Stratification of knee osteoarthritis: two major patient subgroups identified by genome-wide expression analysis of articular cartilage

Abstract

Introduction: Osteoarthritis (OA) is a heterogeneous and complex disease. We have used a network biology approach based on genome-wide analysis of gene expression in OA knee cartilage to seek evidence for pathogenic mechanisms that may distinguish different patient subgroups.

Methods: Results from RNA-Sequencing (RNA-Seq) were collected from intact knee cartilage at total knee replacement from 44 patients with OA, from 16 additional patients with OA and 10 control patients with non-OA. Results were analysed to identify patient subsets and compare major active pathways.

Results: The RNA-Seq results showed 2692 differentially expressed genes between OA and non-OA. Analysis by unsupervised clustering identified two distinct OA groups: Group A with 24 patients (55%) and Group B with 18 patients (41%). A 10 gene subgroup classifier was validated by RT-qPCR in 16 further patients with OA. Pathway analysis showed increased protein expression in both groups. PhenomeExpress analysis revealed group differences in complement activation, innate immune responses and altered Wnt and TGFβ signalling, but no activation of inflammatory cytokine expression. Both groups showed suppressed circadian regulators and whereas matrix changes in Group A were chondrogenic, in Group B they were non-chondrogenic with changes in mechanoreceptors, calcium signalling, ion channels and in cytoskeletal organisers. The gene expression changes predicted 478 potential biomarkers for detection in synovial fluid to distinguish patients from the two groups.

Conclusions: Two subgroups of knee OA were identified by network analysis of RNA-Seq data with evidence for the presence of two major pathogenic pathways. This has potential importance as a new basis for the stratification of patients with OA for drug trials and for the development of new targeted treatments.

Keywords: chondrocytes; disease activity; inflammation; knee osteoarthritis; osteoarthritis.

Figure 1

Unsupervised clustering of gene expression from patients with OA. RNA-Seq data from 44 PLC cartilage samples taken at total knee replacement and 10 non-OA PLC samples were normalised and batch effect corrected. Principal component analysis of the data was performed and the first and second (PC1 and PC2) principal components are shown (A). Network-based NMF or standard NMF was used to cluster the patients with OA into subgroups with variable numbers of predefined clusters (k). Patient co-clustering matrices are coloured to show patient pairwise co-occurrence in the same cluster across the 500 clustering runs (B). Silhouette scores were used to assess the optimal number of clusters (C) and the patient fit to a subgroup (D). NMF, negative matrix factorisation; OA, osteoarthritis; PLC, posterior lateral condyle.

Figure 2

Differentially expressed genes in OA Group A and Group B. Differentially expressed genes (≥1.5 FC and ≤0.1 adjusted P values) were determined for each OA Group A (n=32) and Group B (n=26) compared with patients with non-OA (n=10). OA, osteoarthritis.

Figure 3

PhenomeExpress analysis of OA Group A and Group B. Differential gene expression data between Group A (n=32) and Group B (n=26) were analysed with the PhenomeExpress algorithm to identify dysregulated subnetworks using known disease gene associations. Nodes are coloured by fold change. Heatmaps show the gene expression between Group A, Group B and non-OA for each subnetwork.