with gentle shaking on a rocker plate at room temperature. Tissues were fixed for 5 min at room temperature in freshly prepared 4% PFA. After fixing, the tissues were washed three times in 2 mM MgCl₂, 0.01% sodium deoxycholate, 0.02% Nonidet P-40 and 1× PBS for 20 min at room temperature. The wash solution was replaced with β-galactosidase staining solution (5 mM potassium ferricyanide, 5 mM potassium ferrocyanide, 2 mM MgCl₂ hexahydrate, 0.01% sodium deoxycholate, 0.02% Nonidet P-40, 1 mg ml⁻¹ of β-galactosidase and 1× PBS) for overnight incubation with the plate wrapped in aluminium foil to protect from light. The tissues were then washed three times in 1× PBS and fixed in 4% PFA for long-term storage. Images of embryos were collected using an Olympus DP74 camera mounted on an Olympus MVX10 microscope using Olympus cellSens Standard 2.1 software.

Mouse capture Hi-C sequencing

E18.5 male UGSs were collected, and collagenase-treated samples were crosslinked with 1% formaldehyde (Thermo Fisher Scientific; 28908) for 10 min at room temperature and stored at −80 °C until further processing, as previously described⁷⁶. The SureSelectXT RNA probe design used for capturing DNA was performed using the SureDesign online tool by Agilent. Probes cover the region chr. 2: 72240000–76840000 (mm9) producing twice the coverage, with moderately stringent masking and balanced boosting. DNA fragments were sequenced on an Illumina HiSeq 4000 and processed with HiCUP v.0.9.2 on mm39 with --re1 ^GATC⁷⁷, Bowtie 2 v.2.4.5 (ref. ⁷²) and Samtools v.1.16.1 (ref. ⁷³). The output BAM was converted to a pre-juicer medium format with hic2juicer from HiCUP. The pairs with both mates on chr. 2: 72233000–76832000 were selected, sorted and loaded into a 10-kb bin matrix with cooler v.0.8.11 (ref. ⁷⁸). The final matrix was balanced with the option --cis-only. TADs were computed using HiCExplorer hicFindTADs v.3.7.2 (refs. ^79,80) with --correctForMultipleTesting fdr --minDepth 120000 --maxDepth 240000 --step 240000 --minBoundaryDistance 250000. Data were plotted on mm39 (chr. 2: 73600000–75550000).

Zebrafish Hi-C sequencing

The HiC profiles were derived from a reanalysis of data from previous studies^43,81. The accession numbers are listed in Supplementary Table 5. Reads were mapped on danRer11 in which alternative contigs were removed, and no selection of reads were performed. Valid pairs were loaded into a 10-kb bins matrix. TAD calling parameters were adapted to the smaller size of the genome: --chromosomes "chr9" --correctForMultipleTesting fdr --minDepth 35000 --maxDepth 70000 --step 70000 --minBoundaryDistance 50000. Data were plotted on danRer11 (chr. 9: 1650000–2400000) and on an inverted x axis.

CUT&RUN

Zebrafish samples were processed using a final concentration of 0.02% digitonin (Apollo; APOBID3301). Approximately 0.5 × 10⁶ cells were incubated with 0.1 μg (100 μl)⁻¹ of anti-H3K27ac antibody (Abcam; Ab4729) or 0.5 μg (100 μl)⁻¹ of anti-H3K27me3 (Merck Millipore; 07-449) in digitonin wash buffer at 4 °C. The protein A–micrococcal nuclease was kindly provided by the Henikoff Lab (batch 6) and added at 0.5 μl (100 μl)⁻¹ in digitonin wash buffer. Cells were digested in high-calcium buffer and released for 30 min at 37 °C. Sequencing libraries were prepared with KAPA HyperPrep reagents (07962347001) with 2.5 μl of adapters at 0.3 μM and ligated for 1 h at 20 °C. The DNA was amplified for 14 cycles. Post-amplified DNA was cleaned and size selected using 1:1 ratio of DNA:AMPure SPRI beads (A63881) followed by an extra 1:1 wash and size selection with HXB. HXB is equal parts 40% polyethylene glycol 8,000 (Thermo Fisher Scientific; FIBBP233) and 5 M NaCl. Sequencing was performed at EPFL GECF on an Illumina HiSeq 4000. Raw CUT&RUN paired-end reads were processed with Cutadapt v.4.1 (-a GATCGGAAGAGCACACGTCTGAACTCCAGTCAC -A GATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT -q 30 -m 15) to remove TruSeq adapters and bad-quality bases⁶⁶. Filtered reads were mapped on danRer11, in which alternative contigs were removed with Bowtie 2 v.2.4.5 (ref. ⁷²) with the following parameters: --very-sensitive --no-unal --no-mixed --no-discordant --dovetail -X 1000. Only alignments with mapping quality above 30 were kept (Samtools v1.16.1) (ref. ⁷³). PCR duplicates were removed by Picard v.3.0.0 (http://broadinstitute.github.io/picard/index.html). BAM files were converted to BED with bedtools v.2.30.0 (ref. ⁷⁴). Peaks were called, and coverage was generated by MACS2 v.2.2.7.1 with --nomodel --keep-dup all --shift -100 --extsize 200 --call-summits -B. Coverages were normalized to million mapped reads.

Analyses of conserved sequences

Annotation of orthologous domains was performed using transcription start sites of orthologous genes, as reported in Supplementary Table 6. To identify conserved sequences between mouse and zebrafish, a pairwise alignment was done between the mouse genomic region chr. 2: 73600000–75550000 (mm39) and the zebrafish orthologous region chr. 9: 1650000–2400000 (danRer11) using discontinuous megablast. To reduce false positives, only reciprocal hits were considered. To display multispecies conservation levels, multiple alignment format files were generated between chr. 2 of the mouse genome (mm39) and contig chrUn_DS181389v1 of the platypus genome (ornAna2), chr. 7 of the chicken genome (galGal6), contig chrUn_GL343356 of the lizard genome (anoCar2), chr. 9 of the frog genome (xenTro10), contig JH127184 of the coelacanth genome (latCha1), chr. 9 of the zebrafish genome (danRer11), chr. 1 of the fugu genome (fr3) and the whole lamprey genome (petMar3). Details for the multiple alignment format generation are available on the GitHub repository (https://github.com/AurelieHintermann/HintermannBoltHawkinsEtAl2025; ref. ⁸²). To facilitate visualization, a horizontal line was plotted for each species on each region.

Whole-genome alignments

Whole-genome alignments were performed using Progressive Cactus v.2.6.7 (ref. ⁸³). The cactus command was used with default parameters to obtain the hierarchical alignment format. The hierarchical alignment was then projected on either zebrafish chr. 9 or mouse chr. 2 with cactus-hal2maf⁸⁴ using --chunkSize 500000 and --noAncestor. The genome assemblies are listed in Supplementary Table 6.

Single-cell assay for transposase-accessible chromatin sequencing

The single-cell assay for transposase-accessible chromatin sequencing (scATAC-seq) bigwig files were downloaded from a previous study³⁷ (Gene Expression Omnibus (GEO) GSE243256) where annotations were available. To annotate the cells from the cloaca, the raw matrix of single-cell RNA sequencing (scRNA-seq) from a previous study³⁸ was downloaded from GEO (GSE223922) and stored in a Seurat object. Only 12,424 cells obtained at 14 hpf were kept. The data were normalized, and 3,000 variable features were extracted. Data were scaled. Uniform manifold approximation and projection (UMAP) and t-distributed stochastic neighbour embedding projections were calculated using the first 50 principal components. In parallel, scATAC-seq fragments of cells corresponding to 14 hpf were extracted from the general fragment file provided in a previous study³⁸ on GEO (GSE243256). A new ArchR gene annotation was generated using the Lawson gtf v.4.3.2 (ref. ⁸⁵) to match the scRNA-seq data from a previous study³⁷, and the selected fragments were loaded into an ArchRProject with this genome. Iterative latent semantic indexing was computed with COR-Cut-off of 0.5. The clustering of scRNA-seq was then transferred to scATAC-seq using AddGeneIntegrationMatrix. The profile of the 38 cells whose transferred cluster corresponds to cloaca (endo.31) was generated with getGroupBW.

scRNA-seq

The matrix of the scRNA-seq atlas was downloaded from GEO (GSE223922; ref. ³⁷) and the table with metadata. The matrix was loaded into a Seurat object using Seurat v.4.3.0 (ref. ⁸⁶) in R v.4.3.0. Cells attributed to the ‘tissue.name’ ‘endoderm’ were selected. Normalization and principal component analysis were performed, as described in a previous study³⁷. UMAP was performed on the top 70 principal component analyses and 50 nearest neighbours. UMAP coordinates and hox13 normalized expression of endoderm cells were exported to a file and plotted using ggplot2 v.3.4.4.

Software

The phylogenic tree was generated with http://timetree.org using the following species: Mus musculus, Protopterus, D. rerio, Carcharhinus leucas, Petromyzon marinus and Branchiostoma lanceolatum and subsequently edited using SeaView 4.7. Genomic tracks from next-generation sequencing were plotted using pyGenomeTracks 3.8 using custom gene annotations available at https://doi.org/10.5281/ZENODO.7510796 (ref. ⁶⁸; mm39) and https://doi.org/10.5281/zenodo.10283273 (ref. ⁸⁷; danRer11). RT–qPCR, RNA-seq and domain size quantifications were plotted in R using the ggplot package.

Ethical statement

All experiments involving mice were performed in agreement with the Swiss Law on Animal Protection (Loi sur la Protection des Animaux) under licence no. GE 81/14. For zebrafish, work was carried out either under a general licence of EPFL granted by the Service de la Consommation et des Affaires Vétérinaires of the Canton of Vaud, Switzerland (no. VD-H23) or was either agreed upon by the animal committees of Rutgers University under protocol no. 201702646 or under guidance of the Institutional Animal Care and Use Committee of Boston Children’s Hospital.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.