Integration of 3D genome topology and local chromatin features uncovers enhancers underlying craniofacial-specific cartilage defects

Abstract

Aberrations in tissue-specific enhancers underlie many developmental defects. Disrupting a noncoding region distal from the human SOX9 gene causes the Pierre Robin sequence (PRS) characterized by the undersized lower jaw. Such a craniofacial-specific defect has been previously linked to enhancers transiently active in cranial neural crest cells (CNCCs). We demonstrate that the PRS region also strongly regulates Sox9 in CNCC-derived Meckel’s cartilage (MC), but not in limb cartilages, even after decommissioning of CNCC enhancers. Such an MC-specific regulatory effect correlates with the MC-specific chromatin contacts between the PRS region and Sox9, highlighting the importance of lineage-dependent chromatin topology in instructing enhancer usage. By integrating the enhancer signatures and chromatin topology, we uncovered >10,000 enhancers that function differentially between MC and limb cartilages and demonstrated their association with human diseases. Our findings provide critical insights for understanding the choreography of gene regulation during development and interpreting the genetic basis of craniofacial pathologies.

Lineage-dependent 3D chromatin topology instructs the differential usage of enhancers in cartilages from different body locations.

INTRODUCTION

Cis-acting regulatory elements, such as enhancers, regulate their target genes from a distance and are key determinants for the spatiotemporal gene expression programs during development (1–4). Systematical epigenetic profiling has revealed that the number of putative enhancers in the mammalian genome greatly exceeds the number of genes (5, 6). Different subsets of enhancers can be used during different developmental stages and in different cell types, thereby creating tissue-specific transcriptional regulatory networks (3, 4, 7). Deciphering when and where the enhancers elicit their regulatory functions is thus critical for understanding how the mammalian development processes are orchestrated, as well as how the mutations in noncoding sequences lead to tissue-specific developmental defects (8–10).

The activities of enhancers are usually assessed using a collection of chromatin signatures including high chromatin accessibility, p300 occupancy, histone H3 lysine 27 acetylation (H3K27Ac), and H3 lysine 4 monomethylation (2). However, these local chromatin signatures are not necessarily indicative of the probabilities of enhancer-promoter interactions within the three-dimensional (3D) nuclear space, which are mediated by the cohesin complex and CTCF through a loop extrusion mechanism (11, 12). Therefore, interpreting the functions of the putative enhancers requires the incorporation of 3D chromatin interaction information, which can be measured using the chromosome conformation capture (3C)–based approaches (13, 14). Such a concept has been recently adopted to develop an activity-by-contact (ABC) model (15). The ABC model predicts the enhancer-promoter regulatory relationships across 131 human tissues and cell lines at higher accuracy compared to the traditional local chromatin signature-based methods (9), highlighting the importance of evaluating tissue-specific enhancer functions within the context of 3D genome architecture.

The mammalian SOX9 (SRY-box transcription factor 9) locus provides a fascinating model for dissecting the developmental and pathological functions of tissue-specific enhancers. SOX9 is a critical transcription factor required for the homeostasis of stem cells and the development of cartilage and skeleton (16–18). The large gene desert upstream of the SOX9 gene contains numerous putative cis-regulatory elements. Translocations at the proximal noncoding region of SOX9 interrupt the interactions between the SOX9 promoter and most of its cis-regulatory elements, leading to a semilethal bone dysplasia called campomelic dysplasia (Online Mendelian Inheritance in Man [OMIM] catalog number: 114290) that is characterized by a wide range of clinical manifestations including undergrowth of the lower jaw (micrognathia), sex reversal, and the bowing of long bones (19–21). In contrast, translocations, deletions, and duplications occurring further upstream of SOX9 usually lead to more tissue-restricted phenotypes. Of particular note, structural variations occurring more than 1 Mb upstream of SOX9 have been linked with the Pierre Robin sequence (PRS; OMIM 261800), a severe congenital craniofacial malformation that is characterized by the triad of micrognathia, U-shaped cleft palate, and glossoptosis and can lead to life-threatening breathing and feeding difficulties in infants (22, 23). These findings strongly indicate that the craniofacial-specific regulatory elements for the SOX9 gene are located within its distal upstream region (24, 25).

A recent study provides critical mechanistic insights into understanding the molecular underpinnings of PRS (26). Whereas most cartilage and skeletal tissues in mammals originate from the mesoderm, the craniofacial skeletons arise from the multipotent cranial neural crest cells (CNCCs) (27). During craniofacial morphogenesis, CNCCs migrate to mandibular prominences and differentiate into a pair of rod-shaped cartilages called the Meckel’s cartilage (MC), which eventually mature into the skeleton of the lower jaw. It has been revealed that two clusters of distal enhancers located within the PRS region exhibit high activities in human CNCCs (hCNCCs) (26). These distal SOX9 enhancers are activated in a transient, neural crest–specific manner and are quickly decommissioned in CNCC-derived chondrocytes. On the basis of these observations, it has been proposed that the disruption of the PRS region leads to insufficient expression of SOX9 in CNCCs, thereby perturbing the proliferation and development of CNCCs, ultimately causing mandible-specific malformations. However, deletion of the region orthologous to the strongest neural crest enhancers within the human PRS region led to mild mandibular abnormalities in mice (26), raising the possibility that other cis-elements located within the SOX9 distal upstream region may also contribute to the PRS etiology by regulating the further downstream processes during the development of the lower jaw, such as the formation and growth of MC.

In the present study, we further dissected the intricate tissue-specific transcriptional regulatory mechanisms at the SOX9 locus. We show that the deletion of a 278-kb region orthologous to the human PRS region at the mouse Sox9 locus resulted in evident underdevelopment of the lower jaw. Using this mouse model, we showed that the loss of the PRS region in mice led to substantial down-regulation of Sox9 in mandibular MC but not in limb cartilages, suggesting that the PRS region elicits strong, craniofacial-specific regulatory functions even after the differentiation of CNCCs into cartilage tissues. The MC-specific regulatory functions of the PRS region cannot be attributed to the previously identified neural crest–specific enhancers, whose activities are largely diminished in cartilage tissues. Instead, the PRS region preferentially interacts with the Sox9 promoter in MC, suggesting that such a chromatin topology may enable the functions of weak enhancers within the PRS region in MC. Using an ABC model-based enhancer scoring system that integrates local chromatin signatures and 3D chromatin contact probabilities, we predicted multiple candidate enhancers that preferentially regulate Sox9 expression in MC or limb cartilages and validated their cartilage type–specific functions by performing CRISPR-dCas9–based functional perturbation (28). We further implemented the enhancer scoring system at the genome-wide level to systematically compare the enhancer regulatory landscapes between mouse MC and forelimb (FL) cartilages. We uncovered many cartilage type–specific enhancers and revealed the association of their human orthologs with human diseases and traits. Together, using the Sox9 locus as a model, we show that the lineage-specific 3D chromatin topology and the local enhancer properties may collectively determine the tissue-specific enhancer functions. The compendium of cartilage type–specific enhancers offers valuable resources and insights for interpreting the genetic basis of craniofacial morphological variations and defects.

RESULTS

Deletion of a 278-kb noncoding region far upstream of Sox9 causes craniofacial-specific development defects in mouse

To dissect the genetic underpinnings of the mandibular defects associated with the PRS, we focused on a ~280-kb noncoding region located 1.17 to 1.45 Mb upstream of the human SOX9 gene (Fig. 1A), which represents the overlaps among structural variants detected in multiple PRS patients (23, 25, 29–33). The human PRS region is orthologous to the noncoding region located 1.23 to 0.95 Mb upstream of the mouse Sox9 gene (chr11: 111,555,526 to 111,833,044 in mm10; Fig. 1B), which we refer to as the mouse PRS region from here on.

Fig. 1. Deletion of a 278kb genomic region far upstream of Sox9 in mouse led to underdevelopment of the lower jaw.

(A) The PRS region (shaded) was defined by the overlaps among deletions (23, 25, 29–32) (red) or translocations (23, 33) (blue) detected in multiple PRS patients. SRO, the smallest region of overlap; TBC, translocation breakpoint cluster. (B) The 278-kb region in mouse (chr11: 111,555,526 to 111,833,944, mm10) orthologous of the human PRS region is zoomed and highlighted. Sequence conservation between mouse and human is shown. Arrow, the Peak16 enhancer. (C) 3D reconstruction of mandibles from E18.5 Peak16^−/−, Peak16^+/−, and wild-type (WT) embryos using MicroCT scans. Scale bars, 500 μm. (D) Boxplots summarizing the morphological changes in the mandible in Peak16 deletion embryos. Mandibular body length (2 to 13 and 4 to 9, the numbers correspond to the anatomic landmarks used for quantifying mandibular morphology as indicated in fig. S1B) and coronoid length (7 to 8) remained unchanged in Peak16^−/− embryos (n = 4) compared to Peak16^+/− (n = 6) and WT (n = 2) embryos, while condylar width (12 to 13) showed a significant decrease in Peak16^−/− embryos. (E) 3D reconstruction of mandibles from PRS^−/−, PRS^+/−, and WT mice. Note the visible shortening of the mandible in PRS^−/− mice compared to the mandible in WT as indicated by the dashed lines. (F) Boxplots summarizing the morphological changes in the mandible in PRS region deletion mice. Homozygous deletion of PRS region (PRS^−/−, n = 3) resulted in a significant reduction in mandible length, hypoplasia of coronoid process, and reduced condylar width compared to PRS ^+/− (n = 3) and WT (n = 2). For all boxplots in (D) and (F): center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range. P values (two-tailed t tests): ns, not significant, *P < 0.05, ***P < 0.001, and ****P < 0.0001.

We first examined how the deletion of the previously identified neural crest–specific Sox9 enhancers affects craniofacial development. One such candidate is Peak16 (mm628) (Fig. 1B), which was identified by p300 chromatin immunoprecipitation (ChIP) sequencing from embryonic day 11.5 (E11.5) mouse craniofacial tissue and showed lacZ expression in the mandible (32). The mouse Peak16 enhancer corresponds to a human enhancer located 1.45 Mb upstream of the SOX9 gene (EC1.45) that was highly active in hCNCCs but rapidly decommissioned upon cartilage differentiation (26). To understand the influences of the Peak16 enhancer on mandibular development, we constructed a transgenic mouse model in which a 1170-bp region encompassing the entire Peak16 was deleted (fig. S1A).

We performed micro–computed tomography (MicroCT) analysis on E18.5 embryos obtained from crossing the Peak16 heterozygous deletion mice (Peak16^+/−) to quantitatively assess the effects of Peak16 deletion on mandibular morphology (fig. S1B). We found that the deletion of Peak16, even in a homozygous setting (Peak16^−/−), resulted in only mild alterations in mandibular morphology (Fig. 1, C and D, and fig. S1C). When compared with the wild-type and Peak16^+/− littermates, the Peak16^−/− embryos exhibited a slight reduction in condylar width (Fig. 1, C and D). However, the deletion of Peak16 did not cause significant decreases in either mandibular body length or coronoid length, which are indicators for overall mandibular cartilage development (Fig. 1, C and D). These findings are concordant with the previous report that the deletion of Peak16 in mice only caused changes in mandibular morphology in a sensitized, Sox9 heterozygous deletion background (26). Thus, the deletion of the EC1.45/Peak16 enhancer in mice could not recapitulate the abnormality of patients with PRS.

Having demonstrated that the deletion of a single neural crest–specific Sox9 distal enhancer had negligible effects on mandibular development, we constructed a mouse model in which the entire 278-kb mouse PRS region was deleted (PRS^−/−; Fig. 1B and fig. S1A). Unlike the Peak16^−/− mice, the PRS^−/− mice exhibited observable decreases in mandibular body length, coronoid length, and condylar width compared to their heterozygous (PRS^+/−) and wild-type littermates (Fig. 1, E and F, and fig. S1D), thus more closely recapitulating the micrognathia phenotype in the human PRS patients. The PRS^−/− mice exhibited no detectable defects in limbs and ribcages (fig. S2), suggesting that the mouse PRS region confers craniofacial-specific regulatory function on Sox9 expression.

Notably, despite the evident micrognathia phenotype, the PRS^−/− mice did not recapitulate the other common phenotypes associated with human PRS, such as the cleft palate and glossoptosis, suggesting that the extent of mandible undergrowth in the PRS^−/− mice is not severe enough to cause these secondary defects. The different sensitivity to the reduction in Sox9 dosage in human and mouse may be attributed to the inherent differences in mandibular structure between the two species. Nonetheless, our results confirm that the PRS region harbors critical Sox9 enhancers whose disruption is sufficient to cause mandible malformation.

The deletion of the mouse PRS region led to persistent Sox9 down-regulation in neural crest–derived mandibular cartilage

The PRS^−/− mice provides a valuable model for examining how the distal PRS region influences the expression of Sox9 during craniofacial development. We first examined the expression patterns of Sox9 in different murine embryonic tissues using the publicly available data from The Encyclopedia of DNA Elements (ENCODE) (Fig. 2A). Throughout embryonic development, Sox9 is expressed at high levels in both facial prominences and limb tissues than in liver, consistent with its critical role in regulating cartilage development. Sox9 exhibits the highest expression from E10.5 to E13.5 in facial prominences and from E11.5 to E13.5 in limb. Notably, the expression of Sox9 significantly decreases in both tissues at E14.5, when cartilage has largely formed. These temporal expression patterns imply that Sox9 may be regulated via different schemes in the undifferentiated CNCCs and mesenchymal stem cells versus in cartilage.

Fig. 2. The PRS region confers a strong regulatory effect on Sox9 transcription in mandibular cartilage.

(A) Line graphs summarize the temporal expression changes of Sox9 in facial prominences, limb, and liver during murine embryonic development. The murine transcriptome data are taken from ENCODE. (B to F) RT-qPCR quantification of Sox9 transcripts in different tissues and developmental stages. Data are from three independently replicated experiments. *P < 0.05 and ****P < 0.0001. (B) The Peak16 deletion causes moderate down-regulation of Sox9 in E11.5 mandible tissues. (C) The deletion of the entire PRS region results in down-regulation of Sox9 in E10.5 mandible tissues. Both the E10.5 and E11.5 mandibles are composed of largely undifferentiated ectomesenchyme, while the MC has not emerged. (D) The Peak16 deletion does not affect Sox9 expression in E14.5 MC. (E and F) The deletion of the PRS region causes significant down-regulation of Sox9 in E14.5 MC (E) but not in E14.5 FLs (F).

Previous epigenetic profiling has demonstrated that the human PRS region contains three prominent enhancer clusters in hCNCCs, namely, EC1.45 (homologous to mouse Peak16), EC1.35, and EC1.25 from 5′ to 3′, among which EC1.45 and EC1.25 exhibited enhancer activities in reporter assays in hCNCCs (26). These findings led to a model where the abrogation of the PRS enhancers impairs the expression of Sox9 in undifferentiated CNCCs, thereby affecting their proliferation and downstream development. To explicitly test this model in mouse, we first isolated the E10.5 or E11.5 mandibular prominences, which consist of neural crest–derived ectomesenchyme that has not differentiated into cartilage, muscle, or other connective tissues, from PRS^−/− and Peak16^−/− mice, and quantified the expression of Sox9 in these primary cells using quantitative reverse transcription polymerase chain reaction (qRT-PCR). The E11.5 mandible from Peak16^−/− mice exhibited a mild, ~10% reduction in Sox9 expression compared to their heterozygous and wild-type littermates (Fig. 2B), consistent with the previously reported effect in hCNCCs (26). The deletion of the entire PRS region resulted in a more pronounced, ~25% reduction in Sox9 transcript level in the E10.5 mandible (Fig. 2C), suggesting that the PRS region harbors other functional enhancers besides Peak16.

Because of the stage-specific activities of these enhancers, one would predict that the deletion of the PRS region should have minimal impact on Sox9 expression after the differentiation of neural crest cells into cartilage. To accurately assess how the deletion of the PRS region influences Sox9 expression in mandibular cartilages, we surgically isolated the E14.5 mouse mandible and separated the MC from its surrounding tissues such as muscles and connective tissues (fig. S3, A to C). Immunofluorescence on the isolated MC cells from wild-type mice showed that >90% of cells exhibited detectable Sox9 expression (fig. S3, D to G). We found that E14.5 MC from the Peak16^−/−, Peak16^+/−, and wild-type animals exhibited similar levels of Sox9 transcription (Fig. 2D), suggesting that the activity of the Peak16 enhancer is confined to the early stage of neural crest development.

Unexpectedly, when quantifying the transcript levels of Sox9 in E14.5 MC from the PRS^−/−, PRS^+/−, and wild-type animals, we found that the MC from the PRS^−/− mice exhibited a >40% reduction of Sox9 transcripts (Fig. 2E). In contrast, the Sox9 transcript level in FLs from PRS^−/− mice was largely unaffected (Fig. 2F). Therefore, the deletion of the mouse PRS region resulted in persistent down-regulation of Sox9 in both undifferentiated CNCCs and CNCC-derived mandibular cartilages but not in the limb cartilages.

The mandibular cartilage-specific regulatory function of the PRS region does not result from tissue-specific chromatin accessibility and H3K27Ac

A previous study has suggested that the human PRS enhancer was active in hCNCCs but quickly decommissioned upon cartilage differentiation, as indicated by the marked decreases in chromatin accessibility. To understand why the mouse PRS region still confers a strong regulatory effect on Sox9 expression in mandibular cartilage, we performed Assay for Transposase-Accessible Chromatin with high-throughput sequencing (ATAC-seq) on the isolated MC cells from E12.5 and E14.5 mouse embryos (fig. S4A) and compared the chromatin accessibility landscapes in MC with the previously published ATAC-seq profiles on E8.5 mouse neural crest cells and E10.5 mandibular tissues (34). In both the E8.5 neural crest and E10.5 mandibular cells, the PRS region harbors multiple prominent ATAC-seq peaks (Fig. 3, A and B). Notably, several peaks exhibiting the highest chromatin accessibility occur at the mouse orthologs of the human EC1.45, EC1.35, and EC1.25 enhancers (mEC1.45, mEC1.35, and mEC1.25; Fig. 3, A and B), suggesting overall similar chromatin landscapes at the Sox9 locus in mouse and human. Upon the differentiation of the neural crest, in both E12.5 and E14.5 MC, the chromatin accessibility at all three enhancer clusters was largely diminished and became much lower than the enhancers proximal to the Sox9 promoter (Fig. 3, A to C). Such an observation is concordant with the decommission of the neural crest–specific enhancers observed in human chondrocytes (26). However, we observed that ~40 weak ATAC-seq peaks within the PRS region persist in MC (Fig. 3, A and B), suggesting that some of these previously uncharacterized enhancers may contribute to the regulation of Sox9 during the formation and maturation of mandibular cartilages, thereby contributing to the micrognathia phenotype.

Fig. 3. Mandibular cartilage-specific regulatory function of the PRS region cannot be attributed to differences in local enhancer signatures.

(A) Normalized ATAC-seq and H3K27Ac signals in the 1.3-Mb noncoding region upstream of Sox9. The ATAC-seq data for E8.5 NCC progenitor and E10.5 mandibular tissues are from (34). The ATAC-seq data for E12.5 MC, E14.5 MC, and E14.5 FL, as well as H3K27Ac CUT&Tag data for E14.5 MC and E14.5 FL, are generated from this study. (B and C) Zoomed-in views of ATAC-seq and H3K27Ac tracks in PRS region (B) and proximal TBC (C). (D) Scatterplot showing differential chromatin accessibility at ATAC-seq peaks in E14.5 MC and FL. Each dot represents an ATAC-seq peak within the 1.3-Mb Sox9 genomic neighborhood. Red dots, peaks located in the distal PRS region; blue dots, peaks located in the proximal TBC region; gray dots, peaks not located in the distal PRS region or the proximal TBC region. (E) Boxplot comparing the log₂ ratios of E14.5 MC and E14.5 FL ATAC-seq signals for peaks located in the proximal TBC region (blue) versus PRS region (red). Centerline, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range. n, number of peaks in each region. P value is calculated from the two-tailed t test. (F and G) Comparison of H3K27Ac signal in MC and FL with scatterplot (F) and boxplot (G) as in (D) and (E).

Having shown that the PRS region strongly influences the expression of Sox9 even after the decommission of neural crest enhancers, we next sought to understand why such a strong, long-range regulatory effect only occurs in mandibular cartilage but not in the limb cartilage. In particular, we wonder whether such a cartilage type–specific regulatory effect can be attributed to differential enhancer activities between mandibular and limb cartilages.

To systematically compare the enhancer landscapes between the two cartilage types, we performed ATAC-seq to characterize the genome-wide chromatin accessibility in E14.5 FL (fig. S4A) and also performed genome-wide profiling of active histone mark H3K27Ac in both E14.5 MC and FL (fig. S4B). We found that the two cartilage tissues exhibit overall similar chromatin accessibility landscapes and H3K27Ac profiles, concordant with their similar cellular identities. However, quantitative analyses using DiffBind revealed that 4456 of 49,333 ATAC-seq peaks (fig. S5A) and 844 of 45,840 H3K27Ac peaks (fig. S5B) exhibit higher signals in either MC or FL, suggesting that cartilage type–specific variations in enhancer signatures do exist to some extent.

Within the 1.3-Mb gene desert upstream of Sox9, a total of 233 putative enhancers that overlap with ATAC-seq peaks in either MC or FL were identified. Among the 233 putative enhancers, 20 and 6 exhibited differential chromatin accessibility (fig. S5C) or H3K27Ac signals (fig. S5D) in different cartilage types, respectively. However, the distal PRS region exhibits similar chromatin accessibility and H3K27Ac signals in both MC and FL (Fig. 3, A and B). To evaluate the cartilage type–specific activities of the 233 Sox9 enhancers, we calculated the differences in signals between MC and FL for all 233 putative enhancers within the Sox9 genomic neighborhood. We found that the enhancers located within the PRS region do not exhibit higher MC/FL ratios in either chromatin accessibility or H3K27Ac than the enhancer located within the 285-kb proximal translocation breakpoint cluster (proximal TBC) (Fig. 3, C to G) (33), whose translocation leads to broad cartilage defects. Thus, the differences in the local enhancer signatures were not sufficient to explain the mandibular cartilage-specific regulatory function of the PRS region.

The PRS region specifically interacts with the Sox9 gene in mandibular cartilage

Since enhancers exert their regulatory functions via spatially contacting their target genes, we next investigated whether the mandible-specific functions of the PRS region are caused by differences in chromatin topology in mandibular and limb cartilage. We performed high-throughput 3C (Hi-C) in two biological replicates of purified E14.5 MC and FL cells (table S1). After combining replicates, we generated Hi-C interaction heatmaps binned at 10-kb resolution for each cell type and analyzed the major large-scale chromatin organization features such as A/B compartment and topologically associating domains (TADs) by performing the eigenvector decomposition and insulation index analyses (fig. S6). At a genome-wide level, A/B compartments (fig. S6, A to C) and TADs (fig. S6, D to F) are largely invariant between mandibular cartilage and limb cartilage. The A/B compartment organization at the Sox9 genomic neighborhood also does not exhibit observable differences (Fig. 4A).

Fig. 4. Different 3D chromatin topology at the Sox9 locus in MC and FL.

(A) Eigen1 profiles at 10-kb resolution for the 1.3-Mb Sox9 genomic neighborhood in MC (top) and FL (bottom). (B) Insulation profiles at 10-kb resolution for the Sox9 genomic neighborhood in MC (red) and FL (blue). The local minima on the insulation profiles denote TAD boundaries. The TAD encompassing the Sox9 gene is largely invariant in the two cartilage types. (C and D) Hi-C interaction heatmaps at Sox9 locus in E14.5 MC (C) and FL (D) with the boundaries of the Sox9 TAD indicated by arrows. The enlarged regions indicate the chromatin interaction patterns between the PRS or proximal TBC region and the Sox9 promoter. Note that a genomic bin in the PRS region (arrowheads) exhibits higher interaction frequencies in MC than in FL. Bin size, 10 kb. (E and F) Virtual 4C profiles using the Sox9 promoter as the viewpoint in MC (E) and FL (F). Chromatin loops identified using MUSTACHE are shown above the virtual 4C plots. Blue vertical bars indicate the genomic bin within the PRS region that forms an MC-specific loop with the Sox9 promoter. (G) ATAC-seq tracks show that the MC-specific loop occurs between the mEC1.35 enhancer cluster and the Sox9 promoter.

The structural variations disrupting the TAD organization at the Sox9-Kcnj2 locus have been demonstrated to cause developmental defects such as Cooks Syndrome by rewiring enhancer-promoter interaction landscapes (35, 36). We assessed the TAD organization at the Sox9 genomic neighborhood using the insulation index approach (37). The insulation profiles revealed that the locations and strength of TAD boundaries at either side of Sox9 are the same in the two cartilage types (Fig. 4B). Hi-C analysis on MC cells purified from E14.5 PRS^−/− mice showed that the deletion of the ~280-kb PRS region did not alter the boundaries of the TAD encompassing Sox9 (fig. S7 and table S1). Thus, the greater contribution of the PRS region to the Sox9 expression in MC and the reduction in Sox9 transcription in PRS^−/− mice cannot be attributed to tissue-specific variations in TAD organization.

When taking a closer examination of the Hi-C interaction profiles at the Sox9 locus, we observed a punctate, dot-like signal in MC that corresponds to a long-range contact between the Sox9 promoter and a 10-kb genomic bin (chr11: 111,650,000 to 111,660,000) within the PRS region (Fig. 4C). The PRS-Sox9 interaction was less prominent in FL (Fig. 4D). The stronger PRS-Sox9 contact in MC was evident when comparing the virtual circular chromosome conformation capture (4C) profiles with the Sox9 promoter as the viewpoint in MC and FL (Fig. 4, E and F) and was further confirmed by performing circular chromosome conformation capture followed by sequencing (4C-seq) using the Sox9 promoter as the bait (fig. S8). The anchor of the MC-specific PRS-Sox9 loop coincides with the mEC1.35 enhancer cluster, whose chromatin accessibilities are high in the E8.5 neural crest and E10.5 mandibular mesenchyme but diminish in E14.5 MC and FL cartilages (Fig. 4G and fig. S8A). Thus, the long-range communication between the PRS region and Sox9 may have been established during early neural crest development and persists in neural crest–derived cartilages even after the decommissioning of the strong neural crest–specific enhancers. Together, these findings suggest that the lineage-specific 3D chromatin topology at the Sox9 locus may lead to the preferential usage of enhancers in different cartilage types.

Integration of 3D genome topology and local chromatin features reveals differential functions of Sox9 enhancers in MC and FL

Our results suggest that the accurate assessment of the functional contributions of Sox9 enhancers in MC and FL requires the integration of both local enhancer signatures and long-range enhancer-promoter contact probabilities in a quantitative manner. To this end, we used the recently developed ABC model (15) to evaluate the regulatory contribution of all 233 putative Sox9 enhancers within the 1.3-Mb genomic neighborhood (here named Enh1 to Enh233 from 5′ to 3′) in MC and FL (Fig. 4A). For each of the candidate Sox9 enhancers, an ABC score in either MC or FL was generated by first multiplying the local enhancer activity derived from ATAC-seq and H3K27Ac signals and the enhancer-promoter interaction frequency from Hi-C and further normalizing the score with the sum of scores for all Sox9 candidate enhancers (data S1). In FL, the enhancers exhibiting high ABC scores are enriched in the Sox9 proximal region, while the 41 putative enhancers located within the PRS region exhibit significantly lower ABC scores compared to the Sox9-proximal enhancers (Fig. 5A, blue bars). Notably, the ABC scores of these PRS enhancers increased in MC and became comparable to the Sox9-proximal enhancers (Fig. 5A, red bars). Among the PRS enhancers, Enh51 (chr11: 111,666,408 to 111,666,908) exhibits the highest ABC scores in MC, making it a promising candidate for explaining the MC-specific regulatory function of the PRS region. Although Enh51 exhibits a higher H3K27Ac signal in MC than in FL (fig. S9), the chromatin accessibility and H3K27Ac signal of Enh51 is still considerably lower than most of the Sox9 proximal enhancers even in MC (Fig. 3, A to C, and fig. S9). However, Enh51 overlaps with the anchor of the MC-specific chromatin loop (Fig. 4G and fig. S8A), suggesting that its high ABC score largely results from high contact probability with the Sox9 promoter.

Fig. 5. Identification and validation of cartilage type–specific Sox9 enhancers.

(A and B) ABC scores of putative Sox9 enhancers in E14.5 MC (red) and FL (blue) (A) and their differences (B) revealed several enhancers that exhibit higher activities in MC (asterisk) or FL (arrows), respectively. Red bar, PRS region; blue bar, proximal TBC region; arrowhead, Sox9 promoter. (C) Several Sox9 enhancers are selected for in vitro functional validation based on the ABC score differences between MC and FL. Enhancers that are predicted to be MC- or FL-specific are colored in red and blue, respectively. Negative control enhancers that exhibit low ABC scores and minimal ABC score differences in both cartilage types are colored in gray. (D and E) RT-qPCR quantification of Sox9 transcripts shows that silencing MC-specific enhancers could induce down-regulation of Sox9 expression only in MC (D), whereas inactivating Sox9 FL-specific enhancers could lead to Sox9 reduction specifically in FL (E). Data are from three independently replicated experiments. (F to I) Confirmation of tissue-specific down-regulation of SOX9 by inactivating Enh51 and Enh208 in MC and FL at the protein level by immunofluorescence. (F and H) Representative microscopic fields showing the expression level of SOX9 in MC (F) and FL (H) upon the inactivation of Enh51 or Enh208. Antibody staining signals of SOX9 are shown in green. 4′,6-Diamidino-2-phenylindole (DAPI)–stained DNA is shown in blue. (G and I) Bar graphs summarizing percentages of SOX9-positive cells in MC (G) and FL (I) quantified by microscopic imaging. Cell counting was performed for three independent biological replicates (n = 3). For (D), (E), (G), and (I), error bars are SEM. P values (two-tailed t tests): *P < 0.05, **P < 0.01, and ***P < 0.001.

To pinpoint the enhancers that preferentially exert regulatory function in one cartilage type over the other, we subtracted the ABC score for each Sox9 enhancer in MC and FL (Fig. 5B). We considered an enhancer as MC- or FL-specific when the difference in ABC scores between the two cartilage types is greater than a certain cutoff. We found that the PRS region exhibits an enrichment for MC-specific enhancers when using a differential ABC score cutoff greater than 0.0035 (fig. S9A). When using a lenient cutoff of 0.0035, 4 of 41 putative enhancers within the distal PRS regions, Enh51, Enh59, Enh69, and Enh70, are considered MC-specific enhancers (Fig. 5, B and C, and data S1). When using a stringent cutoff of 0.01, the only MC-specific enhancer within the PRS region is Enh51 (Fig. 5, B and C, ABC score difference = 0.014). Enh200, another enhancer more proximal to Sox9, was also considered MC specific (Fig. 5, B and C, ABC score difference = 0.01). In contrast, all three enhancers that exhibit the greatest FL-MC ABC score differences (Enh194, Enh208, and Enh210) are located within the proximal TBC region (Fig. 5, B and C).

We next sought to validate the cartilage type–specific regulatory activities for Sox9 enhancers using the Kruppel-associated box domain fused to dead Cas9 (dCas9-KRAB) system (28). Three groups of enhancers were selected for dCas9-KRAB perturbation experiments (Fig. 5, B and C, and fig. S9, B and C): (i) Enh51, Enh59, Enh99, and Enh200 were chosen as candidate MC-specific enhancers for validation. These enhancers all exhibit higher ABC scores in MC than in FL by at least 0.0035. Among the four enhancers, Enh51 and Enh59 fall within the PRS region, while Enh200 is located within the proximal TBC region. (ii) Enh194, Enh208, and Enh210, which exhibited the greatest FL specificity according to ABC score differences, were selected as the FL-specific enhancers to validate. (iii) Enh38 and Enh191, which exhibit low ABC scores in both MC and FL and minimal ABC score differences and are located within the distal PRS region and proximal TBC region, respectively, were selected as negative controls.

Using specific single guide RNAs (sgRNAs), we targeted dCas9-KRAB to each enhancer in freshly isolated MC and FL cells from E14.5 mouse embryos, which led to the deposition of silent epigenetic modifications at the enhancer and rendered it inactive. By performing 3C-qPCR, we showed that the 3D contact frequencies between the Sox9 promoter and enhancers were not abrogated upon targeting dCas9-KRAB to the enhancers (fig. S10), suggesting that the local activities of the Sox9 enhancers and chromosome folding topology at the Sox9 locus can be decoupled.

We then evaluated the impact of the enhancer inactivation on Sox9 expression by performing quantitative RT-PCR in both cartilage types. If an MC-specific enhancer was inactivated, then we would expect the expression of Sox9 in MC to be down-regulated, whereas the expression in FL should be affected to a lesser extent.

We found that this was the case for three of four MC-specific enhancers tested. Silencing each of Enh51, Enh59, and Enh99 in MC cells resulted in the down-regulation of Sox9 transcript levels to various extents (Fig. 5D). Of particularly note, silencing Enh51 alone resulted in ~40% down-regulation of Sox9, highlighting its critical role in regulating the mandible cartilage development. On the other hand, silencing Enh200, which is located more proximal to the Sox9 gene, led to no detectable changes in the Sox9 transcript level. Inactivating all four enhancers in FL cells did not cause significant changes in Sox9 transcription, concordant with their very low ABC scores in FL (Fig. 5E). We further confirmed the tissue-specific down-regulation of SOX9 at the protein level by performing immunofluorescence in MC and FL cells in which Enh51 was inactivated (Fig. 5, F to I). Inactivating the negative control enhancers located distal or proximal to Sox9 did not result in significant changes in Sox9 transcription. Together, these results suggest that multiple distal Sox9 enhancers, particularly Enh51, exert MC-specific regulatory function on Sox9.

We further demonstrated that the functional abrogation of Enh194, Enh208, and Enh210 caused significant Sox9 down-regulation in FL but not in MC (Fig. 5, D to I). We found that while the three tested FL-specific enhancers, particularly Enh208, also exhibit high ABC scores in MC (Fig. 5, B and C, and fig. S9B), disrupting each of these enhancers did not cause observable effects in Sox9 expression in MC (Fig. 5, D, F, and G). These results, along with the findings that the inactivation of proximal Enh200 in MC also did not cause significant effects, suggest that the distal enhancers contribute more greatly to the Sox9 regulation than the proximal enhancers in MC. One speculative explanation for this insensitivity to the abrogation of Sox9 proximal enhancers in MC is that the proximal TBC region and the distal PRS region simultaneously contact with the Sox9 promoter in a fraction of MC cells, thus enabling functional compensation between the proximal and distal enhancers in these cells (fig. S11, A and B). In support of this model, we did find that inactivation of Enh200 and Enh208 in MC cells isolated from E14.5 PRS^+/− mice embryos led to a more significant down-regulation of Sox9 (fig. S11C). Thus, the presence of the distal regulatory region may mask the functional contribution of proximal enhancers at the Sox9 locus in MC.

Together, our results suggest that the integration of chromatin topology and local enhancer signature using the ABC model could predict differential functional contributions of Sox9 enhancers in mandibular versus limb cartilages at a reasonable accuracy. Our results reveal that besides the previously annotated neural crest–specific enhancers, other previously uncharacterized enhancers also function in a cartilage type–specific manner to regulate the expression of Sox9 across different body locations.

Genome-wide prediction of cartilage type–specific enhancers in MC and FL

Our functional dissection of the enhancers at the Sox9 locus suggests that many other enhancers may function in a cartilage type–specific manner to regulate gene expression. Having established the effectiveness of the ABC scoring system in interpreting enhancer functions at the Sox9 locus, we used this approach to systematically compare the genome-wide enhancer-gene regulatory landscapes between MC and FL. Taking advantage of our unique Hi-C, ATAC-seq, and H3K27Ac datasets in MC, we identified 28,177 functional enhancers in MC and 23,643 functional enhancers in FL with an ABC score greater than 0.015 (Fig. 6A). These enhancers regulate 10,936 and 10,400 genes in MC and FL cartilages, respectively. Among the 152 genes that have been confirmed to play essential roles in mandible development (38), 76 are target genes of the MC enhancers (Fig. 6B). In contrast, among the 6472 genes that are actively expressed in E14.5 facial prominence (zFPKM > −3) but not regulated by active MC enhancers, only 16 genes are involved in mandible development (Fig. 6B). Thus, the enhancer-regulated genes are more likely to play important functions during mandible development.

Fig. 6. Genome-wide assessment of differential enhancer functions in MC and FL.

(A) Fractions of MC-specific enhancers (left, red) or FL-specific enhancers (right, blue) in all enhancers identified in MC and FL. (B) Target genes of MC enhancers show a higher degree of overlap with known genes involved in mandible development (left) than the actively expressed genes in E14.5 facial prominence not associated with any MC enhancer (nontarget genes of MC enhancers, right). P value is calculated from the chi-square test. (C) Inactivation of MC-specific and FL-specific enhancers of Ctgf, Pbx1, and Dkk1 resulted in down-regulation of target genes in a tissue-specific manner (n = 3). Error bars are SEM. P values (two-tailed t tests): *P < 0.05, **P < 0.01, and ***P < 0.001. (D) GREGOR analysis reveals differential enrichment of the SNPs associated with limb diseases/traits and craniofacial diseases/traits in FL-specific and MC-specific enhancers. SNPs associated with Crohn’s disease are used as control. Enrichment P values are calculated using GREGOR.

We next compared the ABC scores associated with each enhancer in MC and FL. We consider an enhancer as an MC- or FL-specific enhancer if the ABC score difference for the enhancer in the two tissues is greater than 0.01. Notably, the enhancers that are curated as craniofacial-specific enhancers based on the known expression patterns in the VISTA enhancer database (39, 40) exhibited greater ABC scores in MC than in FL by a median difference of 0.0098, suggesting that 0.01 is a reasonable cutoff for assessing the cartilage type specificity of enhancers (fig. S12, A and B). Using these criteria, we found 10,291 enhancers that are preferentially functional in MC (MC-specific enhancers) and 5402 enhancers preferentially functional in FL (FL-specific enhancers) (Fig. 6A and data S2 and S3). As expected, these cartilage type–specific enhancers exhibit higher H3K27Ac and ATAC-seq signals in their respective cartilage types (fig. S13A) while also interacting more frequently with their target genes (fig. S13B).

To further validate the roles of cartilage type–specific enhancers in regulating the expression of key genes for cartilage development, we dissected the regulatory landscapes in MC and FL for three genes, Ctgf, Dkk1, and Pbx1 (Fig. 6C and fig. S14), all of which play critical functions in the skeleton and cartilage development and are linked with human skeletal defects (41–45). We identified enhancers that preferentially regulate Ctgf, Dkk1, and Pbx1 expression in MC or FL (fig. S14) and used the dCas9-KRAB approach to perturb these enhancers in vitro. Silencing the MC-specific enhancers of Ctgf resulted in a near 50% down-regulation of Ctgf in MC but did not cause significant expression changes in FL (Fig. 6C). In contrast, disrupting the FL-specific Ctgf enhancer yielded the opposite effects on gene expression (Fig. 6C). Similar trends were also observed for the cartilage type–specific enhancers associated with the Dkk1 and Pbx1 (Fig. 6C). These enhancers, as well as the aforementioned cartilage type–specific enhancers at the Sox9 locus, exhibit considerable sequence conservation across mammalian species (fig. S15), implicating them in the regulation of human cartilage development and the etiology of cartilage defects.

To further explore the functional significance of the cartilage type–specific enhancers, we turned to genome-wide association studies (GWAS) to interrogate whether the enhancers specifically active in different cartilage types are linked to craniofacial- and limb-associated human diseases and traits. By performing the Genomic Regulatory Elements and GWAS Overlap algorithm (GREGOR) analysis (46), we assessed the enrichment of craniofacial- and limb-associated disease/trait single-nucleotide polymorphisms (SNPs) in the human genomic regions homologous to the mouse ML- and FL-specific enhancers. SNPs associated with Crohn’s disease were included as a control. We observed statistically significant enrichment of craniofacial-associated SNPs only in ML-specific enhancers (Fig. 6D). While the enrichment of limb-associated SNPs could be observed in both the FL-specific and the MC-specific enhancers, the P value for the former is relatively smaller, indicating stronger enrichment of limb-associated SNPs in FL-specific enhancers. No significant enrichment of Crohn’s disease–associated SNPs was observed in MC-specific or FL-specific enhancers (P > 0.01 for both cases; Fig. 6D). Collectively, these findings suggest the cartilage type–specific enhancers may have pathological implications for cartilage and skeleton defects occurring in different body locations.

DISCUSSION

Elucidating the regulatory relationship between the enhancers and the genes is critical for understanding how different cells establish their physiological identities. In this study, we comprehensively characterized enhancer-promoter regulatory landscapes in mandibular and limb cartilages by integrating the chromatin signatures indicative of local enhancer activities and the 3D genome topology. Our efforts revealed that both at the Sox9 locus and across the genome, many enhancers elicit differential regulatory functions in cartilages from different body locations, hence providing important insight for understanding how the variations in noncoding sequences could cause tissue-specific effects.

The Sox9 locus is among the most extensively studied genome regions and provides a fertile ground for dissecting the tissue-specific functions of enhancers. In this study, we focused on understanding how the distal enhancers >1 Mbp upstream of the Sox9 gene contribute to the etiology of PRS, one of the most common craniofacial malformations. By deleting the entire 278-kb PRS region, we constructed a mouse model that partially recapitulates the craniofacial-specific micrognathia phenotype of human PRS patients, thereby providing a valuable resource for investigating the regulatory functions of the PRS region during mandible development. By performing the integrative genomic analysis and the in vitro enhancer perturbation assay, we demonstrated that multiple distal Sox9 enhancers promote Sox9 expression specifically in mandibular MC but not in the limb cartilage. The MC-specific Sox9 enhancers identified by our study differ from the previously identified neural crest–specific enhancers 1.25 Mb (EC1.25) and 1.45 Mb (EC1.45), both of which exhibit transient activities in hCNCCs but are decommissioned in cartilage (26). Therefore, in addition to the proliferation and development of CNCCs, the stage of MC formation and maturation presents another window for PRS-associated defects to manifest.

Our dissection of the putative enhancers at the Sox9 locus suggests that different mechanisms could contribute to the tissue specificity of enhancers. While some Sox9 enhancers do exhibit cartilage type–dependent differences in local chromatin signatures including chromatin accessibility and H3K27Ac levels, the mandibular-specific functions of several distal Sox9 enhancers, particularly Enh51, are more likely caused by an MC-specific chromatin loop that brings the PRS region and the Sox9 promoter to proximity. The anchor of the extremely long-range PRS-Sox9 chromatin loop coincides with a genomic region that is orthologous to the neural crest–specific human EC1.35 enhancer and exhibits high chromatin accessibility in undifferentiated neural crest in mouse. It has been previously shown that the human EC1.35 enhancer harbors a CTCF binding site and interacts weakly with the Sox9 promoter in human ES cells but interacts strongly with the Sox9 promoter in hCNCCs (26). On the basis of these lines of evidence, we propose a model where the activation of neural crest–specific enhancers within the PRS region promotes the establishment of a lineage-specific 3D chromatin topology in CNCCs, which persist in the ensuing stages regardless of the activities of the neural crest–specific enhancers, thereby resulting in the preferential usage of enhancers within the distal PRS regions in CNCC-derived tissues (Fig. 7). Concordant with this model, MC-specific chromatin interactions coinciding with enhancers highly active in the early stage of neural crest development were also observed at other genomic loci (fig. S16). The factors and mechanisms underlying the establishment and maintenance of the lineage-specific chromatin topology, and how the enhancer activities could induce the rewiring of long-range chromatin interactions, will be a subject of our future investigation.

Fig. 7. A model for differential regulatory functions of PRS region in cartilages originating from different cell lineages.

A simplified model that illustrates how the PRS region may interact with the Sox9 promoter in a lineage-dependent manner to instruct the enhancer usage. In neural crest cells, multiple strong enhancers within the PRS region emerge. At the same time, the interaction between the PRS region and the Sox9 gene is established. In neural crest–derived mandibular cartilage, although the activities of the strong neural crest enhancers diminish, the interaction between the PRS region and the Sox9 gene is maintained. As a result, the weak enhancers within the PRS region still exert a strong regulatory function on Sox9 expression. In mesoderm-derived limb cartilage, the enhancers within the PRS play a less significant role in Sox9 regulation due to the absence of long-range chromatin contact.

We note that more experimental evidence is required to fully validate our proposed model. First, while the ABC model uses both the chromatin accessibility and H3K27Ac signals to generate scores to indicate the enhancer strength, it is still unclear to what extent either of these two chromatin signatures can be translated into the regulatory capacity of enhancers. Examining all putative enhancers at the Sox9 locus using the reporter-based assay should lead to a better understanding of how to quantitatively integrate these chromatin signatures and clarify whether the weak enhancers at the Sox9 locus have the ability to regulate Sox9 expression. Second, assays that perturb the enhancer-promoter contacts at the Sox9 locus without abrogating the local enhancer properties, such as deleting the CTCF binding sites near the enhancers, need to be performed to explicitly assess the contribution of chromatin topology to the enhancer functions. Third, while we validated several putative enhancers that were predicted by the ABC model to be the most tissue-specific enhancers using the dCas9-KRAB–based assay, a more thorough evaluation of the rest of the Sox9 enhancers is still needed to derive a quantitative model for the regulation of Sox9. The systematical functional interrogation of all 233 putative enhancers, as well as the CTCF binding sites, at the Sox9 locus by performing not only the dCas9-KRAB–based CRISPRi assay but also the CRISPR-Cas9–based genomic sequence deletion would ultimately elucidate the complex regulatory regime of Sox9. We anticipate that these efforts would lead to a more accurate model that quantitatively integrates the 3D chromatin topology and the local enhancer properties to deduce enhancer functions in the future.

The in vitro perturbation of Sox9 enhancers revealed complex functional relationships between the cartilage type–specific Sox9 enhancers. For instance, the ABC scores integrating both the chromatin interaction probability and local enhancer properties suggest that several Sox9-proximal enhancers, although exhibiting higher function scores in FL than in MC, should be actively involved in the regulation of Sox9 in both cartilage types. However, we found that disrupting each of three Sox9 proximal enhancers only caused a reduction of Sox9 expression in the FL but had negligible effects on Sox9 expression in MC. In contrast, the perturbation of Sox9-distal enhancers in MC led to significant Sox9 down-regulation. These results suggest a model in which the functions of Sox9-proximal enhancers may be compensated by more distal enhancers but not vice versa. One possible explanation for such an apparent epistatic effect between different sets of enhancers is that the Sox9 locus could adopt multiple different folding topologies within the MC cartilage population, which may be heterogeneous to some extent and contain multiple intermediate differentiation states, and certain chromatin topology allows mutual compensation between enhancers. Further verification of such a model would require characterizing the heterogeneous chromatin topologies at the Sox9 locus and correlating them with the transcriptional levels of Sox9 at the single-cell level (47–49). Moreover, multiplexed CRISPRi screening that simultaneously perturbs two or more enhancers can be conducted to elucidate the functional relationship among the proximal enhancers, among the distal enhancers, and between the proximal and distal enhancers (50). Nonetheless, our results corroborate the notion that different enhancers targeting the same genes could function in a cooperative, cumulative, or redundant manner to ensure the precision and robustness of gene regulation.

In summary, our work demonstrates that the cells with highly similar physiological characteristics but originating from different developmental lineages could exhibit substantial differences in enhancer regulatory landscapes, which can be partly attributed to the lineage-dependent chromatin topology. Our findings highlight the importance of integrating higher-order chromatin structure information and local chromatin features to interpret the complex regulatory functions of enhancers. The compendium of cartilage type–specific enhancers generated by our study provides important resources and offers unique insights for understanding the molecular basis of cartilage-related diseases and craniofacial malformations.

METHODS

Animals

C57BL/6J mice were housed under controlled environmental conditions with free access to water and food, with constant ambient temperature (22° ± 2°C) and humidity (55 ± 10%), and an alternating 12-hour light/12-hour dark cycle. Every effort was made to minimize and refine the experiments to avoid animal suffering. All animal experimental procedures were approved by and performed following the guidelines of Shanghai Jiao Tong University School of Medicine.

Deletion of Peak16 enhancer and PRS region in mouse using CRISPR-Cas9

Both the Peak16 enhancer deletion (Peak16^−/−) and PRS region deletion (PRS^−/−) mouse models were generated in C56BL/6 background by Cyagen Biosciences Inc. (Guangzhou, China) using CRISPR-Cas9 technology. Briefly, guide RNA targeting Peak16 enhancer (chr11: 111,568,523 to 111,569,439, mm10) or PRS region (chr11: 111,555,526 to 111,833,044, mm10) was coinjected with Cas9 mRNA into fertilized mouse eggs to generate a targeted knockout offspring. Genotyping of founder mice was performed by PCR, and subsequent sequencing of DNA was extracted from mouse tails. Primers for genotyping are listed in table S2.

Mandibular morphology analysis using MicroCT

The Peak16^−/− mouse embryos were collected at E18.5 and genotyped before fixation in 4% paraformaldehyde. PRS^−/− mice were collected at postnatal day 0 (P0), genotyped, and then fixed in 4% paraformaldehyde. MicroCT scanning of mouse heads was performed using Quantum GX (PerkinElmer, USA). 3D reconstruction of MicroCT data in Dicom format was performed with Mimics Medical 21.0 (Materialise, Belgium). Hemimandibles were segmented, and anatomic landmarks of the mandible were placed following previously reported procedures (51). Distances between mandibular landmarks that represent parameters including mandibular body length, coronoid length, and condylar width were measured, and the statistical significances were evaluated using the t test.

Whole-mount skeletal staining

Mice were collected following euthanasia, washed in PBS, and then fixed in 95% ethanol overnight at room temperature. Staining with Alcian blue solution (0.03%; Sigma-Aldrich) was performed overnight at room temperature. After washing twice in 70% ethanol, mice were then replaced with 1% KOH overnight and subsequently stained with alizarin red solution (0.005%; Sigma-Aldrich) for 3 to 4 hours at room temperature. Alizarin red stain was removed and replaced with 1% KOH for 12 hours for clearing of the samples. Skeletons were transferred to 50% glycerol:50% (1%) KOH solution at room temperature until tissue appears transparent and kept in 100% glycerol for long-term storage (52).

Microdissection and culture of MC and FL

E14.5 MC bars were dissected from wild-type C57BL/6J mouse embryos following the previously published protocols (53, 54) with modifications. Briefly, the head of an embryo was separated from the body under a stereomicroscope (Olympus), and the lower jaw was then removed from the head using fine forceps. The tongue and surrounding connective tissues of MC were also removed to expose MC. Dissected MC was then digested with 0.25% trypsin for 20 min at 37°C in Dulbecco's Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F-12) and was further quenched with fetal bovine serum (FBS). Cartilage was then digested with 0.15% collagenase type II (Sigma-Aldrich) at 37°C with interval pipetting until there was no visible tissue, and the single-cell suspension was acquired by filtering cells through a 40-μm strainer before centrifugation at 1500 rpm for 5 min. The cell pellet was resuspended with DMEM/F-12 supplemented with 10% FBS and 1% penicillin/streptomycin for further application. E12.5 mandibular processes were also dissected with similar procedures. As no cartilage bar structures could be visible at this time point, the entire mandibular arch was separated from the tongue and was then digested with trypsin and collagenase type II to acquire single-cell suspension for downstream library preparations.

E14.5 FL was dissected from wild-type mouse embryos, and the surrounding connective tissues were removed. The dissected FL cartilage was then digested following the same procedure as the digestion of MC. Both MC and FL were plated onto six-well culture plates at a density of 5 × 10⁵ cells per well and were cultured in DMED-F12 containing 10% FBS and 1% penicillin/streptomycin.

ATAC sequencing

ATAC-seq was performed as described previously with the TruePrep DNA Library Prep Kit V2 for Illumina (Vazyme TD501) (55). Briefly, 50,000 cells were collected at 500g for 5 min, resuspended in 50 μl of cold ATAC-seq lysis buffer [10 mM tris-HCl (pH 7.4), 10 mM NaCl, 3 mM MgCl₂, and 0.1% (v/v) Igepal CA-630], and incubated on ice for 10 min. Following centrifugation at 500g at 4°C for 5 min, cells were pelleted, resuspended in 50 μl of transposition mix (10 μl of 5× TTBL, 5 μl of TTE Mix V50, and 35 μl of double-distilled water), and then incubated at 37°C. The reaction was purified with VAHTS DNA Clean Beads and PCR-amplified for 15 cycles. Purified ATAC-seq libraries were sequenced on the NovaSeq platform.

H3K27Ac CUT&Tag

CUT&Tag was performed as previously described (56) with the Hyperactive In Situ ChIP Library Prep Kit for Illumina kit (Vazyme, TD903) following the manufacturer’s instructions. Briefly, a total of 50,000 cells were harvested, washed, and incubated with 10 μl of activated Concanavalin A beads per sample at room temperature for 10 min. Cells bound in beads were then incubated with primary antibodies (H3K27Ac, Abcam; immunoglobulin G rabbit, Abcam) overnight at 4°C, followed by binding with secondary antibody at room temperature for 1 hour. After washing twice with 800 μl of Dig-wash buffer, the pA/G-Tn5 adapter complex was added to the sample in 100 μl of Dig-300 buffer to a final concentration of 1:250 and incubated on a nutator at room temperature for 1 hour. Tagmentation was then performed at room temperature for 1 hour and stopped with 10 μl of 0.5 M EDTA, 3 μl of 10% SDS, and 2.5 μl of proteinase K (20 mg/ml). Template DNA was extracted with DNA extraction beads, followed by PCR amplification and Illumina sequencing.

In situ Hi-C

In situ Hi-C was performed as previously described (57, 58). Briefly, cells were cross-linked in 1% formaldehyde for 10 min at room temperature and quenched with 0.2 M glycine. Cells were pelleted, washed with 1 ml of Hanks’ balanced salt solution (HBSS), then resuspended with 1 ml of ice-cold Hi-C lysis buffer [10 mM tris-HCl (pH 8.0), 10 mM NaCl, and 0.2% Igepal CA-630 with cOmplete Protease Inhibitor Cocktail], incubated on ice for 30 min, and were further dounced with pestles to ensure complete lysis. Lysates were pelleted at 2500g for 5 min and washed with 1× New England Biolabs (NEB) Buffer 3.1, and chromatin was then opened up at the presence of 0.5% SDS at 65°C for 5 min and quenched with Triton X-100 at the final concentration of 1%. Samples were then digested overnight with Dpn II at 37°C on a thermomixer with interval shaking (shaked at 950 rpm for 10 s with 5-min intervals). Dpn II was then heat-inactivated, and 60 μl of fill-in master mix (37.5 μl of 0.4 mM biotin-14–2′-deoxyadenosine 5′-triphosphate,1.5 μl of 10 mM 2′-deoxycytidine 5′-triphosphate,1.5 μl of 10 mM 2′-deoxyguanosine 5′-triphosphate, 1.5 μl of 10 mM 3′-deoxythymidine 5′-triphosphate, 10 μl of Klenow Fragment Exo minus (5 U/μl), 6 μl of 10× NEB Buffer 3.1, and 2 μl of ddH₂O) was added to the samples, which were incubated at 37°C for 4 hours in thermomixer to fill in the restriction fragment overhangs and mark the DNA ends with biotin. Samples were then ligated with T4 DNA ligase at 16°C for 4 hours, and cross-link reversal was performed by proteinase K digestion. 3C libraries were then extracted and purified by phenol-chloroform isoamyl alcohol extraction and ethanol precipitation. The efficiency of digestion and ligation was assessed by gel electrophoresis of quality controls. 3C libraries were then sheared to 200 to 700 bp by Covaris sonication, and Dynabeads MyOne Streptavidin T1 was used to perform biotin pull-down after washing. End repair, dA tailing, and ligation of libraries were performed with the NEBNext Ultra II Ligation Module. Libraries were amplified on the streptavidin beads using Phusion DNA Polymerase, size-selected with AMPure XP beads, quantified by Qubit double-stranded DNA (dsDNA) High-Sensitivity Assay, and then sequenced on the NovaSeq platform.

CRISPR interference of candidate enhancer regions

The puromycin resistance cassette of pLV-hU6-sgRNA-hUbC-dCas9-KRAB-T2a-Puro (Addgene plasmid #71236) was replaced with enhanced green fluorescent protein (eGFP) cassette with PCR (dCas9-KRAB-eGFP plasmid). sgRNA sequences targeting MC-specific and FL-specific enhancers of Sox9, Ctgf, Dkk1, and Pbx1 were designed with CHOPCHOP (59) (http://chopchop.cbu.uib.no/) and were provided in table S3. sgRNAs were cloned into the dCas9-KRAB-eGFP plasmid using Bsm BI sites. Lentivirus production was performed as previously described with modifications (28). Human embryonic kidney (HEK) 293T cells were plated in a 10-cm petri dish at the density of 5 × 10⁶ cells per dish in high-glucose DMEM supplemented with 10% FBS. Eighteen hours after seeding, HEK293T cells were cotransfected with dCas9-KRAB-eGFP lentiviral expression plasmid, psPAX2 (Addgene plasmid #12260), and pMD2.G (Addgene plasmid #12259) with Lipo-8000, and the transfection medium was exchanged with 10 ml of fresh 293T culture medium 6 hours after transfection. The supernatant containing lentivirus was collected 48 hours after transfection, filtered through a 0.45-μm membrane (Millipore), and concentrated with PEG8000 precipitation. The concentrated viral supernatant was snap-frozen in liquid nitrogen and stored at −80°C.

Primary MC and FL chondrocytes were transduced with dCas9-KRAB-eGFP lentivirus in DMEM-F12 supplemented with 10% FBS, 1% penicillin/streptomycin, and polybrene (5 μg/ml). Six hours after transduction, the medium was exchanged for a fresh chondrocyte culture medium, and transduced cells were sorted by flow cytometry. Cells transduced with dCas9-KRAB-eGFP lentivirus without targeting sgRNAs were also sorted as a control for further analysis.

Real-time RT-PCR (RT-qPCR)

Total RNA from primary MC and FL chondrocytes or tissues was isolated using the Eastep Super Total RNA Extraction Kit (Promega) following the manufacturer’s instructions. For isolation of total RNA from E10.5 or E11.5 mandibular prominences, the RNeasy Mini Kit (Qiagen) was used. Complementary DNA synthesis was performed with the PrimeScript RT reagent Kit (Takara). Real-time RT-PCR (RT-qPCR) was then performed using SYBR Premix Ex Taq (Takara) with the LightCycler 96 Instrument (Roche). Primers for RT-qPCR reported in this study are listed in table S2. The expression levels of genes of interest were quantified using the delta-delta CT method with Gapdh as an internal control.

Immunofluorescence

Cells sorted by flow cytometry were plated onto sterile coverslips in 24-well plates, allowed to grow for 24 hours, and then fixed with 4% paraformaldehyde in PBS for 20 min at room temperature. Cells were then permeabilized with 0.2% Triton X-100 for 10 min, exposed to PBS containing 5% bovine serum albumin for 60 min, and subsequently incubated with SOX9 antibody (rabbit anti-mouse, Abcam, ab185230) at 1:200 dilution for 60 min and then with secondary antibodies for 60 min (Invitrogen). Mounting was performed with ProLong Gold Antifade Mountant (Invitrogen), and cells were examined under an Olympus inverted fluorescence microscope.

3C–quantitative polymerase chain reaction

3C-qPCR was performed as previously described (60). 3C libraries for MC of three different conditions (control, Sox9 MC enhancer disrupted, and Sox9 FL enhancer disrupted) were prepared following the cross-linking, digestion with Eco RI, ligation, and reverse cross-linking steps. 3C libraries were quantified with Qubit dsDNA Quantitation Assay kits (Invitrogen) and diluted to the same concentration to ensure an equal amount of input for qPCR, which was performed with SYBR Premix Ex Taq (Takara) on the LightCycler 96 Instrument (Roche). The short-range ligation products using bait primers in combination with a primer for adjacent Eco RI fragments were used for normalizing the relative interaction frequency with Sox9 promoter in the distal region and proximal region among different samples to control for differences in cross-linking and ligation efficiencies as described previously (61).

4C-seq

4C-seq libraries were prepared from microdissected E14.5 MC and FL cartilage as described previously (62). Cells were cross-linked with 1% formaldehyde for 10 min at room temperature and then treated with 0.2 M glycine. The cross-linked cells were pelleted, washed with HBSS, and incubated with lysis buffer [10 mM tris-HCl (pH 8.0), 10 mM NaCl, and 0.2% Igepal CA-630 with cOmplete Protease Inhibitor Cocktail]. Lysates were pelleted at 2500g for 5 min and washed with 1× NEB CutSmart Buffer. Chromatin was then opened up in the presence of 0.5% SDS at 65°C for 5 min and quenched with Triton X-100 at the final concentration of 1%. Samples were then digested overnight with Nlalll at 37°C. The first round of ligation was carried out in a total volume of 1.2 ml with 50 U of T4 DNA ligase at 16°C for 4 hours. The ligated products were then treated with proteinase K (20 mg/ml) overnight to reverse cross-linking and purified by phenol-chloroform isoamyl alcohol extraction and ethanol precipitation. The purified DNA was then treated with Dpn II overnight at 37°C for the second round of restriction enzyme digestion, followed by overnight ligation at 16°C. The DNA was purified again by phenol-chloroform isoamyl alcohol extraction and ethanol precipitation. The clean-up of 4C-seq libraries was then performed using a PCR purification kit (Qiagen). Quality controls of the first and second rounds of restriction enzyme digestion and ligation were performed by gel electrophoresis.

Amplification of 4C-seq libraries was carried out with two rounds of PCR with the specificity of primers checked with test PCR. The first PCR step is to amplify the fragments ligated to the Sox9 promoter viewpoint using the Expand Long Range PCR System (Roche). A total of four PCR reactions with 200-ng 4C-seq template per reaction were performed. Fifty-microliter aliquot of all pooled PCR reactions was purified with AMPure beads at 0.8× volume. The purified PCR products of the first round were amplified by universal primers containing the Illumina adapters and purified using a PCR purification kit (Qiagen). The 4C-seq libraries were sequenced on the NovaSeq platform. Primers for 4C-seq used in this study are listed in table S2.

Data analysis

RNA-seq analysis

Fastq files of RNA sequencing (RNA-seq) from E10.5 to E15.5 facial prominence, E10.5 to E15.5 limb, and E11.5 to E15.5 liver were downloaded from ENCODE (www.encodeproject.org; table S4). Reads were mapped to mm10 reference genome with STAR (version 2.7.3a) (63). mRNA abundance was calculated directly from the alignments with TPMCalculator (version 0.0.4-1) (64). Analysis of RNA-seq data from E8.5 NCC progenitor (GSE89434) was performed following the same pipeline. Transcripts per million values of Sox9 at each embryonic developmental stage and in three different tissues were then compared.

ATAC-seq analysis

E12.5 MC, E14.5 MC, and E14.5 FL ATAC-seq libraries were sequenced on the NovaSeq platform. Trim galore (version 0.6.7) (65) was used to remove adaptor sequences, and cleaned data were then aligned to the mouse genome (mm10) with bowtie2 (version 2.3.4.1) (66) (--very-sensitive -X 2000). Samtools was used for removing duplicated and mitochondrial reads in aligned bam files. For comparison of MC and FL ATAC-seq signal at Sox9 locus, filtered bam files were first normalized with bamCoverage at chr11: 111,000,000 to 112,800,000 with 50-bp bin size according to reads per kilobase per million mapped reads. Normalization was also performed at Pbx1 (chr1: 166,932,169 to 1,699,321,691), Ctgf (chr10: 23,095,441 to 26,095,441), and Dkk1 (chr19: 29,049,496 to 32,049,496). Normalized bigwig files were then compared using bigwigCompare with log₂ ratio. Wig files were generated from bigwig files with bigWigToWig for visualization in the UCSC genome browser. Peak calling was performed with Genrich (https://github.com/jsh58/Genrich, version 0.6, accessed on January 2020) using the parameters -j -p 0.05. Genome-wide differential ATAC-seq peaks in MC and FL were identified with DiffBind (version 2.0.2) (67) based on edgeR analysis. Volcano plots of ATAC-seq peaks were performed by ggplot2 with the threshold of false discovery rate of <0.05 and log₂|fold change| >1 as differential peaks in MC and FL (fig. S5A). Sample correlation of ATAC-seq replicates was also calculated with DiffBind (fig. S4A). For visualization of E8.5 NCC progenitor and E10.5 Md ATAC-Seq data, raw fastq files (GSE89436) were downloaded and analyzed following the same pipeline.

H3K27Ac CUT&Tag analysis

Adaptor sequences in CUT&Tag fastq files were trimmed with trim galore. Alignment of trimmed fastq files was performed with bowtie2 to mm10 with the following parameters: --very-sensitive-local –no-unal –no-mixed –no-discordant –phred33 -I 10 -X 700. Samtools was used for removing duplicated and mitochondrial reads in aligned bam files. The comparison of H3K27Ac signal strength between MC and FL was performed similarly to ATAC-seq. Normalized wig files were visualized in the UCSC genome browser. Peak calling was performed with SEACR (version 1.3) (68) using the parameter 0.01 nonstringent. DiffBind (67) was applied to identify differential H3K27Ac peaks in MC and FL at the whole-genome scale, followed by volcano plotting of differential H3K27Ac peaks using ggplot2 with the same criteria as ATAC-seq (fig. S5B). Sample correlation of H3K27Ac replicates was also calculated with DiffBind.

Generation of a list of putative enhancers at the Sox9 locus

Bam files of chr11 were extracted from processed MC and FL ATAC-seq bam files, respectively. ATAC-seq peak calling was performed using MACS2 (version 2.1.2) (69) (-f BAM -g mm -p .1 –call-summits). MACS2 narrowpeak files for chr11 in MC and FL were then sorted and filtered using the “makeCandidateRegions.py” script provided in the ABC model (https://github.com/broadinstitute/ABC-Enhancer-Gene-Prediction, version 0.2, accessed on July 2022) (15). The top 15,000 strongest ATAC-seq peaks at chr11 in MC and FL were retained as candidate enhancers (--nStrongestPeaks 15,000). Bedtools was then used to combine the candidate enhancers in MC and FL to generate a comprehensive list of 233 putative enhancers within the 1.3-Mb genomic neighborhood upstream of Sox9.

Analyzing and plotting enhancer activity for putative Sox9 enhancers

For fig. S5 (C and D), differential ATAC-seq and H3K27Ac analyses for the 233 putative Sox9 enhancers were performed using DiffBind (67) as described above. For Fig. 2, chromatin accessibility of Sox9 candidate enhancers was calculated by extracting and calculating average coverage from ATAC-seq bam files of MC and FL with megadepth (70) (--op mean). The H3K27Ac signal of each Sox9 candidate enhancer in MC and FL was also extracted with similar procedures. Log₂ MC/FL ratios of ATAC-seq and H3K27Ac signal strength at the putative Sox9 enhancers were then calculated. Scatterplots of chromatin accessibility and H3K27Ac signal of candidate enhancers were generated with ggplot2 in R. Enhancers located in different regions upstream of Sox9 (PRS region, proximal TBC, and other Sox9 noncoding regions) were labeled with different colors, and linear regression was performed on the scatterplots. Statistical comparison of log₂ MC/FL ratio of ATAC-seq and H3K27Ac signal strength in the PRS region and the proximal TBC was performed in R and visualized with boxplot by ggplot2.

Hi-C and virtual 4C analysis

Adaptor-trimmed Hi-C data were mapped to mm10 iteratively with bowtie2. Mapped sequences were parsed into a Python data structure and assigned to Dpn II fragments. Hi-C dataset object was created after fragment filtering, and generated hdf5 files were transformed to cool format using the cooler package (71) with cooler cload hiclib and then balanced with cooler balance. Hi-C analysis in this manuscript was performed using custom scripts based on the cooltools package (version 0.3.1) (72). Both E14.5 MC and E14.5 FL Hi-C data had two biological replicates and were combined, respectively, for further analysis. Heatmap of interaction frequency at Sox9, Pbx1, Ctgf, and Dkk1 locus was plotted after normalization with total interactions in the region of interest. A/B compartments were analyzed by eigendecomposition of the matrix using the call-compartments utility from the cooltools package (version 0.3.1) (72) and visualized using the saddle plots. Insulation scores were calculated using the diamond-insulation utility from the cooltools. Genomic bins with a TAD boundary strength score greater than 0.1 were defined as bona fide TAD boundaries. The TAD boundaries in MC and FL that are within 50-kb from each other were considered unchanged boundaries. Loop calling of MC and FL Hi-C data was performed with MUSTACHE (10-kb resolution, P < 0.1) (73). Virtual 4C analysis was performed by extracting the interaction frequency matrix of viewpoint (Sox9 promoter region) from Hi-C data. Interaction frequency with the viewpoint was then plotted in R.

4C-seq analysis

Reads matching the forward primer sequence were selected from the total fastq files. The primer sequences were then removed from selected reads, which were then mapped to the mm10 reference genome with Bowtie2. For visualization of 4C-seq data, the coverage files in bedGraph format were built with an R script (https://github.com/bbcf/bbcfutils/blob/master/R/smoothData.R, accessed on May 2022) (74, 75), with a range of five fragments used to normalize the data by reads per million mapped reads. The relative signal strength between MC and FL in the PRS-Sox9 loop bin (chr11: 111,650,000 to 111,660,000) was calculated using wiggletools (76) at the window size of 100 bp. MC/FL signal ratios at randomly selected windows in the Sox9 noncoding region other than the PRS-Sox9 loop bin were also calculated as a control for statistical analysis.

Identification of MC- and FL-specific enhancers of Sox9 with the ABC model

The concept and method of the ABC model (15) were followed to identify tissue-specific enhancers of Sox9 in MC and FL. To quantitatively evaluate the effect of an enhancer on Sox9 regulation, its balanced interaction frequency with the Sox9 promoter was multiplied by its enhancer activity (AxC) and was then divided by the sum of the AxC of all candidate Sox9 enhancers to calculate the ABC score. The ABC score of each Sox9 candidate enhancer in MC was then subtracted by its corresponding ABC score in FL. The putative Sox9 enhancers located within the Sox9 genic region or the genomic bins lacking Hi-C contact information were pruned. As a result, ABC scores were assigned to 233 Sox9 putative enhancers (data S1). Enhancers showing the highest differences in ABC score between the two tissues were identified as tissue-specific enhancers for MC or FL.

Genome-wide identification of MC- and FL-specific enhancers

Genome-wide identification of MC and FL enhancers was performed following the pipelines provided by the ABC model (15). Briefly, candidate enhancer regions for MC and FL were defined on the basis of ATAC-seq. Enhancer activity for each candidate enhancer region was calculated according to both ATAC-seq and H3K27Ac CUT&Tag signal, and the ABC score was then computed by adding enhancer-promoter contact from Hi-C data. We apply the ABC score threshold of 0.015 to define enhancer regions in the whole genome. We consider an enhancer as an MC- or FL-specific enhancer if it meets one of the two criteria: (i) The enhancer exhibited an ABC score of greater than 0.015 in only one cartilage type, and (ii) the difference of the ABC scores for the enhancer in the two tissues is greater than 0.01.

Validation of MC- and FL-specific enhancers with VISTA enhancer database

Information on previously verified enhancers was downloaded from the VISTA enhancer browser (39) (https://enhancer.lbl.gov). Bed files containing enhancers with craniofacial activity and limb activity were generated separately. Overlap of MC-specific enhancers and FL-specific enhancers with craniofacial active and limb active enhancers was calculated, respectively, using the bedtools intersect utility. The rate of overlap was then calculated and statistically examined with Fisher exact test.

Enrichment of SNPs of craniofacial- and limb-associated diseases or traits in MC- and FL-specific enhancers

The 203 genome-wide significant signals reported by White et al. (77) were used as SNPs of craniofacial-associated diseases or traits for enrichment analysis. GWAS studies on limb-associated diseases or traits were examined from the GWAS catalog database, and their lead SNPs were used for enrichment analysis. These SNPs were overlaid with MC-specific and FL-specific enhancers, and enrichment was assessed using genomic regulatory elements and the GWAS overlap algorithm (GREGOR; version 1.4.0) (46) (linkage disequilibrium window size = 1 Mb; LD r² ≥ 0.7). As a control, associations for Crohn’s disease were also interrogated for enrichment in MC-specific and FL-specific enhancers.

Supplementary Materials

This PDF file includes:

Fig. S1 to S16

Tables S1 to S4

Other Supplementary Material for this : manuscript includes the following:

Data S1 to S3

View/request a protocol for this paper from Bio-protocol.