Prognostic Genetics in Multiple Sclerosis: Independent Replication Challenges the Clinical Utility of a Proposed Severity Genotype

Genome-wide association studies (GWAS) have mapped multiple sclerosis (MS) susceptibility to hundreds of loci—233 risk-associated single-nucleotide variants (SNVs) are cited as the current benchmark—yet translating genetics into forecasts of long-term disability has remained difficult. This gap matters clinically because disability accumulation (rather than relapse frequency alone) drives much of the lifetime burden in MS. Two recent “severity GWAS” efforts shifted the field by reporting variants linked to progression, most notably rs10191329A reaching genome-wide significance in the International Multiple Sclerosis Genetics Consortium (IMSGC) study, raising the prospect that genotyping could support individualized management. The present Annals of Neurology study directly addresses this translational question: can the putative MS “severity genotype” meaningfully predict real-world disease course in a well-characterized longitudinal cohort?

Study Design and Cohort: A Prospective Registry as a Replication Platform
Kreft and colleagues analyzed 1,455 people with MS from the South Wales MS registry, a prospective cohort with standardized clinical capture extending back to 1985, explicitly excluding individuals included in the original IMSGC or MSBase severity GWAS to preserve independence. Participants met the 2017 McDonald criteria, and analyses were limited to individuals of Caucasian ancestry, aligning ancestry with discovery cohorts to mitigate population stratification. The cohort’s genotype distribution for rs10191329 was typical for Europeans (approximately 2.5% AA and 28.6% CA), and baseline demographics were broadly similar across genotype groups , providing an appropriate setting to test whether previously reported genetic signals generalize to routine clinical populations.

Genotyping and Clinical Outcomes: Anchoring “Severity” to Longitudinal Disability Metrics
Genotyping used established arrays (Immunochip or Illumina CoreExome) with stringent quality control and imputation to a dense genome-wide set of variants; principal components were computed to address stratification. Clinically, disability was operationalized using EDSS assessments performed during in-person examinations (excluding phone/questionnaire EDSS), then transformed into the age-related multiple sclerosis severity score (ARMSS), consistent with the IMSGC severity GWAS framework. The authors interrogated multiple clinically salient endpoints: ARMSS as a continuous severity metric, survival analyses for time to EDSS milestones (4, 6, and 8) and time to secondary progressive MS, plus relapse phenotypes including annualized relapse rate (ARR), time to second relapse, recovery after the first event, and anatomical localization patterns.

The Central Result: rs10191329A Does Not Predict Disability Accumulation in This Real-World Cohort
Across both an IMSGC-like subcohort (older individuals with longstanding disease) and the full 1,455-person dataset, rs10191329A showed no meaningful association with long-term disability as measured by ARMSS (Figures 1–2, pages 4–5). Survival analyses similarly failed to show genotype-dependent differences in time to EDSS 4, 6, or 8, or in conversion to secondary progressive MS (Figure 3, page 6), even when covariate-adjusted Cox models were applied. Importantly, the study also tested whether rs10191329A tracked earlier inflammatory activity: among relapse-onset cases, there were no detectable differences in ARR, time to second clinical event, incomplete recovery after onset, or relapse localization distributions . Collectively, these convergent null findings argue that rs10191329A, as a single marker, is not presently informative for patient-level prognostication in typical clinical practice.

Polygenic Scores: Susceptibility Burden and “Severity wGRS” Also Fail to Stratify Outcomes
Recognizing that complex traits often require polygenic modeling, the investigators evaluated several genetic score constructs. A weighted genomic risk score (wGRS) derived from 181 non-HLA susceptibility loci and an HLA genetic burden score (10 SNVs) were not associated with ARMSS, supporting the concept that MS risk architecture is not simply recycled as MS severity architecture (pages 7–8). They also constructed a “severity” wGRS based on the genome-wide significant SNV and four suggestive IMSGC progression variants available in their data; this score showed no association with first or last ARMSS, age at onset, or time-to-milestone outcomes when comparing extreme quantiles (Figures 2E–H and supplementary analyses described on pages 4 and 8). In practical terms, both single-variant and polygenic approaches derived from currently known loci were insufficient to partition this cohort into reliably different disability trajectories.

What Did Replicate: Modest Associations for Two Suggestive MSBase Severity Variants and Age-at-Onset Effects for HLA-DRB1*1501
While rs10191329A did not replicate, the study did reproduce more modest signals from the MSBase severity GWAS: rs7289446G (and its perfect LD proxy rs1207401) and rs868824C showed associations with longitudinal severity metrics, with effect directions consistent with MSBase and statistically significant results in adjusted models for specific genotype contrasts (Table 2, page 9). These effect sizes were described as modest, reinforcing that even “true positive” severity loci may be individually weak predictors. Additionally, the study assessed rs3135388 as a proxy for HLA-DRB1*1501 and found no association with long-term disability measures (ARMSS/MSSS), but did replicate an association with younger age at onset . This pattern aligns with a broader interpretation that some immunogenetic signals influence disease initiation timing rather than downstream neurodegenerative disability accumulation.

Interpretation and Clinical Implications: Why Independent Replication Matters Before Genotyping Enters Care
The authors emphasize several plausible reasons for discordance with prior findings, including reduced selection bias in a prospective registry (as opposed to case–control enrichment), exclusive use of in-person EDSS measurements to limit measurement bias at lower EDSS ranges, and the possibility that discovery-phase effect estimates may regress toward smaller true effects in replication (Discussion, pages 10–11). They further note intrinsic challenges in using EDSS as a severity anchor—its nonlinearity and emphasis on ambulation at higher scores may undercapture cognition, dexterity, and other disability domains—while also underscoring that relapse measures may not map cleanly onto sustained progression . The overarching conclusion is therefore pragmatic: severity GWAS findings are biologically valuable for understanding mechanisms of MS progression, but current individual genotyping (including rs10191329A and present polygenic severity constructs) is unlikely to support clinical decision-making or counseling at this time; robust replication and phenotype standardization remain essential steps toward clinically actionable prognostic genetics in MS.

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:
Kreft, K. L., Uzochukwu, E., Loveless, S., Willis, M., Wynford‐Thomas, R., Harding, K. E., ... & Robertson, N. P. (2024). Relevance of multiple sclerosis severity genotype in predicting disease course: a real‐world cohort. Annals of Neurology, 95(3), 459-470.