A Choroid Plexus Endophenotype for HLA Burden in MS

Multiple sclerosis (MS) susceptibility is highly polygenic and predominantly driven by immune-related loci, with the strongest single contribution coming from the major histocompatibility complex, where HLA-DRB1*15:01 remains the strongest and most consistently confirmed individual risk allele. Outside that region, hundreds of common variants of small effect add up, and the usual way to summarize them is a polygenic risk score. Those scores have a known ceiling in this disease: they track earlier age at onset and higher relapse activity, but they do not track long-term disability or neurodegeneration. Preziosa and colleagues approached that gap the way an imaging-genetics study would, by looking for an intermediate phenotype that sits closer to immune biology than a clinical endpoint does. Their candidate is the choroid plexus, a key component of the blood–cerebrospinal fluid barrier and one of the major meeting points between the peripheral immune system and the central nervous system. The design has two stages: build a normative lifespan model of choroid plexus volume in 461 healthy controls, then test genetic burden against the deviation from that model in 727 patients with MS.

Constructing a Phenotype That a Genetic Test Can Use
Raw choroid plexus volume makes a poor quantitative trait. It grows during healthy aging, and because the structure sits inside the ventricular system its MRI estimate correlates strongly with brain atrophy and ventricular expansion; ventricular expansion also increases the free CSF around it, improving how well it is visualized and segmented and so inflating the apparent enlargement. Any genetic association measured on the raw volume would therefore be open to confounding by whole-brain anatomy. The normative model exists to strip that out, converting each patient's volume into a z-score, a residual phenotype expressing how far they sit from the value expected for their age, sex and brain morphometry. Imaging came from two 3.0 T Philips scanners between July 2005 and October 2024, with gradient system and head coil configuration held stable within each scanner period. Segmentation ran fully automatically through ASCHOPLEX, an artificial intelligence toolbox built from 27 deep neural network configurations, with every mask visually inspected by readers carrying more than 15 years of segmentation quality control experience and manual editing limited to cases strictly necessary. The 461 controls were 54% women, mean age 41.9 years (SD 15.1), spanning 18.1 to 69.8 and covering the full age range of the patient cohort.

Anatomy Explains More of the Trait Than Age Does
A model built on age terms alone picked up nonlinear, age-dependent enlargement but explained only 16% of interindividual variance, so age by itself is a weak predictor of where any given person sits. Adding brain and ventricular volumes raised explained variance to 54%. In the final scanner-adjusted model, normalized brain volume (β = −1.11×10⁻³, p = 0.004), log-transformed normalized lateral ventricle volume (β = 1.74, p < 0.001) and its squared term (β = −1.53, p < 0.001) were each independently associated with plexus volume, and age dropped out entirely once they were included. In controls, plexus volume correlated with age at r = 0.323, with normalized brain volume at r = −0.453, and with log-transformed lateral ventricle volume at r = 0.713. The plexus trajectory itself stayed flat until roughly age 35 and then rose nonlinearly, with annualized growth climbing from 0.24% at 35 to over 0.7% in later decades, set against brain and thalamic volumes declining quadratically from around ages 30 and 35, cortex and caudate declining linearly at about 0.3% a year, and lateral ventricles expanding from about 0.4% a year at 35 to over 3% by 70.

The Deviation Behaves Like a Stable Quantitative Trait
Three properties make the z-score usable as a phenotype. The 727 patients, 477 relapsing remitting, 160 secondary progressive and 90 primary progressive, median EDSS 2.5 and median disease duration 10.5 years, showed an estimated mean z-score of 0.452 (p-FDR < 0.001). First, it held across clinical phenotypes at 0.442, 0.515 and 0.400 with a heterogeneity p of 0.744, so the trait does not fragment along disease subtype. Second, it was uniform across the entire adult age spectrum with no age-dependent variation (p = 0.957), so it is not an aging artifact. Third, a combined control and patient model retained the same predictors with MS status added, no interaction terms survived, and z-scores derived from the two models agreed at r = 0.996. Disease duration did move the trait: it was already raised one year after clinical onset at 0.252 (p-FDR 0.001), climbed to 0.463 by five years with the estimated slope significant at 0, 1 and 5 years and not after, then plateaued through 0.542 at ten years and 0.302 at forty, a plateau the authors stress is not normalization since the values stayed significantly above controls at every duration.

Which Axis the Trait Loads On
Before the genetic analysis, the trait was tested against the clinical and imaging axes it might be measuring. It associated with normalized T2-hyperintense white matter lesion volume (β = 0.012, SE = 0.004, p < 0.001) and with none of the brain, cortical, thalamic, caudate or lateral ventricle volume z-scores, all at p > 0.089. Disability split by severity: nothing across the full EDSS range, but at a threshold of 3.0, higher z-scores tracked higher disability among mildly impaired patients (β = 0.303, SE = 0.093, p = 0.001) and not among those at EDSS 3.0 or above (β = −0.094, SE = 0.052, p = 0.070), with a significant interaction (p < 0.001). The authors read the lesion result as placing the plexus with focal inflammatory activity rather than with progressive neurodegeneration, noting that once shared morphometric dependencies were accounted for, residual plexus variability related more strongly to inflammatory lesion burden than to global or regional brain volume loss. For a genetic test, that matters because it predicts which side of the MS genetic architecture, immune or neurodegenerative, should carry any signal.

Cumulative HLA Burden, Not Any Single Allele
Whole-genome genotyping on a larger MS cohort used multiple Illumina arrays, and two kinds of score were built from loci established by the International MS Genetics Consortium. Three non-HLA polygenic risk scores were computed as progressively inclusive sets of risk variants, capturing different levels of statistical evidence, alongside a separate HLA Genetic Burden score summing 24 MS-associated HLA loci. Only the HLA burden score associated with higher plexus z-scores (standardized β = 0.097, SE = 0.047, p = 0.038), and it was the sole genetic score to associate with any MRI measure of structural brain damage. The three non-HLA scores returned nothing at any inclusion threshold (p ≥ 0.177), neither score type associated with lesion volume or brain volumetrics (p ≥ 0.231), and a sensitivity analysis pairing the HLA burden score with each polygenic score in turn reproduced the same split. At the tails, patients in the highest HLA burden decile carried plexus z-scores 0.478 above those in the lowest (95% CI 0.071–0.885, p = 0.022), adjusted for age, sex, disease duration, EDSS, phenotype, treatment status and the first five eigenvectors from population substructure analysis, while no non-HLA score produced a decile difference. Tested one locus at a time, none of the 24 survived FDR correction, HLA-DRB1*15:01 included. The allelic architecture the authors infer from that pattern is cumulative rather than single-locus: given the linkage disequilibrium structure of the MHC, multiple correlated variants acting on partly shared immune pathways may collectively modulate immune activation at the blood-CSF barrier, and the strongest individual susceptibility allele carries no detectable effect of its own on this trait. The local biology they cite fits antigen presentation: the plexus contains antigen-presenting cells expressing MHC class II molecules, shows increased HLA-DR expression on epithelial and stromal cells, and shows transcriptomic upregulation of HLA-DRB1 and immune activation pathways in patients compared with controls.

What the Genetic Design Cannot Settle
The genetic limitations the authors name are the ones that bound the inference. There were no genotypes in the healthy controls, so they could not test whether the same HLA effects operate outside the MS context, which leaves open whether this is an MS-specific mechanism or a general property of immune genetic burden. The HLA panel covered 24 of the 32 loci the consortium has identified, so the burden score is incomplete, and other genetic, epigenetic and environmental modifiers remain untested, with EBV serostatus unavailable for this cohort despite published links between EBV and plexus inflammatory change. The cross-sectional design precludes causal inference about the temporal dynamics between plexus enlargement, disease onset and progression. On the phenotype side, the 3D T1-weighted sequence may lack the resolution to separate plexus subregions or to distinguish epithelial hypertrophy from stromal expansion and immune cell infiltration, and gives no access to microstructure, vascularity or permeability; gray matter lesions could not be consistently evaluated and so could not enter as covariates; systemic inflammation, metabolic status, hormonal fluctuations and comorbidities went unassessed; and treatment class and duration may themselves influence plexus volumetry. Without histopathologic validation or CSF biomarker data, the enlargement cannot yet be anchored as a direct marker of immune activity, and without a disease-control cohort the deviation cannot be shown to be specific to MS.

Disclaimer: This blog post is based on the cited study 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:
Preziosa, P., Corazzolla, G., Meani, A., Margoni, M., Storelli, L., Pagani, E., Rubin, M., Clarelli, F., Mascia, E., Sorosina, M., Esposito, F., Filippi, M., & Rocca, M. A. (2026). Lifespan modeling of choroid plexus volume in multiple sclerosis and its dynamic associations with clinical, MRI, and HLA susceptibility. Neurology: Neuroimmunology & Neuroinflammation, 13(4), e200593. https://doi.org/10.1212/NXI.0000000000200593