From GWAS Signals to Cell-Specific Mechanisms in Multiple Sclerosis: A Regulatory and Network-Based Framework

Genome-wide association studies (GWAS) have now produced tens of thousands of robust associations for common traits, yet the mechanistic translation of these signals remains difficult because most lead variants are non-coding, in linkage disequilibrium with many proxies, and likely act through context-dependent regulation rather than direct protein alteration. In this Nature Communications study, the International Multiple Sclerosis Genetics Consortium addresses that bottleneck for multiple sclerosis (MS) by reframing GWAS interpretation as a cell-type–specific systems biology problem: instead of asking “which is the closest gene?”, they infer which genes are most plausibly regulated by MS-associated variation in specific immune compartments, then test whether those genes converge onto coherent protein-interaction modules and pathways. The work leverages a large MS GWAS meta-analysis (47,351 cases and 68,284 controls) with more than 200 non-MHC genome-wide associations, explicitly aiming to connect association signals to actionable cellular processes.

Integrating association signals with cell-resolved regulatory annotations
The analytical framework begins by expanding each independently associated non-MHC signal into sets of linked variants at multiple LD thresholds, then intersecting these variant sets with large-scale functional genomics resources. Regulatory annotations were retrieved from RegulomeDB, which aggregates ENCODE and Roadmap Epigenomics Project (REP) evidence across hundreds of cell and tissue contexts; the authors then consolidate these contexts into biologically motivated “buckets” that are central to MS immunopathology—B cells, T cells, and monocytes—plus CNS as a tissue compartment, with lung included as a negative-control tissue not expected to be primary in MS susceptibility. In total, the pipeline incorporates hundreds of thousands of variants (538,826 SNPs) spanning genome-wide significant (GW), “statistically replicated” (SR), and non-replicated (NR) sets, with the main analysis emphasizing r² ≥ 0.5 proxies and sensitivity analyses probing alternative LD and scoring thresholds.

Predicted Regulatory Effect (PRE) as a quantitative SNP-to-gene prioritization layer
A central contribution is the “predicted regulatory effect” (PRE), a gene-level score computed per locus and per cell type that summarizes the regulatory evidence implicating that gene. Rather than treating all annotations equally, the authors weight SNP–feature–gene relationships by the number of supporting experiments (e.g., transcription factor binding sites, promoter/enhancer marks, DNase hypersensitivity, histone modifications), classify features into activating versus repressing classes, normalize by the overall experimental density of each cell type to mitigate ascertainment bias, and sum these “weighted weights” across all SNPs in LD with the lead signal to obtain a signed PRE per gene. This design is intended to elevate genes that are repeatedly connected to the risk haplotype through multiple regulatory observations, while enabling a single locus to nominate different target genes in different cell types—illustrated by loci where immune-cell PRE highlights genes such as TNFRSF14 broadly in immune cells, versus loci where CD40 is preferentially implicated in B cells.

Cell-specific protein-network convergence reveals shared and distinct immune modules
To move from prioritized genes to biological interpretation, the study projects high-PRE genes onto an experimentally derived human protein interactome (15,783 proteins; 455,321 interactions) and extracts cell-specific subnetworks, focusing on the largest connected component and related connectivity metrics. Statistical significance is assessed by comparing observed network properties to 10,000 size-matched random networks, an approach that tests whether MS-implicated genes cluster in protein space beyond chance expectation. The resulting networks are significantly enriched and connected in T cells, B cells, and monocytes, whereas CNS networks are not significantly different from random—an outcome the authors attribute in part to heterogeneity and sparsity in available CNS epigenomic annotations, which can dilute cell-specific regulatory signals when aggregated. Notably, the immune-cell networks share a core module, and functional enrichment points to signaling systems consistent with MS immunobiology (including JAK/STAT, interferon-γ, interleukin, and integrin-related pathways), while also retaining cell-type–exclusive constituents (e.g., CD28 in T-cell networks, ELMO1 in B-cell networks, and MERTK in monocyte/macrophage networks).

Empirical support: individualized PRE aligns with cell-matched transcriptomic variation
A frequent critique of regulatory scoring frameworks is that they may not correspond to measurable molecular phenotypes; the authors directly test this by computing genotype-dependent PRE at the individual level and correlating it with RNA-seq expression in matched primary immune cell populations. Using FACS-sorted CD4+ T cells and CD14+ monocytes from 25 MS patients, they report that PRE–expression correlations are significantly greater than expected under independence and are consistently stronger when PRE and expression derive from the same cell type (e.g., T-cell PRE correlating more strongly with CD4/CD8 T-cell expression than monocyte PRE does). While correlations are partial—highlighting that PRE captures only a component of expression variance and that regulatory maps are incomplete—the results support PRE as a biologically grounded intermediate phenotype rather than a purely heuristic score.

From population signal to patient heterogeneity: intra-individual risk networks
Beyond locus-level interpretation, the study proposes a personalized systems-level representation of genetic burden by constructing intra-individual, cell-specific “risk networks.” Using genotype data from 2,370 MS patients and 412 controls, the authors recompute PRE per person, then re-embed high-PRE genes into the interactome to quantify per-subject network connectivity (again emphasizing largest connected component edges and related metrics). Under this model, cases exhibit more connected risk networks than controls in the main immune cell types, with monocytes showing the strongest differentiation and the greatest average connectivity, followed by T cells and B cells; CNS remains weaker and often non-significant in these comparisons. The paper further illustrates that individuals can occupy extreme percentiles in one compartment but not others (e.g., highly connected B-cell risk network with comparatively average T-cell and monocyte profiles), arguing that MS “polygenic risk” may be more usefully decomposed into cell-resolved mechanistic burden maps than summarized as a single scalar score.

Implications, limitations, and generalizability of the framework
Conceptually, this work advances MS genetics by operationalizing an “omnigenic” perspective: many variants of small effect may influence a tightly connected core of immune pathways, yet the cellular locus of regulation can differ across variants and individuals, potentially contributing to clinical heterogeneity. Practically, the authors argue that individualized, cell-specific genetic burden profiles could inform hypotheses about therapeutic stratification (for example, elevated B-cell regulatory/network burden might support prioritizing B-cell–targeted approaches, whereas elevated T-cell burden might suggest therapies focused on T-cell function or trafficking), while also providing a reusable blueprint for other complex traits with adequate GWAS power and functional genomics coverage. Key limitations are explicitly acknowledged, particularly the CNS compartment: aggregating diverse CNS cell types and anatomical regions due to sparse epigenomic data may underestimate CNS-relevant regulatory architecture, implying that improved cell-resolved CNS regulatory atlases could materially change conclusions about neuroimmune targeting. Overall, the study exemplifies a scalable post-GWAS strategy that connects non-coding associations to cell-specific regulatory hypotheses, interpretable interaction networks, and patient-level mechanistic heterogeneity maps.

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:
International Multiple Sclerosis Genetics Consortium. A systems biology approach uncovers cell-specific gene regulatory effects of genetic associations in multiple sclerosis. Nat Commun 10, 2236 (2019). https://doi.org/10.1038/s41467-019-09773-y