Cross-Ancestry Performance of a European-Derived Polygenic Risk Score for Multiple Sclerosis: Insights from a British South Asian Cohort
In this 2023 Brain Communications article, Breedon and colleagues evaluate whether a multiple sclerosis (MS) polygenic risk score (PRS) derived from predominantly European genome-wide association studies (GWAS) retains predictive utility in individuals of South Asian ancestry. The question is scientifically and clinically important because PRS are increasingly discussed as tools for risk stratification, trial enrichment, and potentially earlier intervention in preclinical disease phases, yet their transferability across ancestries is a known limitation. The authors explicitly frame this as an equity issue: if PRS are developed and validated mainly in European populations, their underperformance in other groups could widen health disparities rather than reduce them.
Biological and Genetic Rationale
The study is grounded in the complex genetic architecture of MS. MS susceptibility is polygenic, with risk distributed across many loci of modest effect, while the major histocompatibility complex (MHC) region—especially HLA-linked variation—contributes the largest single genetic effect. The paper notes that the MHC (including DRB1*1501-related risk) has a strong influence, whereas many non-MHC loci each confer comparatively small increments in risk. This structure makes MS a natural candidate for PRS approaches, but it also means that portability can degrade when linkage disequilibrium (LD) patterns and allele frequencies differ between ancestry groups.
Cohorts, Phenotyping, and Data Processing Strategy
To test PRS portability directly, the authors analyzed two UK-based cohorts: Genes & Health (G&H), a longitudinal cohort of British–Bangladeshi and British–Pakistani participants, and UK Biobank (UKB), used here as the European-ancestry comparison cohort. In G&H, they started from 44,396 genotyped participants and, after quality control, retained 40,532 individuals, including 42 MS cases and 40,490 controls identified through linked electronic health records using harmonized ICD10/SNOMED coding (with MS defined by ICD10 code G35). Genotyping in G&H used the Illumina Global Screening Array v3, followed by imputation with the TOPMed reference panel and standard variant/sample QC filters. This careful cohort curation is a major strength, even though the eventual case count remains small.
PRS Construction and Comparative Analytical Framework
The PRS were generated using PRSice-2 with a clumping-and-thresholding approach, using external effect sizes from the IMSGC 2019 MS GWAS meta-analysis (14,802 cases and 26,703 controls). The authors harmonized SNPs between discovery and target datasets, restricted to non-palindromic biallelic variants, and explored a broad parameter grid (multiple LD clumping thresholds and P-value thresholds), generating PRS with the MHC included, excluded, and MHC-only variants. Model performance was evaluated using adjusted Nagelkerke’s pseudo-R² (assuming MS prevalence 0.002), logistic regression-based odds ratios across PRS quartiles, AUC, and calibration. Importantly, to address sample-size asymmetry between G&H and UKB, they also performed 1,000 UKB subsampling iterations matched to the G&H case/control counts (42/40,490), which is a rigorous comparative design decision.
Main Findings in the South Asian-Ancestry Genes & Health Cohort
The central result is nuanced: the European-derived MS PRS did show statistically detectable association with MS in G&H, but the effect size and explained liability were modest. The optimal PRS including the MHC explained ~1.1% of liability (adjusted pseudo-R² = 0.011; P = 0.033), while the optimal non-MHC PRS explained ~1.5% (adjusted pseudo-R² = 0.015; P = 0.015). The MHC-only score was not significantly associated with disease status (P = 0.19), and there was no significant difference between the best MHC-including and non-MHC models. In practical discrimination terms, the reported AUCs were ~0.70–0.71 with covariates, but the null model (age, sex, PCs) already achieved AUC 0.664, indicating that the incremental predictive contribution of PRS was limited in this dataset.
Cross-Ancestry Comparison with UK Biobank and Interpretation of Portability Loss
When the same analytical strategy was applied to European-ancestry UKB participants, PRS performance was materially higher. In the full UKB sample, the MHC-inclusive and non-MHC PRS explained 4.4% and 2.3% of liability, respectively; the abstract reports closely related headline values of 4.8% and 2.8% for UKB. In the matched-size permutation/subsampling analysis, the MHC-inclusive PRS remained substantially stronger in UKB than in G&H (UKB adjusted R² 4.3%, 95% CI 1.5–8.5% vs G&H 1.1%; empirical P = 0.01), while the non-MHC difference was directionally similar but less conclusive (UKB 3.2%, 95% CI 0.9–6.9% vs G&H 1.5%; P = 0.10). The authors interpret this pattern as consistent with known PRS portability limitations driven primarily by ancestry-related differences in LD and allele frequency, rather than wholly distinct causal biology.
Limitations, Scientific Significance, and Future Directions
The authors appropriately emphasize several limitations: only 42 MS cases in G&H (yielding wide confidence intervals), possible misclassification from EHR-based phenotyping, no external South Asian validation cohort, and potential overfitting because model fitting and evaluation in G&H occurred in the same dataset. They also note technical non-equivalence between cohorts (different genotyping chips and imputation panels), which can alter SNP composition in constructed scores. Nonetheless, the study makes an important contribution: it demonstrates that European-derived MS PRS retain some signal in a South Asian-ancestry cohort, but with attenuated predictive performance, reinforcing the need for larger, ancestrally diverse genetic studies and better multi-ancestry PRS methods. Scientifically, this is not a negative result; it is a critical translational calibration study showing where the field stands and what must be improved before equitable clinical deployment.
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:
Breedon, J. R., Marshall, C. R., Giovannoni, G., van Heel, D. A., Dobson, R., & Jacobs, B. M. (2023). Polygenic risk score prediction of multiple sclerosis in individuals of South Asian ancestry. Brain Communications, 5(2), fcad041.
