Polygenic Risk Scores for Multiple Sclerosis Across Ancestries: Evidence of Transferability Gaps
Multiple sclerosis (MS) is a complex immune-mediated demyelinating disease of the central nervous system with a multifactorial etiology that includes both environmental exposures and substantial polygenic contribution. In the featured medRxiv preprint, Rivier and colleagues evaluate whether a previously constructed MS polygenic risk score (PRS)—largely informed by genome-wide association studies (GWAS) in European-ancestry cohorts—retains its stratification performance across diverse genetic ancestries in the United States. The scientific motivation is straightforward: if PRS are to inform prevention research, recruitment enrichment for clinical trials, or eventual risk communication, they must generalize beyond the populations in which the discovery GWAS were performed. The authors leverage the “All of Us” Research Program, a large, ethnically diverse resource with whole-genome sequencing (WGS) and linked electronic health records (EHR).
Study Design: Using All of Us to Test Cross-Ancestry Transferability
The investigators conducted a cross-sectional analysis using All of Us data collected between 2018 and 2023, restricting inclusion to participants with both WGS and EHR data. Genetic ancestry was assigned centrally by principal components analysis relative to reference panels, and analyses focused on the three largest strata: European, African, and Latino/admixed American (L/A). From 413,457 All of Us participants, 173,153 met the WGS+EHR inclusion criteria; to enable balanced comparisons, the authors randomly sampled European and African participants to match the L/A sample size (32,428 per ancestry group). MS status was ascertained via ICD-10 and SNOMED codes in the EHR, capturing both prevalent and incident diagnoses relative to baseline.
Constructing the Exposure: A 282-Variant MS Polygenic Risk Score
The PRS was built from 282 independent, common (minor allele frequency >1%) biallelic single-nucleotide polymorphisms meeting genome-wide significance and linkage disequilibrium constraints (reported r² < 0.1), aligning with common best-practice heuristics for risk score construction. For each individual, the score reflects the weighted sum of risk allele counts, where weights are effect sizes reported for MS-associated variants. Within each ancestry-matched sample, the PRS was normalized and discretized into quintiles to form five ordered categories (very low to very high genetic risk). The primary inferential step then assessed whether higher PRS categories corresponded to higher odds of MS diagnosis after adjustment for age, sex, and genetic principal components.
Cohort Characteristics: Disease Prevalence and Baseline Differences
Across the ancestry-matched samples (32,428 individuals per group), MS case prevalence differed by ancestry: 1.0% (327 cases) in the European sample, 0.56% (183 cases) in the African sample, and 0.46% (150 cases) in the L/A sample. In the broader analytic cohort summarized in Table 1, MS cases were disproportionately female (77.3%) compared with non-cases (60.0%), consistent with the known sex skew in MS epidemiology; age at enrollment was similar between groups. The table also reports differences in smoking and body mass index distributions between MS and non-MS groups, which is relevant because PRS captures only inherited genetic propensity and cannot represent environmental or behavioral components of risk. These descriptive findings frame why PRS should be interpreted as one dimension within a broader causal architecture rather than as a stand-alone screening instrument.
Main Findings: PRS Stratifies Risk in European and L/A Groups, Not in African Ancestry
The central result is a clear ancestry-dependent gradient in PRS performance. In adjusted models, individuals in the highest PRS quintile had substantially higher odds of MS than those in the lowest quintile in the European group (odds ratio 2.41, with a strong test-for-trend), and similarly elevated odds in the L/A group (odds ratio 2.56, also with a significant trend). In contrast, the same PRS did not significantly partition the African ancestry group into distinct risk categories (e.g., highest vs lowest quintile odds ratio reported as 1.45, with a non-significant trend). The figures on page 12 visually reinforce this divergence: Figure 1 shows a consistent rise in MS case proportions across PRS quintiles in European and L/A samples, whereas the African sample displays a weaker and less monotonic pattern; Figure 2 similarly shows increasing adjusted odds ratios across quintiles for European and L/A, but attenuated and statistically inconclusive separation for African ancestry.
Interpretation: Why Cross-Ancestry PRS Transfer Often Fails
The authors’ discussion aligns with a broad statistical genetics consensus: PRS portability is limited when discovery GWAS disproportionately represent one ancestry, because effect sizes, linkage disequilibrium structure, allele frequencies, and tag-SNP performance differ across populations. For African ancestry in particular, greater genetic diversity and shorter linkage disequilibrium blocks can reduce the ability of European-derived tag variants to capture underlying causal variation, leading to attenuated prediction even when causal biology overlaps. In addition, true heterogeneity of genetic architecture and gene–environment interplay can further modulate effect estimates across populations, meaning that a “one-size-fits-all” PRS may systematically underperform in groups that were underrepresented during discovery. In practical terms, the paper underscores that observed under-stratification is not merely a technical inconvenience; it is a direct pathway by which precision medicine tools can propagate inequity if deployed without ancestry-aware validation and recalibration.
Limitations and Next Steps: From Demonstration to Methodological Remedies
Several limitations shape how these findings should be operationalized. First, the absolute prevalence of MS is low, so even statistically significant odds ratios may translate into modest absolute risk differences, limiting near-term utility for population screening; second, EHR code-based phenotyping can introduce misclassification and ascertainment heterogeneity across health systems. The authors also note that PRS performance in All of Us appears lower than in some other large resources even among European ancestry participants, implying dataset-specific factors (population heterogeneity, measurement variance, or unmeasured covariates) that can affect transportability beyond ancestry alone. The most actionable scientific direction is explicit: expand non-European representation in MS GWAS, develop ancestry-specific (or multi-ancestry) PRS with appropriate modeling of linkage disequilibrium and local ancestry, and validate performance within each target population prior to clinical translation. Finally, because this work is presented as a preprint, its conclusions should be viewed as provisional pending peer review, albeit methodologically coherent and consistent with established principles of PRS transferability.
Disclaimer: This blog post is based on the provided research article and is intended for informational purposes only. It is not intended to provide medical advice. Please consult with a healthcare professional for any health concerns.
References:
Rivier, C. A., Payabvash, S., Zhao, H., Hafler, D. A., Falcone, G. J., & Longbrake, E. E. (2024). Differential Results of Polygenic Risk Scoring for Multiple Sclerosis in European and African American Populations. medRxiv, 2024-06.
