Decoding Multiple Sclerosis Risk with Polygenic Scores: Validation Across Cohorts and Links to Long-Term Brain Atrophy

Multiple sclerosis (MS) is a chronic inflammatory demyelinating disease of the central nervous system in which inherited susceptibility is strongly polygenic rather than driven by a single locus. The article by Shams and colleagues addresses a central translational question: can a genome-wide polygenic risk score (PRS), derived from large-scale MS genome-wide association study (GWAS) data, robustly stratify disease susceptibility in independent European-ancestry cohorts, and does that inherited liability also relate to clinically meaningful phenotypes such as relapse activity or neuroimaging markers of tissue injury? By focusing simultaneously on prediction performance, integration with conventional risk factors, and phenotype association, the study aims to move PRS from a purely statistical construct toward a clinically informative biomarker framework.

Polygenic Risk Score Construction and Validation Strategy
The investigators constructed an MS-PRS using the largest available MS GWAS summary statistics at the time, applying LD-aware effect size shrinkage with LDpred2 and restricting to high-quality HapMap3 variants that passed stringent quality control across datasets. They optimized hyperparameters in a UK Biobank validation subset (UKBB Phase 1) and then evaluated generalization in an independent UK Biobank test subset (UKBB Phase 2) and in a well-curated case–control cohort from Kaiser Permanente Northern California (KPNC). Scores were standardized and adjusted for ancestry principal components, and performance was quantified using discrimination metrics such as area under the receiver operating characteristic curve (AUC) with bootstrap confidence intervals, enabling a transparent assessment of reproducibility across study designs (population cohort versus case–control).

Susceptibility Stratification and Absolute Risk Gradients
Across cohorts, MS-PRS behaved as a meaningful susceptibility marker with moderate-to-strong discrimination: AUC was ~0.73 in UKBB2 and ~0.80 in KPNC, reflecting both the informativeness of the PRS and the impact of phenotype ascertainment quality (neurologist-confirmed cases and matched controls in KPNC). Importantly, the study emphasizes tail-risk stratification: individuals in the top PRS decile exhibited markedly elevated odds of MS relative to the median decile (reported as >5-fold in UKBB2 and ~15-fold in KPNC), and prevalence increased monotonically with PRS percentile. The authors also examined cumulative incidence by age, observing divergence of risk trajectories after age ~20—consistent with typical MS onset windows—thereby translating relative genetic risk into an interpretable age-dependent absolute-risk gradient for highly predisposed strata.

Added Value Beyond Conventional Risk Factors and a Single HLA Tag SNP
A key translational contribution is the explicit comparison of multivariable clinical risk models with and without PRS. In both UKBB2 and KPNC, adding MS-PRS to baseline models (age, sex, and established conventional risk factors such as infectious mononucleosis and smoking) improved discrimination and reclassification, whereas a model using only a single SNP tagging the well-known HLA-DRB1*15:01 risk haplotype underperformed PRS-inclusive models. The paper reports gains in AUC on the order of ~0.13 (UKBB2) to ~0.26 (KPNC) when PRS is added to age/sex-only models, and additional improvements when combined with conventional risk factors; these analyses frame PRS as an additive, not substitutive, component of multifactorial risk assessment rather than a standalone diagnostic test.

Pathway-Specific PRS Highlights Putative Biological Circuits
To connect polygenic signal with interpretable biology, the authors computed pathway-specific PRS across curated canonical gene sets (MSigDB) and identified multiple pathways whose aggregated genetic burdens were significantly associated with MS risk, with replication across UK Biobank phases after multiple-testing correction. The most prominent signals were concentrated in immune and inflammatory programs—e.g., T-cell receptor signalling, MHC class II antigen presentation, interferon-γ signalling, and interleukin-mediated cascades (including IL-12 and IL-5)—alongside pathways related to infection response, extracellular matrix organization/cell adhesion, post-translational regulation (including glycosylation), and signalling systems such as NOTCH and VEGF. Conceptually, this analysis reframes “polygenicity” as a structured superposition of pathway-level liabilities, generating a shortlist of candidate programs for functional validation rather than treating PRS as a black-box predictor.

Familial Aggregation and Neuroimaging Phenotypes of Tissue Injury
The study further examines two clinically relevant extensions: familial prediction and end-organ phenotype association. In a multicase family dataset (co-affected parent–child pairs and discordant siblings), familial PRS distributions were higher in affected individuals and enabled modest discrimination among siblings (AUC ~0.65), with analyses suggesting that parental genetic load patterns may help contextualize risk in unaffected relatives—though limited variant overlap and small sample size constrain inference. In parallel, the authors tested whether susceptibility PRS associates with MRI-derived brain volume metrics in the UCSF-EPIC longitudinal cohort, finding modest baseline associations and, more notably, a significant association between higher PRS (using a leave-one-out construction to mitigate overlap with discovery GWAS) and accelerated thalamic atrophy over a 10-year follow-up. This links inherited susceptibility to a quantitative marker of neurodegeneration in established MS, consistent with partial overlap between the genetics of susceptibility and mechanisms of tissue loss.

Mendelian Randomization, Disease Activity Signals, and Translational Boundaries
To interrogate causality, the authors used two-sample Mendelian randomization (MR) with genome-wide significant non-MHC MS variants and neuroimaging GWAS summary statistics, observing only weak evidence for an effect of MS liability on thalamic volume in one dataset and noting that sensitivity analyses implicated horizontal pleiotropy rather than a robust causal pathway. Clinically, they also report that higher PRS correlated with relapse activity over a defined interval in UCSF-EPIC, whereas disability worsening based on EDSS did not show a comparable PRS association—underscoring that genetic liability may map more strongly to inflammatory activity and selective neuroanatomical vulnerability than to short-horizon disability trajectories. Overall, the paper supports PRS as a replicable enhancer of risk stratification (especially when combined with conventional factors) and as a research tool for dissecting pathway-level contributors and susceptibility–phenotype coupling, while also delineating limitations: ancestry restriction to Europeans, dependence on phenotype ascertainment, and the low base rate of MS that constrains population-wide screening utility without pretest enrichment (e.g., family history or suggestive symptoms).

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:
Loonstra, F. C., Álvarez Sirvent, D., Tesi, N., Holstege, H., Strijbis, E. M., Salazar, A. N., ... & Uitdehaag, B. (2024). Association of Polygenic Risk Score With Lifetime Risk of Developing Multiple Sclerosis in a Population-Based Birth-Year Cohort. Neurology, 103(7), e209663.