A Multimodal Neuro-Demographic Signature for Immuno-Metabolic Depression

Zhaowen Nie; Simeng Ma; Zipeng Deng; Wei Wang; Enqi Zhou; Lijun Kang; Lihua Yao; Qian Gong; Lihong Bu; Zhili Niu; Zhongchun Liu

doi:10.1016/j.bpsc.2026.01.005

A Multimodal Neuro-Demographic Signature for Immuno-Metabolic Depression

Biol Psychiatry Cogn Neurosci Neuroimaging. 2026 Jan 23:S2451-9022(26)00023-6. doi: 10.1016/j.bpsc.2026.01.005. Online ahead of print.

Authors

Zhaowen Nie¹, Simeng Ma¹, Zipeng Deng¹, Wei Wang¹, Enqi Zhou¹, Lijun Kang¹, Lihua Yao¹, Qian Gong¹, Lihong Bu², Zhili Niu³, Zhongchun Liu⁴

Affiliations

¹ Department of Psychiatry, Renmin Hospital of Wuhan University, Wuhan 430060, P. R. China.
² PET/CT/MRI and Molecular Imaging Center, Renmin Hospital of Wuhan University, Wuhan 430060, P. R. China. Electronic address: bulihongs@whu.edu.cn.
³ Department of Clinical Laboratory, institute of translational medicine, Renmin Hospital of Wuhan University, Wuhan 430060, P. R. China. Electronic address: bulihongs@whu.edu.cn.
⁴ Department of Psychiatry, Renmin Hospital of Wuhan University, Wuhan 430060, P. R. China; Taikang Center for Life and Medical Sciences, Wuhan University, Wuhan 430060, P. R. China; State Key Laboratory of Metabolism and Regulation in Complex Organisms, College of Life Sciences, Wuhan University, Wuhan 430060, P. R. China. Electronic address: bulihongs@whu.edu.cn.

PMID: 41581557
DOI: 10.1016/j.bpsc.2026.01.005

Abstract

Background: The underlying neurobiology of a recently described immuno-metabolic depression (IMD) subtype of major depressive disorder (MDD), characterized by low-grade inflammation and metabolic dysregulation, remains unclear.

Methods: We integrated multimodal neuroimaging (structural/functional MRI) and demographic data from 145 MDD patients and 68 healthy controls (HC). After defining a composite IMD score derived from C-reactive protein, BMI, triglycerides, and high-density lipoprotein cholesterol levels by principal component analysis, we implemented a binary classification task using machine learning to distinguish high IMD score (IMD group, n=37) from low IMD score (nonIMD group, n=37) subgroups. Structural MRI (cortical thickness and gray matter volume), resting-state functional MRI (ReHo/fALFF), and demographic covariates were integrated as predictors.

Results: The multimodal model showed promise in classifying IMD group from nonIMD group (mean cross-validated AUC = 0.826 ± 0.098). Furthermore, its performance appeared somewhat more pronounced for within-MDD subtyping compared to differentiating MDD from HC (mean cross-validated AUCs of 0.647 ± 0.151 for nonIMD group vs. HC and 0.741 ± 0.111 for IMD group vs. HC), indicating subtype specificity. Key predictors included right amygdala volume and functional activity (ReHo/fALFF) in the hippocampus and mid-cingulate cortex. Clinically, the IMD group exhibited significantly higher anhedonia (p = 0.04), but lower somatic symptom scores (p < 0.05) compared to nonIMD group.

Conclusions: Our analysis shows that IMD is characterized by a distinct, multimodal neuro-demographic signature involving cortico-limbic circuitry. This signature demonstrates high specificity for unraveling MDD heterogeneity and is clinically linked to anhedonia, supporting the potential for biologically-informed patient stratification.

Keywords: Anhedonia; Immuno-Metabolic Depression; Machine Learning; Major Depressive Disorder; Multimodal Neuroimaging; Subtyping.