Single‐Cell Analysis of Cartilage‐Derived Cells Identifies Diverse Populations and a Cholesterol/ER‐Stress Chondrocyte Subset in Knee OA

ABSTRACT

This study aimed to elucidate the cellular and molecular mechanisms underlying knee osteoarthritis pathogenesis by analyzing chondrocyte populations and their spatial distribution using single‐cell RNA sequencing (scRNA‐seq). Knee cartilage cells underwent scRNA‐seq. Cellular clusters were identified via gene expression profiles and annotated using gene ontology analyses. Correlations between cluster proportions and patient background characteristics were evaluated. Differentiation trajectories were evaluated using pseudotime analysis. Differentially expressed genes between osteoarthritis and non‐osteoarthritis samples were identified, and their functional relevance was experimentally assessed. Cartilage stratification analysis was also performed. A total of 104,951 cells from 12 patients (9 osteoarthritis, 3 non‐osteoarthritis) were analyzed. Initially, 11 clusters were identified; three non‐chondrocytic clusters (mesenchymal stromal cell‐like cells, blood cells, fibroblasts) were identified and excluded from chondrocyte analysis. Eight chondrocyte subpopulations were annotated, including a novel stress‐responsive metabolic chondrocytes (MetabCs) associated with serum total cholesterol levels and linked to cholesterol metabolism and endoplasmic reticulum stress pathways. Correlation and trajectory analyses revealed two paths: one toward fibrocartilage chondrocytes (FCs) associated with aging and another toward MetabCs associated with osteoarthritis progression. Novel osteoarthritis‐related genes identified included chondroprotective HILPDA and destructive AKR1C2. Stratification analysis indicated that worn cartilage chondrocytes exhibited a hybrid phenotype, shifting toward upper‐zone features while retaining deep‐zone characteristics. In conclusion, a novel chondrocyte cluster was identified as implicated in osteoarthritis pathogenesis, where elevated serum cholesterol levels potentially increase endoplasmic reticulum stress response, promoting MetabC formation. Chondrocytes altered their phenotype in response to spatial distribution changes caused by cartilage thinning.

1. Introduction

Osteoarthritis (OA) of the knee is one of the most prevalent musculoskeletal disorders, affecting approximately 37% of individuals over the age of 60 [1]. It poses a significant public health burden, yet its pathophysiology remains incompletely understood. Recent research has revealed it to be a multifaceted condition influenced by genetic, mechanical, metabolic, and inflammatory factors [2]. A landmark genome‐wide association study (GWAS) identified 100 independent genetic variants significantly associated with OA, highlighting its genetic complexity [3]. Furthermore, recent technological advancements have driven extensive efforts to identify functional genes and genetic loci by integrating genomics with epigenetics, transcriptomics, and proteomics [4]. Among transcriptomic tools, single‐cell RNA sequencing (scRNA‐seq) has emerged as a powerful approach for investigating gene expression at the cellular level, particularly within local tissues.

scRNA‐seq has been conducted on various joint tissues affected by OA, including cartilage, synovium, and meniscus [5]. Traditionally, chondrocytes have been classified based on their developmental and differentiation states [6]. In contrast, scRNA‐seq, through expression analysis at single‐cell level, has identified novel clusters. Ji et al. pioneered scRNA‐seq analysis of human cartilage, categorizing chondrocytes into seven clusters, including three novel populations, thereby challenging conventional histological classifications [7]. Subsequent studies have further identified additional chondrocyte subtypes, with more than 11 clusters currently recognized [8, 9, 10]. Fan et al. conducted scRNA‐seq on over 100,000 cells, identifying two novel clusters: inflammatory (InfC) and pre‐inflammatory (preInfC) chondrocytes [10]. Although new chondrocyte clusters continue to be identified, inconsistencies in clustering criteria and marker genes across studies complicate classification and hinder a comprehensive understanding of OA. Furthermore, the relationship between each cluster and key pathogenic factors of OA, such as sex differences, obesity, and metabolic syndrome, remains unexplored.

Despite the pathophysiological relevance of classical chondrocyte classifications, their relationship to molecularly defined clusters remains unclear. Novel spatial transcriptomics may help bridge this gap; however, technical challenges persist due to the low volume fraction of chondrocytes, which occupy only approximately 2%–5% of the total volume of articular cartilage [11]. To investigate chondrocyte clustering in cartilage, clarify the spatial distribution of clusters in scRNA‐seq relative to traditional histological zones, and identify novel differentially expressed genes (DEGs) and clusters associated with knee OA, we conducted scRNA‐seq on knee cartilage samples. The overall cartilage cell population was first examined with reference to previously reported clusters. Subsequent analyses focused on associations between cluster composition and OA‐related background characteristics, including aging and lifestyle‐related factors. Gene expression profiles from OA and non‐OA cartilage were compared to evaluate the cellular and molecular mechanisms underlying OA progression. In addition, the spatial distribution of chondrocyte clusters was assessed, based on the hypothesis that phenotypic variation may arise in response to structural changes associated with cartilage thinning.

2. Methods

This study was conducted in accordance with the Declaration of Helsinki. It was conducted at Kyoto University Hospital in Kyoto, Japan, and approved by the institutional review board (G1287). Written informed consent was obtained from all patients. A total of 12 cartilage samples were analyzed, including nine from OA patients undergoing total knee arthroplasty and three from non‐OA control (NC) patients undergoing above‐knee amputation for malignant bone tumors (2021–2024). The cohort comprised seven men and five women aged 16–91 years (Figure 1A,B; Table 1). OA samples were obtained from knees showing minimal degeneration in either femoral compartment on preoperative radiographs. NC femoral samples were collected from patients without radiographic OA or tumor invasion in the knee joint. For stratified scRNA‐seq, cells from the last five OA samples were collected by dividing full‐thickness cartilage (International Cartilage Repair Society: ICRS grade 0–1) into top‐ and bottom‐halves and compared with cells from thinned, degenerated cartilage (worn cartilage: ICRS grades 2–3).

Figure 1.

Patients and workflow of this study. (A) Overview of patient cohort comprising nine individuals with osteoarthritis (OA) and three non‐OA controls (NC); age range 16–91 years. (B) Experimental workflow. NC and OA cartilage were collected from knee joints, digested, and analyzed using scRNA‐seq.

Table 1.

Details of clinical cases.

Case number	Age	Sex	Body mass index (kg/m2)	Primary disease	Serum TC level (mg/dL)	Oral statins usage
OA1	71	Male	30.4	Osteoarthritis	NA	No
OA2	81	Female	26.4	Osteoarthritis	156	Yes
OA3	75	Male	24.3	Osteoarthritis	184	No
OA4	83	Female	35.5	Osteoarthritis	NA	Yes
OA5	64	Female	28.2	Osteoarthritis	252	Yes
OA6	74	Male	26.8	Osteoarthritis	248	No
OA7	57	Female	31.2	Osteoarthritis	NA	No
OA8	70	Male	26.4	Osteoarthritis	188	Yes
OA9	74	Female	26.5	Osteoarthritis	187	No
NC1	73	Male	18.3	Dedifferentiated liposarcoma	134	No
NC2	91	Male	19.0	Metastatic bone tumor	148	No
NC3	16	Male	23.2	Osteosarcoma	140	No

Abbreviations: NA, not available; NC, non‐osteoarthritic control; OA, osteoarthritis; TC, total cholesterol.

Cartilage was harvested, minced, and digested in 1% type II collagenase for 4 h [12]. The tissue was dissociated into single cells and used for scRNA‐seq library preparation with the Chromium Next GEM Single Cell 5' Reagent Kits v2 Dual Index (10× Genomics, Pleasanton, California, USA). Libraries were sequenced on the Illumina NovaSeq. 6000 platform. Single‐cell RNA sequencing data were processed using Cell Ranger (v7.2.0) and analyzed in R (v5.1.2) with the Seurat package (v5.1.0). Quality control and doublet removal were performed before downstream analysis. Datasets were integrated using Seurat CCA based on the top 30 principal components (Supporting Information S5: Figure 1A–C). Dimensional reduction and clustering were performed using UMAP and the FindClusters function with a resolution of 0.4. Cluster proportions were calculated as the number of cells in each cluster divided by the total number of chondrocytes.

For external validation, label transfer was performed to publicly available human articular cartilage scRNA‐seq datasets using Seurat's FindTransferAnchors and TransferData functions [7, 8, 10, 13], with dataset of this study as the reference. Cluster proportions were calculated for each dataset, and similarities in cluster composition were evaluated using Spearman's rank correlation analysis. To evaluate functional transcriptional programs in chondrocytes, anabolic, homeostatic, catabolic, and fibrotic gene sets were defined based on review articles [14, 15, 16, 17, 18, 19, 20, 21] (Supporting Information S1: Table 1). Module scores for each gene set were calculated using AddModuleScore function.

Cluster marker genes were identified using FindAllMarkers function. GO enrichment analysis was performed using clusterProfiler package. Gene Set Enrichment Analysis (GSEA) was performed using the fgsea package. Pseudotime trajectories were inferred using Slingshot.

In vitro experiments were performed using TC28a2 human chondrocyte cell lines [22] and primary chondrocytes. Cells were stimulated with cholesterol or tunicamycin, and gene and protein expression were evaluated by RT‐PCR and western blotting. CRISPR/Cas9‐mediated gene knockout was performed for pellet culture.

Detailed methodologies for all experiments are provided in the Supporting Information S6.

3. Results

3.1. Extracting Chondrocytic Clusters From Knee Cartilage Samples

A total of 104,951 cells were integrated and analyzed, leading to the identification of 11 distinct clusters (Figure 2A). To validate the clustering, comparative analyses were performed using publicly available scRNA‐seq datasets of human articular cartilage, using the dataset of this study as the reference. Significant correlations were identified between the present dataset and each of the four public datasets by Spearman's rank correlation analysis (ρ = 0.63–0.87) (Supporting Information S2: Table 2), indicating the reproducibility of the clustering results. First, clusters that were distant from the main cluster were focused. Clusters 9 and 10 demonstrated lower correlation with other clusters (Figure 2B). Moreover, these clusters exhibited low expression of COL2A1, ACAN, and SOX9 (Figure 2C; Supporting Information S3: Table 3). Cluster 9 showed strong expression of HLA‐DRA and PTPRC, indicating the presence of blood cells (Figure 2D). This cluster showed high CD68 (63%) and C1QA (57%) expression, indicating macrophage predominance (Supporting Information S5: Figure 2A,B), consistent with Fan et al.'s InfC cluster (CD68: 62%, C1QA: 54%) [10] (Supporting Information S5: Figure 2C,D). Cluster 10 expressed mesenchymal stromal cell (MSC) markers (THY1, NT5E, and ENG) but expressed low levels of negative markers (PTPRC and HLA‐DRA), suggesting the presence of MSC‐like cells (Figure 2D). Reanalysis of other human cartilage scRNA‐seq datasets consistently identified these two clusters as minor populations (Supporting Information S2: Table 2). In cluster 8, although SOX9 and COL2A1 were expressed, CD55, VCAM1, and PDPN were highly expressed, with low THY1 expression, suggesting thymus cell antigen‐1 negative fibroblast‐like synoviocytes (THY1⁻FLSs) (Figure 2D). Importantly, this cluster was identified exclusively in the OA1 patient and was absent from other published datasets.

Figure 2.

Annotation and validation of non‐chondrocytic subpopulations in knee cartilage. (A) UMAP plot showing 11 distinct cell clusters identified from 104,951 cells. (B) Correlation heatmap of 11 clusters based on gene expression. (C) Dot plot of average and percentage expression of three chondrocyte‐specific genes and the top five DEGs across the clusters 8: FLS, 9: blood cell, and 10: MSC‐like cell. (E) Violin plots and feature plots showing specific marker gene expression in clusters 8, 9, and 10. (E) Immunofluorescence staining for SOX9 (red) and CD45 (green) with nuclear DAPI staining (blue). Dashed lines indicate the cartilage surface boundary. Scale bar = 100 μm. (F) Microscopic images of MSC‐like cells cultured in lineage‐specific differentiation media and stained with Alizarin Red S, Oil Red O, and Alcian Blue. Scale bar = 200 μm. DEG; differentially expressed gene; FLS, fibroblast‐like synoviocyte; MSC, mesenchymal stromal cell; scRNA‐seq, single‐cell RNA sequence; UMAP, uniform manifold approximation and projection.

Although these three non‐chondrocytic clusters may reside in the cartilage, contamination during sample collection could not be ruled out. Immunostaining of frozen OA cartilage sections with CD45 antibody revealed the presence of blood cells around the cartilage surface (Figure 2E). Another possible route of contamination was blood cells observed within cracks (Supporting Information S5: Figure 3A). In the OA cartilage, tidemark disruption allowed blood vessels to invade the cartilage layers (Supporting Information S5: Figure 3B). Low‐density culture of cells obtained from cartilage digestion yielded populations with fibroblast‐like morphology, which were differentiated into adipogenic, osteogenic, and chondrogenic lineages, indicating the presence of an MSC‐like population within the cartilage‐derived cells (Figure 2F). This finding was further supported by fluorescence‐activated cell sorting analysis, which identified MSC‐like cells following negative and positive selections (Supporting Information S5: Figure 4).

3.2. Annotation and Differentiation Trajectories of Chondrocyte Subpopulations

The eight chondrocyte clusters were annotated (Figure 2A), and marker genes for each cluster were identified (Figure 3A; Supporting Information S3: Table 3). Based on previous reports, fibrocartilage chondrocytes (FCs), HTCs, pre‐fibrocartilage chondrocytes (preFCs), proliferative chondrocytes (ProCs), effector chondrocytes (ECs), homeostatic chondrocytes (HomCs), and regulatory chondrocytes (RegCs) were classified. The annotation of these seven clusters was supported by Gene Ontology (GO) enrichment analysis (Supporting Information S5: Figure 5). One cluster that did not match previously reported marker genes was designated as stress‐responsive metabolic chondrocytes (MetabCs), whose top DEGs were strongly related to metabolic processes. Re‐clustering after excluding the three non‐chondrocytic clusters yielded 13 clusters, with the original marker gene expression patterns largely retained, and the MetabC population consistently identified (Supporting Information S5: Figure 6). Further details of the MetabC cluster are provided in a later section. Protein expression of the representative DEGs in all clusters was confirmed by immunohistochemistry in OA cartilage (Figure 3B; Supporting Information S5: Figure 7). When cluster distributions were compared across patients, the young NC03 sample exhibited distinct features, with almost no FCs and a markedly high proportion of ProCs (Figure 3C). Pseudotime trajectory analysis identified bifurcating trajectories originating from HomCs, with RegCs at the branching point, diverging into two distinct directions: toward FCs and MetabCs (Figure 3D). Separate pseudotime analyses of NC and OA chondrocytes showed similar trajectories (Supporting Information S5: Figure 8A,B). DoRothEA analysis of the 30 most variable inferred transcriptional factor (TF) activities revealed distinct activity patterns across chondrocyte differentiation branches (Supporting Information S5: Figure 8C). Along the ProC–preFC–FC branch, the activity patterns of these TFs changed sequentially from ProCs to FCs, whereas the HTC–MetabC branch showed a different pattern. Cluster enrichment analysis using literature‐derived gene sets revealed distinct functional signatures along the trajectory. Increased anabolic and homeostatic gene sets were observed toward the right terminal of the trajectory (EC and MetabC), whereas enrichment was low at the left terminal (preFC and FC). Catabolic gene sets were enriched at the right terminal as well as RegC. Fibrotic gene sets were conspicuously enhanced in FC (Figure 3E).

Figure 3.

Annotation and differentiation trajectories of chondrocyte subpopulations. (A) Dot plot illustrating the top five DEGs for each cluster. Dot size indicates the percentage of expressing cells, and color intensity represents average expression levels. (B) Immunohistochemical staining of representative genes of chondrocyte clusters, and quantification of positive cells in superficial (SZ), middle (MZ), and deep (DZ) zones. Scale bar = 200 μm in the left low‐magnification image and 100 μm in the right high‐magnification image. (C) Bar plots showing the relative proportions of chondrocyte clusters across each sample. (D) Pseudotime trajectory analysis using the Slingshot package, showing bifurcating differentiation pathways from HomCs toward FCs and MetabCs. (E) Enrichment of anabolic, homeostatic, catabolic, and fibrotic gene sets in each chondrocyte cluster. (F) Correlation analysis between cluster proportions and age. DEG, differentially expressed genes; ECs, effector chondrocytes; FCs, fibrocartilage chondrocytes; HomCs, homeostatic chondrocytes; HTCs, hypertrophic chondrocytes; MetabCs, metabolic chondrocytes; NC, non‐OA control; OA, osteoarthritis; preFCs, prefibrocartilage chondrocytes; ProCs, proliferative chondrocytes; RegCs, regulatory chondrocytes; scRNA‐seq, single‐cell RNA sequence; UMAP, uniform manifold approximation and projection.

To investigate the difference of cluster distributions based on patient background, correlations with age, sex, and body mass index (BMI) were examined. The proportions of preFCs and FCs positively correlated with age (r = 0.61, and 0.67, respectively), whereas ProCs negatively correlated with age (r = −0.90) (Figure 3F). No clusters associated significantly with sex or BMI (Supporting Information S5: Figure 9A,B).

3.3. Novel Chondrocyte Cluster Associated With Cholesterol Metabolism and Endoplasmic Reticulum Stress Response

A novel chondrocyte cluster, termed MetabC, lacking enrichment for previously reported marker genes, was identified. Its top markers, MSMO1 and SQLE, are key enzymes in cholesterol biosynthesis (Figure 3A). GO analysis revealed enrichment of alcohol‐ and cholesterol‐related pathways (Supporting Information S5: Figure 5), and GSEA further demonstrated activation of cholesterol metabolism and endoplasmic reticulum (ER) stress–related pathways, including mTORC1 signaling and the unfolded protein response (UPR) (Figure 4A). Induction of ER stress by tunicamycin enhanced cholesterol synthesis–related genes (Figure 4B). Notably, the proportion of MetabCs positively correlated with serum total cholesterol (TC) levels, whereas no such associations were observed for other clusters (Figure 4C; Supporting Information S5: Figure 10). To model cholesterol loading, TC28a2 chondrocytes were stimulated with exogenous cholesterol, which was rapidly taken up by the cells (Supporting Information S5: Figure 11). Western blotting showed dose‐dependent induction of ER stress proteins by TC stimulation (BiP, PERK, CHOP) (Figure 4D), consistent with elevated XBP1 and ATF4 expression in MetabCs (Supporting Information S5: Figure 12). TC stimulation downregulated SOX9 and COL2A1, but upregulated IL1B and MMP13 (Figure 4E). Collectively, it is likely that the PERK‐ATF4‐CHOP axis of the ER stress pathway is activated in OA MetabCs.

Figure 4.

Identification of a Novel Chondrocyte Cluster Associated with Cholesterol Metabolism. (A) GSEA of MetabCs revealed enrichment of cholesterol metabolism pathways and ER stress‐related pathways, including mTORC1 signaling and unfolded protein response. (B) RT‐qPCR analysis of genes related to cholesterol synthesis with TC28a2 cells stimulated with tunicamycin for 3 h. (C) The proportion of MetabCs positively correlated with serum total cholesterol levels. (D) Western blot analysis revealed increased expression of ER stress‐related proteins (BiP, PERK, CHOP) in response to cholesterol stimulation. (E) RT‐qPCR following cholesterol stimulation for 24 h. BiP, binding immunoglobulin protein; CHOP, C/EBP homologous protein; GSEA, gene set enrichment analysis; IRE1, inositol‐requiring enzyme 1; MetabCs, metabolic chondrocytes; NC, non‐OA control; OA, osteoarthritis; PERK, protein kinase RNA‐like endoplasmic reticulum kinase; RT‐qPCR, quantitative reverse transcription polymerase chain reaction.

3.4. DEG Analysis and Novel Genes Associated With OA Pathology

In DEG analysis between NC and OA, due to the nature of scRNA‐seq, where each cell is treated as an individual sample, the p‐values of DEGs were excessively inflated, making it difficult to identify key genes (Figure 5A). To address this issue, pseudobulk analysis was applied [23] (Figure 5B), in which no DEGs were exclusively identified (Figure 5C). From the Volcano plot, among significantly up‐ and down‐regulated genes in OA, two genes, HILPDA and AKR1C2, which have not been extensively linked to OA in previous studies, were identified. Differential gene expression analysis of each chondrocyte cluster demonstrated that OA induced both shared and cluster‐specific transcriptional alterations across chondrocyte subpopulations. (Supporting Information S5: Figure 13). Several genes, including HILPDA and GSTM1, were consistently differentially expressed across multiple chondrocyte subpopulations, suggesting shared OA‐associated transcriptional alterations. In contrast, inflammatory genes showed more restricted expression patterns, including NOS2 upregulation in RegC and HTC and CCL20 upregulation in HTC and MetabC. In addition, some differentially expressed genes were shared among consecutive cell populations within the inferred ProC–preFC–FC progression. Next, to evaluate the effect of gene silencing in chondrocytes on genes maintaining cartilage homeostasis and those related to OA, CRISPR‐Cas9 was used to knock out these genes in TC28a2 human chondrocyte cell lines. Knockouts of HILPDA and AKR1C2 were confirmed by Western blotting. (Figure 5D–G). HILPDA, which was highly expressed in NC chondrocytes, exhibited an unfavorable trend upon knockout, with reduced expression of anabolic markers critical for cartilage homeostasis, such as SOX9 and COL2A1, and increased levels of the catabolic marker ADAMTS5 (Figure 5H). Conversely, AKR1C2, which was highly expressed in OA chondrocytes, showed favorable effects for OA protection upon knockout, leading to increased COL2A1 expression level (Figure 5I). In pellet culture of human primary chondrocytes (Figure 5J), HILPDA‐knockout chondrocytes failed to form spherical pellets, and the pellets were smaller than those of the controls (Figure 5K). In contrast, AKR1C2‐knockout pellets formed spherical structures with strong type II collagen and Safranin‐O staining, as well as a smooth, well‐defined boundary.

Figure 5.

Comparative analysis of DEG functions in NC and OA cartilage. (A) Volcano plot of DEGs using single‐cell analysis with the top five genes identified in NC and OA cartilage. (B) Volcano plot of DEGs using pseudobulk analysis, highlighting overlap with single‐cell findings, including HILPDA and AKR1C2. (C) Venn diagram of DEGs identified by single‐cell and pseudobulk approaches, demonstrating no unique DEGs in pseudobulk analysis. (D–G) Western blot validation of HILPDA and AKR1C2 knockouts (KO) in TC28a2 cells using CRISPR‐Cas9. Quantification shows a significant reduction in HILPDA and AKR1C2 protein levels in KO cell lines (n = 3). (H, I) qRT‐PCR analysis of anabolic (SOX9, COL2A1, ACAN) and catabolic (MMP13, ADAMTS5) gene expression in HILPDA‐KO and AKR1C2‐KO cells. (J) Upper panels: immunohistochemical staining for type II collagen of primary chondrocyte pellet culture, comparing control, AKR1C2‐KO, and HILPDA‐KO groups. Lower panels: Safranin‐O/fast green staining of primary chondrocyte pellet culture. (K) Comparison of the maximum cross‐sectional area of primary chondrocyte pellet culture (n = 3). DEG, differentially expressed genes; IHC, immunohistochemistry; NC, non‐OA control; qRT‐PCR, quantitative reverse transcription polymerase chain reaction.

3.5. Stratification Analysis by Comparing Top‐ and Bottom‐Half Cartilage

From five OA samples (OA5 to OA9), regions graded as ICRS 0‐1 in full‐thickness cartilage were divided into top‐ and bottom‐halves, while regions graded as ICRS 2‐3 in thinned cartilage were collected as degenerated worn cartilage (Figure 6A,B). Then, scRNA‐seq analysis was performed on each sample. When top‐half and bottom‐half were compared, preFCs, RegCs, HomCs, and FCs were dominant in the top‐half, while HTCs and ECs were more abundant in the bottom‐half (Supporting Information S5: Figure 14). These results were partially consistent with the immunohistochemical findings (Figure 3B), which demonstrated an increased presence of preFC marker‐positive cells in the superficial zone and HTC‐ and EC‐positive cells in the deep zone.

Figure 6.

Comparison of cluster proportions and gene expression in different layers. (A) Schematic representation of the experimental workflow. Cartilage samples were collected from ICRS 0‐1 graded OA regions, divided into top‐half (Top) and bottom‐half (Bottom), and from ICRS 2–3 graded degenerated worn cartilage (Worn). scRNA‐seq was performed on each sample. (B) Hematoxylin and Eosin staining of ICRS grade 0 (left) and ICRS grade 2 (right) cartilage to show top‐half, bottom‐half, and worn cartilage. (C) Cluster proportions in the top‐half, bottom‐half, and worn cartilage in OA cartilage. (D) Correlation analysis of gene expression among top‐half, bottom‐half, worn cartilage, and non‐OA control (NC). (E) Comparison of cluster proportions between the bottom‐half and worn cartilage. (F) Violin plots show the expression levels of previously reported superficial zone‐related genes (CHI3L1, CLU, COMP, HSPA8, BGN) in three cartilage layers. (G) Gene set enrichment analysis (GSEA) of HALLMARK pathways in top‐half, bottom‐half, and worn cartilage. ECs, effector chondrocytes; FCs, fibrocartilage chondrocytes; HomCs, homeostatic chondrocytes; HTCs, hypertrophic chondrocytes; MetabCs, metabolic chondrocytes; preFCs, prefibrocartilage chondrocytes; ProCs, proliferative chondrocytes; RegCs, regulatory chondrocytes; scRNA‐seq, single‐cell RNA sequencing.

If worn cartilage resulted merely from loss of the top‐half, it should resemble the bottom‐half; however, it was more similar to the top‐half (Figure 6C). Correlation analysis of gene expression revealed that worn cartilage was more closely related to the top‐half than to the bottom‐half (Figure 6D). A comparison of cluster proportions between bottom‐half and worn cartilage showed differences in five out of eight clusters (Figure 6E). Among the five superficial zone‐specific genes defined by Fan et al. [10], gene expression levels in worn cartilage, despite the loss of the top‐half region including the superficial layer, were comparable to those in top‐half chondrocytes, except for CHI3L1 (Figure 6F). GSEA revealed that inflammation‐ and degeneration‐related pathways such as epithelial–mesenchymal transition, TNF‐α–NFκB signaling, and apoptosis were predominantly upregulated in the top‐half, whereas metabolic and stress‐response pathways, including oxidative phosphorylation and unfolded protein response, were enriched in the bottom‐half (Figure 6G). Worn cartilage exhibited a mosaic pattern of pathway activation derived from both halves, indicating that it is not merely residual deep layer tissue but has acquired molecular traits of the top‐half.

4. Discussion

scRNA‐seq analysis of knee joint cartilage samples revealed several key findings. Comparison of trajectory and proportional distribution suggested two distinct pathways of chondrocyte differentiation: one age‐related and the other disease‐related. A novel chondrocyte phenotype, termed MetabC, related to cholesterol synthesis and ER stress was identified. DEG analysis identified two novel genes, HILPDA and AKR1C2, potentially associated with OA pathology. Finally, stratification analysis of cluster proportions revealed that chondrocytes in degenerated and worn cartilage were not merely residual cells from the deep zone but had acquired superficial characteristics.

Not all cells in the cartilage may be chondrocytes. Two novel clusters, InfC and PreInfC, may represent misidentified hematopoietic and endothelial cells, as suggested by their dominant expression of PTPRC and CD34 [10]. As shown in the current study, three potential sources of contamination were considered: attachment to the fibrillated cartilage surface, entrapment in cracks, and blood invasion beyond the tidemark [24]. Multiple washes alone may be insufficient to completely eliminate cells attached to the cartilage surface. Although contamination from synovial fluid is plausible and likely, the possibility that these cells reside within OA cartilage and potentially play specific functional roles in the microenvironment cannot be excluded. As for MSC‐like cells, they expressed representative MSC marker genes but showed low expression of chondrocyte markers. These MSC‐like cells were possibly contaminants from synovial fluid. However, a recent animal study demonstrated that MSCs, which are capable of tri‐lineage differentiation, reside in articular cartilage and differentiate into articular chondrocytes in response to mechanical stimuli [25]. From this perspective, a small number of MSC‐like cells may persist for decades after the development of articular cartilage, raising another intriguing question for future investigation. Nevertheless, due to the lack of direct scientific evidence in human samples, the origins of these MSC‐like cells were not determined in this study. Cartilage progenitor cells (CPCs), which were not observed as an independent cluster, may be included in the MSC‐like cell cluster, as CPCs and osteo‐chondroprogenitor lineages in mesenchymal cells are expected to express SOX9 [26]. Regarding FLSs, these cells appeared to correspond to the previously reported THY1⁻ FLS population [27]. Because this cluster was observed only in the OA1 sample, sampling error with contamination by synovial infiltration on the cartilage surface was the most likely explanation [28]. Currently, we take great care to avoid harvesting cartilage from the edge of the condyle to minimize contamination from synovial invasion and osteophytic regions. Given that blood cell and MSC‐like cell clusters were consistently observed across previously published datasets [7, 8, 10, 13], careful preselection of cells from digested cartilage was crucial for accurately analyzing chondrocyte clusters in scRNA‐seq.

Articular chondrocytes were classified into eight clusters. Compared with previous scRNA‐seq studies of OA chondrocytes (Supporting Information S4: Table 4), this study did not identify a distinct cluster corresponding to prehypertrophic chondrocytes (preHTCs). As HTCs formed a large cluster, it is likely that preHTCs were overshadowed by the HTC population. In trajectory analysis, two differentiation pathways originating from HomCs were observed, leading toward FCs and MetabCs as the endpoints, and TF activity analysis further supported this concept by identifying distinct TF activity programs in the two lineages. Given that the anabolic, homeostatic, and catabolic gene sets were all enriched, the trajectory toward MetabCs appeared to be associated with disease activity. The proportions of FCs and preFCs were correlated with age. Moreover, the presence of a significant number of FCs in aged non‐osteoarthritic samples suggests that FCs may not be directly involved in OA pathogenesis. Collectively, these findings suggest that the trajectory toward FCs represents an aging process, while the trajectory toward MetabCs is disease‐related.

A novel cluster was identified that did not correspond to previously reported marker genes. Although the proportion of this cluster was very small, this cluster showed a distinct phenotype, exhibiting enrichment in several gene sets, including oxidative phosphorylation, glycolysis, and proteasome pathways, all of which are known to be upregulated in response to cellular stress [29]. Gene sets associated with UPR and mTORC1 signaling, both involved in ER stress, were also enriched, suggesting that MetabCs may represent a stress‐resistant cell population. The UPR is a critical cellular mechanism activated by ER stress and is primarily mediated through three ER transmembrane stress pathways [30]. Serum TC levels correlated with MetabC cluster proportion. In vitro experiments demonstrated that exposure to high cholesterol concentrations induced ER stress via the BiP–CHOP axis, leading to decreased expression of anabolic cartilage genes and increased expression of catabolic genes. Consistently, Lemmer et al. reported that excessive extracellular cholesterol can trigger ER stress and inflammation [31]. Under normal conditions, elevated cholesterol levels trigger negative feedback inhibition of intracellular cholesterol synthesis [32]. However, an enhanced ER stress response can upregulate cholesterol synthesis [33], which is consistent with the enrichment of cholesterol synthesis‐related GO terms in our analysis, as well as with the results of in vitro experiments using tunicamycin stimulation in chondrocytes. These findings suggest that MetabCs may contribute to OA progression through ER stress‐mediated mechanisms, as excessive ER stress exacerbates OA pathology [34, 35]. Although the relationship between cholesterol levels and knee OA remains controversial [36, 37, 38], previous studies have reported elevated cholesterol levels in the synovial fluid of OA patients [39]. Given that cholesterol homeostasis is linked to ER stress, these findings highlight the importance of ER stress mechanisms in OA pathology.

In scRNA‐seq, where each cell is treated as a single sample, p‐values for many genes were inflated to nearly zero, making DEG identification difficult. Thus, pseudobulk analysis was also employed, focusing on DEGs detected by both approaches [40]. From this analysis, two genes, HILPDA and AKR1C2, possibly associated with OA pathology and not highlighted in previous studies, were identified. Within‐cluster differential expression analysis demonstrated that OA induced both shared and subpopulation‐specific gene expression alterations. Notably, HILPDA was consistently downregulated in OA across most chondrocyte subpopulations, suggesting that it represents a broadly shared OA‐associated molecular alteration. Overexpression of HILPDA, which encodes hypoxia‐inducible lipid droplet‐associated, has been reported to promote chondrogenesis in human adipose‐derived stem cells (hADSCs) by increasing the expression of SOX9, ACAN, and COL2A1, while reducing the expression of MMP13 and MMP3 [41]. In the current study, knockout of HILPDA reduced anabolic gene expression, increased catabolic gene expression, and led to small, distorted pellets in primary chondrocyte pellet culture. Considering both the previous findings on enhanced chondrogenesis and the current evidence of impaired chondrocyte homeostasis, HILPDA appears to have a chondroprotective role. AKR1C2 encodes aldo‐keto reductase family 1 member C2, which functions as a 3‐ketosteroid reductase that regulates hormone activity by converting androgens into inactive metabolites [42]. It has been implicated in cancer progression and is linked to inflammatory and metabolic processes [43]. Although its role in chondrocytes was previously unknown, knockdown experiments suggested a chondrodestructive function.

Classically and histologically, cartilage is composed of four layers: superficial, transitional, middle, and deep layers, with an additional calcified layer and subchondral bone beneath them [44]. To determine how clusters defined by scRNA‐seq analysis are localized within these layers, many studies have utilized immunohistochemistry with marker genes to investigate their spatial distribution [7, 9, 10, 45, 46]. However, the drawback of this approach is that marker genes are not cluster‐specific but are expressed in multiple chondrocyte clusters. Fan et al. employed LCM to perform RNA sequencing across four distinct layers of cartilage. Using deconvolution analysis, they estimated the distribution of cellular clusters within these layers [10]. The accuracy of deconvolution analysis largely depends on the reference datasets, tissue type, and data quality [47]. In chondrocytes, the definition of each cluster varies across studies, making it challenging to precisely estimate their proportions. In the current study, cartilage was divided into top‐ and bottom‐halves for scRNA‐seq analysis. These findings are consistent with previous reports, indicating a higher prevalence of relatively undifferentiated chondrocytes in the superficial zone and a greater presence of HTCs in the deep zone [48]. The abundant presence of FCs and preFCs in the superficial zones aligns with previous reports suggesting that fibrosis progresses from the superficial zones [49]. In the worn cartilage, the top‐half was eroded, leaving only the deep zone. However, chondrocytes in worn cartilage differed from those in the bottom‐half of full‐thickness cartilage and exhibited a phenotypic shift toward a top‐half‐like state, expressing known superficial chondrocyte‐specific genes. GSEA indicated that the top‐half cartilage underwent inflammatory and degenerative responses due to direct stress exposure, whereas the bottom‐half showed metabolic and adaptive reactions to maintain homeostasis, suggesting a compensatory role of the deeper layer. Furthermore, GSEA revealed a mosaic pattern in worn cartilage, with upregulated pathways typical of both the top‐ and bottom‐halves. These findings suggest that chondrocytes in OA cartilage dynamically alter their phenotype in response to spatial redistribution and matrix destruction. From a technical perspective, collecting cartilage with sufficient depth is crucial for scRNA‐seq, as sampling only the surface layer may yield misleading gene expression profiles limited to superficial chondrocytes.

This study had several limitations. Although it included one of the largest cell numbers among scRNA‐seq studies of articular chondrocytes, only 12 patients were analyzed. The variability in cell clusters across studies necessitates a larger sample size to reach a universal conclusion. The small number of NC patients and absence of a middle‐aged population should also be addressed. Additionally, enzyme digestion used for scRNA‐seq isolation may have influenced gene expression. Although a combined analysis with other joint tissues, such as synovium and meniscus, could provide a broader understanding of OA pathology [50], crosstalk between chondrocytes and other cell populations within the joint has not been investigated. The origin of some clusters identified as non‐chondrocytic cells remains undetermined, as classification was based on gene expression without thorough in vitro validation. Whether articular cartilage consists solely of chondrocytes remains an important question for future research. Although a novel chondrocyte cluster was identified in this study, the population was small. Its contribution to OA progression is not well elucidated, warranting further investigation.

In conclusion, careful preselection of non‐chondrocytic cells is crucial for accurate scRNA‐seq clustering of chondrocytes. A novel chondrocyte cluster associated with cholesterol metabolism and ER stress was identified and implicated in OA pathogenesis. Worn cartilage chondrocytes were not merely remnants but had acquired phenotypic features of the lost superficial layers.

Author Contributions

Yu Kobori contributed to the analysis and interpretation of the data, drafting the article. Kohei Nishitani contributed to the conception and design, drafting of the article, and final approval of the article. Hiroyuki Yoshitomi contributed to the conception and design, analysis and interpretation of the data, and technical support. Akinori Murakami contributed to analysis and interpretation of the data, and technical support. Kazuki Ichiyanagi, Shigeo Yoshida and Jiu Tanaka contributed to the analysis and interpretation of the data. Shinichi Kuriyama, Shinichiro Nakamura, Akio Sakamoto, and Takashi Noguchi contributed to the provision of study materials or patients. Yugo Morita, Takashi Sono, Takayuki Fujii, and Koichi Murata contributed to the interpretation of the data and critical revision of the article for important intellectual content. Junya Toguchida contributed to the obtaining of funding and interpretation of the data. Hideki Ueno contributed to the interpretation of the data, provision of study materials, and technical support. Yasuhiro Murakawa contributed to the analysis, and technical support. Chikashi Terao contributed to the obtaining of funding and interpretation of the data. Shuichi Matsuda contributed to the provision of study materials or patients, and final approval of the article. All authors have read and approved the final submitted manuscript.

Conflicts of Interest

K.N. serves as the Editorial Review Board Member of the Journal of Orthopaedic Research. The remaining authors declare no conflicts of interest.

Declaration of AI Use

During the preparation of this work the authors used ChatGPT‐5.1 to improve the readability of the English language. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Supporting information

Supporting File 1

Supporting File 2

Supporting File 3

Supporting File 4

Supporting File 5

Supporting File 6

Data Availability Statement

The data that support the findings of this study are openly available in Japanese Genotype‐phenotype Archive at https://www.ddbj.nig.ac.jp/jga/index-e.html, reference number JGAS000880.