A STAT3 Variant Marks White Matter Before MS Begins

Zeng, Khan and colleagues report a multi-ancestry genome-wide association study of 20,831 people with multiple sclerosis (MS) and 729,220 controls, yielding 236 susceptibility variants outside the major histocompatibility complex, four novel loci, and 76 candidate causal genes. The cell-type finding is that inhibitory neurons, not microglia, harbour the most MS-associated expression effects in the brain. The claim that makes this version worth reading, though, runs further than cell type. The STAT3 risk variant turns out to be associated with cognition and white matter integrity in people who do not have MS, and with faster neurofilament rise in people who do. From that the authors propose that MS susceptibility carries a second component alongside the propensity for autoimmunity: how well the central nervous system withstands an inflammatory attack once it comes.

The Genetics, and a Convergence Across Ancestries
The European meta-analysis covers 19,865 cases and 623,043 controls, producing 5,041 non-MHC SNPs past genome-wide significance and 236 independent variants after clumping: 198 already known plus 38 new, falling in four previously unreported loci. The extended MHC (chr6: 25,383,722–33,368,421, GRCh37) was set aside by design, since the question was what happens outside it. MS heritability now comes to 23.2%, up from the consortium's earlier 19.2%. The multi-ancestry arm stays modest: 614 African-American cases against 62,044 controls gave one locus, rs76911648 at an odds ratio of 3.169, whose minor allele frequency is 0.035 there against 0.013 in Europeans; 352 Admixed American cases gave two. The authors say all three need confirmation and note none reaches even nominal significance in Europeans. One detail is worth pausing on: the nearest gene to the African-American locus is SMARCA2, which epigenetically activates STAT3 signalling, and STAT3 is a target gene in the European analysis. Effect sizes correlated across ancestries at r = 0.33 for Europeans against African-Americans and r = 0.42 against Admixed Americans.

Inhibitory Neurons, With the Reproduction Numbers
Colocalization across six brain and eighteen blood cell types put 15 signals in inhibitory neurons against 6 in microglia, with seven unique to inhibitory neurons. That ordering matters because excitatory neurons are the most abundant cortical cell type with the largest transcriptome and therefore the most detectable eQTLs, yet they came second. In blood, naive CD4 T cells led as expected, with NR1D1 and MMEL1 specific to them; the glia contributed KCTD13 and RRAS2 in astrocytes and PHGDH and SYNGR1 in oligodendrocytes. The STAT3 effect held up in three separate single-nucleus datasets, at 424 samples (β = −0.473, P = 2.80×10⁻¹⁷), 148 samples (β = −0.408, P = 6.45×10⁻⁵) and 214 samples (β = −0.299, P = 1.82×10⁻³), and 33 of the 46 colocalized brain cis-eQTLs reproduced in total. Summary-based Mendelian randomization supported a causal relationship between STAT3 expression in inhibitory neurons and MS risk.

The New Result: STAT3 Outside MS
This is the section the published version adds, and it changes what the paper is about. The rs1026916-A risk allele sits in linkage disequilibrium (r² = 0.70) with rs1053004-A, which is associated with reduced cognitive performance and lower neurite density index in white matter among 33,224 UK Biobank participants with MRI data. Those are people without MS. Then, in 3,480 people with MS of European ancestry, the same risk allele tracked with faster accumulation of serum neurofilament light chain, a marker of neuroaxonal injury. In a linear mixed model each A allele was associated with a higher rate (β = 0.05, 95% CI 0.004–0.10, P = 0.03), and comparing AA against GG genotypes the annual change differed by 0.11 (95% CI 0.01–0.21, P = 0.03). Follow-up averaged 1.4 years with 2.7 samples per person. One variant therefore marks white matter structure in the general population and the rate of neuronal damage once the disease is present.

What That Licenses, and What It Does Not
The proposed reframing is that MS susceptibility has two parts: a propensity toward autoimmunity realised in the periphery, and a CNS vulnerability that shapes how the brain handles the resulting inflammatory signals. The authors handle STAT3 itself with appropriate care, writing that its specific role in inhibitory neurons in MS remains unclear and that it may act both as a vulnerability factor influencing brain development and as an early modulator of neuroinflammatory responses. The effect sizes deserve equal attention. Both neurofilament associations sit at P = 0.03, with confidence intervals whose lower bounds are 0.004 and 0.01, in a single cohort followed for an average of 1.4 years. The cognition and neurite density link comes through a proxy variant in linkage disequilibrium rather than the MS variant itself. These are the observations that motivate the reframing; they do not yet establish it.

The Polygenic Score, and the Second AUC
The score was built with PRS-CSx, trained on consortium, UK Biobank and All of Us data, and tested in held-out eMERGE-III participants. In Europeans it reached an odds ratio of 1.79 per standard deviation (1.60–1.99, P = 3.27×10⁻²⁵), with the top 1% carrying 7.10 times the risk of the rest (4.22–11.90). African-American participants gave 1.28 (1.08–1.52) and Admixed American 1.74 (1.32–2.31). The table reports two areas under the curve for each group, and the pair is the informative part: 0.7324 with covariates against 0.6559 crude in Europeans, 0.7442 against 0.5379 in African-Americans, and 0.7636 against 0.5837 in Admixed Americans. The adjusted numbers look uniformly strong because they carry age, sex and ancestry; the score by itself is moderate in Europeans and close to a coin flip in the African-American sample. A phenome-wide scan returned only MS (OR 2.05), other demyelinating diseases of the central nervous system (1.79) and functional disorders of the bladder (1.21), the last a recognised MS symptom. Against MRI in 145 patients, the score without MHC associated with lower white matter volume (P = 3.88×10⁻⁴) while the version including MHC did not (P = 0.12).

The Gap They Name, and an Aside Worth Keeping
Most prioritized variants do not sit in known open chromatin regions of the cell types their expression analysis points to. The authors attribute this partly to sparse data, noting only 92 samples in the single-nucleus chromatin accessibility dataset, and they set the gap in context rather than explaining it away: only 21.0% of rheumatoid arthritis susceptibility loci colocalize with an eQTL, and across autoimmune traits the average is 38.1%. By that standard the unexplained fraction here is ordinary. One locus does converge completely, rs3923387 near PLEC, tying MS susceptibility to microglial chromatin accessibility (PP.H4 = 0.84) and to PLEC expression in the same cell type (PP.H4 = 0.93). A methodological note in the discussion is useful well beyond this paper: their large excess of controls improves effect-size estimation and downstream analyses, but for more prevalent autoimmune disorders, adding controls past a threshold yields diminishing returns, while for less common diseases with few cases, maximizing well-matched controls substantially improves the reliability of association signals.

Disclaimer: This blog post is based on the cited 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.

Reference:
Zeng, L., Khan, A., Fitzgerald, K. C., Lama, T., Chen, J., Li, R., International Multiple Sclerosis Genetics Consortium, Tsai, E. A., Yu, K., Chitnis, T., Weiner, H. L., Le Grand, Q., Debette, S., Wang, G., Fujita, M., Calabresi, P. A., Zipp, F., Taga, M., Kiryluk, K., & De Jager, P. L. (2026). Genome-wide association analyses highlight the neuronal contribution to multiple sclerosis susceptibility. Nature Genetics, 58(9), 2177–2191. https://doi.org/10.1038/s41588-026-02731-7