Peripheral blood DNA methylation changes are predictive of incident radiographic osteoarthritis development: Data and...

The lack of biomarkers to predict knee osteoarthritis (OA) progression is a key unmet need in the OA field. The objective of this study was to evaluate epigenetic changes in baseline peripheral blood cells of healthy individuals who subsequently developed radiographic knee OA (incident OA) as a preclinical biomarker.

Genome-wide DNA methylation data were generated using Illumina Methylation EPICv2 arrays on baseline blood samples from 234 incident OA patients and 234 controls matched by age±5y, sex, ethnicity, and BMI±5 units. Peripheral blood cellular composition was estimated. Lasso was used for CpG selection. Models were first trained using a 40-round Monte Carlo cross-validated generalized logistic modeling approach, then parsimonious models developed using CpG sites selected in ≥10 rounds.

Future incident OA cases were estimated to have increased monocyte frequency (0.089 ​± ​0.0001 vs. 0.083 ​± ​0.0001 mean ​± ​SEM, q ​= ​0.01), although cell composition models were not predictive (AUC ​= ​0.56 ​± ​0.01) Full methylation models accurately differentiated future incident OA patients from controls but suffered from overfitting (development: AUC ​= ​0.91 ​± ​0.001, validation: AUC ​= ​0.64 ​± ​0.006). Parsimonious models developed using 21 top CpG sites performed similarly with less overfitting (development: AUC ​= ​0.91 ​± ​0.001, validation: AUC ​= ​0.87 ​± ​0.003, accuracy 0.79 ​± ​0.004, sensitivity 0.78 ​± ​0.004, specificity 0.80 ​± ​0.006). Previously developed OA progression models did not predict incident OA (AUC ​= ​0.48 ​± ​0.002). Five of 21 CpG sites were located proximate to known genes, including RGS9BP, TOLLIP, UOX, L3MBTL1, and IFI27L1.

Blood-based DNA methylation models may predict incident knee OA in healthy individuals using a small set of CpG sites. Further work should focus on evaluating the pathophysiological role of DNA methylation in incident OA.