Abstract
Category:
Ankle Arthritis; Basic Sciences/Biologics
Introduction/Purpose:
Osteoarthritis (OA) of the ankle is a debilitating condition that impacts millions of people. However, little is known about transcriptomic signatures associated with progression to OA, nor is the synovial contribution to the process established. Our prior work using RNA-sequencing (RNA-seq) has demonstrated unique differences between both ankle and knee OA, as well as ankle OA and non-OA ankles. In these analyses, we have shown visual clustering of ankle OA, ankle non-OA, and knee OA, with subtle transitional regions between each cohort. This study’s objective is to use demographic and patient factors that may contribute to differences in patients within each cohort and evaluate multiple machine learning (ML) clustering methods to better visualize OA subtypes of our complete ankle and knee synovium RNA-seq dataset.
Methods:
RNA-seq of synovium from ankle OA (n=30), knee OA (n=6), and ankle non-OA (n=30) were analyzed. Differential expression of genes and functional enrichments were determined using the Novomagic online platform (Novogene USA, Sacramento, CA) with DESeq2 software. ML methods were employed on the total gene read counts using Python coding language to perform data i) standardization via unit variance, ii) reduction via principal component analysis, t-distributed stochastic neighbor embedding, and uniform manifold approximation and projection (UMAP) techniques, and iii) clustering via k-means, fuzzy c-means, and gaussian mixture model. The number of principal components and clusters were determined via skree plot and elbow method, respectively. The technique with the highest silhouette score was chosen for subsequent analysis; a higher score indicates better separation of the dataset. Resulting data was plotted and color-coded to visualize the data based on original sample information (surgery, prior injury, sex, and age).
Results:
RNA-seq analysis of posttraumatic OA ankles compared to primary OA ankles identified decreased enrichment of MMP- 7, metalloproteinase associated with OA and fibrosis, and increased enrichment of IGKV2D-40, a less characterized gene.
Significant reactome pathways in primary ankle OA included activation of matrix metalloproteinases, collagen formation and degradation, and extracellular matrix organization (Figure 1A). UMAP reduction with fuzzy C-means clustering (n=8 clusters) best separated the dataset. Qualitatively, the data appears separated by surgery type (ankle OA and ankle non-OA, knee OA), with a transitional region between distinct groupings (Figure 1B); a subtle but crucial difference between knee and ankle OA exists.
Increased age appears associated with knee OA and some ankle OA, while patient sex did not display any clear separation in the data.
Conclusion:
Functional analysis of gene transcription displayed enrichment in genes and pathways related to fibrosis, further indicating ankle OA may be the result of a fibrotic signature. This study additionally demonstrated that ML provides unbiased visualization of multiple groups of RNA-seq data from knee and ankle synovium. A unique transitional group of patients who were deemed non-OA yet clustered with OA samples, suggesting some non-OA synovium that could be displaying signs of early OA progression. Ultimately, exploring the impact of patient demographics using multiple data reduction and ML can help identify unique subtype of patients within the context of OA.