2026 ECNP Congress

PS04-4048

A composite cytokine score linked to brain structure predicts subsequent physical and mental health diagnoses

Ed Hutchings — C. Caiafa, J. Solé-Casals, M.E. Lynall, E. Bullmore · University of Cambridge, Psychiatry, Cambridge, United Kingdom

2026 ECNP Congress

Introduction

Dysregulated inflammation is a hallmark of many physical and mental disorders[1], with increasing evidence of a shared component of genetic liability between them[2]. This raises the possibility that immune biomarkers could enhance mechanistic understanding and prediction of disease transdiagnostically. However, widely used biomarkers of inflammation, such as C-reactive protein (CRP) capture only a narrow slice of the immune system’s complexity[1]. We therefore used plasma proteomic data from 46,633 adults[3] to derive principal components reflecting coordinated variation in 104 cytokines. We examined associations between the first principal component (PC1) and regional brain volumes, and compared PC1 and CRP as predictors of incidence of a range of physical and mental disorders.

Methods

We analysed plasma proteomic data[3] collected from 53,013 participants aged 45-85 (53.8% female) in UK Biobank, focusing on 105 cytokines. We excluded proteins with >10% missingness and individuals with >50% missingness or missing covariates, leaving 104 cytokines in 46,633 individuals. Missing protein expression values were imputed using Soft-Impute[4]. We fit a generalised additive mixed model to each protein to generate values residualised for technical, batch, and time-of-day effects. Residuals were z-scored and 10 principal components were derived. CRP values from the same participants were log-transformed and similarly residualised.

Ridge regression was used to model associations between PC1, corrected for age, sex, BMI, and Townsend deprivation index, and volume of 81 grey matter regions. To model associations between PC1 or CRP and disease incidence, correcting for the same covariates, we fitted Cox proportional hazards models with date of attending assessment centre as the origin and 31/12/24 as the censoring date. A range of common disorders were selected as outcomes and participants with a diagnosis of any target phenotype prior to baseline were excluded. Predictors were z-scored, such that hazard ratios (HR) reflect the change in instantaneous incidence rate associated with one standard deviation increase in PC1 or CRP. P-values were corrected for multiple comparisons using the Benjamini-Hochberg procedure.

Results

PC1 explained 15.4% of the total variability in cytokine abundance (other PCs <10%). PC1 scores were highly correlated with mean cytokine expression (Spearman’s r=0.94, p<0.001) and moderately with CRP (Spearman’s r=0.36, p<0.001). PC1 loadings were significantly enriched for Tumour Necrosis Factor (TNF) pathway activity. Elevated PC1 was associated with decreases in inferior frontal and subcortical volumes, most notably in the nucleus accumbens bilaterally. Elevated baseline CRP and cytokine PC1 were significantly associated with elevated hazard in 9/14 and 12/14 diseases, respectively. PC1 HR was larger in 7/8 diseases in which both predictors had HR>1, including lupus, inflammatory bowel disease, type II diabetes mellitus, myocardial infarction, and major depressive disorder.

Conclusions

Principal component analysis of cytokine measurements in a population cohort revealed a latent axis relating to mean cytokine burden, TNF signalling, and fronto-subcortical atrophy. Cytokine PC1 was more strongly related to the hazard of inflammatory, cardiometabolic, and psychiatric disease diagnosis than CRP, highlighting the translational potential of composite immune protein biomarkers for disease prediction, and a potential transdiagnostic inflammatory axis contributing to multiple physical and mental health disorders.

References

  1. Furman D, Campisi J, Verdin E, Carrera-Bastos P, Targ S, Franceschi C, et al. Chronic inflammation in the etiology of disease across the life span. Nature Medicine. 2019;25: 1822–1832.
  2. Lawrence JM, Foote IF, Breunig S, Schaffer LS, Lyons S, Abramowitz SA, et al. Shared Genetic Liability across Systems of Psychiatric and Physical Illness. Nature Communications. 2026;17: 2993.
  3. Sun BB, Chiou J, Traylor M, Benner C, Hsu Y-H, Richardson TG, et al. Plasma proteomic associations with genetics and health in the UK Biobank. Nature. 2023;622: 329–338.
  4. Mazumder R, Hastie T, Tibshirani R. Spectral Regularization Algorithms for Learning Large Incomplete Matrices. J Mach Learn Res. 2010;11: 2287–2322.