All animal protocols were approved by the Animal Care and Use Committee of Wuxi Affiliated Hospital of Nanjing University of Chinese Medicine (approval no. GZR2023032801; Wuxi, China). Mice were housed in groups of ≤5 animals in a room using microisolator technology, and kept at 22-24°C with a 12-h light/dark schedule, with ad libitum access to food and water.

Kdm6a-Flox (a conditional knockout strain where essential exons of the Kdm6a gene are flanked by loxP sites), collagen type II α1 chain (Col2a1)-2A-tamoxifen (Tam)-inducible Cre recombinase (CreERT2) (expressing CreERT2 specifically in CHs, driven by the Col2a1 promoter) and R26-CAG-LSL-tdTomato (a Cre-dependent reporter strain in which a loxP-flanked transcriptional stop cassette prevents the expression of the tdTomato fluorescent protein; upon Cre-mediated recombination, tdTomato is expressed, permanently labeling Cre-active cells and their progeny) mice were purchased from Shanghai Model Organisms Center, Inc. All mutant mice had a C57/B6J background. Col2a1-2A-CreERT2 mice were bred with R26-CAG-LSL-tdTomato mice to generate inducible cartilage-specific labeled mice that carried a Tam-inducible CreERT2 transgene under the control of the Col2a1 gene promoter for CH-specific expression, and the Cre/loxP-inducible Rosa-CAG-LSL-tdTomato reporter to mark the cells after CreER activity was induced by Tam injections. To generate inducible cartilage-specific Kdm6a-knockout mice, Kdm6a^flox/flox mice were bred with Col2a1-2A-CreERT2&R26-CAG-LSL-tdTomato mice to generate Kdm6a^flox/+&Col2a1-2A-CreERT2&R26-CAG-LSL-tdTomato (fx/wt) mice, and these mice were intercrossed or bred with Kdm6a^flox/flox mice again to generate Kdm6a^flox/flox& Col2a1-2A-CreERT2&R26-CAG-LSL-tdTomato (fx/fx) mice and fx/wt mice (44). Fx/wt mice were used as controls. DNA was obtained from tail biopsy samples and genotyping was performed using PCR (2X Taq Plus Master Mix; P212; Vazyme Biotech Co., Ltd.). The conditions for amplification reactions were as follows: 94°C for 3 min, followed by 34 cycles of 94°C for 30 sec, 56°C (Kdm6a-Flox and Col2a1-CreERT2)/58°C (R26-CAG-LSL-tdTomato) for 30 sec and 72°C for 30 sec. The primer sequences are shown in Table SI. The PCR products were separated on a 3% agarose gel, stained with GelStain fluorescent nucleic acid dye (TransGen Biotech Co., Ltd.) and visualized using a fully automatic digital gel imaging analysis system (Tanon Science and Technology Co., Ltd.). As described in Data S1, an example of mouse tail clipping PCR genotype identification results is shown in Fig. S1. A total of 120 male mice were used in the present study. At the start of the experiment, the mice were 6 months old, with body weights ranging between 25 and 30 g. According to their genotype, the mice were divided into two groups: The inducible cartilage-specific Kdm6a-knockout group (fx/fx; n=60) and the control group (fx/wt; n=60). Mice from each group were then randomly allocated to different experimental endpoints as follows: scRNA-seq (n=6 per genotype), histological staining (n=15 per genotype), radiological examination (n=6 per genotype) and cytological experiments (n=33 per genotype). After the primary CHs were extracted from the mouse cartilage callus, KDM6A protein expression in the cells of the two groups was detected by western blotting to verify successful gene knockout (Fig. S2). Live-cell imaging was performed during the in vitro induction of CH transdifferentiation into osteoblasts in the cartilage callus. Imaging was performed directly after transferring the cells into osteogenic induction culture medium. A fluorescence microscope (Carl Zeiss AG) equipped with an environmental control chamber was used to maintain the cells at 37°C with 5% CO₂ during observation. Imaging was performed for 48 h, capturing an image every 30 min. Fluorescence imaging showed that the CHs specifically expressed a clear red labeling signal, and the Col2a1-CreERT2/RosatdTomato model was successfully established (Fig. S3).

Fx/fx and fx/wt mice were intraperitoneally injected with 75 mg/kg Tam (Sigma-Aldrich; Merck KGaA) once daily for 5 days. The fracture model was established on the second day after induction. Mice were weighed and anesthetized by intraperitoneal injection of 0.4% pentobarbital sodium at a dose of 40 mg/kg. After the mice were completely anesthetized, the skin was prepared in front of the left tibia, and hair was removed as cleanly and thoroughly as possible. Mice were placed in the left lateral position. The surgical site on the left lower limb was disinfected three times with iodine, followed by draping with sterile fenestrated drapes. Using high-temperature and high-pressure sterilized instruments, a small incision of ~1 cm was made along the longitudinal axis of the left tibia, from the knee joint to the middle section of the tibia. Hemostatic forceps were used to expose the anterior edge of the tibia to obtain the skin, muscles and fascia. Surgical blade 11 was positioned at the highest point of the anterolateral protrusion of the tibia. A transverse fracture was created by striking the base of the blade vertically with a small hammer and a slight lateral movement was performed to confirm the fracture. A sterile 1-ml syringe needle was inserted into the medulla for fixation, and the exposed needle was removed (45). After disinfection with Aner iodine, the incision was sutured layer-by-layer using a 5-0 medical suture needle and thread. After the operation, 40,000 U penicillin was injected intramuscularly for 3 consecutive days to prevent infection. The mice were raised without any activity restrictions. The incision conditions and activity of the mice were monitored daily. Fractured callus tissues were harvested on days 10 and 21 post-fracture for single-cell sequencing. The samples at each time point were obtained from 6 different mice. Samples of the left tibiae from 12, 12 and 18 different mice were collected on days 10, 21 and 28 post-fracture for radiological and histological analyses.

Bone callus samples were dissociated into single-cell suspensions. The cell concentration and viability were determined using a Countstar Rigel S2 automated cell fluorescence analyzer (Shanghai Ruiyu Biotech Co., Ltd.) with AO/PI staining (RE010212; Shanghai Ruiyu Biotech Co., Ltd.) and Countstar counting slides (CO010101; Shanghai Ruiyu Biotech Co., Ltd.). The concentration was adjusted to 700-1,200 cells/μl. Single-cell capture, barcoding and library preparation were performed using the Chromium Next GEM Single Cell 3' Reagent Kits v3.1 (cat. no. 1000268; 10x Genomics). Cell suspensions were combined with barcoded gel beads and master mix, and encapsulated into gel beads-in-emulsion (GEMs) using a microfluidic system. Within the GEMs, poly (dT) primers bound to mRNA poly-A tails to initiate reverse transcription (RT). Full-length barcoded cDNA was generated by RT using the integrated RT reagents: RT Reagent B (cat. no. 2000165; buffer), RT Enzyme C (M-MLV Reverse Transcriptase; cat. no. 2000085/2000102), Template Switch Oligo (cat. no. 3000228) and Reducing Agent B (cat. no. 2000087) from the Chromium Next GEM Single Cell 3' Reagent Kit v3.1 (10x Genomics). The RT reaction was performed at 53°C for 45 min, followed by enzyme inactivation at 85°C for 5 min. The GEMs were then broken, and the pooled cDNA was amplified by PCR using a high-fidelity DNA polymerase from the Amp Mix (cat. no. 2000047/2000103; 10x Genomics) with the following primers: Forward, 5'-CTACACGACGCTCTTCCGATCT-3' and reverse, 5'-AAGCAGTGGTATCAACGCAGAG-3'. The thermocycling conditions were as follows: 98°C for 3 min; cycling (12-13 cycles, optimized for cell recovery) at 98°C for 15 sec, 63°C for 20 sec and 72°C for 1 min; and 72°C for 1 min. The quality and size distribution of the amplified cDNA were verified using an Agilent 2100 Bioanalyzer (Agilent Technologies, Inc.), showing a main peak of 1,000-2,000 bp. The cDNA was fragmented (~400 bp on average) and sequencing adapters were added. A final sample indexing PCR was conducted using the same Amp Mix and the Dual Index Kit TT Set A primers (cat. no. 1000215; 10x Genomics) containing unique i5 and i7 index sequences. The sequences were: P5 primer (with i5 index), 5'-AATGATACGGCGACCACCGAGATCTACAC-N10-ACACTCTTTCCCTACACGACGCTC-3'; P7 primer (with i7 index), 5'-CAAGCAGAAGACGGCATACGAGAT-N10-GTGACTGGAGTTCAGACGTGT-3' (where N10 represents the unique 10-bp index sequence for each sample). The indexing PCR conditions were: 98°C for 45 sec; cycling (8-16 cycles, optimized for cDNA input) at 98°C for 20 sec, 54°C for 30 sec and 72°C for 20 sec; and 72°C for 1 min. The concentration of the final sequencing library was determined by quantitative PCR (qPCR) to ensure accurate loading for sequencing. The quantification was performed using the VAHTS Library Quantification Kit for Illumina^® (cat. no. NQ101; Vazyme Biotech Co., Ltd.) according to the manufacturer's protocol. The fluorescent dye premix solution and the primers (forward, 5'-AATGATACGGCGACCACCGA-3'; reverse, 5'-CAAGCAGAAGACGGCATACGA-3') were all included in this kit. This method provides absolute quantification of the library. Briefly, a dilution series of the library was prepared and used as template alongside the DNA standards (cat. no. NQ105; Vazyme Biotech Co., Ltd.) to generate a standard curve. Each qPCR reaction was set up in triplicate using the provided premix. Amplification was carried out on a qPCR instrument with a standard cycling protocol appropriate for the kit (95°C for 5 min, followed by 35 cycles of 95°C for 30 sec and 60°C for 45 sec). The library concentration was calculated according to the kit's protocol. A standard curve was generated using the provided DNA standards of known concentrations. The Ct values of the library samples were interpolated from this standard curve to obtain a preliminary concentration. This value was then corrected for the average library length using the following formula: Corrected concentration (pM)=[452 bp/mean library length (bp)] × preliminary concentration (pM). The final, accurate concentration of the original library (nM) was derived by multiplying the corrected concentration by the dilution factor. Following quantification, the pooled libraries were diluted to a final concentration of 200 pM and loaded onto an Illumina NovaSeq 6000 flow cell (Illumina, Inc.) for cluster generation and sequencing. Paired-end 150-bp sequencing was performed on an Illumina NovaSeq 6000 platform (Illumina, Inc.) using the Illumina NovaSeq 6000 S4 Reagent Kit (cat. no. 20028312; Illumina, Inc.). Raw sequencing data were processed using CellRanger (v8.0.1; 10x Genomics). Downstream analysis was performed in R (v4.1.2; https://www.r-project.org/) utilizing the Seurat package (v4.1.1; https://satijalab.org/seurat/).

The raw sequencing data were processed using CellRanger for genome alignment, background cell filtering and unique molecular identifier (UMI) counting of cellular transcripts. Data integration across multiple samples was performed using the anchor-based canonical correlation analysis method as previously described (46). For the cell gene expression matrix generated by CellRanger, Seurat software was used to filter dead or stressed cells by removing cells with abnormal gene or UMI counts, high mitochondrial gene content and potential doublets. The filtering criteria were set as follows: Number of genes per cell, 500-infinity; total UMIs per cell, < infinity; and proportion of mitochondrial gene expression per cell, <25%; and doublets were removed using the DoubletFinder package (version 2.0.4; https://github.com/chris-mcginnis-ucsf/DoubletFinder). The distribution of basic information of each sample cell after cell filtering is shown in Figs. S4 and S5.

After quality control, only the high-quality cells suitable for downstream analyses were retained (Data S1; Tables SII and SIII). During data preprocessing, regression analysis was performed to eliminate the effects of total UMI counts and mitochondrial gene proportion, and expression data were normalized using a scale factor of 1x10⁴. Highly variable genes were selected using the FindVariableGenes algorithm in the Seurat package (parameters: x.low.cutoff, 0.0125; y.cutoff, 0.5), and principal component analysis (PCA) was performed based on these genes. The top 100 significant principal components were selected, and cell sub-clusters were identified using the FindClusters function (derived from Seurat v4.1.1; dims, 1:30; resolution, 0.5). Finally, uniform manifold approximation and projection (UMAP) was employed for dimensionality reduction and visualization of the gene expression data.

Following cell clustering and annotation, the expression patterns of key marker genes across subpopulations were visualized. UMAP plots depicting the expression of individual genes were generated using the FeaturePlotfunction in Seurat, where color intensity corresponds to the expression level of the feature gene in each cell. Violin plots were used to display the distribution of expression for these marker genes across the annotated cell clusters. Additionally, a dot plot was created to summarize the expression of characteristic marker genes across the main CH-osteoblast lineage subpopulations. In this plot, the size of the dot represents the percentage of cells within a cluster expressing the gene, and the color gradient indicates the mean expression level.

Based on marker genes and referencing the CellMarker2.0 database (http://117.50.127.228/CellMarker/), cell type annotation was performed. For enrichment analysis, genes with specific high expression in each cell subcluster (top 10) were subjected to Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis using the clusterProfiler R package (v4.2.2; https://bioconductor.org/packages/clusterProfiler) to interpret potential biological functions. The enrichment results were visualized using the ggplot2 package (v3.4.0; https://cran.r-project.org/web/packages/ggplot2/index.html), with bubble plots and bar charts displaying significantly enriched pathways. Gene set enrichment analysis (GSEA) was conducted using GSEA software (v4.4.0; https://www.broadinstitute.org/gsea/), and the results were visualized using the GseaVis R package (v0.1.1; https://github.com/junjunlab/GseaVis).

Pseudotime trajectories were constructed by integrating highly variable genes from cell subpopulations using Monocle3 (v1.3.4; http://cole-trapnell-lab.github.io/monocle3/). Trajectory visualization was achieved in combination with UMAP. The learn-graph algorithm was employed to define CHs as the root of the trajectory, and dynamic changes in gene expression along pseudotime were analyzed using the plot-genes-in-pseudotime function.

The dataset was analyzed using the CellChat package (version 2.1.2; https://github.com/sqjin/CellChat) in R (version 4.4.1; https://www.r-project.org/). CellChat models the intercellular communication probability by applying the law of mass action and integrating gene expression data with prior knowledge of the interactions among signaling ligands, receptors and their cofactors. Based on cell type annotation and identification of upregulated genes, the computeCommunProb function was used to infer ligand-receptor interaction probabilities, the computeCommunProbPathway function was used to quantify pathway-level communication strength and key interactions were visualized as chord diagrams using the netVisual-chord-gene function.

After fixation in 4% paraformaldehyde at room temperature for 24 h, the samples were decalcified in an EDTA decalcification solution at room temperature. The decalcification solution was changed every 2-3 days. During each solution change, the progress of decalcification was assessed by testing the tissue firmness. Decalcification was considered complete when a 1-ml syringe needle could be inserted into the tibial cortex without resistance. Once decalcification was complete, tibial specimens were rinsed under running water for 4 h, dehydrated, cleared, embedded in paraffin and sectioned. The thickness of sections was 3 μm. The sections were stained with H&E (Beijing Solarbio Science & Technology Co., Ltd.) using hematoxylin solution for 5 min and eosin solution for 1 min, and safranin O-Fast Green (Wuhan Servicebio Technology Co., Ltd.) using Fast Green for 5 min followed by Safranin O for 2 min according to the manufacturer's instructions at room temperature. Images were acquired using a panoramic digital slide scanner (3DHISTECH, Ltd.), which is a high-resolution whole-slide imaging system based on bright-field light microscopy.

The sample fixation and decalcification steps were the same as those for H&E and safranin O-Fast Green staining. The thickness of paraffin sections was 3 μm. The paraffin sections were placed one by one into xylene I and xylene II for dewaxing for 10 min each, then soaked in absolute ethanol I and absolute ethanol II for 5 min each, rehydrated in 95, 85 and 75% ethanol for 3 min each, and finally rinsed in distilled water for 2 min. Subsequently, the paraffin sections were immersed in a 1X antigen retrieval solution and heated at 100°C for 20 min. After cooling to room temperature, the sections were washed with PBS. Permeabilization was performed by incubating the tissues with Triton X-100 (ready-to-use working solution; cat. no. P0096; Beyotime Biotechnology) for 30 min, followed by incubation with 3% BSA (Beijing Biosynthesis Biotechnology Co., Ltd.) at room temperature for 30 min to block nonspecific binding. The sections were then incubated overnight at 4°C with diluted primary antibody [anti-osteocalcin (OCN); 1:400; cat. no. ER1919-20; Hangzhou HuaAn Biotechnology Co., Ltd.] After washing with PBS, the sections were incubated with a fluorescent secondary antibody [anti-rabbit IgG (H+L), F(ab')2 Fragment; Alexa Fluor^® 488 Conjugate; 1:1,000; cat. no. 4412; Cell Signaling Technology, Inc.] in the dark at room temperature for 50 min. After washing with PBS, the nuclei were counterstained with DAPI at room temperature for 10 min. Fluorescence images were acquired using a panoramic digital slide scanner, and analyzed using ImageJ software (1.54p; National Institutes of Health).

On day 10 after modelling, CHs were isolated from the fracture calli of the mice. The mice were euthanized by cervical dislocation, and the left lower limb was removed and placed in a pre-chilled culture dish containing PBS supplemented with penicillin-streptomycin at a 1:9 ratio. Using sterilized instruments, the skin was incised at the knee joint, and the surrounding hair, muscle, fascia and vascular tissues were carefully separated to expose the fracture site. The cartilaginous callus was isolated from the surrounding fibrous tissue, and translucent cartilage was scraped off using a blade, taking care to remove cortical bone and other bony tissues. A randomly selected piece of the removed cartilage tissue was subjected to Alcian blue and Alizarin red combined staining (data not shown) to ensure that the cartilage tissue collected did not contain bone tissue. Briefly, the tissue fragment was fixed in 4% paraformaldehyde at 4°C for 24 h, embedded in paraffin and sectioned at 4 μm thickness. Sections were stained with Alcian Blue (pH 2.5) for 30 min and Alizarin Red S solutions for 5 min at room temperature according to standard protocols, and observed under a light microscope (Olympus BX53; Olympus Corporation).

The collected cartilage was placed in a pre-chilled culture dish containing PBS with penicillin-streptomycin (1:9 ratio) and kept on ice. The cartilage was cut into small fragments of ~1 mm using fine scissors, and the cartilage fragments were transferred into a 15-ml centrifuge tube.

Subsequently, 4 ml of 0.25% trypsin-EDTA solution was added and the sample was incubated at 37°C for 15 min to remove any attached fibroblasts or osteoblasts. During digestion, the tube was manually shaken for 15 sec every 5 min to ensure complete digestion. The digestion was terminated by adding 4 ml complete CH medium [10% FBS (Sigma-Aldrich; Merck KGaA), 1% penicillin-streptomycin and 89% DMEM/F12 (1:1) (HyClone; Cytiva)], and the sample was centrifuged at 200 × g at 4°C for 5 min. The supernatant was discarded and the tissue was washed twice with PBS.

A mixture of 1 ml hyaluronidase (1 mg/ml; ~400 U), 1 ml collagenase IA (2 mg/ml; >250 U) and 1 ml DNase II (0.25 mg/ml; >250 U) was prepared, and 3 ml of this enzyme mixture was added to the tissue. The samples were digested at 37°C for 30 min in an incubator. Digestion was terminated by adding 3 ml complete CH medium, and the sample was centrifuged at 200 × g at 4°C for 5 min. The supernatant from the first 30-min digestion was discarded to remove remaining fibroblasts and osteoblasts.

A new aliquot (3 ml) of the enzyme mixture was then added, and the tissue was digested for an additional 30 min at 37°C. After adding 3 ml complete medium to terminate digestion and allowing the cartilage to settle, the supernatant was collected and filtered through a 70-μm cell strainer, followed by centrifugation at 200 × g at 4°C for 5 min. The supernatant was removed and the cell pellet was gently resuspended in 1 ml complete CH medium.

This process of enzymatic digestion, centrifugation and cell collection was repeated with a freshly prepared enzyme mixture until all CHs were released, as indicated by the absence or minimal presence of cell pellets in the centrifuge tube after spinning the supernatant from the digestion. This protocol can yield highly purified cartilage cell preparations characterized by strong expression of cartilage formation markers and a strong tendency for cells to synthesize a cartilage matrix rich in proteoglycans (47).

After CHs isolated from the fracture callus in both groups of mice had established stable growth to the appropriate size and density under hypoxic conditions (6% O₂), the cells were randomly divided into four groups: fx/wt hypoxia group (continued culture in 6% O₂), fx/wt normoxia group (cultured in 21% O₂), fx/fx hypoxia group (continued culture in 6% O₂) and fx/fx normoxia group (cultured in 21% O₂). The oxygen concentrations of 6 and 21% were selected based on the established oxygen-sensing property of KDM6A. A seminal study demonstrated that the enzymatic activity of its JmjC domain was inhibited below ~5% O₂ (22). Therefore, 6% O₂ represents a point where activity potentially begins to recover, while 21% O₂ (atmospheric level) reflects its fully active state. This range allows interrogation of the functional continuum of KDM6A from inhibition to full activation, thereby enabling a clear analysis of the axis linking oxygen levels, KDM6A-mediated epigenetic regulation and cell fate.

The original CH culture medium [10% fetal bovine serum, 1% penicillin-streptomycin and 89% DMEM/F12 (1:1)] was replaced with osteogenic induction medium [α-Minimum Essential Medium (HyClone; Cytiva) supplemented with 10% fetal bovine serum, 1% L-glutamine, 1% penicillin-streptomycin, 10 mM β-glycerophosphate, 0.2 mM ascorbate-2-phosphate and 100 ng/ml bone morphogenetic protein 2 (BMP-2)] to induce the transdifferentiation of CHs into osteoblasts at 37°C for 7 days. The culture medium was replaced every 2-3 days.

After 7 days of osteogenic induction, the cells seeded in 24-well plates (2.5x10⁴/cm²) were removed from the incubator. The culture medium was discarded and the cells were washed with PBS and fixed with 4% cell fixative (P1110; Beijing Solarbio Science & Technology Co., Ltd.) at room temperature for 15 min. According to the manufacturer's protocols, staining was performed as follows: Alcian blue (E670107; Sangon Biotech Co., Ltd.) staining for 30 min at room temperature, ALP (G1481; Beijing Solarbio Science & Technology Co., Ltd.) staining for 2 h at 37°C and Alizarin red (G1038; Wuhan Servicebio Technology Co., Ltd.) staining for 30 min at room temperature. Images were observed under a light microscope and analyzed using ImageJ software (1.54p; National Institutes of Health).

Total mRNA was extracted from each cell sample using TRNzol reagent (Tiangen Biotech Co., Ltd.), and cDNA was obtained by RT using the PrimeScript™ FAST RT reagent Kit with gDNA Eraser (Takara Bio, Inc.) according to the manufacturer's instructions. Reverse-transcribed cDNA was used as the template, and primers and TB Green Premix Ex Taq II (Tli RNaseH Plus) (Takara Bio, Inc.) were added to perform qPCR. The amplification reaction conditions were as follows: 95°C for 30 sec, followed by 40 cycles of 95°C for 5 sec and 60°C for 34 sec. Finally, β-actin was used as an internal reference, and the relative expression levels of the target genes were calculated using the 2^−ΔΔCq method (48). The mRNA levels of Col2a1, collagen type X α1 chain (Col10a1), aggrecan (Acan), collagen type I α1 chain (Col1a1), secreted phosphoprotein 1 (Spp1), secreted protein acidic and cysteine rich (Sparc) and β-actin were analyzed using the primer sequences listed in Table I.

The protein extraction buffer was prepared with a ratio of RIPA lysis buffer (Jiangsu Kangwei Century Biotechnology Co., Ltd.) to proteinase inhibitor mixture (Jiangsu Kangwei Century Biotechnology Co., Ltd.) of 100:1. Cells were lysed in the protein extraction buffer on ice for 20 min to extract protein samples. When analyzing dimethylation of H3K27 (H3K27me2) and H3K27me3, nuclear lysates were prepared using the Nuclear and Cytoplasmic Protein Extraction Kit (P0028; Beyotime Biotechnology), with histone H3 serving as the nuclear reference. The protein concentration was determined using the BCA method. Equal amounts (20 μg/lane) of protein were loaded onto a sodium dodecyl sulfate-polyacrylamide gel (4-12%) for electrophoresis. After electrophoresis, the proteins were transferred to a PVDF membrane, and the membrane was blocked with a protein-free rapid blocking solution (Boster Biological Technology) at room temperature for 10 min. Primary antibodies, including anti-KDM6A antibody (1:1,000; cat. no. 33510; Cell Signaling Technology, Inc.), anti-Wnt3a antibody (1:1,000; cat. no. 2391; Cell Signaling Technology, Inc.), anti-β-catenin antibody (1:1,000; cat. no. 37447; Cell Signaling Technology, Inc.), anti-RUNX2 antibody (1:1,000; cat. no. ab236639; Abcam), anti-GAPDH antibody (1:2,000; cat. no. ab8245; Abcam), anti-histone H3 (di methyl K27) antibody (1:1,000; cat. no. 9728; Cell Signaling Technology, Inc.), anti-histone H3 (tri methyl K27) antibody (1:1,000; cat. no. 9733; Cell Signaling Technology, Inc.) and anti-histone H3 antibody (1:2,000; cat. no. 4499; Cell Signaling Technology, Inc.), diluted in antibody diluent according to the manufacturer's instructions, were then added for incubation overnight at 4°C with gentle shaking. The membranes were washed with TBS with 0.1% Tween-20 and incubated with the appropriate horseradish peroxidase-conjugated secondary antibodies (1:4,000; cat. no. S0001 or S0002; Affinity Biosciences, Ltd.) at room temperature for 1 h. The proteins were then exposed and stained with ECL reagent. On the same membrane, different primary antibodies were incubated successively. Before changing the primary antibody, the previous round of primary/secondary antibodies was eluted with Stripping buffer (Beyotime Biotechnology). The target bands were analyzed using ImageJ software (1.54p; National Institutes of Health) for gray-value calculations.

ChIP-PCR was used to observe the changes in H3K27me2 and H3K27me3 in the promoter region of Wnt3a, β-catenin and RUNX2 in the fx/wt normoxia group and the fx/fx normoxia group after 3 days of osteogenic induction. The ChIP assay was performed using the ChIP Assay Kit (cat. no. P2078; Beyotime Biotechnology) according to the manufacturer's instructions. The cells were treated with 1% formaldehyde to form covalent cross-links between intracellular DNA-bound proteins and DNA and the cross-linking reaction was quenched with 125 mM glycine. Cells were lysed with 1 mM PMSF in SDS Lysis Buffer (included in the ChIP Assay Kit) (200 μl/10⁶ cells), and the genomic DNA was subsequently fragmented by sonication (20 kHz; 4 sec pulse; 8 sec interval; at a rate of 5 times per min for a total of 5 min in an ice-water bath) to obtain random fragments typically ranging between 200 and 1,000 bp. Specific antibodies against the target proteins were used, including anti-histone H3 (di methyl K27) (1:50; cat. no. 9728; Cell Signaling Technology, Inc.), anti-histone H3 (tri methyl K27) (1:50; cat. no. 9733; Cell Signaling Technology, Inc.) and IgG (cat. no. 2729; Cell Signaling Technology, Inc.). These antibody-chromatin complexes were subsequently captured with 60 μl Protein A+G Agarose beads (included in the ChIP Assay Kit) to enrich the DNA fragments bound by the target proteins. Subsequently, the instructions in the kit (cat. no. P2078; Beyotime Biotechnology) were followed to wash the precipitate. The wash procedure was performed sequentially with Low Salt Immune Complex Wash Buffer, High Salt Immune Complex Wash Buffer, LiCl Immune Complex Wash Buffer and TE Buffer. The cross-linking of proteins and DNA was relieved by centrifugation (1,000 × g; 1 min; 4°C) and heating, and proteins were degraded using proteinase K. The enriched DNA fragments were purified using a DNA purification kit (cat. no. D0033; Beyotime Biotechnology). Purified DNA was used as a template for PCR amplification using PrimeSTAR^® HS DNA Polymerase (Takara Bio, Inc.). The primer sequences are listed in Table II. The amplification reaction conditions were as follows: 30 cycles of 98°C for 10 sec, 60°C for 15 sec and 72°C for 50 sec. The PCR products were separated by 1.5% agarose gel electrophoresis, stained with GelStain fluorescent nucleic acid dye (TransGen Biotech Co., Ltd.) and visualized using a fully automatic digital gel imaging analysis system (Tanon Science and Technology Co., Ltd.).

Micro-CT was used to assess the bone structure and fracture healing (49). At 28 days post-fracture, the mice underwent in vivo micro-CT scanning and three-dimensional reconstruction to observe morphological changes at the fracture site. The mice were initially anesthetized in an induction chamber with 4% isoflurane delivered in air at a flow rate of 1 l/min via a gas anesthesia machine. Once anesthetized, the mouse was transferred to a nose cone for maintenance on 1.8% isoflurane in air at a flow rate of 0.8 l/min throughout the scanning procedure. The positions of the mouse and scanning chamber were carefully fixed, an appropriate scanning bed was selected, and imaging was performed, followed by three-dimensional reconstruction to obtain images for subsequent analysis.

For ex vivo scanning, the bone samples were placed in a small sample holder for scanning and three-dimensional reconstruction to collect data for subsequent analyses. On day 28 post-fracture, tibial specimens were collected and then fixed in 4% paraformaldehyde at 4°C for 48 h. After fixation, the samples were wrapped in gauze and placed in a small ex vivo chamber for standard micro-CT scanning and three-dimensional reconstruction (parameters: tube voltage, 70 kV; tube current, 150 μA; reconstruction algorithm, Feldkamp-Davis-Kress; resolution, 1k × 1k). After reconstruction, the images were segmented and bone analysis was performed using Avatar3 software [2.0.12.0; PINGSENG Healthcare (Kunshan) Inc.]. The region of interest was defined as an area 1 mm above and below the fracture line. Quantitative analyses were performed to determine the trabecular bone volume fraction (BV/TV), trabecular bone surface area fraction (BS/TV), trabecular number (Tb.N) and trabecular thickness (Tb.Th).

Cluster-specific marker genes were identified using the Wilcoxon rank-sum test implemented in the FindAllMarkers function of the Seurat package, with an adjusted P-value <0.05 and |avglog₂ fold change (FC)|>0.25 as thresholds. Differentially expressed genes between groups were explored using the FindMarkers function with the built-in Wilcoxon rank-sum test, with an adjusted P-value <0.05 and |avglog₂FC|>0.15. The experimental data were obtained from at least three independent replicates. Statistical analyses of all data were performed using GraphPad Prism (version 8.0.2; Dotmatics). Data were normally distributed. Comparisons between two groups were analyzed using a two-tailed, unpaired Student's t-test. For comparisons among multiple groups, one-way ANOVA was performed, followed by Tukey's post hoc test. Data are presented as the mean ± SD. P<0.05 was considered to indicate a statistically significant difference.

First, six fx/wt (WT) mouse fracture callus samples were analyzed, including three samples collected at 10 days post-fracture (WT1_10, WT2_10 and WT3_10) and three samples collected at 21 days post-fracture (WT1_21, WT2_21 and WT3_21). Following quality control and data filtering, 56,189 high-quality deeply sequenced cells were obtained and 25 distinct clusters were identified. Each cluster was defined by comparing its expression profile with that of known lineage markers or canonical gene signatures.

During cell annotation, dimensionality reduction and unbiased clustering were performed based on gene expression profiles and classical markers of the cells, resulting in the identification of 19 clusters (Fig. 1A). There were distinct differences in the UMAP plots between callus samples collected 10 and 21 days after fracture, indicating different cellular compositions at these stages (Fig. 1B). The present study focused particularly on six CH subpopulations: i) Proliferative CHs, characterized by high expression levels of Ucma, C1qtnf3, Cnmd, Itm2a and Scrg1; ii) pre-HTCs, marked by elevated levels of Cdkn1c, Hapln1, Ppa1, Pth1r and Snorc; iii) HTCs, expressing Col10a1, Cst3, Spp1, Ihh and Ddit4l; iv) CH-derived osteoprogenitors (CDOPs), distinguished by high expression levels of Mmp13, Spp1, osteomodulin (Omd), matrix Gla protein (Mgp) and MAF bZIP transcription factor B (Mafb); v) osteoblasts, identified by the expression of Bglap, Bglap2, Ifitm5, Ibsp and Col1a1; and vi) osteoclasts, which specifically expressed Acp5, Ctsk, Ckb, S100a4 and Mmp9 (Fig. 1C).

Proportion plots were generated for each cell type (Fig. 1D). At 10 days post-fracture, CHs in the tissue were in a rapid proliferative phase, with pre-HTCs, HTCs and CDOPs constituting a larger proportion. Osteoblasts were already present but at relatively low levels. By day 21, the proportion of CHs had markedly decreased, indicating that hypertrophy and transdifferentiation were near completion, whereas the numbers of endothelial cells and osteoclasts had increased, accelerating vascular invasion and cartilage matrix degradation.

The volcano plot demonstrated that a special CDOP subset was distinguished by the co-expression of a specific set of marker genes, including Mmp13, Spp1, Mgp, Omd and Mafb (Fig. 1E). The expression profiles of key representative genes during CH transdifferentiation, including Col2a1, Sox9, Col10a1, Mmp13, RUNX2, Spp1, Sp7 (Osx) and Col1a1, were examined and visualized using UMAP and violin plots (Fig. 1F and G). Dot plot analysis revealed that CDOPs co-expressed Col2a1 and Col1a1, and lacked HTC marker Col10a1, suggesting active matrix mineralization, remodeling and proliferation (Fig. 1H) (35).

Systematic functional analysis of endochondral ossification-related differentially expressed gene clusters in CDOPs revealed that the core differential genes, according to GO functional annotation, were significantly enriched in key biological processes such as 'protein binding' (molecular function), 'chondrocyte differentiation' (biological process) and 'skeletal system development' (cellular component). Notably, these genes were highly associated with 'ossification' and 'bone development' (Fig. 1I). KEGG pathway enrichment analysis further demonstrated that the differential genes were primarily involved in the 'FoxO signaling pathway', 'TGF-β signaling pathway', 'MAPK signaling pathway', 'ECM-receptor interaction' and 'PI3K-Akt signaling pathway' (Fig. 1J). These pathways collectively mediate the establishment of the osteogenic microenvironment through the coordinated regulation of cell cycle progression, proliferative and differentiation dynamics, migratory and adhesive behaviors, and matrix remodeling capacity. The significant enrichment of 'signaling pathways regulating pluripotency of stem cells' (Fig. 1J) suggested that CDOPs may possess the potential for mesenchymal stem cell differentiation toward the osteogenic lineage.

Based on trajectory inference analysis of single-cell transcriptomic data, the present study utilized the Monocle3 algorithm to perform pseudo-time modeling of six relevant cell subpopulations involved in the transdifferentiation process of CHs. UMAP nonlinear dimensionality reduction (Fig. 2A) was used to construct a continuum of cellular states, followed by integration with RNA velocity to resolve the topological structure of the cell differentiation trajectories. Using CHs as the starting point of the trajectory (pseudotime t=0), the results revealed a pronounced bifurcation in cell fate with notable bidirectional heterogeneity: The primary branch (branch 1) culminated in osteoblasts as the terminally differentiated state, while the secondary branch (branch 2) terminated in HTCs. Notably, osteoblasts were strictly localized to the terminal region of branch 1 along the pseudotime axis, with no significant temporal overlap in transcriptomic features with intermediate cell states, such as the CDOP subpopulation, suggesting that this subpopulation may undergo a rapid terminal differentiation program. By contrast, HTCs and CDOPs exhibited partially overlapping segments along the pseudo-time axis, indicating that they may share transitional regulatory networks. Further analysis examined the temporal changes in the expression of characteristic genes at various time points and across different cell subpopulations (Fig. 2B and C). Pseudotime analysis showed that 1700025G04Rik was highly expressed during the early stages of CH transdifferentiation, suggesting its potential role in driving the differentiation process. Collagen type III α1 chain (Col3a1) is primarily distributed in blood vessels, marking vascular ingrowth during differentiation, and together with Col1a1, helps maintain tissue tensile strength (50). Collagen type V α2 chain (Col5a2) serves a regulatory role in the assembly of Col1a1, thereby influencing collagen fiber diameter and organization of the extracellular matrix (51). Serpin family E member 2 (Serpine2) inhibits plasmin and thrombin, thereby regulating ECM remodeling, axon guidance and synaptic plasticity (52). Myosin IB (Myo1b), an unconventional myosin, modulates the interplay between the cell membrane and the cytoskeleton through its association with actin. Myo1b participates in vesicle trafficking and cell migration, and serves a crucial role in the regulation of diverse cellular functions, including cell proliferation and apoptosis (53). During the transdifferentiation process, Col5a2, Myo1b and Serpine2 expression progressively increased, peaking at the terminal stage, whereas Col3a1 expression remained stable during the CH phase and was gradually downregulated upon osteoblast commitment. Taken together, these findings suggested that CDOPs are the precursor subpopulations of osteoblasts, whereas conventional HTCs maintain paracrine regulatory roles during endochondral ossification.

GO functional enrichment analysis of the top 100 differentially expressed genes (Fig. 2D) revealed the following: First, the core biological themes were skeletal development and remodeling, as evidenced by terms such as 'endochondral ossification', 'skeletal system development' and 'ossification'. There were also implicit connections, such as strong enrichment of 'ECM proteoglycans' and the 'PID INTEGRIN1 PATHWAY', indicating that dynamic remodeling of the extracellular matrix was a driving force in bone formation. Second, crosstalk between signaling pathways was apparent: The enrichment of the 'PID INTEGRIN1 PATHWAY' along with the 'response to mechanical stimulus' suggested that integrins may regulate CH differentiation by sensing mechanical signals (such as the piezoelectric effect in bones). The 'PID SYNDECAN 4 PATHWAY' and its association with ECM proteoglycans may mediate cell matrix adhesion and coordinate with growth factors (such as cell adhesion molecule and BMP). Third, enrichment of 'ECM proteoglycans' and 'keratan sulfate biosynthesis' underscored the critical role of glycosaminoglycan modifications in establishing and maintaining a CH-to-osteoblast transition microenvironment. The term 'negative regulation of molecular function' likely reflected the precise control of matrix degradation by protease inhibitors, such as Serpine2. Fourth, microenvironmental homeostasis mechanisms were highlighted by terms such as 'hemostasis' and 'blood vessel development', indicating the coupled roles of angiogenesis, coagulation factors and the oxygen microenvironment in bone repair processes. Enrichment of the 'apoptotic signaling pathway' and 'positive regulation of execution phase of apoptosis' likely reflected the contribution of programmed death of HTCs to the progression of ossification.

Safranin O/Fast Green staining of callus tissues harvested from the fracture sites at 10 and 21 days post-fracture was performed (Fig. 2E). On day 10, the robust formation of cartilaginous calluses was observed in all groups. The extracellular matrix of HTCs, rich in proteoglycans, was intensely stained red by Safranin O, indicating that the cartilage tissue constituted the majority of the callus at this stage, with only a small proportion of gray-blue-stained osteogenic areas present. A marked increase in the number and proportion of tdTomato⁺/OCN⁺ double-positive cells indicative of CH-derived osteoblasts was evident from the immunofluorescence images comparing fracture calli at day 10 and day 21 post-fracture (Fig. 2E).

Given that KDM6A acts as an oxygen sensor to regulate cellular differentiation (22), the present study investigated how CH-to-osteoblast differentiation during endochondral ossification is altered under different oxygen levels. CHs derived from fx/wt and fx/fx mice were subjected to transdifferentiation induction. After 7 days of induction, Alcian blue staining (Fig. 3A and B) revealed that the fx/wt-normoxic group exhibited a significantly lower amount of acidic mucopolysaccharides than the fx/wt-hypoxic group (P<0.05), while the fx/fx-normoxic group showed a significant increase in acidic mucopolysaccharides compared with the fx/wt-normoxic group (P<0.01). ALP staining (Fig. 3C and D) demonstrated that the number of ALP-positive cells was markedly higher in the fx/wt-normoxic group than in the fx/wt-hypoxic group (P<0.001), whereas the number of ALP-positive cells in the fx/fx-normoxic group was significantly lower than that in the fx/wt-normoxic group (P<0.001). Alizarin red staining (Fig. 3E and F) indicated an increased number of mineralized nodules in the fx/wt-normoxic group compared with the fx/wt-hypoxic group (P<0.05), whereas the fx/fx-normoxic group had fewer mineralized nodules than the fx/wt-normoxic group (P<0.05).

qPCR analysis of CH marker gene expression 3 days after induction (Fig. 3G) showed that the mRNA levels of Col10a1, Acan and Col2a1 were reduced in the fx/wt-normoxic group compared with the fx/wt-hypoxic group (P<0.05, P<0.01 and P<0.0001, respectively). By contrast, fx/fx-normoxic mice exhibited increased expression of all three genes compared with fx/wt-normoxic mice (P<0.05), with a particularly significant upregulation of Col2a1 and Acan (P<0.0001). Furthermore, analysis of the relative mRNA levels of osteogenic differentiation-related genes (Fig. 3H) revealed that Col1a1, Spp1 and Sparc were significantly upregulated in the fx/wt-normoxic group compared with the fx/wt-hypoxic group (P<0.001, P<0.0001 and P<0.05, respectively). However, the expression levels of all three genes were markedly reduced in the fx/fx-normoxic group compared with the fx/wt-normoxic group (P<0.0001, P<0.0001 and P<0.05, respectively).

scRNA-seq was also performed on chondral callus cell samples from fx/wt (WT) and fx/fx mice (FX), 10 days post-fracture. The analysis focused on CHs, pre-HTCs, HTCs, CDOPs and osteoblasts (Fig. 4A). Violin plots of gene expression (Fig. 4B) revealed that the mean expression levels of Col2a1 in the WT group were lower than those in the FX group, whereas Col10a1, RUNX2 and Col1a1 were more highly expressed in the WT group than in the FX group. The expression levels of Acan and Bglap were similar in the two groups. The present study specifically focused on the CDOP subpopulation. The volcano plot of differentially expressed genes showed that osteogenic genes, including Col1a1, Sp7, RUNX2 and Spp1, were significantly upregulated in CDOPs of the WT group compared with those of the FX group (Fig. 4C). GO analysis indicated that the upregulated genes were enriched in biological processes, such as 'cell adhesion', 'chondrocyte differentiation', 'skeletal system development', 'endochondral ossification' and 'osteoblast differentiation'. Pathway analysis revealed that the 'Wnt signaling pathway' was the most prominent osteogenic pathway enriched in our dataset (Fig. 4D and E). GSEA (Fig. 4F) indicated that the 'Wnt signaling pathway' was significantly enriched in the WT group (enrichment score, 0.276; P=0.0063; false discovery rate, 0), suggesting that global Wnt pathway activity was higher in WT cells.

CellChat analysis identified intercellular communication among the 16 clusters detected in callus samples on day 10 post fracture, revealing extensive molecular interactions (Fig. 4G and H). Sankey plots showed how 13 sender cell populations (acting as signaling sources) coordinated with one another and with specific signaling pathways to drive communication, and how 12 receiver cell populations (acting as targets) coordinated to respond to incoming signals (Fig. 4I and J). Notably, CHs and CDOPs exhibited the highest frequency of intercellular interactions via the Wnt signaling pathway (Fig. 4K). Furthermore, the hierarchical plot (Fig. 4L) delineated the sending and receiving roles of different CH subsets within the Wnt signaling network, highlighting CDOPs as key signaling hubs.

Western blot analysis was performed to assess the histone methylation levels in CHs from fx/wt-normoxic and fx/fx-normoxic cells during transdifferentiation. H3K27me2 and H3K27me3 levels were normalized to those of total histone H3 (loading control) to quantify relative changes in these histone modifications independently of total protein variation. As shown in Fig. 5A, the expression levels of H3K27me2 and H3K27me3 proteins were significantly higher in the fx/fx-normoxic group than in the fx/wt-normoxic group (P<0.001). ChIP-PCR (Fig. 5B) showed that enrichment of H3K27me2 and H3K27me3 at the Wnt3a promoter region was markedly increased in the fx/fx normoxic group compared with the fx/wt-normoxic group. H3K27me3 enrichment at the RUNX2 promoter was also higher in the fx/fx-normoxic group than in the fx/wt-normoxic group. Additionally, the protein expression levels of Wnt3a, β-catenin and RUNX2 (Fig. 5C and D) were reduced in the fx/fx-hypoxic group compared with the fx/wt-hypoxic group (P<0.05). In addition, the fx/wt-normoxic group exhibited significantly higher expression levels of β-catenin and RUNX2 compared with the fx/wt-hypoxic group (P<0.001), while Wnt3a expression was also increased (P<0.01). By contrast, the expression levels of all three proteins were significantly decreased in fx/fx-normoxic mice compared with fx/wt-normoxic mice (P<0.001). Differences in the mRNA expression levels of Wnt3a, β-catenin and RUNX2 (Fig. 5E) among the groups were in accordance with their respective protein expression patterns.

The present study subsequently investigated the impact of KDM6A deletion on fracture healing in mice. H&E staining (Fig. 6A) showed that at 10 days post-fracture, both fx/wt and fx/fx mice exhibited calluses comprising a small amount of fibrous tissue and a large amount of cartilage tissue, with woven bone already formed in both groups. Notably, the fx/wt group displayed a greater amount of woven bone than the fx/fx group. By day 28 post-fracture, the bone entered the remodeling phase; the woven bone in the fx/fx group appeared loose and irregular, whereas the fracture ends in the fx/wt group were smoother. In the fx/wt group, the woven bone transformed into lamellar bone, with a relatively continuous lamellar bone observed, suggesting a better degree of healing than in the fx/fx group.

At 28 days post-fracture, in vivo micro-CT scanning and three-dimensional reconstruction (Fig. 6B) were performed to assess the morphology of the fracture ends. In the fx/wt group, a cortical bone bridge was formed, with no obvious peripheral callus observed, and partial recanalization of the medullary cavity was evident. By contrast, in the fx/fx group, a bone bridge was present, but the peripheral callus was not completely resorbed. Ex vivo micro-CT standard scanning and three-dimensional reconstruction (Fig. 6B) were performed on the harvested mouse tibiae to assess bone microarchitecture at the fracture ends. Three-dimensional views from the lateral, medial and dorsal aspects of the tibia, as well as two-dimensional sagittal and coronal sections, revealed that both groups exhibited bony union at the fracture site, with varying degrees of endosteal and periosteal callus resorption. Calli in the remodeling phase were observed in the outer cortex. On two-dimensional sagittal and transverse sections, irregular block-like intramedullary calluses were visible at the fracture ends, which were more pronounced in the fx/fx group. Quantitative analyses were conducted for BV/TV, BS/TV, Tb.N and Tb.Th within 1 mm of and below the fracture line (Fig. 6C). In the fx/fx group, both BV/TV and BS/TV were significantly higher than those in the fx/wt group (P<0.01), whereas Tb.N was also higher in the fx/fx group (P<0.05); no significant difference in Tb.Th was observed (P>0.05).

Immunofluorescence staining for OCN at the fracture callus at 10 and 21 days (Fig. 6D) was performed to assess the proportion of CH-to-osteoblast transdifferentiation. KDM6A deletion resulted in a significant reduction in transdifferentiation efficiency on day 10 (Fig. 6E; P<0.01) and a further decrease by day 21 (Fig. 6G; P<0.001). Safranin O/Fast Green staining (Fig. 6F) on day 10 post-fracture revealed active cartilage callus formation in both groups, with HTC-rich matrix stained bright red (proteoglycan content) and cartilage predominating in the callus, while a few gray-blue-stained osteogenic regions were visible. The fx/wt group had a greater osteogenic area than the fx/fx group, and the latter had fewer gray-blue regions than the former. At 21 days post-fracture, the red-stained cartilage area in the fx/wt group was decreased, indicating active ossification, and the gray-blue osteogenic area was increased; the red cartilage alternated with the loose gray-blue woven bone. In the fx/fx group, abundant red cartilage regions persisted, and the gray-blue woven bone was sparser, indicating a lower degree of ossification compared with the fx/wt group.