Хроническое низкосортное воспаление, генетическая восприимчивость и риск множественных склерозов: перспективное когортное исследование.

Introduction

Multiple sclerosis (MS) is a chronic inflammatory autoimmune disease that traditionally affects young adults aged 20 to 40, with females accounting for nearly three-quarters of cases ( 1 , 2 ). However, recent evidence points to a growing prevalence of MS among older populations in developed regions, a trend likely attributable to demographic aging, advances in diagnostic capabilities, and the introduction of high-efficacy disease-modifying therapies that improve disease stability and extend patient longevity ( 3 ). The majority of patients initially present with a relapsing-remitting course (RRMS); however, as disease duration increases, incomplete recovery from relapses leads to cumulative disability and eventual transition to secondary progressive MS (SPMS) ( 3 ). This progressive decline is compounded by the accumulation of age-related comorbidities, which further exacerbate motor, cognitive, and psychiatric impairment ( 2 , 4 , 5 ). Given the expanding burden of MS in aging populations, identifying modifiable risk factors relevant to this demographic is of considerable clinical and public health importance.

Current evidence suggests that MS is a multifactorial disease influenced by genetic predisposition and environmental factors ( 1 ). Tissue damage in MS results from a complex and dynamic interplay involving the immune system, glia, and neurons ( 1 ). Findings from the International Multiple Sclerosis Genetics Consortium (IMSGC) have demonstrated that the majority of genetic risk loci for MS implicate immune-pathway genes, underscoring the central role of immune dysregulation in disease pathogenesis ( 6 ). Critically, the aging immune system is characterized by two interrelated phenomena—immunosenescence and chronic low-grade inflammation (LGI)—both of which may act upon this genetic substrate to alter MS risk ( 7 ) ( 8 ). Beyond biological aging, various contemporary lifestyle factors—including obesity, microbiome dysbiosis, dietary changes, and environmental toxin exposure—further promote a state of systemic LGI ( 9 ).

The LGI state, as an overproduction of peripheral pro-inflammatory mediators in the absence of overt clinical symptoms, has garnered growing interest in the neurology field ( 10 – 13 ). Gialluisi et al. demonstrated that a state of systemic LGI had a significant effect on symptoms and mortality in major depression ( 14 ). A meta-analysis revealed that elevated interleukin-6 (IL-6) and C-reactive protein (CRP) levels were associated with increased susceptibility to all-cause dementia ( 15 ), and systemic LGI has been linked to functional impairment in patients with Alzheimer’s disease ( 16 ). With respect to MS specifically, recent studies have reported a correlation between elevated neutrophil-to-lymphocyte ratio (NLR) and both the onset and progression of MS ( 17 , 18 ). However, other studies have yielded inconsistent results ( 19 , 20 ). These discrepancies may arise from ethnic or demographic differences, residual confounding, or methodological limitations—most existing studies are constrained by small sample sizes, retrospective designs, and assessment of inflammatory markers only at the time of diagnosis or hospital admission ( 21 ).

From a mechanistic standpoint, peripheral inflammatory cytokines can cross the blood–brain barrier (BBB), activate microglia and astrocytes, and disrupt synaptic plasticity and neuronal integrity ( 22 ). Consistent with this pathway, elevated inflammatory markers have been associated with reduced subcortical and cortical volumes in population-based studies ( 23 , 24 ), and neuroimaging research in MS patients demonstrates widespread brain atrophy—particularly affecting the thalamus, caudate nucleus, striatum, and cerebellar cortex ( 25 – 27 ). These converging lines of evidence suggest that LGI may contribute to MS-related structural brain changes, yet this hypothesis remains to be fully examined in large population-based cohorts.

Building upon these considerations, we leveraged the UK Biobank resource to conduct a prospective cohort study evaluating the association between LGI and incident MS risk. LGI level was quantified using a composite inflammatory index, the INFLA-score, alongside its individual biomarkers. We further assessed whether genetic susceptibility potentially modulates this relationship and investigated the links between LGI and MS-related brain structural measures. These findings are intended to provide longitudinal epidemiological evidence that may facilitate risk stratification and inform the potential development of targeted intervention strategies for MS, particularly within the context of an aging population.

Methods

Study design and participants

The UK Biobank is a large cohort study that recruited about 500,000 individuals aged 40 to 69 in England, Scotland, and Wales. Baseline data were collected at 22 centers using touchscreen questionnaires, interviews, physical measurements, and genetic assessments, with informed consent, from 2006 to 2010. A linked electronic healthcare record is available for most participants, allowing researchers to analyze data from various sources. The follow-up period extended from the date of the baseline assessment until the earliest date of MS diagnosis, death, or the end of follow-up (December 19, 2022), whichever transpired first. This study was conducted under approval from the North West Multi-center Research Ethics Committee as a Research Tissue Bank approval. (ref. 21/NW/0157). Our study was conducted under Application Number 99001 (Approval date: 2023-01–31 to 2026-01-31).

In our study, we screened 502,369 participants from the UK Biobank. Participants were excluded if they met the following criteria: 1) pre-existing MS at baseline, identified via self-report or hospital inpatient records; 2) residence in rural areas — an exclusion necessitated to minimize potential confounding from urban-rural lifestyle variations highly correlated with LGI and ensure the completeness of medical records; 3) missing data on plasma samples, important covariates, or follow-up; 4) failure to meet genetic quality control standards, including a genotyping call rate <95%, excessive heterozygosity, or a mismatch between genetic and reported sex. Finally, 377,487 participants were included in the main analysis. Furthermore, a total of 32,117 participants were involved in the brain imaging analysis after excluding those with missing or abnormal MRI data.

Low-grade inflammation score

We computed the INFLA-score as an aggregated measure of LGI. The INFLA-score, containing CRP, white blood cell (WBC), platelets (PLT), and NLR, has been reported to be associated with diet and mental health in previous prospective studies ( 28 , 29 ) ( Supplementary Figure 1 ). To compute the INFLA-score, for all four components, laying in the highest deciles (7th to 10th) were assigned values from +1 to +4; while biomarker levels laying in the lowest deciles (1st to 4th) were given values from -4 to -1. The INFLA-score, ranging from -16 to +16, is the sum of these scores, with higher values indicating greater LGI ( 30 ). Scores are categorized into low-level and high-level based on the median INFLA-score. More details are displayed in Supplementary Methods and Supplementary Table 1 .

Definitions of outcomes

The primary outcome in the study is the incidence of MS. We derived MS disease status from the UK Biobank “summary diagnoses” fields (UKB category 2002), which contains summary fields relating to diagnoses made during hospital inpatient admissions. Most data-fields in this category have a corresponding field showing the date each diagnosis code was first recorded in the inpatient data. MS cases were defined as participants with the International Classification of Diseases (ICD) (ICD-9: 340 & ICD-10: G35).

The secondary outcome of the study is the volumes, thickness and area of key brain structures. Brain imaging acquisition commenced in 2014, neuroimaging data were collected on a standard Siemens Skyra 3T scanner with a 32-channel head coil. Quality-controlled T1-weighted neuroimaging data were processed with FreeSurfer, surface templates were utilized to extract imaging derived phenotypes referring to atlas regions’ surface volume. According to previous studies based on MS, we selected six brain regions, including caudate nucleus, putamen, thalamus, lateral orbitofrontal cortex, supramarginal gyrus, and superior temporal gyrus. Meanwhile we calculated mean thickness and area of three cortical structures, including lateral orbitofrontal cortex, supramarginal gyrus, and superior temporal gyrus. MRI variables were adjusted for skull size and converted to Z scores for interpretation and comparison. The imaging data used in this study were collected between 2014 and 2020. More details of the above brain imaging phenotypes are displayed in Supplementary Methods and Supplementary Table 2 .

Polygenic risk score

The procedure for genotyping and imputation in the UK Biobank has been described in detail previously( 31 ). To construct the PRS, we used 110 single nucleotide polymorphisms (SNPs) reported by the IMSGC ( 32 ). However, due to restricted access to individual-level genotyping data, genotyping array and genetic ancestry principal components were not included as covariates in the PRS model. According to the number of risk alleles, each SNP was recorded as 0, 1 or 2 according to the number of risk estimate (β coefficient) on MS. The PRS was calculated using the equation:

, where M is the total number of SNPs, and β j is per allele log odds ratio (OR) for MS associated with SNP j . Then participants were categorized into three groups: low (lowest tertile), intermediate (intermediate tertile), or high (highest tertile) genetic risk. The details of the SNPs and calculation of PRS are displayed in Supplementary Table 3 .

Covariates

Relevant covariates were measured at baseline. Demographic variables included age, sex, index of multiple deprivation (IMD) scores, waist-to-hip ratio (WHR). Healthy lifestyle included never smoking (yes vs no), never drinking (yes vs no), regular physical activity, healthy sleep, and healthy diet. Regular physical activity was defined as at least 150 minutes of moderate activity per week or 75 minutes of vigorous activity per week ( 33 ). Sleep healthy scores were used to assess sleep behaviors ( 34 ), which included 5 aspects: chronotype, sleep duration, insomnia, snoring, and excessive daytime sleepiness. Each sleep factor was coded 1 if meeting the healthy criterion and 0 if not. We categorized it into “healthy sleep pattern” when the score ≥4. Healthy diet was based on consumption of at least 4 of 7 commonly eaten food groups, in accordance with dietary guidelines designed to enhance cardiometabolic health ( 33 ). Chronic disease conditions included hypertension, diabetes, and stroke. More details of variables are shown in Supplementary Table 4 .

Statistical analysis

For differences in baseline characteristics across INFLA-score categories (low and high), we used the chi-squared (χ2) test for the variables. Baseline characteristics of these subsets are presented as the number (percentage).

The incidence density of MS was calculated for the overall cohort, as well as separately for women and men, and across three genetic risk groups (low, moderate, high). We used a Cox proportional hazards model to quantify associations of INFLA-score and its components (CRP, WBC, PLT, NLR) with incidence risk of MS with years of follow-up as the time metric, reporting hazard ratios (HR) and 95% CIs. Schoenfeld residuals were used to test the proportional hazard assumption. Three sets of models were applied. Model 1 did not adjust for any variables. Model 2 adjusted for age, sex, and IMD scores. Model 3 further adjusted for WHR, healthy lifestyle, and chronic disease conditions. Based on the model with full adjustments for potential confounders, subgroup analyses were further conducted by sex (male and female), PRS category (low, moderate, high), age (<65 years and ≥65 years), IMD scores (low and high), regular physical activity (yes and no), healthy sleep pattern (yes and no), hypertension (yes and no) and diabetes (yes and no). The likelihood ratio test between models with and without the interaction term was conducted to estimate the significance of the interaction effects.

Additionally, we used multiple linear regression analyses to estimate the associations of LGI and the inflammation biomarkers with brain structural traits related to MS (including volume of caudate nucleus, putamen, thalamus, lateral orbitofrontal cortex, supramarginal gyrus and superior temporal gyrus, area of lateral orbitofrontal cortex, supramarginal gyrus and superior temporal gyrus, and mean thickness of lateral orbitofrontal cortex, supramarginal gyrus and superior temporal gyrus) in the model with full adjustments (including age, sex, IMD scores, WHR, no smoking, no drinking, regular physical activity, healthy sleep pattern, healthy diet, and chronic health conditions such as hypertension, diabetes, and stroke). The estimates are shown as β s and 95% CIs. As all brain structural measures were standardized via Z-transformation, these β coefficients represent the mean difference in standard deviation (SD) units for participants in the high-level group compared with the low-level reference group. Multicollinearity assumptions were tested by the variance inflation factor (VIF), and the value of all variables was below five. Subgroup analyses were also conducted for brain structural traits. Interaction significance was determined via likelihood ratio tests, with nominal and false-discovery rate (FDR) -corrected P values reported to distinguish robust associations from potential multiple testing artifacts.

Furthermore, we performed a series of sensitivity analyses: 1) We exclude the non-White participants from the cohort, as the non-White participants only constitute approximately 3% of the dataset. 2) We observed the association of the COVID-19 pandemic with mortality ( 35 , 36 ), and moved the follow-up deadline forward to December 31, 2019. 3) We changed the covariate of never drinking (yes vs no) to moderate drinking (yes vs no), and moderate drinking was defined as the average daily intake ≤16g of pure alcohol (2 units of alcohol) for both men and women. 4) To address potential delayed case ascertainment or reverse causation, we conducted lag analyses excluding participants diagnosed with MS within the first two or five years of follow-up.

All statistical analyses were performed with R software, version 4.2.1, we used package ‘survival’ for Cox proportional hazards regression, ‘stats’ for logistic regression and other statistical analyses, and ‘forestmodel’ for the creation of forest plots. Two-sided p <0.05 was considered statistically significant.

Result

Social demographic characteristics of participants

A total of 377,487 participants were included in this study, with a median follow-up of 13.8 years, amounting to 5,081,861 total person-years of observation ( Figure 1 ). During the follow-up period, 432 participants developed MS, with a mean age at diagnosis of 61.4 (± 8.8) years. The overall incidence rate was 8.5 per 100,000 person-years; the sex-specific rates were 5.8 per 100,000 person-years for males and 10.8 per 100,000 person-years for females ( Supplementary Table 5 ). Table 1 presents the baseline characteristics of the participants in this study by LGI. We found that participants with the higher level of LGI tended to be more likely observed among people aged more than 65 years, females, and those with high index of IMD scores, poor WHR. Participants with low-level LGI were more likely to have regular physical activity, healthy sleep, and healthy diet.

Study flowchart. (A) Population flowchart. (B) The calculation of immune biomarkers and an aggregated inflammation score (INFLA-score). (C) The main results of the study. IMD, Index of Multiple Deprivation.

Table 1 Population characteristics included in the study. Characteristic All samples (N = 377487) Low-grade inflammation a Low-level (N = 200578) High-level (N = 176909) P value Age <0.001  <65 306637 (81.2%) 165723 (82.6%) 140914 (79.7%)  ≥65 70850 (18.8%) 34855 (17.4%) 35995 (20.3%) Sex <0.001  Female 203980 (54.0%) 103660 (51.7%) 100320 (56.7%) IMD scores b <0.001  Low 188948 (50.1%) 105375 (52.5%) 83573 (47.2%) WHR c 0.001  Ideal 191019 (50.7%) 112526 (56.2%) 78493 (44.5%) Lifestyle  Never smoking 205645 (54.8%) 115642 (57.9%) 90003 (51.2%) <0.001  Never drinking 30012 (8.0%) 14673 (7.3%) 15339 (8.7%) <0.001  Regular physical activity d 165827 (54.4%) 94382 (57.1%) 71445 (51.2%) <0.001  Healthy sleep pattern e 199333 (52.8%) 110332 (55.0%) 89001 (50.3%) <0.001  Healthy diet f 143838 (38.1%) 83233 (41.5%) 60605 (34.3%) <0.001 Chronic disease condition  Hypertension 108411 (28.7%) 49300 (24.6%) 59111 (33.4%) <0.001  Diabetes 19789 (5.2%) 8513 (4.2%) 11276 (6.4%) <0.001  Stroke 6090 (1.6%) 2598 (1.3%) 3492 (2.0%) <0.001 Cases of MS 432 (0.11%) 184 (0.09%) 248 (0.14%) <0.001 IMD scores, index of multiple deprivation scores; WHR, Waist-to-hip ratio; MS, multiple sclerosis. a Low-grade inflammation was categorized into low-level and high-level according to the median of INFLA-score; b IMD scores were categorized into low and high levels by the median. c Waist-to-hip ratio was classified into ideal and poor categories with 0.85 for women and 0.90 for men as the cutoff value. d Healthy sleep pattern was defined as healthy sleep score ≥4, and healthy sleep score include 5 aspects: chronotype, sleep duration, insomnia, snoring, and excessive daytime sleepiness, healthy sleep factors were defined as early chronotype (“morning” or “morning than evening”); sleep 7–8 h/day; reported never/rarely or sometimes insomnia symptoms; no self-reported snoring; and no excessive daytime sleepiness (“never/rarely” or “sometimes”). Each sleep factor was coded 1 if meeting the healthy criterion and 0 if not. e Regular physical activity was defined as meeting the American Heart Association recommendations of at least 150 minutes of moderate activity per week or 75 minutes of vigorous activity per week (or an equivalent combination) or engaging in moderate physical activity at least 5 days a week or vigorous activity once a week. f Healthy diet was based on consumption of at least 4 of 7 commonly eaten food groups following recommendations on dietary priorities for cardiometabolic health.

Population characteristics included in the study.

IMD scores, index of multiple deprivation scores; WHR, Waist-to-hip ratio; MS, multiple sclerosis. a Low-grade inflammation was categorized into low-level and high-level according to the median of INFLA-score; b IMD scores were categorized into low and high levels by the median. c Waist-to-hip ratio was classified into ideal and poor categories with 0.85 for women and 0.90 for men as the cutoff value. d Healthy sleep pattern was defined as healthy sleep score ≥4, and healthy sleep score include 5 aspects: chronotype, sleep duration, insomnia, snoring, and excessive daytime sleepiness, healthy sleep factors were defined as early chronotype (“morning” or “morning than evening”); sleep 7–8 h/day; reported never/rarely or sometimes insomnia symptoms; no self-reported snoring; and no excessive daytime sleepiness (“never/rarely” or “sometimes”). Each sleep factor was coded 1 if meeting the healthy criterion and 0 if not. e Regular physical activity was defined as meeting the American Heart Association recommendations of at least 150 minutes of moderate activity per week or 75 minutes of vigorous activity per week (or an equivalent combination) or engaging in moderate physical activity at least 5 days a week or vigorous activity once a week. f Healthy diet was based on consumption of at least 4 of 7 commonly eaten food groups following recommendations on dietary priorities for cardiometabolic health.

Association between LGI and the incidence risk of MS

The association between LGI and inflammation markers with MS occurrence is shown in Figure 2 , Supplementary Tables 6 , 7 . Sensitivity analyses confirmed the robustness of our main results ( Supplementary Table 8 ). When adjusting for multiple factors using proportional Cox regression models (model 3), the LGI was significantly associated with an increased incidence risk of MS (HR = 1.38, 95% CI: 1.10–1.72). Among the individual inflammation biomarkers, NLR (HR = 1.42, 95% CI: 1.14–1.76), WBC (HR = 1.41, 95% CI: 1.13–1.76), and PLT (HR = 1.37, 95% CI: 1.10–1.71) remained significantly associated with MS risk after adjusting for all covariates. However, the association between CRP and MS incidence was attenuated and no longer statistically significant in the fully adjusted model (HR = 1.09, 95% CI: 0.87–1.36).

Forest plot of the associations between low-grade inflammation and inflammation biomarkers with the incidence risk of multiple sclerosis. Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional hazards regression models, with the low-level group as reference. Dots with error bars colored in light to deep blue were hazard ratios with 95% CIs for model 1, model 2 and model 3, respectively. The vertical dashed line indicates the null value (HR = 1.00). Model 1 did not adjust for any variables. Model 2 adjusted for age, sex, and index of multiple deprivation (IMD) scores. Model 3 adjusted for age, sex, IMD scores, waist-to-hip ratio, no smoking, no drinking, regular physical activity, healthy sleep pattern, healthy diet, and chronic disease conditions including hypertension, diabetes, and stroke. LGI, low-grade inflammation; CRP, C-reactive protein; WBC, white blood cell count; PLT, platelet count; NLR, neutrophil-to-lymphocyte ratio; IMD, Index of Multiple Deprivation; CI, confidence interval.

Subgroup analyses were conducted to examine potential effect modification of the above associations ( Table 2 ; Supplementary Table 9 ). No significant interactions were observed between any inflammation biomarker and either sex, PRS category, healthy sleep patterns or regular physical activity (all P for interaction > 0.05), indicating that the associations were generally consistent across these subgroups. In contrast, significant effect modification was detected by age (<65 vs. ≥65 years), socioeconomic deprivation, hypertension, and diabetes status. When stratified by age (<65 vs. ≥65 years), a significant interaction was detected only for PLT ( P for interaction = 0.047), with the association being significant in participants aged <65 years (HR = 1.49, 95% CI: 1.18–1.88) but not in those aged ≥65 years (HR = 0.72, 95% CI: 0.37–1.42). For IMD scores, significant interactions were found for LGI (P for interaction = 0.035) and WBC ( P for interaction = 0.048). The association between elevated LGI and incident MS was stronger among individuals with higher deprivation (HR = 1.73, 95% CI: 1.27–2.35) than among those with lower deprivation (HR = 1.07, 95% CI: 0.78–1.48). Similarly, the WBC–MS association was more pronounced in the high IMD group (HR = 1.75, 95% CI: 1.30–2.37) compared to the low IMD group (HR = 1.10, 95% CI: 0.80–1.52). For hypertension, significant interactions were observed for LGI ( P for interaction = 0.002) and CRP ( P for interaction = 0.005). The LGI–MS association was markedly stronger among participants with hypertension (HR = 2.67, 95% CI: 1.63–4.35) compared to those without (HR = 1.12, 95% CI: 0.87–1.44). Likewise, elevated CRP was significantly associated with MS risk only in the hypertension subgroup (HR = 1.83, 95% CI: 1.19–2.82) but not in those without hypertension (HR = 0.89, 95% CI: 0.68–1.16). For diabetes, a significant interaction was detected for CRP ( P for interaction = 0.010). The CRP–MS association was stronger among participants with diabetes (HR = 1.81, 95% CI: 1.16–2.82) than among those without (HR = 0.90, 95% CI: 0.66–1.22).

Table 2 Associations of low-grade inflammation and the inflammation biomarkers with incidence risks of multiple sclerosis across subgroups. Subgroup LGI a CRP b WBC c PLT d NLR e HR (95% CI) P for interaction HR (95% CI) P for interaction HR (95% CI) P for interaction HR (95% CI) P for interaction HR (95% CI) P for interaction Sex Male 1.35 (0.93, 1.96) 0.911 1.13 (0.77, 1.65) 0.803 1.76 (1.21, 2.57) 0.153 1.49 (1.02, 2.17) 0.608 1.43 (0.99, 2.07) 0.971 Female 1.39 (1.06, 1.82) 1.07 (0.81, 1.40) 1.26 (0.97, 1.65) 1.32 (1.01, 1.72) 1.42 (1.09, 1.85) PRS category Low 1.09 (0.70, 1.72) 0.468 0.80 (0.50, 1.29) 0.107 1.46 (0.93, 2.29) 0.42 1.42 (0.90, 2.23) 0.537 1.03 (0.65, 1.63) 0.281 Moderate 1.38 (0.96, 2.01) 0.96 (0.66, 1.40) 1.17 (0.80, 1.69) 1.59 (1.10, 2.29) 1.50 (1.04, 2.16) High 1.56 (1.11, 2.18) 1.41 (1.01, 1.98) 1.62 (1.16, 2.26) 1.20 (0.86, 1.68) 1.62 (1.16, 2.25) Age <65 years 1.43 (1.13, 1.81) 0.362 1.08 (0.85, 1.38) 0.903 1.47 (1.16, 1.86) 0.319 1.49 (1.18, 1.88) 0.047 1.39 (1.11, 1.75) 0.619 ≥65 years 1.05 (0.56, 1.95) 1.13 (0.61, 2.10)   1.05 (0.56, 1.96) 0.72 (0.37, 1.42) 1.65 (0.88, 3.08) IMD scores f Low 1.07 (0.78, 1.48) 0.035 0.95 (0.68, 1.32) 0.262 1.10 (0.80, 1.52)) 0.048 1.40 (1.02, 1.92) 0.856 1.29 (0.94, 1.77) 0.403 High 1.73 (1.27, 2.35) 1.22 (0.90, 1.65) 1.75 (1.30, 2.37) 1.35 (1.00, 1.81) 1.55 (1.15, 2.08) Hypertension Yes 2.67 (1.63, 4.35) 0.002 1.83 (1.19, 2.82) 0.005 1.80 (1.18, 2.76) 0.187 1.74 (1.15, 2.64) 0.186 1.37 (0.91, 2.07) 0.838 No 1.12 (0.87, 1.44) 0.89 (0.68, 1.16)   1.29 (1.00, 1.67) 1.25 (0.97, 1.62) 1.44 (1.12, 1.86) Diabetes Yes 1.88 (1.21, 2.92) 0.132 1.81 (1.16, 2.82) 0.01 1.58 (1.07, 2.32) 0.569 1.83 (1.25, 2.68) 0.074 1.37 (0.94, 1.99) 0.808 No 1.27 (0.97, 1.66) 0.90 (0.66, 1.22) 1.38 (1.05, 1.80)   1.20 (0.92, 1.57) 1.45 (1.11, 1.89) LGI, low-grade inflammation; HR, Hazard ratios; CI, confidence intervals; CRP, C-reactive protein; WBC, white blood cell; PLT, platelets; NLR, neutrophil to lymphocyte ratio; PRS, polygenic risk scores; IMD scores, index of multiple deprivation scores. a The reference is low-level LGI; b The reference is low-level CRP; c The reference is low-level WBC; d The reference is low-level PLT; e The reference is low-level NLR; f IMD scores were categorized into low and high levels by the median.

Associations of low-grade inflammation and the inflammation biomarkers with incidence risks of multiple sclerosis across subgroups.

LGI, low-grade inflammation; HR, Hazard ratios; CI, confidence intervals; CRP, C-reactive protein; WBC, white blood cell; PLT, platelets; NLR, neutrophil to lymphocyte ratio; PRS, polygenic risk scores; IMD scores, index of multiple deprivation scores.

The reference is low-level LGI; b The reference is low-level CRP; c The reference is low-level WBC; d The reference is low-level PLT; e The reference is low-level NLR; f IMD scores were categorized into low and high levels by the median.

Association between LGI and brain structures

In fully adjusted models with FDR correction, LGI was negatively associated with the volumes of all six MS-related brain regions identified a priori from the literature ( Figure 3 ; Supplementary Table 10 ), including thalamus ( β: -0.057, 95% CI: -0.079, -0.035); superior temporal gyrus ( β : -0.053, 95% CI: -0.076, -0.031); lateral orbitofrontal cortex ( β : -0.050, 95% CI: -0.072, -0.027); putamen ( β : -0.047, 95% CI: -0.069, -0.025); caudate nucleus ( β : -0.041, 95% CI: -0.064, -0.017); supramarginal gyrus ( β : -0.036, 95% CI: -0.058, -0.013). Higher LGI levels were also associated with lower mean thickness of the superior temporal gyrus ( β : -0.039, 95% CI: -0.062, -0.015) and smaller surface areas of the superior temporal gyrus ( β : -0.045, 95% CI: -0.067, -0.023), lateral orbitofrontal cortex ( β : -0.042, 95% CI: -0.064, -0.020) and supramarginal gyrus ( β : -0.028, 95% CI: -0.050, -0.006). Consistent with the associations observed for the composite LGI score, higher CRP and WBC were associated with lower volumes across these imaging phenotypes. Notably, while CRP demonstrated inverse associations with both cortical thickness and surface area, WBC counts showed significant relationships exclusively with the surface areas of the three cortical structures ( Figure 4 ; Supplementary Table 10 ).

Associations of low-grade inflammation and the inflammation biomarkers with the volumes of brain structures. Multiple linear regression analyses were employed to estimate the associations of low-grade inflammation (LGI) and the inflammation biomarkers with volume of six brain structures related to multiple sclerosis. The inflammation biomarkers included C-reactive protein (CRP), white blood cell count (WBC), platelets (PLT) and the neutrophil to lymphocyte ratio (NLR). The model was adjusted for age, sex, index of multiple deprivation scores, waist-to-hip ratio, no smoking, no drinking, regular physical activity, healthy sleep pattern, healthy diet, and chronic health conditions such as hypertension, diabetes, and stroke. Data were presented as the β s. *p value FDR < 0.05. CRP, C-reactive protein; WBC, white blood cell count; PLT, platelet count; NLR, neutrophil-to-lymphocyte ratio.

Associations of low-grade inflammation, CRP, and WBC with the thickness and area of brain structures. (A) Associations between low-grade inflammation and brain structure. (B) Associations of CRP and WBC with brain structure. Multiple linear regression analyses were utilized to estimate the associations of low-grade inflammation (LGI), C-reactive protein (CRP) and white blood cell count (WBC) with the thickness and area of six brain structures related to multiple sclerosis. The model was adjusted for age, sex, index of multiple deprivation scores, waist-to-hip ratio, no smoking, no drinking, regular physical activity, healthy sleep pattern, healthy diet, and chronic health conditions such as hypertension, diabetes, and stroke. Data were presented as the bs. *p value FDR < 0.05. CRP, C-reactive protein; WBC, white blood cell count.

Subgroup analyses indicated that the negative associations between LGI and regional brain volumes were more pronounced in males than in females, with significant interactions observed for the putamen (male: β : -0.077, 95% CI:-0.109, -0.046; female: β : -0.020,95% CI: -0.050, 0.009, P for interaction = 0.036), thalamus (male: β : -0.088, 95% CI:-0.120, -0.057; female: β : -0.030,95% CI: -0.060, 0.000, P for interaction = 0.034), lateral orbitofrontal cortex (male: β : -0.080, 95% CI:-0.112, -0.048; female: β : -0.023,95% CI: -0.053, 0.008, P for interaction = 0.038), and superior temporal gyrus (male: β : -0.087, 95% CI:-0.119, -0.054; female: β : -0.024,95% CI: -0.054, 0.007, P for interaction = 0.030) ( Supplementary Table 11 ). Similarly, the negative correlation between LGI and the mean cortical thickness of the superior temporal gyrus (male: β : -0.073, 95% CI: -0.107, -0.039; female: β : -0.008,95% CI: --0.040, 0.024, P for interaction = 0.030) and supramarginal gyrus (male: β : -0.067, 95% CI: -0.102, -0.033; female: β : 0.015,95% CI: -0.018, 0.047, P for interaction = 0.008) were stronger in males participants than in females participants. No significant sex-based interactions were observed for the associations of CRP or PLT with brain structural measures (all P for interaction > 0.05). Meanwhile, the associations of NLR with brain volume, mean thickness, and surface area were in opposite directions between males and females—inverse in males and positive in females—except for caudate nucleus volume and supramarginal gyrus surface area ( Supplementary Tables 12, 13 ).

In genetic risk-stratified analyses, we observed consistent negative associations between high-level LGI and subcortical volumes across strata, particularly in the thalamus (low genetic risk: β :−0.050, 95% CI: −0.088, −0.012; moderate: β : −0.062, 95% CI: −0.100, −0.025; high: β : −0.060, 95% CI:−0.098, −0.022), caudate nucleus (moderate: β : −0.052, 95% CI: −0.092, −0.012; high: β : −0.059, 95% CI: −0.099, −0.019), and putamen (low: β : −0.050, 95% CI: −0.088, −0.013; moderate: β : −0.057, 95% CI: −0.094, −0.020). For cortical volumes, high-level LGI was consistently associated with the superior temporal gyrus (moderate: β : −0.058, 95% CI: −0.096, −0.020; high: β : −0.067, 95% CI: −0.105, −0.029), lateral orbitofrontal cortex (moderate: β : −0.051, 95% CI: −0.089, −0.013; high: β : −0.058, 95% CI: −0.096, −0.020), and supramarginal gyrus (high: β : −0.051, 95% CI: −0.089, −0.013). Regarding cortical surface area, significant negative associations were observed only in the high genetic risk group for the lateral orbitofrontal cortex ( β : −0.054, 95% CI: −0.092, −0.017) and superior temporal gyrus ( β : −0.058, 95% CI: −0.096, −0.021). No significant associations were detected for cortical thickness in any subgroup. Despite the pattern of more pronounced associations among individuals with higher genetic risk, no statistically significant interactions between LGI and genetic risk strata were detected for any brain structural measure (all P f or interaction > 0.05, Supplementary Table 14 ).

Discussion

In this large prospective cohort study, we found that high-level LGI was associated with an increased risk of incident MS, and comparable associations were observed for PLT, WBC, and NLR individually. This association was more pronounced among females and among individuals with high genetic risk, greater socioeconomic deprivation, or comorbid hypertension. Furthermore, cross-sectional neuroimaging analyses revealed significant inverse associations between systemic LGI and the volumes, mean thickness, and surface area of six MS-related brain regions. These associations were notably more pronounced among male participants.

Our finding that high-level LGI was associated with an increased risk of incident MS is consistent with prior evidence linking peripheral inflammatory markers to a range of neurological and psychiatric conditions, including dementia ( 15 ), Alzheimer’s disease ( 16 ), and major depression ( 14 , 37 ). Whereas most previous studies relied on single biomarkers assessed cross-sectionally, the present study employed a composite inflammation score measured prospectively in a large, initially MS-free population, thereby reducing the potential for reverse causation and providing more robust evidence for the observed association. Beyond methodological considerations, this association is also supported by biological plausibility. Genome-wide analyses by the IMSGC have demonstrated that the majority of MS susceptibility loci implicate genes within innate and adaptive immune pathways ( 6 ), and Shams et al. further identified inflammatory signaling mediation, oxidative damage responses, and epigenetic regulation as key contributors to MS genetic susceptibility ( 38 ). Collectively, these genetic findings support a plausible biological framework in which chronic peripheral immune dysregulation may interact with underlying genetic architecture to influence MS risk. Nevertheless, whether this interaction reflects a causal pathway or shared upstream determinants warrants further investigation through mechanistic and interventional studies.

The association between LGI and incident MS was significantly modified by socioeconomic deprivation and hypertension, but not by sex, polygenic risk score, or behavioral lifestyle factors (sleep patterns and physical activity), suggesting that peripheral inflammation–driven MS risk may be potentiated by socioeconomic and metabolic factors rather than by genetic predisposition or other lifestyle behaviors. Regarding sex specifically, the incidence density of MS was higher among women than among men in the present study, consistent with the well-documented female predominance in MS ( 39 ). However, this sex difference in MS incidence did not translate into a differential effect of systemic inflammation on MS risk, this difference in background susceptibility did not appear to amplify the inflammatory pathway to MS onset. Regarding IMD scores, our analysis revealed that low IMD scores amplified the association between LGI and incident MS, suggesting that socioeconomic factors—such as dietary quality, living conditions, air quality, and other unmeasured environmental exposures—may potentiate the effect of peripheral inflammation on MS development ( 40 , 41 ). This finding is of particular relevance given the evidence that area-level deprivation also limits access to disease-modifying therapies for MS in England ( 42 ), highlighting the need for targeted interventions in socioeconomically disadvantaged populations. Regarding pre-existing cardiometabolic conditions, both hypertension and diabetes appeared to strengthen the LGI–MS association. Mechanistically, endothelial dysfunction likely represents a central shared pathway, given that MS is characterized by significant endothelial impairment, as evidenced by lower reperfusion hyperemia index (RHI) values in patients compared to controls ( 43 ). Hypertension may compound this impairment through immune-mediated vascular remodeling, chronic vascular inflammation, and impaired cerebral perfusion ( 44 – 47 ). Furthermore, hypertension-driven oxidative stress can activate matrix metalloproteinases that compromise blood–brain barrier (BBB) integrity ( 48 ), thereby facilitating the central infiltration of systemic LGI mediators and potentiating MS-related neuroinflammatory processes. With respect to diabetes, we identified a significant interaction between CRP levels and pre-existing diabetes in relation to MS risk, consistent with prior evidence linking diabetes to elevated MS susceptibility (HR = 1.59) ( 49 ). This observation supports the hypothesis that hyperinsulinemia amplifies cytokine responses to inflammatory stimuli, further exacerbating the systemic inflammatory burden relevant to MS pathogenesis ( 50 ). Although the observational nature of this study precludes causal inference, these findings suggest that individuals with pre-existing hypertension or diabetes may represent a particularly vulnerable subgroup in whom monitoring and management of systemic low-grade inflammation could be of greater relevance for MS risk mitigation.

Based on the prospective association between LGI and risk of MS, we hypothesized that inflammation-related structural damage to vulnerable brain regions may constitute a key mechanism in MS pathogenesis. To explore this, we leveraged the UK Biobank imaging subsample to examine the associations between LGI and volumetric measures in six brain regions previously reported to undergo atrophy in MS ( 25 – 27 ). Notably, this analysis was conducted in the general imaging cohort, and therefore the results reflect population-level associations rather than MS-specific pathology. Indeed, our results revealed a significant negative association between LGI and the volume of all six MS-related brain regions, lending indirect support to the biological plausibility of this pathway. This finding is consistent with accumulating evidence. Elevated IL-6 and CRP levels have been correlated with reduced gray matter and hippocampal volume as well as increased white matter hyperintensity burden in older adults ( 51 ), and a large-scale population study demonstrated that elevated peripheral inflammatory markers are linked to cognitive decline mediated by total gray and white matter atrophy ( 52 ). A recent study has also shown a correlation between lipid-related biomarkers and brain atrophy, suggesting potential shared metabolic–inflammatory pathways that could be explored in future research ( 53 ). Critically, the fact that LGI was inversely associated with volume in all six regions specifically selected for their known vulnerability in MS—rather than in only a subset—strengthens indirect mechanistic support for the prospective association between LGI and MS risk. This consistency suggests that chronic peripheral inflammation may preferentially target brain regions inherently susceptible to inflammatory insult, potentially lowering the threshold for MS-related neurodegeneration. A previous longitudinal study in progressive MS further supports this directionality, indicating that systemic inflammation temporally precedes cervical cord atrophy ( 54 ), although confirmation in brain parenchymal regions requires more prospective MS-specific imaging cohorts.

Furthermore, the inverse associations between LGI and brain structural phenotypes appeared notably more pronounced among male participants. This finding contrasts with the comparable LGI–MS risk observed in both sexes, suggesting that the structural susceptibility of MS-related brain regions to peripheral inflammation may be sex-dependent. A parallel pattern of male vulnerability is observed in the clinical trajectory of MS, where men typically progress to disability milestones more rapidly than women ( 39 ) and are more prone to transitioning to secondary progressive MS ( 55 ). This progressive phase is characterized by extensive axonal loss and brain atrophy, driven predominantly by compartmentalized inflammatory and neurodegenerative processes within the CNS rather than the peripheral immune-mediated mechanisms that dominate the early relapsing phase ( 56 , 57 ). Sex differences in microglial reactivity, blood–brain barrier permeability, and the neuroprotective effects of estrogen have all been proposed as potential mechanisms underlying such male vulnerability ( 58 , 59 ), though these hypotheses remain to be directly tested in future studies. However, we found no significant interaction between MS-PRS and LGI for any brain structural measure, indicating that the inverse association between systemic inflammation and brain volumes in MS-related regions was comparable across different levels of genetic risk. Thus, this relationship does not appear to be limited to genetically susceptible individuals but is broadly observed in the general population, further highlighting the potential importance of monitoring and managing systemic inflammation for brain health at the population level.

Additionally, our decomposition of cortical measures across the three cortical regions revealed that the negative association between LGI and lateral orbitofrontal cortex volume was primarily attributable to smaller surface area, whereas for the supramarginal gyrus and superior temporal gyrus, higher LGI was associated with both lower cortical thickness and smaller surface area. The region-specific patterns observed here suggest that LGI may affect cortical morphology through different mechanisms depending on regional cytoarchitecture and vascular characteristics ( 60 , 61 ). These population-level findings are consistent with cortical abnormalities documented in the MS literature: lower cortical thickness has been consistently reported in patients, with region-specific reductions correlating with poorer cognitive performance ( 62 ), and smaller cortical surface area has been associated with shallower sulcal depth in sensory, cognitive, and motor regions ( 63 ). The convergence between these population-level patterns and MS-related cortical pathology suggests that certain cortical regions may be inherently susceptible to both systemic inflammatory insults and MS-related neurodegeneration. Because this study relied on a single imaging time point, prospective studies with serial imaging are needed to clarify the longitudinal impact of LGI on cortical morphology.

Risk stratification strategies identifying individuals with heightened susceptibility are essential for personalized health management and may offer greater efficacy than universal approaches. LGI represents a chronic, subclinical state, due to its insidious nature, LGI often remains undetected during routine clinical assessments, leading to significant gaps in early risk identification despite robust evidence linking this persistent inflammatory burden to the pathogenesis and progression of various chronic conditions ( 9 ). Systemic inflammation is shaped by a range of modifiable lifestyle and behavioral factors, including dietary patterns, physical activity levels, sleep adequacy, and psychological wellbeing ( 64 – 66 ). While our observational findings do not permit causal inference, the observed associations between LGI and both MS incidence and unfavorable brain structural phenotypes, coupled with the modifiable nature of many inflammatory determinants, provides a rationale for future interventional studies to examine whether reducing systemic inflammation may favorably influence MS-related outcomes.

This study has several strengths. Leveraging the large-scale UK Biobank cohort with a substantial number of healthy controls to ensure adequate statistical power, we constructed survival models that adjusted for potential confounders and explored mediation pathways linking inflammation to MS. Additionally, we integrated genetic and neuroimaging data to characterize the structural brain correlates associated with inflammation in the context of MS, providing preliminary evidence on potential biological pathways. However, this study also has several limitations. First, the UK Biobank lacks granular MS-specific clinical and radiological data—such as relapse rates, treatment responsiveness, and lesion load. Furthermore, the absence of information on disease progression (primary vs. secondary) limits our ability to explore potentially divergent inflammatory mechanisms across different MS phenotypes. Second, the generalizability of our findings may be limited by the composition of the UK Biobank, which is predominantly composed of individuals of White European ancestry (approximately 97%) and is subject to a recognized ‘healthy volunteer bias’. Furthermore, our analysis was restricted to urban residents. While this exclusion was necessitated to minimize potential confounding from urban-rural lifestyle variations highly correlated with LGI and to ensure the completeness of medical records, it may further restrict the broad applicability of these results to rural or more socioeconomically diverse populations. Third, incident MS cases were identified via hospital inpatient records, which may lack temporal precision regarding symptom onset and potentially lead to delayed case ascertainment. However, the consistency of our primary findings across sensitivity analyses (employing two-year and five-year lag periods to mitigate reverse causality) reinforces the robustness of the observed associations. Fourth, MS cases remaining undiagnosed or managed exclusively in private healthcare settings would not be captured through hospital inpatient linkage, potentially resulting in outcome misclassification that biases associations toward the null. Fifth, due to data access limitations, genetic ancestry principal components and genotyping array were not adjusted for in the PRS model, which may introduce residual population stratification bias. However, the UK Biobank cohort is predominantly of European ancestry, and a sensitivity analysis restricted to White participants (excluding approximately 3% of non-White individuals) yielded consistent results, partially mitigating this concern. Sixth, although brain MRI acquisition (from 2014) postdated the baseline assessment (2006–2010), providing some degree of temporal ordering, the imaging data were cross-sectional. Longitudinal studies with repeated neuroimaging are needed to establish a formal causal relationship between LGI and brain structural changes. Seventh, retrospectively collected confounder data are susceptible to recall bias, and inflammatory markers measured at a single time point may not reflect chronic inflammatory burden. However, the large UK Biobank cohort should mitigate the impact of any resulting non-differential misclassification. Finally, hypertension and diabetes—both linked to MS incidence and MRI measures—typically present later in life than MS; their effects may therefore differ between young-onset and late-onset MS, a distinction our study did not evaluate, and which warrants further investigation.

Conclusion

In summary, this large-scale prospective cohort study identified that LGI is robustly associated with an increased risk of incident MS. This association was notably amplified by socioeconomic deprivation and cardiometabolic conditions, particularly among individuals with high IMD scores and those with hypertension at baseline, but was not modified by sex, polygenic risk, or lifestyle factors (sleep and physical activity).Furthermore, higher LGI levels were consistently linked to lower volumes across six MS-related brain regions among the study participants, a relationship that appeared more prominent in males but remained independent of individual genetic risk scores. Given the rising prevalence of MS within aging populations, our findings provide valuable longitudinal evidence for risk stratification. While the observational nature of this study precludes definitive causal inference, these results highlight systemic inflammation as a significant correlate of neural integrity and disease susceptibility, warranting further research to determine the potential efficacy of managing subclinical inflammatory states in clinical practice.