Where MS Risk Genes Act Inside the Brain

Almost everything genome-wide work has found for multiple sclerosis (MS) susceptibility points at immune cells. That is a solid result and it leaves an obvious question open, because the disability comes from the brain. If the risk variants act on lymphocytes, what are they doing in neural tissue, and can anyone show it rather than assume it? Three papers from 2026 attempt the demonstration from different angles. One takes a single risk gene shared between MS and schizophrenia and asks where it is expressed and what machinery it belongs to. One instruments plasma proteins genetically and tests them against brain imaging phenotypes, then checks the hits in brain tissue. And one deletes a receptor from astrocytes alone and looks at the hippocampus.

A Risk Gene With Two Jobs
Korkmaz and Everest took KIF21B, which sits in a locus shared between MS and schizophrenia, and asked whether that sharing is biologically plausible rather than statistical coincidence. The approach is entirely computational, using public resources: GTEx for tissue-specific expression, STRING for protein interaction networks, and published association data for the phenome-wide work. Expression came out significantly enriched in both brain and immune tissues relative to other tissue types, which is the first thing a dual-function risk gene would need to show. The protein interaction network then put KIF21B inside microtubule-based transport machinery essential for axonal function, which is a job description from the neural side rather than the immune side. For a gene that reached the MS literature through an immune association scan, that is an unexpected result.

What the Co-expression Pointed At
The other half of the analysis runs the opposite way. Transcriptome-wide co-expression put KIF21B in significant correlation with immune-related genes, including an established MS susceptibility gene, and pathway enrichment on that set returned T cell activation and lymphocyte differentiation. So the same gene reads as axonal transport by protein interaction and as T cell biology by co-expression, depending on which analysis you run. Phenome-wide association then confirmed pleiotropy across multiple autoimmune and psychiatric conditions. The authors are careful about what this is: a case that the gene is biologically plausible as a shared risk factor whose dual neural and immune functions may contribute to susceptibility for both kinds of disease. It is a hypothesis built from public data, and they say so.

From Genetically Instrumented Proteins to Brain Structure
Ayubcha and colleagues built the route differently, with genetics as the instrument rather than the target. In a preprint, they took protein quantitative trait loci from the two largest plasma proteomic resources, the UK Biobank Pharma Proteomics Project and the deCODE dataset, and tested them against genome-wide association results for brain imaging-derived phenotypes in UK Biobank, using Mendelian randomization with colocalization alongside. The design is what makes it a brain study rather than a blood study: the exposure is a circulating protein under genetic control, the outcome is a measured brain phenotype, and the instrument makes the direction of the arrow arguable rather than assumed. They report 120 cis and 20 trans associations between plasma proteins and imaging phenotypes.

The Step That Matters Most
What they did next is the part worth copying. The imaging associations were validated using brain tissue-derived proteomic and transcriptomic datasets, which asks whether a protein linked to a brain phenotype through blood is actually present and expressed in the organ it was linked to. Multivariable Mendelian randomization then identified eleven plasma proteins with significant direct effects on disease risk across Alzheimer's disease, Parkinson's disease, multiple sclerosis, bipolar disorder and schizophrenia, and single-cell expression with pathway enrichment gave cell-type-specific effects for those protein-disease pairs. Four layers in sequence: a genetic instrument, a brain imaging outcome, confirmation in brain tissue, and a cell type. Each layer answers a different objection, which is why the chain is worth more than the 120 associations on their own. The paper is a preprint and has not been through peer review.

Deleting a Receptor From One Cell Type
Itoh and colleagues supply the arm the other two cannot: a genetic manipulation with the brain examined directly. Midlife female mice that had been ovariectomized showed cognitive deficits on the Morris Water Maze with glial activation and synaptic loss in the dorsal hippocampus, and their astrocytes carried raised Enolase 1, a protein central to glucose use, with higher levels correlating with worse performance. Then the subtractive test: deleting oestrogen receptor beta from astrocytes in midlife females whose ovaries were intact reproduced the deleterious effects of ovariectomy. Running it forward, estriol given from 11 to 12 months lowered astrocyte Enolase 1, reduced glial activation and synaptic loss, and improved performance. One receptor, one cell type, one brain region, measured.

The MS Result Underneath It
That work sits on a human MS finding with a tissue endpoint: oral estriol had previously improved cognitive processing speed and reduced cerebral cortex atrophy at treatment month twelve in women with MS. Cortex atrophy is the kind of outcome the first two papers are trying to reach by inference, and here it was measured in patients. The new human component is weaker and the authors say so, being a case series of twenty menopausal women with self-reported cognitive improvements at month twelve, no control arm, and a stated conclusion that the treatment warrants further investigation. Taken together the three show the field's current position plainly: one gene with a plausible neural job, 120 protein-to-imaging links needing brain tissue to be believable, and one receptor deletion where the brain was opened and looked at.

Disclaimer: This blog post is based on the cited studies 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:
Korkmaz, E. D., & Everest, E. (2026). Investigating the Pleiotropic Role of KIF21B in Schizophrenia and Multiple Sclerosis: A Bioinformatics Analysis. Clinical Neuropsychiatry, 23(2), 136–143. https://doi.org/10.36131/cnfioritieditore20260202
Ayubcha, C., Dennis, E., Bhattacharyya, U., John, J., Lam, M., Lencz, T., Ge, T., & Chen, C.-Y. (2026). Characterizing the impact of plasma protein levels on human brain structure and disorders leveraging integrative multi-omics analysis. medRxiv preprint. https://doi.org/10.64898/2026.07.13.26358006
Itoh, N., Itoh, Y., Patil, R., Till, C., Smith, M., Schwartz, J., Davis, B., Oberman, E., Romeroso, B. D., & Voskuhl, R. R. (2026). Targeting hormone treatment for the brain in menopause. Scientific Reports, 16(1). https://doi.org/10.1038/s41598-026-65603-4