Liebenberg syndrome severity arises from variations in Pitx1 locus topology and proportion of ectopically transcribing cells

Abstract

Enhancer hijacking, a common cause of gene misregulation linked to disease, occurs when non-matching enhancers and promoters interact ectopically due to genetic alterations. While the concept of enhancer hijacking is well understood, the reasons behind the variation in phenotypic severity remain unexplored. In this work, we expand on the ectopic activation of the hindlimb-specific transcription factor Pitx1 by one of its own enhancers, Pen, in forelimb tissues that causes the Liebenberg syndrome. Using a series of inversions and relocations we show that reduction in Pitx1-Pen relative genomic positioning leads to increased proportions of Pitx1 forelimb-expressing cells and more severe phenotypical outcomes. We demonstrate in ectopically expressing cells that the Pitx1 locus assumes an active topology and that its promoter generates consistent transcription levels across different alleles. Finally, we show that changes in 3D chromatin structure and enhancer-promoter contacts are not the result of Pitx1 transcription.

Introduction

The restriction of enhancer-promoter contacts is a fundamental feature of gene regulation. This was shown to be mediated by domains of preferential interactions called topologically-associating domains (TADs). Indeed, TADs foster high internal chromatin interactions while reducing interactions with external regions. Biophysically, TADs are believed to be formed by a process called loop extrusion where cohesin molecules extrude chromatin until reaching CTCF which induces a temporary stalling of the process^1,2. Changes in CTCF binding therefore impact the 3D architecture of loci and enhancer-promoter contacts³. Moreover, tissue-specific chromatin interactions can actively control enhancer-promoter communications in a spatiotemporally-defined manner, enabling the activation of associated genes^4,5,6,7.

Alterations in this organized process can lead to the wrongful connection between non-matching enhancers and promoters, leading to gene de-repression and expression in ectopic tissues, in a process named “enhancer-hijacking”. In particular, structural variants (SVs) that impact the topological organisation of loci have been shown to lead to congenital malformations in such a way^8,9,10. Although the patho-mechanism of SV-induced enhancer-hijacking has been documented across numerous loci, these accounts often overlook the influence of variations in SV breakpoints on disease outcomes or severity¹¹. Furthermore, the precise relationship between distinct SVs and subsequent changes in the 3D genome architecture, chromatin modifications, and ectopic gene transcription is yet to be fully elucidated.

This is what happens at the Pitx1 locus, where different SVs underlying the Liebenberg syndrome, a congenital malformation associated to a partial arm-to-leg transformation, are associated with variable morphological changes^12,13,14,15. During normal development, the Pitx1 gene is specifically expressed in developing hindlimb, and not in forelimbs, where it controls hindlimb outgrowth and differentiation into a leg^16,17,18. So far, three limb enhancers have been identified at the locus: PelB, RA4 and Pen^4,19. Notably, another enhancer, PDE, has been described to contact the gene and as being strongly marked with H3K27ac in hindlimb, however, in reporter assays, the region only displays activity in the developing mandible^4,20. Importantly, both RA4 and Pen display a fore- and hindlimb activity when assayed in transgenic reporter approaches, and indeed, in the Liebenberg syndrome, the Pitx1 gene gets endo-activated, i.e. ectopically activated by one of these two enhancers, Pen, in developing forelimbs⁴. This activation results from SVs that re-arrange the locus and generally bring Pen, normally located 400 kb away from Pitx1, in a closer genetic proximity to Pitx1. Patients with SVs that slightly reduce the Pitx1-Pen genetic distance show rather mild malformation features, yet, patients where Pitx1-Pen linear distance is strongly reduced display more severe ones (Supplementary Fig. 1, Supplementary Data 1)^12,13,14,15.

Here, we combine a previously developed in-embryo cell-tracing approach with engineered Liebenberg structural variants and Pen relocations to measure and isolate Pitx1-expressing cells in mouse forelimbs²¹. In this context, we explore how structural variants can cause different degrees of phenotypic manifestations by identifying their link to the proportion of ectopically expressing cells and transcriptional activities. Moreover, we investigate how de-repression or targeted activation of Pitx1 can impact transcriptional activities and the locus topology.

Results

Pitx1-Pen relative genomic position affects the proportion of Pitx1 ectopically expressing cells

To address how differential SVs breakpoints lead to gene mis-activation, we took advantage of the previously described Pitx1^EGFP/EGFP (referred to as Pitx1^EGFP) sensor allele that allows for the tracking and sorting of Pitx1 active and inactive cells from developing tissues (Rouco et al., 2021). We re-engineered in the Pitx1^EGFP background a previously published inversion leading to Liebenberg syndrome in mice: Pitx1^{EGFP;Inv1/EGFP;+} (referred to as Pitx1^EGFP;Inv1), as well a larger one Pitx1^{EGFP;Inv2/EGFP;+} (referred to as Pitx1^EGFP;Inv2) (Fig. 1A-C)⁴. These inversions place Pen at the positions of RA4 and PDE, located 225 kb and 116 kb from Pitx1, respectively. At these locations, Capture-HiC (C-HiC) reveals that baseline interactions with Pitx1 are stronger than Pitx1-Pen interactions in control Pitx1^EGFP mouse embryonic stem cells (mESCs) (Fig. 1A). To measure how inversions perturb the locus poised 3D organisation, we performed C-HiC in Pitx1^EGPF;Inv1 and Pitx1^EGFP;Inv2+/- mESCs. In Pitx1^EGFP;Inv1, we observed a similar structure as in control mESCs (Fig. 1B). In contrast, in Pitx1^EGFP;Inv2+/-, we observed several differences in the locus topology, with increased contact between Pitx1, Pen, and Neurog1 (Fig. 1C).

Fig. 1: Inversions at the Pitx1 locus lead to increased mis-activation of the gene.

A C-HiC analysis of the Pitx1 locus in Pitx1^EGFP mESCs (n = 1). Upper right corner: quantification of interactions between Pitx1 and RA4/PDE/Pen. B A 113 kb inversion Pitx1^EGFP;Inv1 (Inv1) that swaps the relative position of Pen and RA4 shows a relative decrease in Pitx1-Pen interactions (n = 1). C A 204 kb inversion Pitx1^EGFP;Inv2 (Inv2) shows an overall increase of contacts between Pitx1 and Pen (n = 1). D Fluorescence microscopy of Pitx1^EGFP, Inv1 and Inv2 E12.5 embryos. Note the EGFP mis-expression in Inv1 and Inv2 developing forelimbs. Scale bars: 2 mm. Forelimbs (FL) are delineated with a dotted white line and indicated with a white arrow. In total, 43 Pitx1^EGFP embryos, 19 Inv1 and 9 Inv2 E12.5 embryos have been analysed and showed reproducible patterns. E Histogram of EGFP signal and quantification of the proportion of mis-expressing cells. The grey and green areas show the delimitation of gating for EGFP- and EGFP+ cells, respectively, in the three alleles. The dotted red line in histograms indicates the upper limit of fluorescence. Each graph represents the catenation of 2-3 independent experiments. F Normalised FPKMs of EGFP and Pitx1 in E12.5 wildtype bulk forelimbs, EGFP+ cells of Inv1 and Inv2 forelimbs and EGFP+ cells of Pitx1^EGFP hindlimbs (Supplementary Data 2). Black squares indicate each biological replicate (n = 2). Note a plateau in Pitx1 and EGFP expression in both inversions, which is significantly lower than in EGFP+ cells from Pitx1^EGFP hindlimbs. Source data are provided as a Source Data file.

We then derived Pitx1^EGFP, Pitx1^EGFP;Inv1 and Pitx1^EGFP;Inv2 E12.5 embryos through tetraploid complementation and characterised forelimb EGFP fluorescence through microscopy and fluorescence activated cell sorting (FACS) (Fig. 1D, E; for gating strategy see Supplementary Fig. 2)²². We could measure in Pitx1^EGFP;Inv1 forelimbs 6.4% of EGFP+ expressing cells in contrast to 0% in Pitx1^EGFP control (Fig. 1E). This number rose to 27% in Pitx1^EGFP;Inv2 (Fig. 1E) suggesting that the variation in SV size can alter the proportion of cells ectopically expressing Pitx1 in the forelimb.

Interestingly, we observed an upper limit of EGFP fluorescence in both inversions, suggesting that the abundance of EGFP in active cells was similar between alleles (Fig. 1E). To confirm that hypothesis, we measured transcription in EGFP+ cells of both Pitx1^EGFP;Inv1 and Pitx1^EGFP;Inv2 forelimbs using RNA-seq. We observed similar transcription levels of both Pitx1 and EGFP across the two alleles (Fig. 1F, Supplementary Data 2). In fact, the ectopic transcriptional activity was only 1.5x lower than the one found in wildtype Pitx1^EGFP EGFP+ cells from hindlimbs (Fig. 1F). This minor difference might be the result of the heterozygous state of both inversions in the Pitx1^EGFP homozygous background, suggesting that the transcriptional activity per allele in forelimb EGFP+ cells is comparable to that in wildtype hindlimb EGFP+ cells. Yet, as the inverted intervals of both Pitx1^EGFP;Inv1 and Pitx1^EGFP;Inv2 contain CTCF sites (Supplementary Fig. 3A) and other Pitx1 enhancers, the interpretation of the results can be confounding. Therefore, alternative approaches to solely measure the effect of the relocation of Pen were further developed.

A series of Pen relocations induce varying proportions of Pitx1-expressing cells

To rule out the positional effect induced by the inverted genomic interval, we devised a parallel approach where we solely re-mobilized the Pen enhancer itself in a Pitx1^{EGFP;ΔPen/EGFP;ΔPen} (referred to as Pitx1^EGFP;ΔPen) homozygous deleted background. Here, we inserted Pen at the same locations as in the inversions, at RA4 (Pitx1^{EGFP;ΔPen;Rel1/EGFP;ΔPen;+}: Pitx1^{EGFP;ΔPen;Rel1}) and at PDE(Pitx1^{EGFP;ΔPen;Rel2/EGFP;ΔPen;+}: Pitx1^{EGFP;ΔPen;Rel2}) (Fig. 2A, B). Moreover, we also introduced Pen 7.7 kb upstream of the Pitx1 promoter (Pitx1^{EGFP;ΔPen;Rel3/EGFP;ΔPen;+}: Pitx1^{EGFP;ΔPen;Rel3}), in a similar genetic distance (10.5 kb enhancer-promoter distance) as the one found in the most severe case of Liebenberg syndrome described (Fig. 2C, Supplementary Fig. 1)¹⁵. Of note, with each relocation reducing the genetic distance between Pitx1 and Pen, there is also a consequent reduction in the number of CTCF binding sites separating these two elements (Supplementary Fig. 3B).

Fig. 2: Relocation of Pen through the locus ectopically activates Pitx1.

A Illustration of Pitx1^{GFP;ΔPen;Rel1} (Rel1) where Pen is inserted, at RA4, 216 kb away to Pitx1. B Illustration of Pitx1^{GFP;ΔPen;Rel2} (Rel2) where Pen is inserted at PDE, 125 kb away from Pitx1. C Illustration of Pitx1^{GFP;ΔPen;Rel3} (Rel3) where Pen is inserted 7.7 kb away from Pitx1. D Fluorescence microscopy of Rel1, Rel2 and Rel3 E12.5 embryos. Note the mis-expression of EGFP and in developing forelimbs of E12.5 embryos. Forelimbs (FL) are delineated with a dotted white line and indicated with a white arrow. Scale bars: 2 mm. In total, 17 Rel1, 5 Rel2 and 8 Rel3 E12.5 embryos have been analysed and showed reproducible patterns. E Histogram of EGFP signal and quantification of the proportion of mis-expressing cells. The grey and green areas show the delimitation of gating for EGFP- and EGFP+ cells, respectively, in the three alleles. The dotted red line in histograms indicates the upper limit of fluorescence, each graph represents the catenation of 2-3 independent experiments. F Normalised FPKMs of EGFP and Pitx1 in E12.5 wildtype bulk forelimbs, EGFP+ cells of Rel1, Rel2 and Rel3 forelimbs and EGFP+ cells of Pitx1^EGFP hindlimbs (Supplementary Data 2). Black squares indicate each biological replicate (n = 2). Note the consistent Pitx1 expression level between relocations, similar to inversion (See Fig. 1). Source data are provided as a Source Data file.

Similar to Pitx1^EGFP;Inv1, Pitx1^{EGFP;ΔPen;Rel1} E12.5 forelimbs showed 2% EGFP+ cells, suggesting that at this location the inversion and relocations bear a similarly mild transcriptional effect on Pitx1 and the EGFP sensor (Fig. 2D, E). In contrast, in Pitx1^{EGFP;ΔPen;Rel2}, we measured 59% of EGFP+ forelimb cells and observed a clear bimodal EGFP signal distribution (Fig. 2D, E). This is twice the proportion observed when Pen was positioned at the same location in Pitx1^EGFP;Inv2 forelimbs, where only 27% of cells were EGFP+ (Fig.1E). This difference suggests that the alterations in CTCF relative positioning and binding site directionality within the inverted interval might restrict the capacity of Pen to induce Pitx1 in Pitx1^EGFP;Inv1 forelimbs. Indeed, in contrast to Pitx1^{EGFP;ΔPen;Rel2}, the Pitx1^EGFP;Inv2 allele causes the relocation and inversion of a Pitx1-convergent CTCF binding site at PDE, to the telomeric inversion breakpoint (Supplementary Fig. 3A-B). However, it cannot be excluded that the effect is also mediated by the interplay between Pen and PDE. Finally, in the E12.5 forelimb of the most proximal relocation, Pitx1^{EGFP;ΔPen;Rel3}, we observed a bimodal distribution of the EGFP signal and measured 62% of EGFP+ cells (Fig. 2D, E). Overall, the similar proportion of EGFP+ cells in Pitx1^{EGFP;ΔPen;Rel2} and Pitx1^{EGFP;ΔPen;Rel3}, shows that repositioning the enhancer either in the PDE region or a few kb upstream of the gene promoter induces a similar effect on Pitx1 mis-activation (Fig. 2D, E, Supplementary Fig. 3B).

As inversions showed a similar transcription level between alleles in EGFP+ cells, we wanted to confirm this in the context of the relocations. We therefore performed RNA-seq in EGFP+ cells and found that Pitx1 and EGFP expression is similar in all the active cells (Fig. 2F, Supplementary Data 2). Overall, this data shows that the ability of Pen to contact Pitx1 defines the proportion of cells in which the gene will be ectopically activated, yet, it does not strongly affect Pitx1 transcription level per allele.

Increase in Pitx1 ectopically expressing forelimb cells associate with worsened skeletal defects

As changes in Pen positioning lead to a different proportion of cells ectopically activating Pitx1, the phenotypic effect of these variations is unknown. To test whether an increase in affected cells is linked to a worsened phenotype, we analysed mutant skeletons of E18.5 embryos and scored forelimb malformations. We decided to compare wildtype to Pitx1^EGFP;Inv1, Pitx1^EGFP;Inv2 and Pitx1^{EGFP;ΔPen;Rel3} skeletons as these three precisely showed a progressive increase in EGFP+ cell proportions with 6.4%, 27%, and 62%, respectively. Weakly overexpressing forelimbs from Pitx1^EGFP;Inv1 resulted in a mild phenotype, specifically with a slight bowing of the radius and ulna (Fig. 3A, B, Supplementary Data 3). Notably, the same allele showed a stronger phenotype when bred to homozygosity and assayed in adult mice (Kragesteen, et al., 2018). Pitx1^EGFP;Inv2 forelimbs, where 27% of cells are EGFP+ at E12.5, showed more striking bowing of the radius and ulna (Fig. 3C, Supplementary Data 1, 3). Furthermore, we noted a significant reduction of the deltoid crest, a structure characteristic of the forelimb, accompanied by a mildly hypoplastic olecranon. Additionally, there was a noticeable broadening of the distal head of the humerus and the proximal head of the radius, a phenotype that aligns with previous descriptions in patients (Fig. 3C, Supplementary Data 1, 3). Finally, Pitx1^{EGFP;ΔPen;Rel3} forelimbs, where 62% of cells are EGFP+ at E12.5, exhibited the most severe phenotype. This included the recurring bowing of the long zeugopodal bones, strong reduction of the deltoid crest, broadening of the distal humerus and proximal radius and notably, in all analysed Pitx1^{EGFP;ΔPen;Rel3} skeletons, an aplastic or severely hypoplastic olecranon, a feature not observed in other alleles, but often in Liebenberg syndrome patients (Fig. 3D, Supplementary Data 1, 3). Finally, we observed a relative thinning of the ulna compared to its radius counterpart, in a similar way as the fibula is thinner than the tibia, underlining the arm-to-leg transformation. Most of the phenotypes scored across alleles were highly penetrant, with some variability in expressivity, even when only a limited number of cells were affected by Pitx1 misexpression (Supplementary Data 3). Overall, our analysis shows that an increase of Pitx1 ectopically activating cells, and in fact a change in the balance between the proportion of active and inactive cells, has a positive correlation with the accumulation of defects in the developing forelimb skeleton.

Fig. 3: Increasing proportions of Pitx1 ectopically-expressing cells correlates with severity of skeletal defects.

A Alizarin red and alcian blue staining of wildtype E18.5 forelimbs. Black arrows pinpoint to sca: scapula, hu: humerus, dc: deltoid crest, ole: olecranon, ra: radius, ul: ulna. Alizarin red and alcian blue staining of mutants (B) Pitx1^EGFP;Inv1, C Pitx1^EGFP;Inv2 and (D) Pitx1^{EGFP;ΔPen;Rel3} E18.5 forelimbs. Arrows point to dysplastic skeletal features: bowing of the radius and ulna, reduction of the deltoid crest, reduction of the olecranon, broadening of distal humerus and proximal radius and ulna, relative thinning of ulna. Number of independent limbs analysed and relative phenotypes are displayed in Supplementary Data 3.

Pitx1 forelimb endo-activation retains mesenchymal specificity

Liebenberg-associated SVs have been described to lead to arms assuming various skeletal and soft tissue features of legs^4,12,23. To understand to what extent SV-induced Pitx1 forelimb transcription resembles its normal hindlimb activity, we performed 10X single-cell RNA-seq (scRNA-seq) on stage-matched E12.5 Pitx1^Inv1/+ forelimbs and compared to wildtype fore- and hindlimbs. We selected this particular SV because it is a well-established model for Liebenberg syndrome and closely matches the rearrangement size of three SVs that have been independently identified in patients^4,12,14,21. The first level of clustering revealed six main limb clusters: muscle, neuron, immune cells, epithelium, endothelium and mesenchyme (Supplementary Fig. 4A, Supplementary Data 4). We noticed that Pitx1 expression was restricted to the mesenchyme in wildtype hindlimbs but also in Pitx1^Inv1/+ forelimb, although at lower expression levels (Supplementary Fig. 4A-B). We then subclustered the mesenchyme to obtain more definition to quantify Pitx1 expression across sub-populations (Fig. 4A, Supplementary Data 4). Here, we identified nine mesenchymal populations comparable to the ones previously characterised in E12.5 limb mesenchyme²¹. Four clusters showed proximal identity: Proximal Proliferative Progenitors (PPP), Tendon Progenitors (TP), Irregular Connective Tissue (ICT) and Proximal Condensations (PC). An additional four clusters showed distal identity: Distal Proliferative Progenitors (DPP), Distal Progenitors (DP), Early Digit Condensations (EDC) and Late Digit Condensations (LDC). Finally, we identified a Mesopodium (MS) cell cluster, neither proximal nor distal. In Pitx1^Inv1/+ forelimbs, at least one Pitx1 transcript could be detected in 34% of cells (compared to 2% in wildtype forelimbs), a higher percentage than the 6% of EGFP+ cells observed by FACS. This discrepancy can be attributed to the gating strategy, which excluded cells with weak GFP signals (See Figs. 1–2 and Supplementary Fig. 2), and the inherent sensitivity of both approaches. In wildtype hindlimbs and Pitx1^Inv1/+ forelimbs, we observed Pitx1 expression in all mesenchymal clusters showing that the forelimb gain of expression occurred with a similar specificity than in hindlimbs. However, the variation in Pitx1 expression between clusters was more pronounced in wildtype hindlimbs compared to Pitx1^Inv1/+ forelimbs (Fig. 4B, Supplementary Data 4). This observation indicates that the mesenchymal specificity of Pitx1 expression is preserved in mutant forelimbs when compared to wildtype hindlimbs, albeit not to its full extent across mesenchymal subpopulations.

Fig. 4: Pitx1 forelimbs ectopic expression is restricted to mesenchymal clusters.

A UMAP of mesenchymal cell clusters present in E12.5 wildtype fore- and hindlimbs and E12.5 Pitx1^Inv1/+ forelimbs. B Pitx1 expression by cell cluster in wildtype fore- and hindlimbs and Pitx1^Inv1/+ forelimbs. Note the overall similarity of expression between wildtype hindlimbs and Pitx1^Inv1/+ forelimbs. C Selected marker genes enrichment across EGFP+ population of Pitx1^EGFP hindlimbs, as well as Pitx1^EGFP;Inv1 (Inv1), Pitx1^EGFP;Inv2 (Inv2), Pitx1^{EGFP;ΔPen;Rel1} (Rel1), Pitx1^{EGFP;ΔPen;Rel2} (Rel2), Pitx1^{EGFP;ΔPen;Rel3} (Rel3) forelimbs compared to wildtype bulk forelimbs. Source data are provided as a Source Data file.

To assay whether these expression specificities are a general feature of Pitx1 endo-activation, we analysed the enrichment of marker genes in Pitx1^EGFP;Inv1, Pitx1^EGFP;Inv2, Pitx1^{EGFP;ΔPen;Rel1}, Pitx1^{EGFP;ΔPen;Rel2}, Pitx1^{EGFP;ΔPen;Rel3} forelimbs and control Pitx1^EGFP hindlimb EGFP+ cells compared to wildtype forelimbs. Generally, we observed homogenous marker gene enrichment among mutants, showing high similarity between EGFP+ cells (Supplementary Data 2). More specifically, all the EGFP+ populations showed a depletion of genes linked to non-mesenchymal cell identity (Mrc1, Ttn, Krt14, Dlk2, Cldn5) and an enrichment for mesenchymal markers (Prrx1, Lhx9) confirming that Pitx1 endo-activation specifically occurs in mesenchymal cell types (Fig. 4C). We also observed enrichment of proximal (Shox2 and Tbx15), tendon (Egr1) and chondrogenic markers (Sox9, Runx2), corroborating the previous findings obtained from scRNA-seq. Furthermore, we also found that cells expressing Pitx1 were enriched for cell division markers as JunB and JunD in line with the tissue outgrowth properties associated to Pitx1^21,24 (Fig. 4C). This shows the cell-specificity of Pitx1 endo-activation in forelimbs mirrors to a certain extent its physiological expression in wildtype hindlimbs.

Finally, to understand whether Pitx1 endo-activation can induce a wider hindlimb-like transcriptional program, we compared bulk Pitx1^{EGFP;ΔPen;Rel2} and Pitx1^{EGFP;ΔPen;Rel3} to wildtype forelimbs transcriptome. Here, we detected that the hindlimb-specific gene Tbx4 was upregulated in mutant forelimbs, indicating that Pitx1 expression could induce its transcription (Supplementary Data 5)²⁵. We also noted an increase in cartilage and chondrogenesis related markers such as Sox9, Foxc1 and Gdf5 suggesting an increased chondrogenic program in mutant forelimbs (Supplementary Data 5)¹⁸. We then explored whether the Pitx1^Inv1/+ forelimb scRNA-seq data showed a similar trend and found a comparable, though weaker, effect, with Tbx4 and Gdf5 significantly upregulated, while Foxc1 and Sox9 remained unchanged (Supplementary Fig. 4C). Altogether, these findings underline that Pitx1 endo-activation can establish, in the forelimb counterpart of hindlimb Pitx1 expressing cell-types, features of hindlimb transcriptional programs.

SV-induced Pitx1 endo-activation is linked to discrete topological changes

Hindlimb cells transcriptionally active for Pitx1 adopt a fundamentally different 3D locus topology than their inactive counterparts²¹. Consequently, it is plausible that SVs-induced Pitx1 endo-activation leads to topological changes in transcriptionally active cells. To test this hypothesis, we initially performed C-HiC on Pitx1^EGFP;Inv1, comparing EGFP+ and EGFP- forelimb cells. We found that Pitx1 contacts Pen as well as PelB, PDE, and RA4 more frequently in EGFP+ cells than in EGFP- cells (Fig. 5A). Conversely, in EGFP- cells, the repressive contact between Pitx1 and Neurog1 was more prevalent than in EGFP+ cells (Fig. 5A). These differences are strikingly similar to those observed between hindlimb EGFP+ and EGFP- cells (Rouco et al., 2021), indicating that this inversion facilitates the formation of an active topology specifically in transcriptionally active cells.

Fig. 5: Topological changes at the locus diminish as Pitx1-Pen contact probabilities increase.

C-HiC of the Pitx1 locus in EGFP+ (red maps) and EGFP- (blue maps) cells from (A) Pitx1^EGFP;Inv1 forelimbs, B Pitx1^EGFP;Inv2 forelimbs, C Pitx1^{EGFP;ΔPen;Rel2} forelimbs and (D) Pitx1^{EGFP;ΔPen;Rel3} forelimbs. Darker red or blue bins indicate stronger interaction frequencies as shown on the scale bars. For each panel, the lowest map is a subtraction of the two above where preferential interactions in EGFP+ cells are shown in red, while the ones in EGFP- cells are shown in blue. Contacts between Pitx1 and Pen are shown with a green arrow, Pitx1 contacts with PelB, PDE or RA4 are shown with grey arrows, the Pitx1-Neurog1 contact is shown with a red arrow. All subtraction scales were homogenized for comparison purposes. All experiments were performed once.

We next investigated whether different active-inactive topologies would also be present in the other alleles described or if this was a specific feature of Pitx1^EGFP;Inv1 forelimbs. Thus, we generated C-HiC maps of EGFP+ and EGFP- cells obtained from Pitx1^EGFP;Inv2, Pitx1^{EGFP;ΔPen;Rel2} and Pitx1^{EGFP;ΔPen;Rel3} forelimbs. In Pitx1^EGFP;Inv2, despite a higher proportion of Pitx1-expressing cells (See Fig. 1D-E), we observed fewer changes in interaction between EGFP+ and EGFP- cells. Here, only the interaction between Pitx1 and Pen was strongly increased in EGFP+ cells and, to a lesser extent, that between Pitx1 and PDE (Fig. 5B). Similarly, in Pitx1^{EGFP;ΔPen;Rel2}, EGFP+ cells showed a clear gain of contacts between Pitx1 and PDE, where the Pen enhancer is relocated, but not with other regions (Fig. 5C). These results consistently highlight strengthened Pitx1-Pen contact in transcriptionally active cells, suggesting that increased physical proximity is essential for transcription. Lastly, Pitx1^{EGFP;ΔPen;Rel3} EGFP+ and EGFP- forelimb cells exhibited limited topological changes (Fig. 5D). Here, due to the short 7.7 kb interval between Pitx1 and Pen, the contact frequency between the two elements was very high in both active and inactive cells. Yet, we noted a relatively stronger contacts in EGFP- cells, a phenomenon already observed for active short-range regulatory contact (Fig. 5D)²⁶. In conclusion, across the different gain-of-function alleles, we observe that fewer locus-wide topological changes associate with Pitx1 transcription, i.e. in GFP+ vs GFP- cells, when Pen is brought closer to Pitx1.

We further explored whether changes in chromatin topology are associated with changes in cis-regulatory element activities by performing H3K27ac Chromatin Immunoprecipitation (ChIP-seq) in Pitx1^EGFP;Inv1, Pitx1^{EGFP;ΔPen;Rel2} and Pitx1^{EGFP;ΔPen;Rel3} EGFP+ cells, as well as in Pitx1^EGFP;Inv1 EGFP- cells as a control (Supplementary Fig. 5). As expected in Pitx1^EGFP;Inv1 EGFP- cells, we did not observe H3K27ac at Pitx1 promoter, nor at several enhancers aside of Pen. In contrast, we observed a strong enrichment of H3K27ac at the Pitx1 promoter in EGFP+ cells across alleles. In both Pitx1^EGFP;Inv1 and Pitx1^{EGFP;ΔPen;Rel2} EGFP+ cells, there was an increase in H3K27ac coverage at PDE, a region interacting with Pitx1 in both alleles. In Pitx1^{EGFP;ΔPen;Rel3} EGFP+ cells, only the region adjacent to the Pen relocation showed a clear acetylation signal (Supplementary Fig. 5). Finally, we also noted that in the two relocation alleles, the loss of Pen at its endogenous genomic location resulted in decreased H3K27ac spreading around it, while an increase around the Pen-relocated region was observed, showcasing the spreading potential of the histone mark. In summary, the increased chromatin contacts observed in C-HiC data involved regions marked by H3K27ac and therefore imply a type of homotypic interaction.

Targeted activation of Pitx1 does not induce topological change

In the context of SV-induced Pitx1 endo-activation, the relocation of the Pen enhancer associates with changes in transcriptional activity and genome topology (See Fig. 5 and Supplementary Fig. 5). Because both events occur in the same cells, it is unclear whether it is the transcription of the locus that induces the 3D topological changes or whether these occur independently. To assay whether ectopic activation of Pitx1 is sufficient to induce changes in the locus topology, we first developed an in vivo dCas9-P300 activator targeted to the Pitx1 promoter. To achieve specific expression of the activator in cell clusters permissive to Pitx1 expression (See Fig. 4), we integrated the dCas9-P300 transgene preceded by a minimal promoter, as a sensor, upstream of the Shox2 gene promoter to produce Shox2^dCas9P300/+ mESCs (Fig. 6A). We selected Shox2 because of its similar expression specificity with Pitx1 in developing hindlimbs (correlation coefficient=0.577, p-value = 0.001, where the p-value is the probability for the correlation coefficient to be negative). To direct the dCas9-P300 activator to Pitx1, we integrated two sgRNAs that target the Pitx1 transcriptional start site (TSS) at the ColA1 locus to produce Shox2^dCas9P300/+;ColA1^TSSsgR/+ ESCs (Fig. 6A)²⁷.

Fig. 6: dCas9P300 induces Pitx1 expression in forelimbs without topological changes.

A A dCas9-P300 cassette was inserted as a sensor upstream of the Shox2 promoter, two sgRNAs to target dCas9 activity were integrated at the ColA1 safe harbour locus through an FRT-mediated recombination. B RT-qPCR of dCas9P300 in wildtype and Shox2^dCas9P300/+;ColA1^TSSsgR/+ E12.5 forelimbs (FL) and trunk (TR) tissues. The values represent a log2 fold change compared to wildtype forelimb that was set to 1. C Normalised Pitx1 FPKMs in Shox2^dCas9P300/+;ColA1^+/+ and Shox2^dCas9P300/+;ColA1^TSSsgR/+ (Supplementary Data 6). Black squares indicate each biological replicate (n = 2). Source data are provided as a Source Data file. D Whole Mount In-Situ Hybridization (WISH) of Pitx1 in wildtype and Shox2^dCas9P300/+;ColA1^TSSsgR/+ forelimbs. Note the proximal gain of Pitx1 expression (black arrow; pattern observed in 4/4 embryos). Scale bars: 1 mm. E Individual UMAPS of scRNA-seq data from wildtype and Shox2^dCas9P300/+;ColA1^TSSsgR/+ forelimbs showing the distribution of Shox2, Pitx1 and dCas9P300 expressing cells as well as the respective percentage of expressing cells in proximal forelimb (proximal) and distal forelimb (distal). F Subtraction of Shox2^dCas9P300/+;ColA1^TSSsgR/+ (TSS sgRNA) and Shox2^dCas9P300/+;ColA1^+/+ (no sgRNA) E12.5 proximal forelimbs C-HiC maps. Contacts that are more frequent in TSS sgRNA are colored in red, and those more frequent in no sgRNA are colored in blue (See scale bar on the left). Contacts between Pitx1 and Pen are shown with a green arrow, Pitx1 contacts with PelB, PDE or RA4 are shown with grey arrows, the Pitx1-Neurog1 contact is shown with a red arrow. Note the absence of visible change. Corresponding individual C-HiC maps are shown in Supplementary Fig. 7 (n = 1).

We then derived E12.5 Shox2^dCas9P300/+;ColA1^TSSsgR/+ embryos using tetraploid aggregation²². Using RT-qPCR we could detect dCas9-P300 transcripts in forelimbs but not in the embryonic trunk, confirming the expression specificity of the sensor (Fig. 6B). Using RNA-seq, we measured Pitx1 expression in Shox2^dCas9P300/+;ColA1^+/+ and Shox2^dCas9P300/+;ColA1^TSSsgR/+ forelimbs and could detect a 15-fold upregulation of the gene in the latter (Fig. 6C, Supplementary Data 6). As observed by whole mount in-situ hybridization (WISH), the expression pattern of Pitx1 was localized to the proximal forelimb and reminiscent of Shox2 expression in E12.5 forelimbs (Fig. 6D). Single-cell RNA-seq revealed that Pitx1 was expressed in 9% of ^dCas9P300/+;ColA1^TSSsgR/+ forelimb mesenchyme compared to 2% of wildtype counterparts²¹. Moreover, we could generally observe that, Pitx1 and Shox2 expression domains colocalized in proximal clusters (Pitx1-Shox2 correlation in the entire Shox2^dCas9P300/+;ColA1^TSSsgR forelimb=0.441 p-value = 0.0005, where the p-value is the probability for the correlation coefficient to be negative Fig. 6E, Supplementary Fig. 6).

We next tested whether the gain of Pitx1 transcription would elicit a change in 3D conformation of the locus. To enriched for Pitx1 transcriptionally active cells, we micro-dissected E12.5 proximal forelimbs of Shox2^dCas9P300/+;ColA1^TSSsgR/+ (14.9% of Pitx1-expressing cells) and Shox2^dCas9P300/+;ColA1^+/+ (2.1% of Pitx1-expressing cells as defined in wildtype forelimb scRNA-seq) and performed C-HiC (Supplementary Fig. 7). By comparing Shox2^dCas9P300/+;ColA1^TSSsgR/+ and Shox2^dCas9P300/+;ColA1^+/+ C-HiC maps we did not observe change in locus interactions (Fig. 6F), suggesting that pronounced increases in topological contacts with PelB, PDE, RA4, and Pen are not driven by direct Pitx1 activation. It is nevertheless possible that subtle changes may have been missed due to the small proportion of active cells in the sample studied.

Loss of PRC2 induces Pitx1 forelimb transcription without increasing enhancer-promoter contacts

Because the targeted activation of Pitx1 affected a limited proportion of forelimb cells, subtle topological changes could be missed. Therefore, we used a different approach to activate Pitx1 transcription and asked whether removal of PRC2-mediated polycomb repression would be a more effective method. PRC2 is a multiprotein complex made of several subunits including the H3K27me3 reader EED which enables the spreading of the mark over chromatin domains²⁸. Here we exploited a conditional Eed floxed allele combined to a full Eed knock out, and a limb-specific mesenchymal CRE driver (Prx1-CRE;Eed ^flox/-) to assess the effect of its loss on both Pitx1 transcription and locus structure^29,30,31.

Through WISH, we could observe a strong gain of Pitx1 expression in proximal Prx1-CRE;Eed ^flox/- E12.5 forelimbs (Fig. 7A). We then re-analysed RNAseq data and observed a 27-fold upregulation of Pitx1 in mutant forelimbs compared to wildtype littermates (Fig. 7B, Supplementary Data 7)³⁰. As expected, in proximal E12.5 forelimb, a decrease in H3K27me3 could be detected throughout the locus (Fig. 7C)³². It is also interesting to note that despite the loss of H3K27me3 at Neurog1, the gene, unlike Pitx1, was not ectopically transcribed in forelimb cells, underlying cell-specificity as a requirement for mis-activation of genes (Fig. 7C). The decrease of H3K27me3 at Pitx1 also coincided with the accumulation of the active H3K27ac mark at the gene promoter and PDE (Fig. 7C, black arrows)^30,32. This shows that, at Pitx1, the removal of PRC2 repression results in the activation of the locus.

Fig. 7: PRC2 loss activates Pitx1 transcription in forelimbs without increasing enhancer-promoter contacts.

A Pitx1 Whole Mount In-Situ Hybridization (WISH) of E12.5 wildtype and Prx1-Cre;Eed^flox/- (Eed^-/-) forelimbs. Note the strong proximal gain of Pitx1 expression (black arrow). Scale bars: 1 mm. B Normalised FPKMs of Pitx1 in E12.5 wildtype and Prx1-CRE;Eed^flox/- (Eed ^-/-) forelimbs. Black squares indicate each biological replicate (n = 2) (Supplementary Data 7). Source data are provided as a Source Data file. C ChIP-seq of H3K27ac (first two tracks) and H3K27me3 (two last tracks) show an accumulation of H3K27ac at the Pitx1 locus (black arrows) in proximal Prx1-Cre;Eed^flox/- (Eed^-/-) compared to wildtype (WT) forelimbs and an overall reduction of H3K27me3 signal. D Subtraction of wildtype and Prx1-Cre;Eed^flox/- (Eed^-/-) E12.5 proximal forelimbs C-HiC maps. Contacts more frequent in Eed^-/- are colored in red, and those more frequent in widtype are colored in blue (See scale bar on the left). Contacts between Pitx1 and Pen are shown with a green arrow, Pitx1 contacts with PelB, PDE or RA4 are shown with grey arrows, the Pitx1-Neurog1 contact is shown with a red arrow. Corresponding individual C-HiC maps are shown in Supplementary Fig. 8 (n = 1).

We then explored whether the loss of PRC2 repression results in a change of topological organisation of the locus (Supplementary Fig. 8). First, as expected from previous work³³, we observed a reduction of the Pitx1-Neurog1 PRC2-associated contact in Prx1-Cre;Eed ^flox/- proximal forelimbs compared to wildtype (Fig. 7D). However, similarly to the dCas9-P300 C-HiC data, we did not observed a gain of interactions between Pitx1 and its enhancers in Prx1-Cre;Eed ^flox/- forelimbs (Fig. 7D). In fact, we observed a relative loss of the contacts with PelB, PDE, RA4 and Pen, which suggests that PRC2 loss leads to a disorganisation of the locus topology. We concluded that loss of PRC2 leads to Pitx1 activation independently from increased enhancer-promoter interactions.

Discussion

In this work, we show that changes in the relative positioning between Pitx1 and its Pen enhancer associate with a variable proportion of overexpressing cells in developing forelimbs. Within this active cell population, the levels of Pitx1 expression do not increase with enhancer-promoter proximity but rather reach a conserved threshold of activation. This finding aligns with previous studies showing that enhancers primarily regulate transcription by increasing bursting frequency rather than bursting size^34,35. Moreover, it suggests that once activation is achieved at the Pitx1 locus it is done so at its full transcriptional potential where Pitx1 promoter activity is saturated. However, the bimodal distribution of active and repressed cells contrasts with findings from other studies in mESCs, where enhancer repositioning led to a unimodal gene activation with varying transcriptional intensity, and that rarely left a proportion of cells in a repressed state^36,37. This difference could stem from the heterogeneity of in vivo tissues or the specific nature of the Pen/Pitx1 functional interactions.

Changes in Pitx1-Pen distance and its associated variation in the proportion of cells ectopically expressing Pitx1, but not in Pitx1 transcription per allele, provide a mechanistic framework to account for the variation in Liebenberg syndrome severity among cases described so far. Here, we have shown that the more a SV reduces the Pen-Pitx1 distance, and consequently the number of intermediate CTCF sites, the higher the proportion of forelimb Pitx1 overexpressing cells will be and the more severe the skeletal defects. Similarly, patients with SVs inducing a short genomic distance and few intermediate CTCF binding between Pitx1 and Pen displayed more severe malformations (Supplementary Fig. 1 and 9). In general, variability in rare disease severity has already been described in a few cases. For instance, several overlapping deletions at the Epha4 locus, that induce rewiring of enhancers toward the Pax3 gene, result in brachydactyly and variable hand defects⁹. Here, the proportion of cells affected by the Pax3 overexpression in the distinct SVs could explain the variability in phenotypical outcome. In another reported case, different duplications at the Ihh locus, leading to variable increases of gene expression in developing limbs, were also shown to result in variable syndactyly phenotypes. Moreover, LacZ analysis of Ihh in the mutants indicated broadened expression domains of the gene, suggesting that increase in expression could be due to more cells ectopically activating Ihh³⁸. Therefore, although our data provides a mechanism for variation in the Liebenberg syndrome, it could be applied to other syndromes linked to ectopic gene transcription.

In a previous study, we have shown that the homozygous loss of Pen did not result in a full Pitx1 loss-of-function in hindlimbs, but in a 30% reduction of Pitx1 transcription (Rouco et al., 2021). This hindlimb loss was mostly the result of a fraction of cells from all mesenchymal clusters, without further specificity, not displaying any Pitx1 transcription. It was therefore hypothesised that Pen acts as a “support” enhancer enabling the robust Pitx1 transcriptional initiation in the mesenchyme. In this perspective, other regions would act to provide more cell-type specificity, such as RA4 that was recently described as a chondrogenic enhancer³⁹. This is similar to what happens during endo-activation, where Pen activates Pitx1 in all forelimb mesenchymal clusters without further specificity. Together, Pen-dependent loss and gain of Pitx1 expression pinpoint to the same role for Pen: to act as a pan-mesenchymal enhancer with the ability to trigger robust transcriptional onset at the Pitx1 locus. As in hindlimbs other enhancers are required to further define Pitx1 cell-type specific expression, it remains to be shown whether other local enhancers, such as RA4 which is also active in forelimbs, contribute to the final Pitx1 endo-activated expression in forelimbs.

By comparing the locus 3D topology in active and inactive cells, we observed that alleles driving Pitx1 expression in a limited proportion of cells displayed the most extensive topological changes. Specifically, in the smallest inversion, Pitx1^EGFP;Inv1, multiple enhancer-promoter contact are observed in transcriptionally active cells involving interaction between Pitx1 and Pen but also with PelB, PDE and RA4. This configuration is similar to the previously described stack configuration that occur, in fact, only in a fraction of Pitx1-expressing hindlimb cells⁴⁰. In the other extreme, when the Pen enhancer was introduced directly upstream of Pitx1, in Pitx1^{EGFP;ΔPen;Rel3}, topologies were very similar between inactive and active cells. Together these data suggest that genetic configurations that reduce the searching space of the Pitx1 promoter to find Pen, i.e. where the Pen-Pitx1 contact is a very probable choice, are more likely to initiate transcription and therefore to result in Pitx1 expression in a larger proportion of cells.

These changes in topology can be either the consequence of Pitx1 transcriptional activation or an independent process required to activate and sustain Pitx1 transcription. Here, our dataset can exclude the first possibility as the ectopic transcriptional activation of the Pitx1 promoter via an exogenous dCas9-P300 activator or via the alteration of PRC2 activities did not result in a gain of enhancer-promoter contacts. This finding is consistent with similar results from targeted transcriptional activator approaches at other loci^26,41, reinforcing our results and raising further questions about how enhancer-promoter contacts are established in transcriptionally active cells. A recent study focusing on the Pitx1 locus suggested that this process is controlled by changes in the loop extrusion process⁴⁰. In fact, such changes might be directly influenced by the presence of RNA polymerase II at promoters and enhancers⁴², a model that would account for the observed topological changes at Pitx1. However, this hypothesis seems unlikely, as neither ectopic transcription of Pitx1 nor the endogenous activity of the Pen enhancer in forelimbs, both involving RNA polymerase II binding, is sufficient to drive these contacts. It is therefore plausible that other mechanisms related to Pitx1 transcriptional activity influence loop extrusion. An alternative explanation could involve the formation of micro-compartments⁴³, independently of loop extrusion, facilitated by transcription factors. This model is supported by the observed loss of enhancer-promoter interactions in Hoxc genes or Pitx1 knockout limbs⁴. In summary, our results demonstrate that transcription does not induce enhancer-promoter contacts at the Pitx1 locus. Yet, such contacts, whether formed by loop extrusion or micro-compartments, could be essential to initially alter the state of the Pitx1 promoter through its de-repression, and potentially to sustain its expression later on.

Methods

Animal procedures

Animal work performed in Geneva adheres to all relevant ethical regulations of the University of Geneva and follows procedures approved by the animal care and experimentation authorities of the Canton of Geneva, Switzerland (animal protocol numbers GE/89/19 and GE192A). Animal work performed at the Institut de Recherches Cliniques de Montréal (IRCM) was reviewed and approved by the IRCM animal care committee (protocols 2020-01 and 2021-04).

Genetically engineered alleles

Engineered alleles using CRISPR/Cas9 technology were created in accordance with the methodology outlined in ref. ⁴⁴. sgRNAs were designed using the Benchling software, selecting them based on predicted on-target and off-target scores. Detailed information on all sgRNAs and their corresponding genomic locations for CRISPR–Cas9 can be found in Supplementary Data 8. The sgRNAs were sub-cloned into the pX459 plasmid from Addgene, with 8 μg of each vector utilized for the transfection of mESCs. Standard procedures for mESCs culture and genetic editing, were followed. The Pitx1^EGFP mESCs clone used was previously described in ref. ²⁰. Requests for transgenic G4 ESCs clones can be accommodated. C2 cells were engineered by frt-flippase system for recombination at the ColA1 locus. C2 cells were transfected with 8 μg of recombination plasmid and 1 μg of FLP-recombinase plasmid. Transfection was done with Lipofectamine LTX-Plus (Thermo Fischer, 15338030) following the manufacture’s transfection protocol. All primers used to characterise the alleles can be found in Supplementary Data 8.

Culture of mESC cells and aggregation

G4 (129/sv x C57BL/6 F1 hybrid, male) and C2 (C57B/6 and 129svJae F1 hybrid, male) mouse ESCs, obtained from the Nagy laboratory (http://research.lunenfeld.ca/nagy/?page=mouse%20ES%20cells), were cultured on male and female CD1 feeders with complete growth medium supplemented with 1000 U/mL LIF (Murine Leukaemia Inhibitory factor ESGRO^TM(10⁷U/ml, Chemicon #ESG1107). Embryos were derived by tetraploid complementation from G4 and C2 ESCs. The mESCs were thawed and grown for two days. Donor tetraploid embryos were provided from in vitro fertilisation using the c57bl6J x B6D2F1 background. Aggregated embryos were transferred into CD1 foster females. Sample size calculation was not performed as there was no prior knowledge of percentages of GFP positive cells in forelimbs and numbers of embryos harvested per tetraploid aggregation. Therefore sample sizes were chosen relative to the numbers of GFP positive cells needed to perform experiments.

Skeletal preparation

Skeletal preparation was performed as follows. E18.5 sacrificed foetuses were heat-shocked in H2O at 70 °C for 30” and skin and viscera were removed. This was followed by fixation in 100% EtOH at room temperature overnight and then in 100% acetone overnight at room temperature. Embryos were then stained in Alcian Blue (150 mg/l Alcian Blue 8GX Sigma-Aldrich) overnight at room temperature. Alcian Blue was removed and foetuses were washed with 100% EtOH and placed in Alzarin Red (50 mg/l Sigma Aldrich) in 0.2% KOH over two days. Finally, the remaining tissues were digested in 1% KOH with visual inspection and skeletons were stored in 0.2%KOH-30% glycerol for imaging and then long-term in 60% glycerol.

Whole mount in situ hybridization

Pitx1 WISH was performed on E12.5 embryos with a digoxigenin-labelled Pitx1 antisense probe designed from a cloned antisense probe (PCR DIG Probe Synthesis Kit, Roche 11636090910). Experimental procedure followed the protocol outlined in ref. ⁴.

Imaging

Embryos were imaged in PBS and skeletons in 0.2%KOH-30% glycerol on an Axio Zoom V16 (ZEISS) microscope. GFP laser exposure was set to 3000 ms.

Preparation of single-cell limb suspension

E12.5 limb tissues were microdissected in cold PBS and pooled for processing. To maintain efficiency in downstream experiments, no more than 6 limbs were pooled together at a time. The tissues were dissolved in 400 μL Trypsin-EDTA and 40 μL 2.5% BSA (Sigma Aldrich, A7906-100G) over 12 min at 37 °C in a Thermomixer set at 1500 rpm, with a brief resuspension at the 6-min mark. Trypsin was quenched by adding 400 μL 2.5% BSA, and the homogenised tissue was passed through a 40μm cell strainer. An additional volume of 2.5% BSA was passed through to collect any remaining cells. The collected cells were then centrifuged 5’ at 4 °C and 400 x g, followed by resuspension in 1% BSA. If H3K27ac ChIP was planned as a downstream experiment, 5 mM NaButyrate (Sigma Aldrich, 303410) was added to the 1% BSA.

Preparation for single-cell RNA-seq and library construction

Following the preparation of a single-cell limb suspension, cells were counted using an automated counter and resuspended to achieve a concentration of 1400 cells/μL. 50 μL of this suspension were provided to the iGE3 Genomic Platform for 10X Library Preparation. The platform performed library preparation for Pitx1^Inv1+/- using the Chromium Single Cell 3’ GEM, Library & Gel Bead Kit v3.0 following the manufacturer’s protocol. Libraries were pair-end sequenced on an Illumina HiSeq 4000 with approximately 8029 cells loaded on a Chromium Chip. For Shox2^dCas9P300/+;Cola1^TSSsgR/+ library preparation was done using the Chromium Single Cell 3’ GEM, Library & Gel Bead Kit v3.1 following the manufacturer’s protocol. Libraries were pair-end sequenced on an Illumina NovaSeq 6000 with approximately 10,141 cells loaded on a Chromium Chip.

Cell sorting

Fluorescence-activated cell sorting (FACS) was employed to identify and sort distinct cell populations in this study, utilizing the Biorad S3 with GFP laser (excitation wavelength 488 nm). To eliminate debris from the analysis, FCC/FCS settings were established between 30/40 and 230/220. The viability stain Draq7 was employed to distinguish live cells, and standard protocols were applied to select for singlets. For each sample, a negative control tissue, the embryo’s tails, was included to ensure the purity of the GFP- positive population. Moreover, the gating of GFP- positive populations was consistently applied across multiple experiments to ensure the selection of uniform populations and mitigate variability in GFP intensity over time. FlowJoTM Software was utilized for exporting the analysis in histogram format.

Cell processing for ChIP-seq and C-HiC

After sorting, cells were suspended in 1% BSA and then centrifuged 5’ at 400 x g at 4 °C in a tabletop centrifuge. The supernatant was discarded, and cells were resuspended in 10% FCS/PBS before fixation at room temperature. For ChIP, 1% formaldehyde was used, and for C-HiC, 2% formaldehyde was applied, both for a duration of 10’ with rolling. Fixation was quenched by adding 1.45 M cold glycine, followed by centrifugation at 1000 x g, 8’, 4 °C. Cells were then resuspended in cold lysis buffer (10 mM Tris, pH 7.5, 10 mM NaCl, 5 mM MgCl2, 1 mM EGTA, Protease Inhibitor (Roche, 04693159001)). After 10’ of incubation on ice, fixed nuclei were isolated through a 3-min centrifugation at 1000 x g at 4 °C, followed by washing in cold 1 x PBS buffer (1000 x g, at 4 °C for 1 min). The PBS was removed, and nuclei were stored at -80 °C.

Cell processing for RNA-seq and library preparation

For bulk limb analysis, two independent limbs were microdissected and snap-frozen at -80 °C for subsequent total RNA extraction using the RNEasy Mini Kit (QIAGEN, 74134) following protocol. RNA quantification was performed with Qubit 2.0 (LifeTechnologies) and the RNA Broad Range Assay (Q10210).

For GFP population studies, after sorting, at least two replicates of 2.5 ×105 cells were pelleted 5’ at 400 x g, 4 °C. After removal of 1% BSA, cells were snap-frozen at -80 °C for total RNA extraction. RNA extraction was carried out with the RNEasy Micro Kit (QIAGEN, 74004) following the manufacturer’s instructions. Quantification was performed with Qubit and RNA High Sensitivity Assay (Q32852).

Library preparation and sequencing were conducted at the iGE3 Genomic Platform. RNA integrity was assessed with a Bioanalyzer (Agilent Technologies). The SmartSeq v4 kit (Clontech) was used for reverse transcription and cDNA amplification, following the manufacturer’s instructions, with 5 ng RNA as input. Library preparation followed with a 200 pg cDNA input, using the Nextera XT kit (Illumina). Libraries were assessed by Tapestation and Bioanalyzer with a DNA High Sensitivity Chip, 2 nM were pooled and sequenced on an Illumina NovaSeq 6000 sequencer using SBS TruSeq chemistry with an average of 35 million reads (single-end 50 bp) per library.

RT-qPCR

RNA was extracted from forelimbs and trunks using the RNEasy Mini Kit (QIAGEN, 74134) and was converted into cDNA using the SuperScriptII RT kit (Invitrogen #18064-014), starting with 200 ng of RNA. The process utilized random hexamer primers (Thermo Scientific #S0142), DTT (Invitrogen Y00147), and RNaseOUT (Invitrogen 100000840). The amplification of cDNA was carried out with PowerUP™ SYBER Green Master Mix (Applied Biosystems A25742) and conducted on the QuantStudio 1 RT-PCR System from Applied BioSystems. The average CT values were standardized against the Gapdh housekeeping gene. The RT-qPCR primers for dCas9-P300 and Gapdh are detailed in Supplementary Data 8.

Immunoprecipitation

Nuclei were sonicated to an average size of 200-500 bp fragments on a Bioruptor Pico Sonicator (Diagenode) for 8’ 30”ON/OFF cycles at 4 °C. Immunoprecipitation was performed with the α-H3K27Ac antibody (Diagenode C15410174) at a 1/500 dilution, 5 mM of Na-Bu was added to all buffers. Before sonication, magnetic beads were pre-cleared with 30 μL of Protein G beads (for H3K27ac – Invitrogen 10003D) and 0.25% BSA in PBS. After the addition of the antibody, the beads were left to rotate at 4 °C for at least 4 h. Unbound antibodies were removed, and following sonication, the chromatin was added to the beads and incubated rotating overnight at 4 °C. Unbound chromatin was then removed by seven washes in RIPA buffer and one in TE buffer. Chromatin was eluted and de-crosslinked overnight with the addition of 5 μL Proteinase K (10 mg/mL, Promega V3021). RNase A (4 μL, 10 mg/mL, ThermoFisher EN0531) treatment followed, and then phenol:chloroform:IAA extraction and precipitation. Chromatin was eluted in 50 μL H₂O.

Library preparation and sequencing

Library preparation was performed by the iGE3 Genomic Platform. The Illumina ChIP TruSeq protocol was followed with a < 10 ng DNA input, and libraries were sequenced as 50 bp single-end reads with the Illumina NovaSeq 6000 sequencer. Libraries were validated on Tapestation and Qubit fluorimeter, pooled as 2 nM, and sequenced with TruSeq SBS chemistry.

Capture-HiC and library preparation

C-HiC experiments were conducted as singlets using an average of 1×10⁶ fixed nuclei for sorted cells and 3×10⁶ mESC cells. The experiments adhered to the protocol outlined in Kragesteen et al., 2018, and Paliou et al., 2019. In this process, chromatin underwent digestion with the DpnII enzyme (1000U total; NEB, R0543M) at 37 °C overnight, supplemented with 20% SDS and 20% Triton X-100. Subsequent ligation was carried out with 100U of ligase in a 1.15% Ligation buffer (ThermoFisher, EL0012) at 16 °C for 4 h, followed by 30 min at room temperature. The decrosslinking step occurred overnight at 65 °C with the addition of 30 μL Proteinase K (10 mg/mL, Promega V3021). RNAse A treatment (30 μL, 10 mg/mL, ThermoFisher EN0531), 45’ at 37 °C, was followed by phenol:chloroform:IAA extraction and an overnight precipitation. After precipitation, the DNA pellet was reconstituted in 150 μL Tris pH7.5. Total DNA quantification was performed using the Qubit High Sensitivity DNA Assay (Q32851).

Preparation of 3 C library and sequencing

Libraries were prepared by the iGE3 Genomic Platform. In brief, chromatin was sheared, and adapters were ligated following the manufacturer’s protocol for Illumina sequencing (Agilent). Libraries underwent pre-amplification and hybridization on custom Sure Select beads spanning the chr13: 54,000,001–57,300,000 region, indexed for sequencing as 50 bp paired-end reads (Agilent). Once again, 2 nM of libraries were clustered for sequencing on an Illumina Novaseq 6000 with SBS TruSeq chemistry.

Data analysis

RNA-seq.⁴⁵(Brawand et al. ⁴⁵)RNA-seq reads were processed using CutAdapt v1.18 to trim low-quality bases and NextSeq sequencing adapters (-a CTGTCTCTTATACACATCTCCGAGCCCACGAGAC, quality cutoff -q30 and minimum length required -m15). Unstranded reads were mapped to the relevant GRCm39/mm39 custom genome filtered GTFs (see Custom Genomes section below) with using the STAR 2.7.2b mapper with settings allowing for accurate gene quantification (--outSAMstrandField intronMotif--sjdbOverhang ‘99’ -- sjdbGTFfile $gtfFile--quantMode GeneCounts--outFilterType BySJout--outFilterMultimapNmax 20 -- outFilterMismatchNmax 999--outFilterMismatchNoverReadLmax 0.04--alignIntronMin 20 -- alignIntronMax 1000000--alignMatesGapMax 1000000--alignSJoverhangMin 8 -- alignSJDBoverhangMin 1). Output BigWig files were displayed on the UCSC genome browser. Counts were compiled from STAR counts using R 3.6.2, and FPKM were computed through Cufflinks 2.2.1 using the filtered GTFs created for this study (--max-bundle-length 10000000 -- max-bundle-frags 100000000 -- multi-read-correct--library-type “fr-firststrand” --no-effective-length-correction -M MTmouse.gtf). The code is available on https://github.com/bompadreolimpia/Bompadre_etal_2024.

Count normalisation and differential expression analysis was done following published pipelines (https://github.com/lldelisle/rnaseq_rscripts). Briefly, normalized FPKM values were calculated by first determining coefficients extrapolated from a set of 1,000 housekeeping genes known for their stable expression as defined from the comparison of a series of RNA-seq (Brawand et al. ⁴⁵). The coefficients obtained were then applied to adjust the respective FPKM values. Differential expression analysis utilized the DEseq2 R package (version 1.38.3), with the Wald test for comparisons across samples and multiple test correction using the FDR/Benjamini-Hochberg test. Each analysis included two biological replicates per condition. Fold-enrichment of Pitx1 and was calculated using DEseq2’s normalization by size factor. Custom Genomes. For RNA-seq analysis, custom mm39 genomes were generated using STAR 2.7.2b, incorporating an additional chromosome to accommodate the custom sequences of EGFP and SV40 polyA tails. The gft file was modified to specify these sequences as coding genes and exons. Cell Ranger 6.1.2 was utilized for single-cell RNA-seq analysis, creating a custom mm39_dCas9P300 genome by adding an extra dCas9P300-containing chromosome and customizing the reference gtf file. The code is available on https://github.com/bompadreolimpia/Bompadre_etal_2024.

ChIP-seq. Reads from ChIP-seq data generated for this study were pre-processed withCutAdapt v1.18 to trim low quality bases and TruSeq adapters (-a GATCGGAAGAGCACACGTCTGAACTCCAGTCAC, -q30 and -m15). Reads were then mapped to the reference GRCm39/mm39 genome using Bowtie2 2.3.5.1 with default settings. Only reads with mapping quality score (MAPQ) of 30 or above were retained by filtering with SAMtools v1.10. For coverage and peak analysis reads were extended by 200 bp and processed with MACS2 v2.2.7.1 (--broad --nolambda --broad-cutoff 0.05 --nomodel --gsize mm --extsize 200 -B 2). Coverage normalization was performed by MACS2, normalized by the number of million tags used by MACS2. BedGraphToBigWig v4 was used to convert files into BigWig format for visualization in the UCSC browser. Datasets of CTCF ChIP-seq of mouse embryonic E11.5 forelimb (Andrey et al., 2017) and human fetal limbs day 58/59 (ENCODE, annotation file set ENCSR191WSJ) were visualized in the UCSC browser. The code is available on https://github.com/bompadreolimpia/Bompadre_etal_2024.

Capture-HiC. Capture-HiC data analysis was done as follows. Paired-end reads were mapped against the reference NCBI37/mm9 genome using Bowtie2 v2.3.4.2 for mapping of short reads. Filtering, de-duplication, and processing of valid and unique di-tag pairs were performed with HiCUP v0.6.1 with default parameters for configuration file, but adding Nofill: 1 as an additional parameter⁴⁶. Valid and unique read pairs were then processed with Juicer Tools v1.9.9 to produce binned contact maps with MAPQ ≥ 30 and maps were normalised using Knights and Ruiz matrix balancing, considering only the genomic region chr13: 54,000,001–57,300,000 (length 3,300,000 bp), and exported at 5 kb resolution^46,47,48. All maps were produced on the wildtype reference genome in order to use the same genomic coordinates across all samples Subtraction maps of the KR normalised maps were scaled together across their subdiagonals to normalise for distance-dependant signal decay. Here, each subdiagonal vector in a matrix is divided by its sum and multiplied by the average of the sums in the two matrices. All maps were visualised as heatmaps where values above the 99^th percentile were truncated for visualisation purposes. For further details can be found in ref. ⁴. The codes are available on https://github.com/bompadreolimpia/Bompadre_etal_2024.

Single Cell RNA-Seq. Sequenced reads were mapped to the custom genome mm39_dCas9P300 and corresponding GTF file using the 10X Genomics Cell Ranger 6.1.2 software. Data filtering (nFeature_RNA > 200 & nFeature_RNA < 5000 & percent.mt <5 & nCount_RNA > 1000 & nCount_RNA < 26000), quality control, normalization, scaling, dimensional reduction, and doublet identification were performed using Seurat 4.3.0 and DoubletFinder 2.0.3. Cells were further filtered to exclude blood cells present in our dataset (percent.mt > 1 & percent.mt <5).

Merging and Normalization. Following individual dataset filtering and normalization, the two wildtype forelimb and hindlimb replicates and the two Pitx1^Inv1/+ forelimb replicates were merged as single Seurat objects. To account for potential variance due to cell-cycle variations, cell cycle regression was implemented using the CellCycleScoring method with a predetermined list of marker genes (Tirosh et al., 2016). The dataset underwent additional normalization through SCTransform with standard parameters, incorporating the scored cell-cycle and the dCas9P300 feature as regressed variables (Hafemeister et al., 2019).

Clustering of Whole Limbs and Mesenchyme. The cells were clustered after cell cycle and dCas9P300 regression using the SCTransform Seurat package. For clustering, PCA (50 npcs) and UMAP (50 dims) were utilized, and the closest neighbors of each cell were calculated. The Seurat FindClusters function was employed with a resolution of 0.1, defining 9 clusters. Cluster identification was performed with the FindMarkers function, enabling the selection of differently expressed gene markers among clusters (ident.1, only.pos=TRUE). Three mesenchymal cell clusters and two epithelial cell clusters were merged, with 6 final clusters remaining, where the FindMarkers function was re-run (Supplementary Data 4).

Given the exclusive expression of Pitx1 and Shox2 in the mesenchymal cells of the limb, downstream analysis focused on these populations. The 3 mesenchymal cell populations were merged and reclustered. PCA of 20 npcs and UMAP of 20 dims were applied, and closest neighbours were calculated for each cell. Using Seurat FindClusters, 10 clusters were defined with a resolution of 0.3. FindMarkers was then run for each cluster, selecting gene markers (ident.1, only.pos=TRUE). Two Irregular Connective Tissue clusters were merged and FindMarkers was re-run on the final 9 clusters (Supplementary Data 4). Analysis of Pitx1 and dCas9P300 expression in wildtype and Shox2^dCas9P300/+;ColA1^TSSsgR/+ forelimbs was restricted to these limbs only which were subset after cluster analysis. UMAP density plots were obtained using the R package Nebulosa v1.8.0 and scTransform v0.4.1. The codes are available on https://github.com/bompadreolimpia/Bompadre_etal_2024.

Expression correlation. To calculate the correlation of expression of two genes in a sample from single-cell-RNAseq data we employed baredSC v2.0.0 (Lopez-Delisle, et al., 2022). Here, the confidence interval of correlation is given as a percentage and the p-value, where p is the probability for the correlation coefficient to be negative, is the mean probability with the estimated standard deviation of this mean probability. The code is available on https://github.com/bompadreolimpia/Bompadre_etal_2024.

Reporting summary

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