Single-cell mass cytometry reveals cross-talk between inflammation-dampening and inflammation-amplifying cells in osteoarthritic cartilage

Abstract

Aging or injury leads to degradation of the cartilage matrix and the development of osteoarthritis (OA). Because of a paucity of single-cell studies of OA cartilage, little is known about the interpatient variability in its cellular composition and, more importantly, about the cell subpopulations that drive the disease. Here, we profiled healthy and OA cartilage samples using mass cytometry to establish a single-cell atlas, revealing distinct chondrocyte progenitor and inflammation-modulating subpopulations. These rare populations include an inflammation-amplifying (Inf-A) population, marked by interleukin-1 receptor 1 and tumor necrosis factor receptor II, whose inhibition decreased inflammation, and an inflammation-dampening (Inf-D) population, marked by CD24, which is resistant to inflammation. We devised a pharmacological strategy targeting Inf-A and Inf-D cells that significantly decreased inflammation in OA chondrocytes. Using our atlas, we stratified patients with OA in three groups that are distinguished by the relative proportions of inflammatory to regenerative cells, making it possible to devise precision therapeutic approaches.

Fig. 1. High-dimensional profiling of normal and OA chondrocytes using mass cytometry.

(A) Schematic outlining the procedures used to profile chondrocytes by mass cytometry. Briefly, cells are dissociated from cartilage tissue, stained with metal-conjugated antibodies, and analyzed using cyTOF. The resulting data are then gated for live, SOX9/CD44-positive chondrocytes that are used for downstream analyses, including identifying clusters with FlowSOM. (B) tSNE projections of the normal (blue) and OA (red) chondrocytes where each cell is represented by a dot. Each group was downsampled randomly to 9000 cells. (C) Normal chondrocytes colored by patient sample, downsampled to 9000 cells. (D) OA chondrocytes colored by patient sample, downsampled to 9000 cells. (E) tSNE plots of 9000 normal chondrocytes, colored by the expression of two chondrogenic markers (SOX9 and CD44), the cell surface receptor NOTCH1, and pNF-κB. Expression is set at the max of each channel and is comparable between (E) and (F). (F) tSNE plots of 9000 OA chondrocytes, colored by the expression of two chondrogenic markers (SOX9 and CD44), the cell surface receptor NOTCH1, and pNF-κB. Expression is set at the max of each channel and is comparable between (E) and (F).

Fig. 2. Normal and OA cartilage landscape consists of both abundant and rare subpopulations.

(A) Abundance of each of the 20 clusters called by FlowSOM analysis in normal samples. Each point represents a single sample. (n = 5). (B) Abundance of each of the 20 clusters called by FlowSOM analysis in OA samples (n = 20). Each point represents a single sample. (C) Expression of cell surface receptors used for delineating the 20 clusters. Expression is averaged between all cells of a given cluster ID. Color is scaled to 1 for each protein between all the clusters. Dendograms were drawn using complete-linkage hierarchical clustering. (D) Table of the cluster IDs that are enriched, depleted, or similar between OA and normal samples. Colors in the enriched section correspond to the tSNE projection on the right. The tSNE projection contains cells from clusters that are enriched in OA compared to normal, sampled to 9000 cells. Enrichment, depletion, or similarity between the ranked means of normal (n = 5) and OA (n = 20) cluster abundance was tested using an unpaired, two-tailed Mann-Whitney test with Bonferroni correction (α = 0.0025). Adjusted P values for all enriched or depleted clusters are 0.002. (E) Coefficient of variation (mean divided by SD) for each cluster in normal or OA samples. (F) Shannon’s diversity index (H) calculated for each normal and OA sample (see Methods). Theoretical max H value is 2.99. Equality between the means H values for OA (n = 20) and normal (n = 5) samples was tested using a two-tailed Mann-Whitney test. ***P = 0.001. (G) Hierarchical clustering of normal and OA samples by cluster abundances. Abundance is scaled to 1. Samples belonging to the three designated groups are labeled at the bottom. (H) Average cluster abundance in normal and group A, B, and C patients with OA. Each color designates a cluster ID.

Fig. 3. Patients with OA are differentially enriched in types of CPCs.

(A) Expression of the 13 CPC markers among the clusters that are enriched for them. Expression is scaled to 1 between all clusters. (B) tSNE projections of the type I (depleted), type II (similar), and type III (enriched) CPCs in OA, colored by cluster ID, where each cluster ID has a different color. Cells are sampled to 9000 when possible. (C) Cell cycle analysis for each cluster. Cell cycle stages were analyzed for each cell, and then, the proportion of the population in G₀ and in the cell cycle was calculated for each cluster. The percentage in the cell cycle is given to the right of each bar graph. (D) Cell signaling and other intracellular and cell surface receptor markers for the CPC clusters. Expression is scaled to 1. (E) Cluster abundance for each sample in the OA groups and normal cells. Significance is tested with a multiple-test corrected Welch’s t test. ns, not significant. (F) Correlation between abundance of each cluster, labeled on each axis. Each point represents a patient with OA. The full matrix of correlations between clusters is plotted in fig. S3A. *P = 0.05, **P = 0.01, and ***P = 0.001.

Fig. 4. Identification of a rare immune recruiting population in OA cartilage.

(A) tSNE projection of normal cells (gray) with clusters 15 and 20 colored, sampled at 9000 cells. (B) tSNE projection of OA cells (gray) with clusters 15 and 20 colored, sampled at 9000 cells. (C) A magnified projection of clusters 15 and 20 from normal and OA samples, ****P = 0.0001. (D) Quantification of the abundance of clusters 15 and 20 in normal and OA samples. Significance is tested using Welch’s t test. Each point represents a sample. (E) Magnified projection of clusters 15 and 20 depicting expression of the two cell surface receptors, TNFRII and IL1R1, and of intracellular HIF2A. Expression is scaled to max value in dataset for each protein and is comparable across normal and OA samples. Heatmap below the tSNE depicts quantification of average expression in representative chondrocytes (cluster 5) in comparison to clusters 15 and 20. (F) Single-cell RNA (scRNA) sequencing data from (25) reanalyzed. Cells expressing TNFRII and IL1R1 were sorted in silico, and their transcriptome was compared to the rest of the OA cells and used for Gene Ontology (GO) term and STRING analyses. (G) Same as in (E), for signaling markers pJNK1/2, pNF-κB (H), and pSMAD1/5 (I). (J) Fold change in cytokines from human 62-plex Luminex assay between dimethyl sulfoxide (DMSO) and JNK inhibitor treatment. (K) Fold change in cytokines from human 62-plex Luminex assay between DMSO and NF-κB inhibitor treatment. (L) Fold change in cytokines from human 62-plex Luminex assay between DMSO and Alk inhibitor treatment. (M) Raw mean fluorescence intensity (MFI) values for cytokines that were significantly altered between DMSO- and JNK-treated samples in at least five of six tested OA samples. Significance was first tested for using analysis of variance (ANOVA) with multiple corrections for the 62 comparisons, and then, t test with Tukey’s correction was applied for each comparison on a patient-by-patient sample. Each point represents an independent technical treatment and cytokine analyses for the same patient (n = 6 patients with OA). AU, arbitrary units. (N and O) Same as in (M) but with NF-κB and Alk inhibitors, respectively (n = 3 patients with OA). *P = 0.05, **P = 0.01, and ***P = 0.001.

Fig. 5. A CD24⁺ subpopulation mitigates inflammation in OA cartilage.

(A) tSNE projection of normal and OA cells (gray) with clusters 17 and 18 colored, sampled at 9000 cells each. (B) Abundance of each cluster per sample. Differences between the means were tested using Welch’s t test. (C) Heatmaps of chondrogenic markers SOX9 and CD44, as well as CD24. Expression is scaled to the highest expressing cell in the group. (D) scRNA sequencing data from (25), reanalyzed. Cells expressing CD24 with a high Col2a1/Col1a1 ratio were sorted in silico, and their transcriptome was compared to the rest of the OA cells and used for GO term and STRING analyses. (E) Hierarchical clustering of OA samples based on clusters 15, 17, 18, and 20. Abundance is scaled to one for each cluster. Groups are labeled along the x axis. (F) Violin plots of abundance of clusters 17, 18, 15, and 20 in low and high Inf-D groups. Each sample is represented as a point. (G) Correlation between the abundance of cluster 20 with clusters 17 + 18. Ninety-five percent confidence interval is shown in gray dashed line. Slope of line tested is significantly nonzero. (H) tSNE projection of OA cells, with clusters 15, 20, and 19 labeled, sampled at 9000 cells. (I) Heatmaps of the average expression of each marker in the given cluster. (J) Fold change in cytokines from human 62-plex Luminex assay between control and 3-isobutyl-1-methylxanthine (IBMX) treatment. (K) Fold change in cytokines from human 62-plex Luminex assay between control and a combined IBMX and JNK inhibitor treatment. (L) Percent change in cytokine MFI between control and the combined IBMX/JNK inhibitor treatment.