134 Plasma Proteins Linked to Osteoporosis Risk Identified in 42,000-Person Study

Osteoporosis affects roughly 1 in 5 adults globally and causes debilitating fragility fractures, yet early detection remains inadequate. Circulating plasma proteins offer a window into disease biology that could enable earlier intervention, but prior proteomic studies have been limited by small samples, cross-sectional designs, and narrow protein panels.

This prospective cohort study enrolled 42,325 adults from the UK Biobank who were free of osteoporosis at baseline. Of these, 1,477 developed osteoporosis during follow-up through November 2022. Plasma proteomics used the Olink Explore Proximity Extension Assay, quantifying 2,919 proteins after quality control. Cox proportional hazards regression screened all proteins for longitudinal associations with osteoporosis onset, adjusting for demographics, lifestyle, and comorbidities. Two-sample Mendelian randomization (MR) using protein QTLs from UKB-PPP and GWAS summary data from FinnGen (8,017 cases, 391,037 controls) tested causal directionality. Machine learning (LightGBM) then ranked protein predictors, and Weighted Gene Correlation Network Analysis (WGCNA) identified co-expressed protein modules.

Among 2,919 proteins tested, 134 were significantly associated with osteoporosis after Bonferroni correction. The strongest associations involved SOST (sclerostin), a canonical Wnt signaling inhibitor that suppresses bone formation; ADIPOQ (adiponectin), an adipokine with roles in bone metabolism; and CKB (creatine kinase B-type). Twelve proteins showed significant associations with bone mineral density T-scores at the femoral neck, lumbar spine, and total body. MR analysis confirmed causal relationships for 17 proteins, with FSHB (follitropin subunit beta), SOST, and ADIPOQ emerging as top predictive features in machine learning. The 10-protein predictive model achieved AUC = 0.803 for onset within 5 years, outperforming a model using traditional risk factors alone.

WGCNA network analysis identified three osteoporosis-related protein modules: one enriched in immune and complement pathways, one in lipid metabolism, and one centered on FSH regulation. Mediation analyses showed these modules acted as biological intermediaries between established risk factors—smoking, sleep, physical activity, polygenic risk score, and menopause—and osteoporosis development. This finding suggests that diverse environmental and genetic exposures converge on a limited set of molecular pathways to drive bone loss.

The study's large scale, prospective design, and integration of causal inference methods represent major strengths. However, the population is predominantly of European ancestry, limiting generalizability. Protein measurements were taken at a single baseline timepoint, precluding assessment of protein trajectory dynamics. Additionally, osteoporosis was ascertained via hospital records and ICD-10 codes, which may undercount mild or community-diagnosed cases.