From Risk to Outcome: Longitudinal Genomics and the Biology of Multiple Sclerosis Severity
In Not all roads lead to the immune system: the genetic basis of multiple sclerosis severity (Brain, 2023), Jokubaitis and colleagues address a long-standing gap in neuroimmunology: while hundreds of loci robustly influence multiple sclerosis (MS) susceptibility, the biological determinants of severity—the pace and extent of disability accumulation after onset—remain poorly resolved. The study is motivated by a clinically consequential observation: relapse-onset MS exhibits substantial heterogeneity in long-term outcomes, yet prognostication at diagnosis is limited, constraining rational escalation of disease-modifying therapy (DMT) and stratification for trials. The authors therefore ask two connected questions: whether longitudinally defined severity phenotypes can reveal novel genetic modifiers of outcome, and whether combining genome-wide common variation with baseline clinical variables can produce clinically useful predictions of future severity.
Cohort Architecture and Longitudinal Severity Phenotyping
A major strength of the work is its use of prospectively collected, multi-centre registry data from MSBase, enabling a longitudinal characterization of disability trajectories rather than reliance on single cross-sectional scores. The investigators assembled 5,851 relapse-onset MS cases of European ancestry meeting minimum inclusion criteria, with 1,813 individuals ultimately genotyped and analysed (representing 22,884 patient-years of follow-up within the genotyped subset). Severity was operationalized using relapse-independent Expanded Disability Status Scale (EDSS) assessments to derive median longitudinal Age-Related MS Severity Scores (l-ARMSS) and median longitudinal MSSS (l-MSSS), and outcome “extremes” were defined by the top and bottom quintiles to enrich contrast between mild and severe disease courses. Notably, l-ARMSS and l-MSSS were highly correlated (r≈0.90), supporting convergent validity while still permitting complementary analyses.
GWAS Signals: No Major-Effect Variants, but a Detectable Polygenic Component
Using genome-wide association testing adjusted for ancestry principal components and clinically relevant covariates (including therapy exposure and HLA-DRB1*15:01 allele count), the primary analyses did not identify any single nucleotide variants (SNVs) exceeding conventional genome-wide significance for continuous l-ARMSS or l-MSSS. The strongest (still sub-threshold) association mapped intronically to SEZ6L on chromosome 22 (e.g., rs7289446 with β≈−0.49, P≈2.7×10⁻⁷), mirrored by a second SEZ6L intronic variant in perfect linkage disequilibrium in the l-MSSS analysis. The authors then quantify the aggregate genetic contribution via SNP-heritability, estimating ~0.19 (SE 0.15) for l-ARMSS and ~0.29 (SE 0.14) for l-MSSS (with some cross-method instability for l-MSSS), collectively supporting a model in which severity is influenced by many loci of small effect rather than a small set of large-effect modifiers.
Machine Learning as a Prognostic Layer on Polygenicity
Recognizing that conventional GWAS is underpowered to exploit weak, distributed effects, the study implements a non-linear gradient boosting model (xgboost) using SNVs with GWAS P≤0.01 (62,351 variants) together with baseline variables such as age at onset (AAO) and a weighted genetic risk score (wGRS). Critically, the model’s feature weighting corroborates the GWAS inference: no single SNV carries substantial importance (weights remain very small), but the collective signal is informative when aggregated. When trained to classify mild versus severe outcome extremes defined by l-ARMSS, the genetic-plus-baseline model achieves AUROC ≈0.835–0.836, with sensitivity ~89%, specificity ~79%, positive predictive value ~80%, and negative predictive value ~88%. In contrast, an otherwise comparable model restricted to clinical/demographic variables available at onset performs near chance (AUROC ≈0.54), underscoring that the added value derives primarily from genome-wide variant patterns rather than from baseline clinical features alone.
Sex-Stratified Associations and Time-to-Disability Validation
Although primary analyses lacked genome-wide significant hits, secondary sex-stratified GWAS identified two loci meeting genome-wide thresholds, providing evidence—albeit requiring replication—for sex dimorphism in severity genetics. In females, rs10967273 (intergenic) was associated with l-MSSS-defined severity (β≈0.83, P≈3.5×10⁻⁸), while in males, rs698805 (intronic to CAMKMT) was associated with l-MSSS severity (β≈−1.54, P≈4.35×10⁻⁸), with minimal cross-sex effects. To strengthen inferential confidence, the authors also perform survival analyses on “hard” disability milestones—time to irreversible EDSS 3 and 6—showing that several top GWAS-prioritized loci track clinically meaningful progression, including rs7289446 (SEZ6L) with adjusted hazard ratios around 0.77 for irreversible EDSS 3 and 0.72 for irreversible EDSS 6, and additional signals near RCL1 and SUCLA2; sex-specific milestone associations are also reported for loci such as MTSS1 and TCF7L2.
Tissue and Pathway Enrichment Point to CNS-Forward Biology
A central conceptual contribution of the paper is the mechanistic divergence it implies between risk biology and severity biology. Tissue enrichment analyses identify overrepresentation of genes expressed in central nervous system compartments, particularly the cerebellum (with cerebellar signals more prominent than whole blood), aligning with clinical observations that cerebellar involvement predicts poorer outcomes. Pathway-level enrichment implicates processes plausibly relevant to progression and repair: mitochondrial function, synaptic plasticity and synaptopathy, oligodendroglial biology and remyelination-related signalling (including Wnt pathway components), cellular senescence, calcium signalling, and heteromeric G-protein receptor pathways. The discussion leverages these enrichments to argue that part of severity heterogeneity may reflect differences in neural resilience, plasticity, and remyelination capacity—axes that are biologically distinct from the immune-mediated mechanisms that dominate susceptibility GWAS.
Translational Implications, Constraints, and the Path to Clinical Utility
The study’s translational proposition is pragmatic: while severity is not governed by common variants of moderate-to-large effect, genome-wide SNV clusters can be combined with readily available baseline variables to meaningfully improve prognostic classification at diagnosis, potentially informing earlier therapeutic intensity once independently validated. At the same time, the authors are explicit about key limitations that temper immediate clinical deployment, including the absence of an equivalent external cohort for direct replication and the potential confounding introduced by historical differences in DMT availability (with many severe cases diagnosed earlier, before modern DMT access), even though analyses attempted to adjust for therapy exposure. The work therefore sets a clear agenda for the field: larger, similarly deep longitudinal cohorts, harmonized severity definitions, rigorous external validation of predictive models, and functional follow-up to connect implicated loci (often intronic/tagging) to causal mechanisms in cerebellar, synaptic, mitochondrial, and oligodendroglial pathways. Source article:
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:
Jokubaitis, V. G., Campagna, M. P., Ibrahim, O., Stankovich, J., Kleinova, P., Matesanz, F., ... & Butzkueven, H. (2023). Not all roads lead to the immune system: the genetic basis of multiple sclerosis severity. Brain, 146(6), 2316-2331.
