Related Experiment Video
Updated: Aug 5, 2026
Standardized Histomorphometric Evaluation of Osteoarthritis in a Surgical Mouse Model
Published on: May 6, 2020
Continuous multi-domain phenotyping of knee osteoarthritis identifies independent structural and symptom axes and
Yosef Sourougeon1, Gilad Nesher1, Sagy Apterman1
1Chaim Sheba Medical Center, Division of Surgery, Orthopedic Department, Ramat Gan, Israel; School of Medicine, Faculty of Medical and Health Sciences, Tel Aviv University, Tel Aviv, Israel.
Objectives:
To test whether knee osteoarthritis (OA) partitions into discrete subtypes across six biological domains, characterise structural and symptom axes, quantify the information gap between deep and routine clinical phenotyping, and identify invisible subgroups at elevated arthroplasty risk.
Methods:
Multi-Omics Factor Analysis (MOFA) was applied across six domains (886 variables) in 600 knees from the FNIH OA Biomarkers Consortium. Discrete clustering was tested using four algorithms with bootstrap stability assessment. A Clinical Shadow model projected MOFA factors into 8697 validation knees from the Osteoarthritis Initiative. Total knee replacement (TKR) over 12 years was the primary outcome.
Results:
Four clustering algorithms (GMM, Ward, HDBSCAN, k-means) failed to identify stable discrete subtypes (the best silhouette score was 0.178). MOFA revealed continuous, largely independent structural and symptom axes (r = -0.117, p = 0.004). Routine clinical assessment accounted for only 19% of the multi-domain phenotypic variance. Median splits defined four quadrants; Silent Progressors (high structure, low symptoms) comprised 23.5% of discovery and 9.7% of validation knees. Within Silent Progressors (discovery cohort), TKR rates ranged from 10.6% to 44.7% across MOFA risk-score tertiles (p=0.0002). Structure-symptom interaction was super-additive on the additive (risk-difference) scale (RERI=0.139, 95% CI 0.013-0.247, p=0.02); the odds-ratio-scale RERI (5.05, 95% CI -1.97 to 12.08) is inflated and is shown for comparison. The factor model achieved an AUC of 0.836 for TKR prediction in external validation.
Conclusions:
Knee OA showed no stable discrete subtypes and instead varies continuously along largely independent structural and symptom axes. Nearly one-quarter of deeply phenotyped knees (Silent Progressors) show high structural burden with low symptoms and may be under-recognised by symptom-triggered monitoring. External validation tested the generalisation of the continuous TKR gradient, not the replication of discrete subtypes.
