Importance  Clinical decision-making in thrombotic disorders is impeded by long-standing uncertainty regarding the magnitude of venous and arterial thrombosis risk associated with low protein S. Population-scale multiomic datasets offer an unprecedented opportunity to answer questions regarding the epidemiology and clinical impacts of protein S deficiency.

Objective  To evaluate the risk associated with protein S deficiency across multiple thrombosis phenotypes.

Design, Setting, and Participants  Cross-sectional study using longitudinal population cohorts derived from the UK Biobank (n = 426 436) and the US National Institutes of Health All of Us (n = 204 006) biorepositories. UK Biobank participants were enrolled in 2006-2010 (last follow-up, May 19, 2020) and underwent whole exome sequencing, with a subset (n = 44 431) having protein S levels measured by high-throughput plasma proteomics. Recruitment for All of Us began in 2017 and is ongoing, with participants receiving germline whole genome sequencing. Both cohorts include individual-level data on demographics, laboratory measurements, and clinical outcomes.

Exposure  Presence of rare germline genetic variants in PROS1, segmented by functional impact score (FIS), an in silico prediction of the probability that a genetic variant will disrupt protein activity.

Main Outcomes and Measures  Firth logistic regression and linear regression modeling were used to evaluate the thrombosis risk associated with low plasma protein S levels and PROS1 variants across a range of FIS ratings.

Results  The UK Biobank cohort was 54.3% female, with a median age of 58.3 (IQR, 50.5-63.7) years at enrollment. Most participants (95.6%) were of European ancestry, and 18 011 had experienced a venous thromboembolism (VTE). In this population cohort, heterozygosity for the highest-risk PROS1 variants with an FIS of 1.0 (nonsense, frameshift, and essential splice site disruptions) was rare (adjusted prevalence, 0.0091% in the UK and 0.0178% in the US) and associated with markedly increased risk of VTE (odds ratio [OR], 14.01; 95% CI, 6.98-27.14; P = 9.09 × 10⁻¹¹). Plasma proteomics (n = 44 431) demonstrated that carriers of these variants had total protein S levels that were 48.0% of normal (P = .02 compared with noncarriers). In contrast, less damaging missense variants (FIS ≥0.7) occurred more commonly (adjusted prevalence, 0.22% in the UK and 0.20% in the US) and were associated with marginally reduced plasma protein S concentrations and a smaller point estimate for VTE risk (OR, 1.977; 95% CI, 1.552-2.483; P = 1.95 × 10⁻⁷). Associations between PROS1 and VTE at both FIS cutoffs were independently validated in the All of Us cohort with similar effect sizes. No association was detected between the presence of coding PROS1 variants and 3 forms of arterial thrombosis: myocardial infarction, peripheral artery disease, and noncardioembolic ischemic stroke. The presence of PROS1 variants correlated poorly with low plasma protein S levels, and protein S deficiency was significantly associated with VTE and peripheral artery disease regardless of PROS1 variant carrier status. The elevated risk of VTE associated with germline loss of function in PROS1 was evident in Kaplan-Meier survival analysis and appeared to persist throughout life (log-rank P = .0005).

Conclusions and Relevance  True inherited loss of function in PROS1 is rare but represents a stronger risk factor for VTE in the general population than previously understood. Acquired, environmental, or trans-acting genetic factors are more likely to cause circulating protein S deficiency than coding variation in PROS1, and low plasma protein S is associated with VTE.