Introduction
Multiple sclerosis (MS) is a chronic inflammatory demyelinating disease of the central nervous system that arises from a complex interplay of genetic and environmental factors.1–3 Clinically, MS typically follows a relapsing-remitting course, and patients may experience diverse impairments in sensory, motor, visual, and cognitive functions. Although environmental exposures contribute to disease risk,4–9 genetic susceptibility plays a substantial role, with twin studies estimating heritability at approximately 25%–30% and genome-wide association studies identifying hundreds of susceptibility variants.10 Among these, the HLA-DRB1*15:01 locus is the strongest known risk allele and confers an approximately 3-fold increase in MS risk in carriers of at least 1 copy.2 Because most common variants individually exert only modest effects, their cumulative impact is commonly summarized using a polygenic risk score (PRS).11 In the UK Biobank (UKB), individuals in the top 10% of MS-PRS distribution have been reported to have more than 5-fold greater risk of developing MS than those at average risk.12 Similarly, in the Dutch 1966 birth cohort, individuals in the top 10th percentile of MS-PRS had a significantly higher lifetime risk of MS onset compared to those in the lowest 30th percentile,13 whereas MS-PRS was not significantly associated with age at symptom onset or time to secondary progressive disease.14 Together, these findings suggest that MS-PRS primarily reflects disease susceptibility rather than clinical course.
In addition to physical disability, cognitive impairment is a common and clinically important feature of MS.13 Cognitive dysfunction can occur across the disease spectrum, from clinically isolated syndrome (CIS) to secondary progressive MS, and most consistently affects information processing speed and memory, whereas basic language, semantic memory, and attention are relatively less affected.15–17 These deficits are increasingly recognized as major determinants of quality of life and functional outcome in patients with MS.
Due to its sensitivity to MS-related abnormalities, MRI is central to the diagnosis and monitoring of MS, but conventional MRI is relatively insensitive to diffuse microstructural abnormalities in normal-appearing white matter (WM). Diffusion tensor imaging (DTI) and neurite orientation dispersion and density imaging (NODDI) provide quantitative markers of such changes. In MS, WM lesions typically exhibit elevated mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AD), as well as reduced fractional anisotropy (FA).18 Elevated MD in the corpus callosum (CC) has also been reported in CIS, suggesting that diffusion MRI can detect WM abnormalities at an early stage of disease.19,20 Besides, cognitively impaired patients with MS show decreased FA and increased MD and RD in multiple fiber pathways, including the CC, cingulum bundle, longitudinal fasciculi, corticospinal tract, fornix, and uncinate fasciculus, supporting a link between WM microstructural damage and cognitive dysfunction.21
Despite these observations, it remains unclear whether MS genetic susceptibility is associated with subtle WM microstructural variation and cognitive performance in neurologically healthy individuals without MS. Clarifying this question may help identify subclinical imaging and cognitive correlates of MS genetic risk before clinical disease onset. Therefore, in this study, we used a large UKB cohort of neurologically healthy individuals to investigate the associations among MS-PRS, diffusion MRI metrics, and cognitive function.
Materials and Methods
Study participants
Data were obtained from the UKB, a large prospective cohort study that includes phenotypic, genotypic, and neuroimaging data from more than 500000 participants recruited across the United Kingdom between 2006 and 2011.22 The UKB study has received ethical approval from the Research Ethics Committee (reference 11/NW/0382), and the present study was conducted under UKB application 135200.23 Among UKB participants, we included 35952 neurologically healthy individuals aged 45–83 years who had available data on cognitive function tests, diffusion MRI, and standard MS-PRS.11 Diffusion MRI indices and MS-PRS were downloaded from the data provided by UKB and used in this study. Participants with a history of demyelinating disease, including MS, neurodegenerative disease, dementia, brain tumor, head trauma, or cerebrovascular disease, as identified using ICD-10 codes, were excluded. Figure 1 shows the patient inclusion/exclusion flowchart.
MS-PRS calculation
The MS-PRS used in this study were generated by UKB (field ID: 26254) based on a fixed-effect inverse variance-weighted meta-analysis of 2 large-scale, publicly available Genome Wide Association Study (GWAS) datasets that excluded any UKB participants from the analysis.11 The combined GWAS data included 10820 MS cases and 234517 controls,11 with no overlapping samples between the 2 source datasets. Variants included in the PRS model were filtered for quality and consistency across studies, and PRS algorithms were built from trait-specific meta-analyses using a Bayesian approach, where appropriate combining data across multiple ancestries and related traits. The individual PRS values were calculated as the sum over all variants of the posterior effect size multiplied by the individual’s allele dosage.11
MRI acquisition and processing
MRI data was acquired on a 3T Siemens Skyra system equipped with a standard 32-channel receive head coil as part of the UKB imaging protocol.24 Acquisition parameters for 3D T1-weighted imaging and diffusion-weighted imaging are summarized in Table 1. For details on preprocessing of diffusion MRI data and DTI/NODDI analysis, please refer to the previous paper by Alfaro-Almagro et al.24 After preprocessing, mean values of FA, MD, AD, RD, intracellular volume fraction (ICVF), isotropic volume fraction (ISOVF), and orientation dispersion index (OD) were extracted for each participant from 48 WM tracts defined by the Johns Hopkins University ICBM-DTI-81 atlas.25 Intracranial volume (ICV) was calculated from the 3D T1-weighted images using FreeSurfer version 6.0. (http://surfer.nmr.mgh.harvard.edu/).
Further details of the UKB imaging pipeline are available at: https://www.ukbiobank.ac.uk/enable-your-research/about-our-data/imaging-data.
Statistical analysis
All statistical analyses were performed using R software ver.4.2.2. general linear models (GLMs) were employed to evaluate the associations between MS-PRS and 11 cognitive measures derived from 9 cognitive tests. The 9 cognitive tests were Reaction Time (RT), Trail Making Test (TMT), Numeric Memory Test, Prospective Memory, Symbol Digit Substitution, Paired Associate Learning (PAL), Fluid Intelligence, Matrix Pattern Completion, and Picture Vocabulary (PV). The TMT yielded 3 measures for analysis: TMT-A, TMT-B, and TMT-B minus A.
Lower scores indicate better performance for RT, TMT-A, TMT-B, and TMT-B minus A, whereas higher scores indicate better performance for the other cognitive measures.
For each cognitive measure, the following model was fitted:
cognitive score = β0 + β1 × MS – PRS + β2 × age + β3 × age2 + β4 × sex + β5 – 14 × genetic principal components (PC1 – PC10) β15 × ICV + ε.
We first examined the associations between MS-PRS and all 11 cognitive measures in the full sample. All variables, including MS-PRS, cognitive measures, and covariates, were standardized using Z-score transformation (mean = 0, standard deviation = 1). The same covariates were included in all cognitive models.
A 2-sided FDR-corrected P < 0.05 was considered statistically significant. Next, participants were divided into 4 groups based on MS-PRS quartiles, and the order from highest to lowest MS-PRS group was Q4 > Q3 > Q2 > Q1.26 We extracted the highest MS-PRS (Q4) and the lowest MS-PRS (Q1) groups to re-examine associations between MS-PRS and cognitive performance within these subgroups.26 These subgroup analyses were conducted as post-hoc exploratory analyses following the absence of significant associations in the overall analysis. To avoid arbitrary selection of cutoff values, participants were stratified by MS-PRS quartiles, a commonly used approach in PRS studies.26,27 Because the 11 cognitive measures were derived from 9 cognitive tests, with TMT-A, TMT-B, and TMT-B minus A representing related measures from a single test, Benjamini-Hochberg false discovery rate (FDR) correction was applied across the 9 cognitive tests.
Furthermore, in the subgroup analysis based on MS-PRS quartiles, FDR correction was applied to these cognitive tests, similar to the full sample. However, because multiplicity was not controlled across the entire analytical framework, the risk of false-positive findings cannot be fully excluded. Therefore, this subgroup analysis should be regarded not as a confirmatory analysis, but as a hypothesis-generating exploratory analysis aimed at identifying potential associations in specific high-risk groups.
Additionally, to examine whether the effects of MS-PRS on cognitive functions were modulated by age or sex, we tested the interaction terms (MS-PRS × age and MS-PRS × sex) within the regression models in all participants and the highest (Q4) or lowest (Q1) MS-PRS group.
Furthermore, we conducted a follow-up analysis of cognitive function indicators that showed a significant association with MS-PRS in the highest MS-PRS group (Q4). Since only RT met this criterion, subsequent image analysis focused on WM regions associated with RT in previous reports.28–30 Specifically, the superior longitudinal fasciculus, superior corona radiata, external capsule, posterior limb of the internal capsule, posterior thalamic radiation, and the genu, body, and splenium of the CC, which are associated with RT, were selected based on previous reports.28–30 Using GLMs adjusted for the same covariates, we investigated the associations between MS-PRS and DTI/NODDI metrics in these regions within the highest MS-PRS group (Q4). Overall, because the 56 statistical tests in this study addressed independent research questions and hypotheses, FDR correction was applied separately within each conceptual unit of analysis rather than across all tests.
A 2-sided FDR-corrected P < 0.05 was considered statistically significant.
For brain regions showing significant associations in both the cognitive and imaging analyses, we performed mediation analyses with the corresponding MRI metrics as the mediator, MS-PRS as the independent variable, and cognitive performance as the dependent variable. All mediation models included the same covariates as the primary GLMs. We estimated the total effect of MS-PRS on cognition, the direct effect (controlling for the mediator), and the indirect effect via the MRI metric.
Results
In this study, a total of 35952 UKB participants were included. However, due to missing data in imaging or cognitive measures, the final analytic sample size varied across analyses. Demographic and cognitive characteristics stratified by MS-PRS quartiles are summarized in Table 2.
While age, sex, ICV, and 9 cognitive test scores were broadly comparable across groups, significant differences were observed in MS-PRS, age, ICV, and selected cognitive measures such as PAL (Table 2).
In the full sample, MS-PRS was not significantly associated with any cognitive measure after FDR correction (Table 3). The smallest nominal association was observed for PAL (β = −0.012, SE = 0.006, P = 0.038), but this association did not remain significant after FDR correction (FDR-corrected P = 0.42). RT was also not significantly associated with MS-PRS in the full sample (β = 0.006, SE = 0.005, P = 0.21, FDR-corrected P = 0.73).
Participants were then stratified by MS-PRS quartiles for subgroup analyses. In the highest MS-PRS (Q4) group, higher MS-PRS was significantly associated with longer RT (β = 0.048, FDR-corrected P = 0.047; Fig. 2).
In contrast, no other cognitive measure met the threshold for significance after FDR correction. No significant association was observed between cognitive measures and MS-PRS in the lowest MS-PRS group (Q1).
In an additional analysis of interaction terms to examine whether the effect of MS-PRS on the dependent variable is modified by age and sex, neither the MS-PRS × age nor the MS-PRS × sex interaction was significant in full sample or the highest MS-PRS group (Q4) (Supplementary Tables 1, 2). In contrast, a significant association was found only between PV and MS-PRS × age in the lowest MS-PRS group (Q1), even after FDR adjustment (β = 0.107, FDR-corrected P = 0.024, Supplementary Table 3). However, this result was detected in isolation within a post-hoc exploratory analysis that examined numerous cognitive indicators, and the possibility of false positives due to multiple comparisons cannot be completely ruled out.
Furthermore, because RT was the only cognitive measure significantly associated with MS-PRS in the highest MS-PRS group (Q4), follow-up diffusion MRI analyses were performed in WM regions previously associated with RT.28–30 Of the DTI/NODDI indices examined in the 8 selected brain regions, only MS-PRS and MD in the splenium of the CC showed a significant association (β = 0.021, FDR-corrected P = 0.024; Fig. 3).
Mediation analysis in the highest MS-PRS group (Q4) showed that MD in the splenium of the CC partially mediated the association between MS-PRS and RT. The total effect of MS-PRS on RT was 0.048 (P = 0.004), and the direct effect remained significant after adjustment for the mediator (c′ = 0.047, P = 0.006). The estimated indirect effect was 0.0014 (95% confidence interval [CI], 0.0002–0.0032), corresponding to a mediation effect of 3.0% (Fig.
4).
Discussion
The main finding of this study is that, in a large general population cohort of about 35000 individuals (full sample), genetic susceptibility to MS did not show an overall association with cognitive function after multiple-comparison correction. This result suggests that, at the level of the general population, the genetic risk of MS in question does not have a direct and widespread impact on macro-level cognitive function. On the other hand, a weak association between MS-PRS and RT mediated by the MD of the splenium of CC was observed only in the top 25% of individuals using MS-PRS in this study.
These results do not negate the null effect in the primary analysis but should be interpreted as exploratory findings suggesting latent and local susceptibility that only becomes apparent when considering populations with high genetic risk of MS. Crucially, the subtle alterations observed in our sample should be conceptualized not as a clinical biomarker for detecting “preclinical MS,” but rather as a neural endophenotype reflecting the subtle, continuous impact of genetic susceptibility on brain morphology within the general population. It has been previously reported that the GWAS summary statistics used to construct the MS-PRS did not include samples from the UKB, ensuring complete independence between the datasets.11 Therefore, any potential bias or overfitting arising from sample overlap is avoided in this study.
In this study, the association between higher MS-PRS and longer RT in the highest MS-PRS group is consistent with the established vulnerability of information processing speed in MS. RT is a clinically relevant index of processing speed and attentional efficiency, both of which are frequently impaired in MS. Previous studies have shown that poorer performance on processing-speed measures is closely associated with WM disruption in structures such as the CC and thalamic pathways.29 Akaike et al. showed that FA of the body of the CC and cingulate gyrus, normalized volumes of the brain parenchyma, cortex, thalamus, putamen, globus pallidus, caudate, and normalized lesion volume had significant correlations with neuropsychological tests in patients with MS.30 Then, Akaike et al. suggest that cognitive dysfunction is confined to impaired information processing, resulting in secondary dysfunction in reasoning, working memory, memory, and other cognitive domains in MS.30 Our findings extend this literature by suggesting that similar, albeit much subtler, associations may already be detectable in neurologically healthy individuals with a high genetic burden for MS.
This interpretation is also broadly consistent with a recent study showing that higher polygenic susceptibility to MS was associated with poorer working memory performance in low-performing healthy young adults.31 At the same time, this finding should be interpreted cautiously, because the association was observed in subgroup analyses rather than in the full sample.
A second notable finding was the association between higher MS-PRS and higher MD in the splenium of the CC among healthy individuals in the highest MS-PRS group. However, the significant association between RT and WM microstructure in the top 25% group of MS-PRS in this study, and the fact that it was limited to the splenium of the CC rather than the general frontal lobe network (the genu and body of the CC), requires careful interpretation. Generally, the association between processing speed and executive function is often emphasized as being related to the frontal lobe WM,32 but the splenium of the CC is a group of fibers that mainly connect the left and right occipital and parietal lobes and plays a central role in the initial processing of visual information and the interhemispheric integration of visuospatial attention.33,34 The task used in this study strongly demanded reaction speed, suggesting that the efficiency of bottom-up visual-motor mapping by the posterior WM network, rather than top-down control by the frontal lobe, may have determined individual differences in RT.
Indeed, Madden et al. (2004) reported that a decrease in processing speed involving visual attention and delayed RT specifically correlated with the structural integrity of the splenium of the CC, and the results of this study are consistent with these findings.35 Nevertheless, the present findings should be interpreted cautiously, and further validation is warranted. Furthermore, elevated MD is a nonspecific finding and may reflect a variety of microstructural processes, including age-related change, vascular-related change, altered extracellular water content, or other subtle tissue alterations.
The lack of significant differences in RD and FA suggests that the observed MD finding is not accompanied by the diffusion pattern more typically associated with overt demyelination or marked axonal injury. Furthermore, the lack of a significant association between MS-PRS and ISOVF suggests that an increase in extracellular fluid is unlikely to account for the observed changes in MD. Therefore, the association between MS-PRS and MD in the splenium of the CC among individuals with highest MS-PRS should not be interpreted as evidence of irreversible WM destruction.
Rather, it may reflect subtle and nonspecific microstructural variation that cannot be captured by FA, RD, and ISOVF alone. However, the pathological basis of this finding remains uncertain and should be interpreted cautiously.
This study showed that MD in the splenium of the CC partially mediated the association between MS-PRS and RT in the highest MS-PRS group. This finding suggests that WM microstructural change may account for a small proportion of the relationship between MS genetic susceptibility and cognitive function. However, the mediating effect observed in this study (3.0%) is unlikely to have immediate clinical relevance and should instead be interpreted as a modest population-level association.
In a previous study, de Mol et al. found no statistically significant association between high MS-PRS and total brain volume, white or gray matter volume, or cognitive function in a large cohort of healthy children.36 However, in a secondary analysis within the same study, children with particularly high MS-PRS showed associations between WM microstructural variation and specific cognitive measures.36 These findings may suggest that higher MS-PRS is associated with subtle variation in brain structure and cognition even in apparently healthy individuals. In addition, Zou et al. reported that NODDI metrics partially mediated the relationship between Alzheimer’s disease PRS and cognitive function in healthy UKB participants.37 Taken together, these findings support the possibility that disease-related polygenic burden may be associated with subtle brain phenotypes before overt clinical disease, although the effect sizes appear to be very small.
Several limitations of this study should be acknowledged. First, the cross-sectional design precludes conclusions regarding temporal or causal relationships among MS-PRS, diffusion MRI measures, and cognitive performance. Although individuals with higher MS-PRS showed subtle differences in imaging metrics and cognitive performance, it remains unclear whether these changes are associated with a higher likelihood of developing clinically manifest MS in the future.
Additionally, since this study utilized a cross-sectional design, caution is required when interpreting the causal pathways in the mediation analysis. Because the independent variable, mediator, and dependent variable were measured simultaneously, we cannot establish temporal precedence or rule out the possibility of reverse causality. Therefore, the observed mediating effects should be interpreted as statistical associations rather than definitive causal relationships.
Future longitudinal studies are needed to confirm these causal directions. Second, while brain imaging data for the UKB was collected at multiple sites, the same Siemens 3T MAGNETOM Skyra scanners of the same model, 32-channel head coils, and imaging protocols are used at all facilities to minimize inter-site differences.38 Furthermore, in the official UKB-dedicated pipeline, the same preprocessing methods, such as eddy and topup corrections for eddy currents and magnetic field inhomogeneities, were performed to ensure data quality before calculating DTI/NODDI indices.38 However, many studies have pointed out that inter-institutional differences exist even when MRI equipment and protocols are standardized.39 Ideally, applying harmonization would increase the likelihood of statistically removing differences in meaning and variance between institutions, but this study did not implement harmonization. Harmonization is recommended for future studies using UKB.
Third, diagnosing the observed alterations as an indicator of “preclinical MS” faces epidemiological limitations. Based on previous literature and meta-analyses of incidental MRI findings in healthy populations, the prevalence of radiologically isolated syndrome—the asymptomatic precursor to clinical MS—is estimated to be between 0.1% and 0.5% in the general population. Applying this rate to our cohort of approximately 35000 individuals, the estimated number of potential preclinical MS cases would be at most 35 to 175 individuals, accounting for less than 0.5% of the entire sample.
Given this very small proportion, the potential impact of undiagnosed preclinical cases on our statistical outcomes is tightly constrained and unlikely to confound the main findings. Therefore, rather than being driven by a small number of undiagnosed preclinical MS cases, our results reinforce the paradigm that genetic susceptibility may modulate brain WM wiring and related cognitive phenotypes (as an endophenotype) across the broader healthy population. Fourth, although this study included a large sample and controlled for multiple covariates, residual confounding by unmeasured lifestyle, environmental, vascular, or epigenetic factors remains possible.
In addition, educational years and vascular risk-related factors, including hypertension, diabetes, smoking history, and WM hyperintensity (WMH) burden, were not included as covariates in the primary analyses. These factors are known to influence WM microstructure, particularly in the splenium of the CC, as well as cognitive function, and may contribute to age- and vascular-related brain changes.40 In the present study, fluid-attenuated inversion recovery (FLAIR) – derived WMH measures were not available for all participants included in the diffusion MRI analyses, and restricting the sample to participants with complete WMH data would have reduced the sample size and statistical power. Our primary aim was to characterize the association between MS polygenic risk and WM microstructure without additionally conditioning on variables that might partly reflect downstream vascular or neurodegenerative burden.
At the same time, because some of these factors may act as confounders whereas others may also lie along pathways related to brain microstructural changes, the present cross-sectional data do not allow a clear distinction between confounding and mediation. Therefore, these variables were not included in the primary models. Nevertheless, residual confounding by educational attainment, vascular burden, and WMH cannot be excluded, and the observed findings, particularly the MD finding in the splenium of the CC, should be interpreted with caution.
Future studies incorporating these factors will be important to clarify the specificity and clinical significance of the present findings. Fifth, no significant association was found in the full-sample analysis after FDR correction, whereas a significant finding emerged only in a specific test within the post-hoc subgroup analyses. Although FDR correction was also applied within the subgroup analyses, the increased multiplicity across the overall study means that false-positive findings cannot be entirely ruled out.
Therefore, the association observed in the highest-quartile group should not be interpreted as conclusive, but rather as a hypothesis-generating finding that requires validation in future independent studies. Finally, diffusion MRI metrics such as MD are sensitive but nonspecific markers of tissue microstructure and cannot definitively distinguish among demyelination, axonal damage, and other biological processes. In addition, because the principal positive findings emerged in the highest-risk subgroup rather than in the full sample, independent replication and longitudinal validation will be essential.
Conclusion
Among neurologically healthy individuals in the highest MS-PRS quartile, higher genetic susceptibility to MS was associated with slower RT and higher MD in the splenium of the CC. MD in the splenium showed a small partial mediation effect on this association. These findings support the possibility that subtle WM microstructural variation is linked to cognitive performance in individuals at elevated genetic risk for MS, but further longitudinal and mechanistic studies are needed to clarify the biological significance of these associations.