Multi-omic profiling reveals distinct microrna, long non-coding RNA, circular RNA, and gene interactomes in late radiographic stage knee osteoarthritis patients

Purpose: Osteoarthritis (OA) is a painful condition affecting mobility and quality of life. Molecular mechanisms which underlie the development and manifestation of this disease are not well understood. Current evidence indicates that OA is heterogenous in presentation, and there may be different molecular phenotypes which contribute to these observations. Elucidating phenotypes could assist in the development of precise therapies targeted towards specific patients. The synovium is an established mediator of joint inflammation and OA pathology, yet there are few studies performing deep phenotyping in this tissue. Our objective was to develop comprehensive molecular profiles of patients with advanced radiographic knee OA using total RNA sequencing of synovial tissue paired with miRNA sequencing of blood and characterize common and differing mechanisms between patients. Methods: Synovium samples were obtained from 50 late radiographic-stage knee OA patients (KL 3/4) and matched plasma samples were obtained from the same individuals. Patients with post-traumatic, crystalline or inflammatory arthritis were excluded. These matched samples were subjected to multi-omic approaches (RNA and miRNA sequencing) paired with computational and rigorous biostatistical analyses. Total RNA sequencing was performed on synovial tissue samples using Illumina Truseq-Stranded Total RNA kits on the NextSeq 550. After filtering, 19,857 genes were expressed in synovium. 4267 genes with mean, and variance greater than 30 across all samples were kept for further downstream analysis. Cluster number was objectively identified by plotting the mean silhouette width. It peaked at 3, indicating that this was the optimal cluster number. Genes differentially expressed between clusters were used for pathway analysis. Identified pathways were annotated if they had a q-value less than 0.01 and a gene-ratio greater than 0.05. Networks were constructed based on interactions among genes collected using the Integrated Interactions Database (University of Toronto). Pearson correlation was calculated in each cluster for long-noncoding (lnc) and circular (circ) RNA from differentially expressed genes between clusters. Only correlations with absolute rho greater than 0.95 (lncRNA) or 0.85 (circRNA) were retained and used to construct Protein-Protein Interaction networks. MicroRNAs targeting genes differentially expressed between clusters in the synovium were predicted using mirDIP 4.1 (University of Toronto). MiRNA from the plasma of 50 matched late stage knee-OA patients was measured using miRNA sequencing (QIAseq miRNA library kit, Illumina NextSeq 550). Hypergeometric testing and correlation analyses was used to identify whether predicted miRNA from synovium was correlated to miRNA expressed in the blood of the same patients. Results: Cluster analysis of RNA seq data indicates three clusters of patients distinguished by unique interactomes. Common signaling pathways between clusters include innate immune system, Epidermal Growth Factor Receptor Signaling (EGFR1), and adaptive immune system. Differentially regulated pathways between clusters 2 vs 1 include: signaling by receptor tyrosine kinase, cytokine signaling, and T-cell receptor pathway (TCR). Pathways which are differentially regulated between cluster 3 vs 2 include: neutrophil degranulation, skeletal muscle, EMT regulation, apoptosis, membrane trafficking, vesicle mediated transport, toll-like receptor signaling and hemostasis. We have also identified several long-noncoding and circular RNA that interact with genes in each of these distinct pathways and have defined which cluster they are associated with as well as their direction of interaction (positive or negative). Further, we have found that hundreds of miRNA are predicted to target genes expressed in the synovium. When we compared the predicted miRNA expressed in synovium, to expressed miRNA measured in the blood (through miRNA-seq) of the same patients; we found that these were significantly correlated (p=0.003, cluster 3 vs 2, p=3.3xe-09, cluster 2 vs 1). Out of these, 13 miRNA were associated with specific pathways annotated in our clusters. Conclusions: We have demonstrated that endogenous molecular signatures derived from the synovium of advanced radiographic knee OA patients cluster into 3 sub-groups which share common and distinct actively transcribed pathways. Identified long-noncoding and circular RNA contribute to the complex regulation of unique molecular mechanisms differentially expressed between clusters. Systemically measured miRNA demonstrate significant correlation to predicted miRNA based on genes expressed in synovium, indicating potential roles for these miRNA as disease mediators. Though whether these miRNA are released from the joint, or whether they are derived systemically and target the joint remains unknown. Further understanding the regulation of endogenous molecular mechanisms through the elucidation of patient-specific interactomes may help develop more precisely targeted therapies.