A Causal Map of MS and the Cortex
Imaging studies have reported cortical changes in multiple sclerosis (MS) for years, but they disagree with one another and share a structural weakness: an observational scan cannot tell whether the disease reshaped the cortex or whether the two simply travel together. Sun and colleagues address that directly, and by their account this is the Mendelian randomization study of the causal relationship between MS and brain structure. They take MS as the exposure from the International Multiple Sclerosis Genetics Consortium, 14,498 cases against 24,091 controls, and cortical structure as the outcome from the ENIGMA Consortium, 51,665 individuals. The area of the cortical sheet (SA) and its thickness (TH) were tested for the whole brain and for all 34 regions of the Desikan-Killiany atlas, weighted and unweighted, across 138 analyses. What the study delivers is a systematic map and a shortlist of six regions worth pursuing.
Why This Design Answers Something Observation Cannot
The reverse-causation problem here is not hypothetical. Cortical structure varies between people before any disease begins, so an association between MS and a thinner cortex could run either way. Mendelian randomization sidesteps this because genetic variants are fixed at conception and cannot be altered by disease onset or progression. The authors do not simply assert that advantage, they test for it: every surviving estimate passed the Steiger test, which checks that the instrument explains more variance in the exposure than in the outcome and so confirms the direction rather than assuming it. Instruments were filtered to an F-statistic above 10 to guard against weak-instrument bias, and the work follows the STROBE-MR reporting guidelines.
A Systematic Map Rather Than a Targeted Look
Testing every region of the atlas, rather than the handful that earlier imaging work had flagged, is what makes the results interpretable as a pattern. It also produces an informative negative: there was no evidence for a causal effect of MS on global cortical TH or SA. Reported rather than omitted, that null does real work, because it says that whatever MS does to the cortex is regional rather than diffuse. A study that had looked only at the frontal and temporal regions of interest would have had no way to establish that, and no way to know whether it had simply found the regions it went looking for.
Six Regions, and Why They Are Anatomically Plausible
Six estimates survived the full analysis: lingual SA (beta 5.7127, P = 0.0342), parahippocampal SA (1.5577, P = 0.0224), rostral middle frontal SA (−9.0301, P = 0.0154), cuneus TH (−0.0020, P = 0.0418), lateral orbitofrontal TH (0.0030, P = 0.0281), and lateral orbitofrontal TH unweighted (0.0029, P = 0.0417). The convergence with clinical phenomenology is the reason to take this list seriously. The parahippocampal gyrus connects to multiple regions of the prefrontal cortex and is tied to spatial processing and episodic memory, and olfactory impairment is reported in 20 to 45% of MS patients. The lingual gyrus and cuneus sit in the occipital lobe and are closely related to vision, which offers a route to some visual impairment in MS beyond optic neuritis. The frontal and orbitofrontal findings line up with observational reports of cortical thinning in frontal and temporal lobes. These are not regions picked to fit a story; they emerged from testing all 34.
The Sensitivity Work Is the Part Worth Copying
Nine regions reached nominal significance in the primary analysis and the authors discarded three of them. Lingual SA unweighted was removed after the MR-Egger intercept detected directional pleiotropy (−2.51, se 1.19, P = 0.04). A PhenoScanner search then identified instrument SNPs associated with body mass index, alcohol consumption frequency, schizophrenia, Parkinson's disease, total cholesterol and self-reported hypertension; the authors removed those variants, re-estimated, and dropped postcentral SA and superior temporal TH when the estimates no longer held. Leave-one-out analysis confirmed that no surviving estimate depended on a single SNP, and MR-PRESSO and Cochran's Q were applied throughout. Discarding a third of your own positive findings before publishing them is the behaviour that makes the remainder worth attention.
How to Read the Numbers, and What They Set Up
The authors set a demanding bar for themselves, correcting for 138 comparisons to give a threshold of P < 0.05/138 = 0.0004. None of the six reached it; the smallest is 0.0154. This is why the abstract describes the findings as suggestive evidence, and that is the right word for them. The magnitudes are correspondingly modest: a thickness effect of 0.0030 mm sits against a cortex of roughly 2.5 mm, about one part in a thousand. The directions also vary, with lingual and parahippocampal SA and lateral orbitofrontal TH raised while rostral middle frontal SA and cuneus TH are lowered; the authors call the increases unexpected and offer compensatory cortical reorganisation or brain edema as candidate explanations, marking both as speculation. Read together, these define the follow-up rather than undermining the work. A study powered for these six regions specifically would carry a fraction of the multiplicity burden and could clear a far easier threshold, and that study is now possible to design because this one narrowed the field from 34 regions to 6.
The Agenda the Paper Sets
Enrichment analysis of the causal loci points to leukocyte proliferation, T cell activation, JAK-STAT signalling, NF-κB signalling and cytokine-cytokine receptor interaction, with genes including ELMO1, SOCS1, STAT3, IL7R, IL2RA and TNFRSF1A. The authors draw two useful consequences: that the current emphasis on anti-inflammatory disease-modifying therapy is well founded, and that Janus kinase inhibitors, already established in rheumatic disease, are worth evaluating for cortical outcomes in MS. One refinement would sharpen that inference considerably. Because the gene set derives from the MS instrument itself, and MS risk variants are known to cluster near immune genes, the enrichment is consistent with immune mediation without yet separating it from MS susceptibility in general; formal mediation analysis or cell-type-specific expression colocalization would distinguish the two. Alongside that sit the extensions the authors name themselves: non-European cohorts, and individual-level data that would allow relapsing and progressive disease to be analysed separately, which summary statistics cannot support. What the paper contributes is a prioritized, systematically derived shortlist where the field previously had an inconsistent observational literature, and a method for testing it.
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:
Sun, D., Wang, R., Du, Q., Zhang, Y., Chen, H., Shi, Z., Wang, X., & Zhou, H. (2024). Causal relationship between multiple sclerosis and cortical structure: a Mendelian randomization study. Journal of Translational Medicine, 22, 83. https://doi.org/10.1186/s12967-024-04892-7
