Genetic interaction between Adgrg6 and Sox9 reveals a feedforward mechanism for postnatal spinal stability

Abstract

Adolescent idiopathic scoliosis (AIS) is the most common spinal disorder in children and is best understood as a complex polygenic condition. Genome-wide association studies (GWAS) have identified several AIS risk loci, including regions near ADGRG6 and SOX9, yet the functional mechanisms underlying AIS heritability remain poorly defined. Here, we use spatial transcriptomics to characterize altered gene expression in the spine of a conditional Adgrg6 mutant mouse model of AIS (Adgrg6-cKO), revealing reduced expression of Sox9 and several components of extracellular matrix organization in the intervertebral disc. We further show that SOX9 occupies regions of open chromatin within the Adgrg6 locus in cells isolated from the mouse intervertebral disc. Finally, we demonstrate a strong genetic interaction between Adgrg6-cKO and a hypomorphic Sox9 allele that increases both the penetrance and severity of AIS-like pathology in mice. Collectively, these findings support a self-reinforcing feedforward regulatory circuit, where Adgrg6 and Sox9 are co-regulated to maintain extracellular matrix gene expression in the annulus fibrosus and paraspinal tissues. These findings provide mechanistic insight into the functional significance of AIS-associated GWAS loci near ADGRG6 and SOX9 and establish combined Adgrg6–Sox9 insufficiency as a tractable model of polygenic scoliosis susceptibility.

INTRODUCTION

Adolescent idiopathic scoliosis (AIS) is the most common pediatric musculoskeletal disorder, affecting approximately 3% of school-age children worldwide^1,2. AIS typically presents as spine curvature around the onset of puberty, without obvious axial patterning or vertebral body fusion defects. Notably, the need for surgical correction or bracing is more frequent in females than in males. Population-based genome-wide association studies (GWAS) meta-analysis have identified several AIS susceptibility loci, predominantly within non-coding regions of the genome, supporting a complex polygenic etiology^3–8. In addition, a polygenic burden of rare variants across extracellular matrix genes likely contributes significantly to AIS risk^4,9,10. Taken together, an emerging model suggests that AIS results from the combined effects of complex polygenic variation, hormonal influences, and mechanical stresses encountered during rapid adolescent growth.

Maturation and homeostasis of a healthy, functional spine require the coordinated integration of multiple musculoskeletal tissues, including bone, cartilage, connective tissue, muscle, and the peripheral nervous system. Consistent with this, AIS-like spinal curvatures are observed in patients with connective tissue disorders featuring joint hypermobility, such as Marfan syndrome and subtypes of Ehlers-Danlos syndrome, as well as in neuromuscular conditions including muscular dystrophy, Rett syndrome and pediatric spinal cord injury. This suggests that subclinical defects in connective tissue integrity or neuromuscular regulation may underlie susceptibility to AIS. In support of this, animal models with targeted disruption of neural circuits involved in proprioception¹¹ or central pattern generation^12,13, spinal cord central canal physiology^14–16, muscle physiology^17–20 or connective tissue gene expression and mechanics²¹ all recapitulate AIS-like phenotypes.

Generalized joint hypermobility is a recognized risk factor for AIS²², and postnatal dysregulation of connective tissue homeostasis has emerged as a recurring theme in human genetic studies of scoliosis²³. Multiethnic GWAS have identified intronic variants in GPR126/ADGRG6 as strongly associated with human AIS^7,24. ADGRG6 belongs to a large class of adhesion G-protein coupled receptors (aGPCRs) that act as mechanosensors through force-based mechanisms, whereby dissociation of the N-terminal fragment from the seven-transmembrane domain elicits downstream signal transduction^25–27. Our prior studies demonstrated that conditional loss of Adgrg6 in mice recapitulates AIS, alters the biomechanical properties of tendons, and reduces SOX9 protein expression in the intervertebral disc (IVD) through disruption of CREB signaling^21,28.

Sox9 is a key transcription factor that controls skeletal development and postnatal homeostasis of the spine^29,30. Notably, the postnatal loss of Sox9 causes reduced skeletal growth, scoliosis, and leads to altered IVD gene expression, including a significant reduction in Adgrg6/Gpr126 expression³⁰. GWAS have identified variants in the SOX9 locus associated with human AIS^8,31, and coding variants in the transactivation middle (TAM) domain have been independently implicated in human scoliosis^32,33. Consistent with this, our prior studies showed that a microdeletion in the Sox9 TAM domain causes late-onset scoliosis in mice, accompanied by reduced expression of SOX9, ADGRG6, and several ECM proteins in the annulus fibrosus of the IVD³³.

The complex polygenic architecture underlying AIS susceptibility and severity remains poorly understood. To address this, we performed unbiased spatial transcriptomic profiling and RNA in situ hybridization in the conditional Adgrg6 mouse model of AIS^21,28, documenting reduced expression of Sox9 and ECM genes in the IVD. We further analyzed chromatin accessibility and SOX9 occupancy in Sox9-expressing skeletal cells, revealing that SOX9 binds putative regulatory regions within Adgrg6. Finally, we show that genetic interactions between conditional Adgrg6 and Sox9 TAM domain mutations increase the penetrance and severity of scoliosis in mice. Taken together with our prior work²¹, these findings indicate that Adgrg6 and Sox9 operate in a feedforward regulatory loop to maintain gene programs essential for postnatal spinal stability.

RESULTS

Spatial Transcriptomics Reveals Reduced Expression of Genes Related to Cartilage Development and Extracellular Matrix Organization in the Intervertebral Discs of the Adgrg6-cKO Mutant Mouse Model of AIS.

We previously showed that conditional loss of Adgrg6 in osteochondral progenitors (Col2a1-Cre; Adgrg6^f/f; hereafter Adgrg6-cKO) or in connective tissue (ScxCre; Adgrg6^f/f) models adolescent idiopathic scoliosis (AIS) in mice²¹. To generate a comprehensive, unbiased map of how Adgrg6 contributes to the molecular regulation of the spine, we performed spatial transcriptomics on P20 spine sections from wild-type and Adgrg6-cKO mice. Unsupervised clustering of the entire spine section identified nine transcriptionally distinct populations, encompassing nucleus pulposus cells of the intervertebral disc (IVD), osteoblasts/osteoclasts, paraspinal muscle and adipose tissue, paraspinal connective tissue, and patches of dorsal root ganglion. Because osteoblast-lineage–driven CRE recombination of Adgrg6 does not induce scoliosis in mice²¹, all downstream differential expression analyses were restricted to capture spots overlaying IVD tissues (Fig. 1a–d).

Figure 1. Spatial transcriptomics and differential expression of Adgrg6-cKO intervertebral disks (IVD).

Hematoxylin and Eosin (H&E) staining of IVD sections from control and Adgrg6-cKO mice (a–b). Spatial DimPlots projecting transcriptomic clusters on control and Adgrg6-cKO samples (c–d). Uniform Manifold Approximation and Projection (UMAP) visualization of the first 9 principal components (e). Cells are colored by cluster identity and annotated by cell type (right). Volcano plot showing global differential expression (DE) between control and Adgrg6-cKO conditions (f). Significance was determined using a Wilcoxon Rank Sum test (adjusted p-value < 0.05, log2 fold change > 0.25). Gene Set Enrichment Analysis (GSEA) shows that cartilage development (red line, GO:0051216) and extracellular matrix organization (teal line, GO:0030198) are significantly downregulated in the Adgrg6-cKO compared to control (g). Scale bar = 1 mm.

The spine section is a structurally heterogeneous tissue in which metabolically active paraspinal muscle and adipose depots are in direct proximity to the fibrous connective tissues of the IVD. Because high-UMI transcripts released from these adjacent tissues can diffuse into neighboring capture spots, we implemented a per-spot muscle-score quality control filter as part of our standard preprocessing pipeline, excluding spots with greater than 1% muscle-transcript contribution prior to clustering and differential expression analysis. To validate that this filtering strategy was correctly identifying contamination rather than genuine IVD signal, we performed orthogonal RNA in situ hybridization for canonical muscle markers (Myh2, Ryr1, Apobec2) and immunofluorescence for UCP1 protein in Adgrg6-cKO and wild-type spine sections. Neither muscle transcripts nor UCP1 protein showed detectable expression within the IVD in either genotype (Supp. Fig. 1a–h), confirming that capture spots enriched for these signals reflected transcript diffusion from adjacent paraspinal tissues rather than genuine IVD expression differences. We note that Ryr1 was of independent interest given recent reports linking RYR1 mutations to proprioceptive defects and scoliosis in mice and humans^10,20; however its absence from IVD tissue by in situ analysis suggests that its differential expression is more related to technical challenges of using spatial transcriptomics in heterogenous tissue sections rather than any real interaction with Adgrg6 signaling. Taken together, these findings underscore that paraspinal tissues exhibit greater RNA release and diffusion than the dense fibrous IVD, and that muscle-score filtering is a necessary and effective QC step for spatial transcriptomics in heterogeneous spinal tissue.

Following quality control at the spot level (feature counts, UMI counts, and mitochondrial content), the muscle-filtered dataset comprised 730 spots and 17,480 genes (Control: 371 spots; Mutant: 359 spots), with comparable QC metrics across conditions indicating high-quality IVD data suitable for robust clustering and differential expression analyses. Approximately 9 principal components captured the major variation in the dataset, yielding clear separation into seven biologically coherent clusters (Fig. 1e, Supp. Fig. 2, Table 1): cartilage/annulus fibrosus (AF) tissue (e.g., Clec3a, Chad, Col2a1, Comp; Supp. Fig. 2); nucleus pulposus (NP) tissue (e.g., Krt19, T, Krt8, Arhgef16)^34,35 (Supp. Fig. 4); adipocytes (Lgals12, Ppara, Tmem79, Cidec); broadly expressed genes (Stfa3, Usp1, Bbip1, Cks2); immune-related genes (Fanca, Malt1, Lin28a, Tonsl); neuromuscular/bone-associated genes (Mobp, Prg2, Casq1, Agt); and dorsal root ganglion markers (Caly, Hand1, Epha10, Gprin1). Differential expression analysis within IVD-restricted, muscle-filtered capture spots revealed significant downregulation of genes associated with cartilage development (GO:0051216) and extracellular matrix (ECM) organization (GO:0030198) in Adgrg6-cKO samples (Fig. 1f–g, Table 2). Collectively, these data indicate that Adgrg6 modulates postnatal gene expression in the IVD, including maintenance of Sox9 expression and regulation of ECM genes such as Acan, Col9a1, Col2a1, Matn3, Col11a2, Col27a1, and Col15a1.

Adgrg6-cKO Mutant Mice Display a Significant Reduction of ECM Genes in the Inner Annulus Fibrosus.

To validate and spatially resolve the transcriptomic changes identified by spatial transcriptomics, we performed RNA in situ hybridization for a panel of ECM and cartilage-related genes found to be downregulated in Adgrg6-cKO mice. Genes fell into three broad categories based on the magnitude and spatial character of their expression changes. A first set — including Comp, Chad, and Col27a1 — showed no obvious change in pattern or expression level between wild-type and Adgrg6-cKO sections (Fig. 2a–d’, 3e–f’). A second set — Acan, Sox9, and Creb3l2 — showed markedly reduced but detectable expression in Adgrg6-cKO mice relative to wild-type controls (Fig. 3a–d’, g–h’). A third set of genes — Grep1, Frzb, Fmod, Col6a2, and Col14a1 — was profoundly decreased or nearly undetectable in Adgrg6-cKO sections despite robust expression in wild-type tissue (Fig. 2e–n’).

Figure 2. Genotype-specific expression of key ECM-regulatory genes in the intervertebral disc.

RNA in situ hybridization (RNAscope) was used to validate the expression of extracellular matrix-associated genes identified as significantly decreased in the spatial transcriptomic analysis. The following genes were examined in the intervertebral disc (IVD): Comp, Chad, Grep1, Frzb, Fmod, Col6a2, and Col14a1 (a-n). Representative images demonstrate gene-specific punctate signals within the intervertebral disc (a-n) and higher-magnification views of the annulus fibrosus (a’-n’). Columns correspond to individual genes as follows: Comp (a-b’), Chad (c-d’), Grep1 (e-f’), Frzb (g-h’), Fmod (i-j’), Col6a2 (k-l’), and Col14a1 (m-n’) assessed in controls (a, c, e, g, i, k, m) and Adgrg6-cKO mutant (b, d, f, h, j, l, n) mice. Images are representative of n > 2 biologically independent samples per genotype. Scale bars = 100 μm in (a-n) and 50 μm (a’-n’).

Figure 3. Expression of Sox9 and Sox9-related genes in the intervertebral disc across genotypes.

RNA in situ hybridization (RNAscope) was performed to validate spatial transcriptomic findings and examine the expression of select Sox9 and Sox9-related genes within the intervertebral disc. Expression of Sox9 (a-b’), Acan (c-d’), Col27a1 (e-f’), and Creb3l2 (g-h’) were evaluated. Representative images demonstrate gene-specific punctate signals within the IVD (a-h), along with corresponding higher-magnification views of the annulus fibrosus (a’-h’) in wild-type controls (a, c, e, g) and Adgrg6-cKO (b, d, f, h). Images are representative of n > 2 biologically independent samples per genotype. Scale bars = 100 μm in (a-h) and 50 μm (a’-h’).

Among this third set, Grep1 (also known as 1520401A03Rik in mouse) showed the most pronounced downregulation in the spatial transcriptomics dataset (average logFC = −3.96) and a striking spatial specificity by in situ analysis confined to the inner annulus fibrosus (AF) in wild-type mice (Fig. 2e’), which was almost entirely absent in Adgrg6-cKO mutants (Fig. 2f’). Frzb/Sfrp3, a secreted Wnt antagonist, was strongly expressed in the cartilaginous endplate with minor expression in the AF of wild-type spines and was severely decreased in Adgrg6-cKO mice (Fig. 2g–h’). Col6a2 and Fmod were broadly expressed throughout the AF in wild-type sections, with expression severely reduced in mutants (Fig. 2i–l’). Col14a1 was strongly expressed in a thin layer of paraspinal (periosteum-like) tissue surrounding the length of the spine (Fig. 2m,m’; Supp. Fig. 5) and this expression was largely absent in Adgrg6-cKO mice (Fig. 2n, n’, Supp. Fig. 5). Together, these spatial expression patterns reveal that Adgrg6 loss affects transcriptional programs across multiple discrete tissue compartments within and surrounding the IVD — including the inner AF, cartilaginous endplate, and paraspinal connective tissue — rather than reflecting a uniform suppression of ECM gene expression.

Notably, several of the ECM genes downregulated in spatial transcriptomics analysis of Adgrg6-cKO mice, including Acan, Matn3, Col9a3, and Creb3l2, have been identified as direct SOX9 targets by ChIP-seq in mouse chondrocytes³⁶ (Supp. Fig. 3). Consistent with this, in situ hybridization confirms that Sox9, Acan, and Creb3l2 display overlapping expression in the endplate and AF (Fig. 3a–d’, g–h’) and are also reduced in Adgrg6-cKO mutants. Col27a1, despite being identified as a gene downregulated in Adgrg6-cKO mouse IVD by spatial transcriptomics, was not confirmed as reduced by in situ analysis (Fig. 3e–f’), which may reflect a subtle change below the sensitivity of the in situ hybridization approach. Collectively, these findings confirm that Adgrg6 is required to sustain Sox9 expression in the IVD and further suggest that SOX9 in turn drives transcription of a broad set of ECM targets with distinct spatial distributions across the AF and paraspinal tissues.

Sox9 Occupies Accessible Chromatin Regulatory Elements Within the Adgrg6 Locus.

Our data above demonstrate that Adgrg6 loss reduces Sox9 expression in the IVD, while prior work has shown that postnatal Sox9 deletion significantly reduces Adgrg6 expression in the same tissue³⁰. Together with evidence that Adgrg6-dependent CREB signaling is required for maintenance of SOX9 protein expression²¹, these observations support a self-regulatory feedforward loop to sustain gene expression in the IVD. However, whether SOX9 directly occupies regulatory elements at the Adgrg6 locus has not been established.

To test this, we asked whether SOX9 binds accessible chromatin regions within the Adgrg6 locus in IVD-resident cells. We used fluorescence-activated cell sorting (FACS) to isolate SOX9-EGFP⁺ cells from the IVDs of P7 Sox9ires^EGFP/+ embryos (Fig. 4a–b) and profiled chromatin accessibility genome-wide by ATAC-Seq³⁷. To identify direct SOX9 occupancy at accessible sites, we performed SOX9 and IgG control Cut&Run-Seq on the same cell population³⁸. ATAC-Seq identified multiple accessible regions within both the Col2a1 locus which is a well-established SOX9 transcriptional target³⁹ used here as a positive control and the Adgrg6 locus (Fig. 4c–d). Cut&Run-Seq confirmed SOX9 occupancy at several of these ATAC peaks, with the strongest enrichment at the promoter and 5′ coding region of Col2a1 and at a peak within the second intron of Adgrg6, a region with high vertebrate evolutionary conservation (Fig. 4d). These findings demonstrate that SOX9 directly binds putative regulatory elements at the Adgrg6 locus in IVD-resident cells, establishing a transcriptionally direct arm of the Adgrg6–Sox9 feedforward loop and providing a molecular mechanism through which SOX9 sustains Adgrg6 expression to promote postnatal spinal homeostasis.

Figure 4. Sox9 directly binds to putative regulatory elements surrounding the Adgrg6 locus.

Direct fluorescence of the lumbar spine/IVD region isolated from postnatal day 7 (P7) Sox9^EGFP mouse (a) (Left). DIC image or direct fluorescence of cryosectioned lumbar intervertebral disc isolated from P7 Sox9^EGFP mouse (a) (Right). The top panel is a coronal cryosection of the IVD; the bottom panel is a transverse cryosection of the IVD (a). Scale bar equals 200 microns. FACS profile of Sox9-EGFP+ cells isolated from the developing spine/IVD of P7 Sox9^EGFP mouse (b). Analysis of chromatin regulation Sox9-EGFP+ cells isolated from spine/IVD of P7 Sox9^EGFP mouse shown as Integrated Genome Views (IGV) of ATAC-Seq, Sox9-Cut&Run-Seq, or IgG-Cut&Run-Seq (c, d), demonstrating SOX9 binding to several differentially accessible regulatory regions surrounding the Col2a1 (c) and Adgrg6 loci (d). Regions of overlap between ATAC-Seq peaks and SOX9 enriched-Cut&Run peaks are indicated (gray shadows), with evolutionary conservation in 60 vertebrate genomes displayed.

Genetic Interactions Between Adgrg6 and Sox9 Increase Scoliosis Penetrance and Severity in Mice.

SOX9 dosage sensitivity is well established, with partial reductions in SOX9 levels producing measurable differences in target gene expression and phenotypic outcomes⁴⁰. Given the feedforward co-regulatory relationship between Adgrg6 and Sox9 demonstrated above, and the independent association of both loci with human AIS^{8,21,24,31–33}, we hypothesized that reducing Sox9 dosage in the spine would exacerbate scoliosis onset and severity in Adgrg6-cKO mice. To test this, we utilized a hypomorphic Sox9 allele encoding an in-frame Asp272 deletion within the Sox9 TAM domain (hereafter Sox9^del). Homozygous Sox9^del mutants are viable and display mild skeletal phenotypes reminiscent of acampomelic dysplasia — including bilateral loss of floating ribs and late-onset scoliosis (>5 months) — without craniofacial malformations or respiratory distress³³. We generated Col2Cre; Adgrg6^f/+;Sox9^del/+ and Adgrg6 ^f/+;Sox9^del/+ breeders and performed longitudinal microCT imaging and analysis of spine morphology at P40 and P120 across all genotypes. Body weights remained comparable across groups at both time points, indicating no general health differences that could confound interpretation of skeletal phenotypes (Supp. Fig. 6).

Genetic interactions between Adgrg6-cKO and Sox9^del alleles were evident at both timepoints. Wild-type controls displayed straight spines with normal thoracic architecture at P40 and P120 (Fig. 5a, a’). Consistent with prior work, homozygous Sox9^del mutants exhibited bilateral loss of the T13 floating ribs (90%; n=11) (Fig. 5b, b’, asterisk; Supp. Fig. 7a), but did not develop scoliosis at either timepoint (0%; n=11; Fig. 5e, f). Adgrg6-cKO single mutants showed progressive single thoracic scoliosis by P120 (20%; n=10) (Fig. 5c, c’, f; Table 3). In contrast, Adgrg6-cKO; Sox9^del/del double mutants demonstrated a marked increase in scoliosis penetrance as early as P40 (54.5%; n=11; Fig. 5d, e; Table 3), which further increased by P120 (63.6%; n=11), with 9% of double mutants displaying double thoracic curves (Fig. 5d’, f; Table 3). A gene-dosage effect was also observed in compound Adgrg6-cKO; Sox9^del/+ heterozygous mutants, which exhibited increased scoliosis incidence (30%; n=11) and severity (20%; n=11) relative to Adgrg6-cKO alone (Supp. Fig. 7d), demonstrating that a single copy of the Sox9 microdeletion allele is sufficient to modify scoliosis susceptibility. Notably, transheterozygous Col2Cre; Adgrg6^f/+; Sox9^del/del mice did not develop scoliosis by P120 (0%; n=12), implying that heterozygous loss of Adgrg6 is not sufficient to sensitize the spine to reduced Sox9 gene dosage. Fisher’s exact tests with Monte Carlo simulation (B = 5000) for pairwise severity counts at P120 confirmed significant genotype-dependent differences, with Adgrg6-cKO; Sox9^del/del double mutants differing significantly from all other experimental genotypes and wild-type controls (Bonferroni-adjusted p < 0.01; Supp. Fig. 7d). Collectively, these data demonstrate that dose-dependent reduction in Sox9 function is sufficient to enhance both the penetrance and severity of scoliosis in the Adgrg6-cKO background, establishing combined Adgrg6–Sox9 insufficiency as a tractable model of polygenic scoliosis susceptibility.

Figure 5. Genetics interaction between Adgrg6 and Sox9 mutations contribute to increased scoliosis penetrance and severity.

Reconstructed microCT images from control (a-a’), Sox9^del/del (b-b’), Adgrg6-cKO (c-c’), and Adgrg6-cKO;Sox9^del/del (d-d’) reveal skeletal defects at postnatal day (P)40 and (P)120. Wildtype mice show straight spines and natural barreling of the chest cavity at both time points (n=10) (a, a’). Sox9^del/del mice present with loss of T13 rib pair (asterisk, 90%) and no scoliosis at either time points (0%, n= 11) (b, b’). Adgrg6-cKO mice show single thoracic scoliosis in 20% of animals at P120 (n=10) (c, c’). Adgrg6-cKO;Sox9^del/del mice presented with pectus excavatum (black arrow, 45.5%), complete loss of T13 rib pair (asterisk, 100%), and increased scoliosis incidence and severity as evidenced by no scoliosis (27.3%), single thoracic curvatures (63.6%), double thoracic curvatures (9.1%), and scoliosis concurrent with pectus excavatum in 36.4% of mice at P120 (n=11) (d, d’). Adgrg6-cKO, Adgrg6-cKO;Sox9^del/+, and Adgrg6-cKO;Sox9^del/del mice showed age-dependent progression in scoliosis penetrance and severity (e-f). Scale bar = 2 mm.

Adgrg6-Sox9 Genetic Interactions Increase Pectus Excavatum Penetrance in Mice.

Pectus excavatum (PE) is a congenital chest wall deformity characterized by inward depression of the sternum and ribs⁴¹. We previously reported PE in Adgrg6-cKO mutant mice, occurring either independently or concurrently with AIS-like scoliosis²⁸. In the present cohort, PE was observed in Sox9^del mutants and all double mutant combinations, but was absent in Adgrg6-cKO single mutants (Supp Fig. 7b, c). Sox9^del homozygotes displayed low PE penetrance at P120 (18.2%; n=11) without concurrent scoliosis (Supp. Fig. 7b, c).

A clear gene-dosage effect on PE penetrance was evident across double mutant genotypes. Transheterozygous Col2Cre; Adgrg6^f/+; Sox9^del/+ mice exhibited similarly low PE penetrance to Sox9^del homozygotes (16.7%; n=12; Supp. Fig. 7b). Adgrg6-cKO; Sox9^del/+ compound heterozygotes showed a modest increase in PE incidence (30%; n=11), while Adgrg6-cKO; Sox9^del/del double mutants displayed the highest PE penetrance (54.5%; n=11; Supp. Fig. 7b) and the greatest co-occurrence of PE and scoliosis (45.4%; n=11; Fig. 5d, d’, arrow; Supp. Fig. 7b, c). Fisher’s exact test for pairwise comparisons of PE incidence at P120 confirmed that only Adgrg6-cKO; Sox9^del/del double mutants differed significantly from all other experimental genotypes and controls (p < 0.05; Supp. Fig. 7b). Concurrent incidence of PE and scoliosis at P120 showed no significant genotype-dependent differences (Supp. Fig. 7c), suggesting that these two phenotypes arise through at least partially independent pathogenic mechanisms. Taken together, these data indicate that the sternum and chest wall may be more sensitive to graded reductions in Adgrg6 and Sox9 function than the IVD and annulus fibrosus and further underscore the dose-dependent nature of the Adgrg6–Sox9 genetic interaction in maintaining axial skeletal integrity.

Loss of Sox9 exacerbates the incidence of growth plate defects and ectopic cartilage regions in vertebrae.

To evaluate if structural morphological changes in spinal tissues contributed to this increased susceptibility of scoliosis, P20 spines were analyzed using Alcian Blue Hematoxylin Eosin Orange-G staining (Supp Fig. 8a–d). We observed an increase of atypical growth plate extensions and ectopic cartilaginous regions in the vertebrae of Adgrg6-cKO, Sox9^del, and Adgrg6-cKO; Sox9^del/del double mutant spines, that were never observed in wild-type controls (Supp Fig. 8). To quantify the incidence and severity of this phenotype, we calculated the percentage of affected IVDs per spine and number of growth plate extensions per affected IVD (Supp. Fig. 8e, f). Consistent with prior reports of conditional Adgrg6 deletion in IVD-associated tissues⁴², Adgrg6-cKO mice exhibited growth plate defects with incomplete penetrance (33%; n = 3; Supp. Fig. 8c, e). In contrast, Sox9^del spines showed a fully penetrant appearance of growth plate extensions with increased severity compared to Adgrg6-cKO (100%, n = 5; Supp. Fig. 8b, e, f). Adgrg6-cKO; Sox9^del/del spines exhibited an increased severity of growth plate extensions and ectopic cartilage regions in the vertebrae compared to Adgrg6-cKO (85%, n = 6; Supp. Fig. 8d–f). Together, these findings suggest that reduction in Sox9 expression is a major regulator that drives the growth plate defects. Moreover, these data suggest that defects in bone mineralization may be involved in the increased incidence and severity of AIS phenotypes in Adgrg6-cKO; Sox9^del/del double mutant mice.

Increased scoliosis incidence and severity are associated with decreased bone mineral density in thoracic vertebrae.

Although conditional deletion of Adgrg6 in osteoprogenitor cells does not cause scoliosis²¹, Adgrg6-cKO; Sox9^del/del double mutant mice showed severe growth plate extensions (Supp. Fig. 8), prompting us to examine whether combined Adgrg6–Sox9 insufficiency alters vertebral microarchitecture. Bone density-mapped reconstructions with Otsu-based threshold analysis⁴³ of thoracic spines at P40 revealed a general reduction in bone mineralization in the high-density range (0.286–0.551 g/cm³) within the region of the scoliotic curve in double mutants compared to controls (Fig. 6a, d). To further characterize this, we segmented the T4 vertebra (Fig. 6a, d, arrow), as this region typically corresponds to the apex of thoracic scoliotic curves. Quantitative microCT analysis of T4 across all genotypes revealed significant reductions in both bone volume and bone mineral density (BMD) in Adgrg6-cKO; Sox9^del/del mice compared to controls (p < 0.05; Fig. 6g, h). Otsu analysis further confirmed a significant reduction in high-density bone volume fraction in double mutants (p < 0.05; Fig. 6i), consistent with the density-mapped reconstructions of T4 vertebrae (Fig. 6c, f).

Figure 6. Adgrg6-cKO;Sox9^del/del double mutant mice exhibit microarchitectural alterations in a thoracic vertebra at P40.

MicroCT analysis of thoracic vertebral bone architecture at P40. Density-mapped reconstructions of the thoracic spine reveal differences in overall mineral distribution between straight control and scoliotic Adgrg6-cKO;Sox9^del/del spines (a, d). Arrows indicate the T4 vertebra analyzed in subsequent panels. Three-dimensional segmentations of T4 illustrate overall vertebral morphology in control and double mutant mice (b, e). Corresponding density-mapped reconstructions of representative T4 vertebrae demonstrate altered mineral distribution throughout the vertebrae in double mutant mice (c, f). One way ANOVA and t-test analysis revealed a significant difference in bone volume and BMD in control and Adgrg6-cKO;Sox9^del/del T4 vertebrae (p < 0.05) (g, h). Two way ANOVA Otsu-based threshold 3D bone volume stratifying bone into low-(0–0.0808 g/cm³; grey), mid- (0.0809–0.285 g/cm³; magenta), and high-density (0.286–0.551 g/cm³; green) fractions revealed a significant loss of high-density distribution in double mutant mice (p < 0.05) (i). Each dot represents an individual mouse vertebrae, with mean ± SEM overlaid. Scale bar = 0.5 mm (b,c, e, f).

To determine whether these bone changes are specific to spinal regions susceptible to scoliosis, we assessed L5 vertebral microarchitecture in the same cohort at P40 (Supp. Fig. 9). MicroCT analysis of L5 revealed no significant differences in overall bone morphology or BMD across genotypes (Supp. Fig. 9g, h), consistent with representative cross-sectional images showing grossly normal inner bone architecture (Supp. Fig. 9a’–f’). However, Otsu analysis revealed a significant reduction in low-density bone volume in Adgrg6-cKO; Sox9^del/del double mutants relative to controls (Supp. Fig. 9i), showing that subtle compositional differences are present in lumbar vertebrae which are not directly part of scoliotic regions. Collectively, these findings indicate that while mild alterations in bone composition extend throughout the spine, the most pronounced microarchitectural defects are concentrated in thoracic vertebrae, likely reflecting the combined effects of Adgrg6–Sox9 insufficiency and the region-specific mechanical demands imposed by the onset and progression of scoliosis.

DISCUSSION

Adolescent idiopathic scoliosis is best understood as a complex polygenic condition in which multiple susceptibility alleles of individually modest effect combine to cross a phenotypic threshold during the period of rapid pubertal growth. GWAS have reproducibly identified non-coding variants near ADGRG6 and SOX9 as AIS risk loci in human populations, yet the functional relationship between these two genes in the postnatal spine has remained undefined. Here, we integrate spatial transcriptomics, chromatin accessibility profiling, SOX9 chromatin occupancy, and multi-allelic mouse genetics to demonstrate that Adgrg6 and Sox9 operate in a feedforward regulatory circuit in which each gene has a self-reinforcing role in co-regulation to maintain a broad program of ECM gene expression in the IVD and paraspinal connective tissues. Critically, we show that combinatorial reduction of Adgrg6 and Sox9 function, which individually produce incompletely penetrant scoliosis phenotypes, dramatically increases both the incidence and severity of AIS-like pathology in mice. These findings provide a mechanistic framework for understanding how co-variation at the ADGRG6 and SOX9 loci may interact to modulate scoliosis susceptibility in human populations.

The central molecular finding of this study is the identification of a feedforward loop between Adgrg6 and Sox9 in IVD and paraspinal tissues. Prior work established that Adgrg6-dependent CREB signaling is required for maintenance of SOX9 protein expression in the IVD²¹, and that postnatal Sox9 deletion significantly reduces Adgrg6 expression in the same tissue³⁰. The present study extends this bidirectional relationship by demonstrating, first, that Adgrg6 loss reduces Sox9 transcript levels and putative SOX9-target gene expression in the IVD, and second, that SOX9 directly occupies several more accessible and conserved chromatin regions surrounding the Adgrg6 locus, including a region of the second intron of this gene in IVD-resident cells. Together, these observations define a self-reinforcing transcriptional circuit in which Adgrg6 signaling sustains SOX9 protein levels, while SOX9 in turn directly promotes Adgrg6 transcription by directly binding to conserved regulatory elements. Feedforward architectures of this kind confer robustness to transient perturbations under normal conditions but render a system acutely sensitive to sustained or combinatorial disruption, which is consistent with the dose-dependent exacerbation of scoliosis we observe in Adgrg6-cKO; Sox9^del/del compound mutants. We propose because Adgrg6 can act as a force-sensing mechanoreceptor²⁵ that this circuit represents a core regulatory node for postnatal IVD homeostasis in response to stress and strain of the spine, and that partial disruption of either arm through non-coding variants affecting regulatory element activity in the IVD and paraspinal tissues or through coding variants affecting protein function may be sufficient to shift an individual toward AIS susceptibility when combined with additional genetic or environmental risk factors.

The spatial transcriptomic and in situ hybridization data reveal that Adgrg6 loss does not produce a uniform suppression of ECM gene expression in the IVD but rather affects discrete subsets of targets with distinct spatial distributions and with apparent sensitivity thresholds, which is a modality well-established in response to alterations in SOX9 dosage and its regulatory role on target gene expression⁴⁰. Using RNA in situ hybridization we showed that many genes such as Comp, Chad, and Col27a1 had no detectable change in expression, while Acan, Sox9, and Creb3l2 were reduced but the patterning of expression was maintained, and a third set — Grep1, Frzb, Fmod, Col6a2, and Col14a1 — were profoundly depleted. This graded response likely reflects differences in promoter architecture, SOX9 occupancy affinity, or dependence on co-activators whose activity is selectively disrupted downstream of Adgrg6. Among the most severely downregulated targets, Grep1 is of particular interest. Grep1 encodes a secreted ECM protein with homology to human Glycine-rich extracellular protein 1 and has been implicated in regulating GDF15 expression⁴⁴. Moreover, Grep1 shows highly specific expression in the inner AF of wild-type mice, and near-complete absence in Adgrg6-cKO mutants, predictive of a spatially restricted role in maintaining inner AF identity that warrants further investigation. Frzb (Sfrp3), a secreted Wnt antagonist expressed in the cartilaginous endplate, is similarly notable given emerging evidence for Wnt pathway dysregulation of cartilage homeostasis and correlated expression with disc degeneration⁴⁵. Col14a1 expression in a periosteum-like layer surrounding the spine, and its marked reduction in Adgrg6-cKO mice, extends the transcriptional consequences of Adgrg6 loss beyond the IVD proper to the paraspinal connective tissue envelope — a compartment increasingly recognized as relevant to spinal biomechanics and AIS pathogenesis^21,23. Collectively, the spatial heterogeneity of these expression changes underscores that Adgrg6 coordinates transcriptional programs across multiple anatomically distinct tissue compartments, and that its loss disrupts IVD homeostasis in a spatially and quantitatively nuanced manner.

A key translational implication of this work is that the ADGRG6 and SOX9 GWAS loci, while typically analyzed as independent AIS risk factors, may in fact converge on a shared regulatory pathway whose output is sensitive to combinatorial genetic perturbation. We identified a putative enhancer in the second intron of mouse Adgrg6. This is compelling given that the strong ADGRG6 AIS-associated variant (rs6570507) is also within the 2^nd intron of human ADGRG6²⁴. This suggests that these putative enhancer regions are instructive for spatially defined regulation of gene expression within the AF and paraspinal tissues. Future studies employing reporter assays or CRISPR-based mutagenesis of these putative regulatory elements in human cell lines or in vivo models will be needed to test this hypothesis directly. More broadly, our double mutant data provide a functional analog to the proposed polygenic burden model of AIS³. Our genetic interactions demonstrated that neither heterozygous Adgrg6 nor heterozygous Sox9^del loss alone is sufficient to cause scoliosis, but their combination on the Adgrg6-cKO background crosses a phenotypic threshold in a dose-dependent manner. This mirrors the epidemiological observation that AIS risk is not determined by any single variant but by the aggregate burden of partially penetrant alleles across interacting pathways^4,10, and positions Adgrg6–Sox9 compound insufficiency as a tractable experimental model for dissecting the genetic architecture of polygenic scoliosis susceptibility.

The microarchitectural bone phenotype observed in Adgrg6-cKO; Sox9^del/del double mutants provides additional insight into the mechanisms by which Adgrg6–Sox9 insufficiency promotes scoliosis progression. The partial reduction in BMD in double mutants, and its spatial restriction to thoracic rather than lumbar vertebrae, argues against a primary osteoblast defect, consistent with the prior demonstration that osteoprogenitor-specific Adgrg6 deletion does not cause scoliosis²¹. Instead, we favor a model in which failure of cartilage endplate maintenance, evidenced by growth plate defects and ectopic cartilage formation within the vertebrae body, secondarily compromises vertebral bone quality. Supporting this model, ablation of Adgrg6 Adgrg6 in osteochondral progenitor cells causes delayed formation of the secondary ossification center and growth palate dysplasia in the long bone⁴⁶, which may alter bone mechanics in the vertebrae. Furthermore, Adgrg6 loss in the cartilaginous endplate elevates STAT3 activation and MMP13 expression alongside reduced SOX9 expression, promoting growth plate herinations that resemble Schmorl’s nodes⁴². SOX9 is a well-established regulator of cartilage endplate formation and IVD compartment identity³⁰, and its combined reduction with Adgrg6 loss may further destabilize growth plater integrity below the threshold required for normal vertebral development. Such growth plate disruption could impose asymmetric mechanical loading across the thoracic spine, driving the aberrant bone remodeling and curve progression characteristic of AIS⁴⁷. The growth plate defects observed in Sox9^del and double mutant spines at P20 support this interpretation and raise the possibility that endplate pathology is an early initiating event in the disease cascade, rather than a secondary consequence of established curvature.

In summary, this study demonstrates that Adgrg6 and Sox9 are co-regulated through a self-regulating feedforward transcriptional circuit, and that their combinatorial insufficiency is sufficient to model the polygenic architecture of AIS susceptibility in mice. By establishing a direct molecular link between two of the most replicated AIS GWAS loci, these findings move beyond statistical co-association to provide a mechanistic basis for understanding how non-coding variation at ADGRG6 and SOX9 might interact to modulate disease risk in human populations. More broadly, this work supports an emerging view of AIS as a condition arising from the convergence of subtle defects across interconnected regulatory circuits governing IVD homeostasis, connective tissue integrity, and spinal biomechanics — and suggests that the Adgrg6–Sox9 feedforward loop represents one functionally validated node within this broader network. Identifying additional interacting loci within and beyond this circuit and determining how their combined perturbation maps onto the spectrum of AIS severity observed in human patients, remains an important goal for future investigation.

METHODS AND MATERIALS

Mouse strains.

All animal studies were approved by the Institutional Animal Care and Use Committee at the University of Texas at Austin (AUP-2024–00222) and in accordance with national animal welfare guidelines. All mouse strains were previously described, including Col2a1-Cre⁴⁸, Adgrg6^f/f (Taconic #TF0269), Sox9^{del 33}. All mouse positioned in the scanner with the incisors secured in the nose cone and maintained on isoflurane at low flow during acquisition. Respiratory rate was continuously monitored throughout scanning using a tape marker on the dorsal thoracic region. In vivo microCT scans were acquired at 55 kV, 200 μA, 10 μm voxel size, 4K resolution, continuous mode, with 1 mm aluminum filter on Skycan 1276 (Bruker). Reconstructions were performed using the following parameters: smoothing = 5, ring artifact correction = 6, beam hardening = 10, and threshold = 0.065. Images were visualized and analyzed using CTVox, Dataviewer, CTAn Bruker software and 3D Slicer software (Version 5.6.1) including SlicerMorph⁴⁹. Cobb angles were measured in thoracic spine segments using 3D slicer. Curvature was quantified by measuring the angle from the apical vertebra to the most inferior tilted vertebra. P40 and P120 thoracic cavity morphometrics were visualized using 3D Slicer segmentation to colorize the thoracic column, including the ribs from the costal cartilage and sternum. Morphometric analysis of P40 T4 and L5 vertebrae were segmented in CTAn and sequentially straightened in DataViewer. Straightened datasets were re-opened in CtAn, and the vertebral midline was defined by identifying the most dorsal and most ventral slices containing the vertebral body. From the midline, 10 um slices were used to extracted to generate inner bone visualizations in 3D Slicer. All vertebral microCT datasets were processed using CtAn with standardized batch tasklists. Separate tasklists were used for whole vertebral body segmentation, BMD histogram analyses, and Otsu multi threshold-based analysis. Thresholds and ROI handling steps were held constant across all samples within each analysis type. Whole vertebral body processing was performed using a two-step thresholding approach. Tissue was first segmented using a grayscale threshold of 55–255. A bitwise operation was then performed to define the ROI (ROI = Copy Image), followed by image reload to apply the ROI mask. Bone was subsequently segmented using a threshold of 80–255, and standard 3D analysis was performed. BMD measures were obtained from histogram analysis in three-dimensional space within the defined volume of interest (VOI). Otsu-based thresholding analysis was performed as previously described⁴³. Spine and individual vertebral segmentations were subject to colorimetric threshold mapping to visualize the Otsu-based thresholding.

Histological Analysis of Mice

For staining and immunohistochemistry protocols, P20 thoracic spines were histologically processed as previously described by Liu et al²¹. Alcian Blue Hematoxylin Eosin Orange G staining was completed using standard protocols from Center of Musculoskeletal Research, University of Rochester. Immunohistochemistry analysis against UCP1 (1:200) (Thermofisher, #PA1–24894) was performed using a DAB chromogenic kit (Vector SK-4105). Slides were baked at 60°C, deparaffinized in xylene, and rehydrated through graded ethanol to water. Antigen retrieval was performed using proteinase K (Invitrogen #25530–015) for UCP1 and pepsin (Sigma P-7000) in HCl for SNORC. Endogenous peroxidase activity was quenched, and sections were blocked with normal serum prior to overnight incubation with primary antibodies at 4°C. The following day, the signal was developed using a DAB chromogenic detection kit. Slides were rinsed, dehydrated, and cover slipped for brightfield imaging.

RNAScope Staining for Spine Sections

RNA in situ probes including Mm-Acan (439101), Mm-Apobec2 (482001), Mm-Chad (484881), Mm-Col14a1 (581941), Mm-Col27a1 (520001), Mm-Col6a2-C1 (1084001-C1), Mm-Comp (480711), Mm-Creb3l2-C1 (1079301-C1), Mm-Fmod (479421), Mm-Frzb (404861), Mm-Grep1-C1 (1755071-C1), Mm-Myh2 (401401), Mm-Ryr1-C1 (1120301-C1), and Mm-Sox9 (401051) were purchased from Advanced Cell Diagnostics (ACD). Thoracic spines were histologically processed following a protocol specific for skeletal tissue for RNA analysis⁵⁰. P20 spine tissues were fixed in 4% paraformaldehyde (PFA) at room temperature for 24 hours, washed 3X 15 minutes in RNAase-free 1X PBS, decalcified in a Morse’s Solution for 24 hours at room temperature, and washed again 3X 15 minutes in RNAase-free 1X PBS. Samples were immediately processed in a 70%, 80%, 90%, 95%, 100%, and 100% ethanol series followed by two changes of xylene, and two changes of paraffin. Samples were paraffin-embedded and sectioned at a 4 μm thickness. RNA in situ analysis was performed using RNAScope 2.5 HD Detection Reagent – RED kit (ACD Bio) following the manufacturer’s instructions, with modifications optimized for skeletal tissues as previously described⁵⁰.

Spatial Transcriptomics Cohort, Data, and Quality Control

Mouse thoracic spine sections from an Adgrg6 mutant and a wildtype control (n=1 per condition) were profiled using 10x Visium v1 spatial transcriptomics with our previously established workflow⁴⁶. To ensure disc centric signal and minimize off target high UMI contamination from adjacent tissues (e.g., muscle), intervertebral disc (IVD) regions were manually delineated in 10x Genomics Loupe Browser v8.0.0 Loupe Browser, and downstream analyses were restricted to these ROIs. Genomics Space Ranger (v3.0.0) was used to compute raw data, and the outputs were imported into Seurat (v.4.2.0) in R (v.4.2.1); spot level QC metrics (feature counts, UMI counts, mitochondrial percentage) were computed, and high quality IVD spots were retained using 200<nFeature_Spatial<8000 and %mt<5%.

Mitigating Sampling Bias: Muscle-filtering Strategies in Spatial Clustering

Differential sampling of muscle adjacent to the IVD can inflate global signals and confound disc-intrinsic DE. We therefore applied a gene-based exclusion bias-mitigation strategy using Muscle-score filtering. A per-spot “muscle-ness” score was computed from canonical muscle genes, and spots with Muscle percentage >1% were excluded. The cleaned dataset was re-normalized and re-clustered to recalibrate variance. Joint analysis was performed on the cleaned dataset using SCTransform residuals. Principal component analysis (PCA) was computed, and the effective dimensionality was set to PCs≈9 based on elbow shape and cumulative variance criteria. Downstream analysis utilized Seurat (v.4.2.0) build-in functions. UMAP embeddings were generated on PC 1:9, followed by k-NN graph construction and Louvain clustering across a resolution grid 0 to 1. Cluster stability was inspected with clustree, and a final working resolution of r=0.5 was selected for interpretability and modularity. Spatial context was obtained by overlaying cluster assignments on H&E images with SpatialDimPlot. Cluster annotation used FindAllMarkers (Wilcoxon; thresholds log 2 FC>0.25, p adj<0.05) and curated IVD marker sets to assign biological identities. For DE, count layers were unified and analyzed on the Spatial assay with log normalization to yield interpretable log2 FC and p-values (Wilcoxon test; thresholds log 2 FC>0.25, p adj<0.05).

Cluster Marker Identification

After muscle-score filtering and re-clustering, cluster identities were set. We conducted One-vs-all marker discovery used Seurat’s FindAllMarkers on the SCT assay. To emphasize cluster specificity, stringent markers were additionally defined as detected in more than 50% of spots in the target cluster and detected in fewer than 10% of spots outside that cluster. The top 20 up-regulated markers per cluster were ranked by fold change. Visualization included per-cluster volcano plots (one vs rest), and a heatmap of the top 15 significant markers per cluster.

Assessing Disc Programs: Key Genes and Pathway Analysis

To assess the disc program switch, spatial and UMAP feature plots were generated for Sox9 and key ECM/cartilage genes using log-normalized data from the Spatial assay to visualize regional expression differences between conditions. Gene Ontology over-representation analysis (GO-ORA; Biological Process) was performed with clusterProfiler on “Down in mutant” gene sets, using Benjamini–Hochberg adjustment; focus terms included cartilage development (GO:0051216) and extracellular matrix organization (GO:0030198). Gene Set Enrichment Analysis (GSEA) was conducted with clusterProfiler’s GSEA function using a ranked list created from the muscle-filtered Adgrg6-cKO vs Control comparison, ranking all genes by average log fold change (ranging from the most upregulated to the most downregulated genes). Cluster-specific marker gene sets were defined via FindAllMarkers on the re-clustered dataset, filtered at avg_logFC > 0.25 and p_adj<0.05. Pathway-level enrichment for “cartilage development” and “extracellular matrix organization was visualized showing the enrichment score “mountain” curve and the ranked-list “barcode.”

Fluorescence detection

Lumbar spine/IVD from P7 Sox9^EGFP mouse was dissected and fixed in 4% paraformaldehyde (in PBS) at 4 °C overnight, washed with PBS, and incubated in 30% sucrose at 4 °C overnight. Tissues were embedded in OCT, and frozen sections were cut at 16 μm using a cryostat. Slides were washed in PBST (Phosphate Buffered Saline with 0.1% Tween 20) 3 × 15 minutes at room temperature; and mounted under a coverslip with Aqua-mount (13800; Lerner Labs). EGFP fluorescence in intact spine/IVD was directly imaged, using a dissection Leica fluorescent microscope. EGFP fluorescence in cryosections was directly imaged at the Core for Imaging Technology & Education at Harvard Medical School.

Fluorescence-activated cell sorting from the spine/IVD of postnatal day 7 Sox9^EGFP mouse.

To determine whether Sox9 directly regulates the expression of Adgrg6 by binding to its regulatory elements in the spine/IVD, we employed FACS to isolate Sox9-expressing cells (EGFP+) and Sox9 non-expressing cells (EGFP-) from P7 Sox9^EGFP mice. The spine/IVD was dissected, and the tissues surrounding the vertebral body and intervertebral disc were removed under a fluorescent dissecting microscope. The dissected spine/IVD tissues were further digested in 2.5 % collagenase type II (Worthington, Cat#: LS004177) plus 0.5% collagenase type I (Worthington, Cat#: LS004196) at 37 °C in a thermomixer (Eppendorf ThermoMixer C, 700 rpm) for 2.5 hours to 3 hours. The homogenate was filtered with a 35 μm cell strainer (Falcon, Cat#: 352235). The cells were pelleted by centrifugation for 5 min at 300 g, washed with PBS at least four times, and resuspended in FACS sorting buffer (PBS plus 1% fetal bovine serum; DRAQ7 dye or DAPI was added to the solution at 1/1000 dilution). DRAQ7 dye (Novus Biologicals Cat# NBP2–81126) is a DNA dye that fluoresces at far-red upon DNA binding, and DAPI (4’,6-diamidino-2-phenylindole) is a DNA dye that fluoresces at blue; both are only capable of penetrating dead cells with compromised membrane structure. The cells were passed through the cell strainer (Falcon, Cat#: 352235) immediately before putting them in the FACS. The digested single cells were sorted using a SONY SH800 cell sorter to separate different populations that express or don’t express EGFP. For cell sorting, all particles (events) passing over the laser were first evaluated for forward scatter (FSC) and back scatter (BSC) to eliminate aggregates and debris by gating for those cells that were in the center of the FSC/BSC graph. The doublets were also eliminated by gating for those cells in the linear range of forward scatter height (FSC-H) and forward scatter area (FSC-A). Subsequently, the dead cells were eliminated by gating for those that were negative for DRAQ7 DNA dye or DAPI. After the gates were set, the laser power, gain, and all other parameters of the cell sorter were kept the same throughout the experiment. Although the same laser parameters were used across experiments, the process of gating was assessed for every litter that was sorted to ensure the accuracy of the gates.

ATAC-Seq and Cut&Run-Seq

After FACS, cells were washed twice in ice-cold PBS. 50,000 cells (per sample) were used to perform ATAC-Seq, and 500,000 cells (per sample) were used to perform Cut&Run-seq, following the detailed protocols described in our most recent work⁵¹.

Statistics

Statistical analysis and graphs were generated using R (v.4.2.1) and GraphPad Prism (version 10.2.1). Statistical significance was determined by a p-value of less than 0.05.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies and RNASCOPE probes
UCP1	Thermofisher	#PA1-24894
Mm-Frzb	ACD	404861
Mm-Col14a1	ACD	581941
Mm-Fmod	ACD	479421
Mm-Grep1	ACD	1755071-C1
Mm-Argef16-C1	ACD	1747211-C1
Mm-Sox9	ACD	401051
Mm-Acan	ACD	439101
Mm-Col6a2-C1	ACD	1084001-C1
Mm-Ryr1-C1	ACD	1120301-C1
Mm-Apobec2	ACD	482001
Mm-Myh2	ACD	401401
Mm-Comp	ACD	480711
Mm-Chad	ACD	484881
Mm-Creb3l2-C1	ACD	1079301-C1
Mm-Col27a1	ACD	520001
2.5HD Reagent Kit- RED	ACD	322350
Custom Pretreatment Catalog	ACD	300040
Sox9 antibody	Millipore	Cat#: AB5535
IgG antibody	Epicypher	Cat#:13-0042
CUTANA™ ChIC/CUT&RUN Kit	Epicypher	Cat#: 14-1048
Oligonucleotides
Cre genotyping primers forward: 5’- GCAGAACCTGAAGATGTTCGC -3’	IDT	N/A
Cre genotyping primers reverse: 5’- ACACCAGAGACGGAAATCCATC -3’	IDT	N/A
Adgrg6f/f genotyping primers reverse: 5’- CATTGAGCTCTCCCTGTATGG -3’	IDT	N/A
Adgrg6f/f genotyping primers reverse: 5’- CCTGCAGGTACCCACACATG -3’	IDT	N/A
Sox9del genotyping primers reverse: 5’- GTCTTTCTCTTTTATGGCCTGC -3’	IDT	N/A
Sox9del genotyping primers reverse: 5’- TGGCAAGTATTGGTCAAACTCA -3’	IDT	N/A
Chemicals, peptides, and recombinant proteins
DRAQ7 DNA dye2-81126	Novus Biologicals	NBP2-81126
Cell Strainer, 35 μm	Falcon	352235
Collagenase Type I	Worthington	LS004196
Collagenase Type II	Worthington	LS004177
Aqua-Mount Mounting Medium	Larner Labs	13800
SDS 10%	VWR	E719-100ML
Bluing Reagent	Epredia	6769001
Coplin Jars	VWR	100500-232
RNAase Away	Molecular BioProducts	7002
Tissue Marking and Margin Dye Kit- Black	Lieca Biosystems	3801753
ColorBond Tissue Marking Dye Mordant	StatLab	CBM-6
Foam Biopsy Sponges	StatLab	SL30B/T CS
Cytoseal	Thermo-Scientific	8310-16
Orange G	Sigma-Aldrich	O3756-25G
Alcoholic Eosin	Sigma-Aldrich	HT1100116-500ML
Alcian Blue	Sigma-Aldrich	B8438-250ML
Hydrochloric Acid	J.T.Baker	5621-02
Gills II Hematoxylin	Sigma-Aldrich	GHS232-1L
Harris Hematoxylin	Sigma-Aldrich	HHS16-500ML
Super HT PAP Pen Hydrophobic Slide Marker	Research Products International Corp.	195505
Immedge Hydrophobic Barrier Pen	Vector Laboratories	H-4000
Cover Glass	VWR	48393-195
VectaMount AQ Aqeous Mounting Media	Vector Laboratories	H-5501-60
SuperFrost Plus Microscope Slides	Fisher Scientific	12-550-15
Paraffin Type 6	Thermo Scientific	8336
Ethanol	KOPTEC	64-17-5
Xylene	Surgipath	3803665
Formic Acid	Thermo Scientific	64-18-6
Formic Acid Bone Decalcifier (Immunocal)	StatLab	1414-1
DEPC (Diethyl Pyrocarbonate)	Research Products International	1609-47-8
Accu-Edge Low Profile Blades	Feather Safety Razor Co.	4689
Histosette II Embedding Cassettes	Simport	M493-2
Sodium Citrate	Ward’s Science	470302-528
Acetone	Fisher Chemical	A38-212
Agarose	VWR	N605
DNase I	NEB	M0303A
EcoRV-HF	NEB	R3195S
Ethanol	Decon Labs, Inc.	2716
Glycerol	Sigma Aldrich	G7893-500ML
GoTaq Green	Promega	M7123
Instant Ocean Sea Salt	Instant Ocean	Instant Ocean
Low Melt Agarose	IBI Scientific	IB70056
Methanol	Fisher Chemical	A454-1
Methylcellulose	Sigma	M0387
Neutral Buffered Formalin, 10%	VWR	89370-094
Paraformaldehyde	Thermo Scientific	043368.9M
Phosphate Buffered Saline (10X)	Corning	20-031-CV
Pronase	Roche	10 165 921 001
Sodium Hydroxide (NaOH)	Fisher Chemical	S318-1
SpeI	NEB	R3133S
TBE (10X)	VWR	E442
Tricaine-methanesulfonate	Syndel	200-266
Tris-HCl	Sigma	T3253
TRIzol Reagent	Ambion by Life Technologies	15596026
Tween20	Promega	H5151
Isoflurane	Piramal Critical Care	NDC 66794-013-25
DAB Chromogenic detection kit	Vector	SK-4105
Vectastain Elite Universal Kit	Vector	PK-6200
Proteinase K	Invitrogen	25530-015
Pepsin	Sigma	P-7000
Experimental models: Organisms/strains
B6.Cg-Tg(Col2a1-cre)3Amc/RgrayJ	The Jackson Laboratory	RRID:IMSR_JAX:039049
Sox9 Asp272del	33	Sox9del
Adgrg6-floxed	Taconic Biosciences	TF0269
Sox9 IRES-EGFP transgenic mice	52	JAX: 030137
Software and algorithms
3D Slicer	53	v5.6.2
3D slicer morph	49	V5.15
DataViewer	Bruker	v1.6.0.0
GraphPad Prism	GraphPad	https://www.graphpad.com/features
ImageJ/FIJI	54	v2.14.0
Seurat	55	v.4.2.0
R	https://www.r-project.org/	v.4.2.1
LoupeBrowser	10X Genomics	v8.0.0
SpaceRanger	10X Genomics	v3.0.0
Other
Keyence Microscope	Keyence	BZX-700
Confocal Microscope	Nikon	CSU-W1 Yokogawa
MicroCT Instrument	Bruker	Skyscan 1276

• Adgrg6 maintains Sox9 and extracellular matrix gene expression in the intervertebral disc

• SOX9 directly occupies open chromatin at the Adgrg6 locus in SOX9⁺ skeletal cells

• Adgrg6 and Sox9 genetically interact to increase scoliosis incidence and severity

• Combined loss of Adgrg6 and Sox9 drives ectopic cartilage formation and reduced vertebral bone mineral density in the thoracic spine

Statistics Code Availability

All code used to generate spatial transcriptomics results has been deposited on Github (https://github.com/xuziyi0909/Mouse_Spine_Intervertebral_Disc_SpatialTranscriptomics).