Introduction
Mental health disorders constitute a major global disease burden, accounting for 155.42 million disability-adjusted life years (DALYs) in 2021 and ranking as the sixth leading contributor to global DALYs (GBD 2021 Diseases and Injuries Collaborators, 2024). The frequently co-occurring depression-anxiety comorbidity represents the most prevalent psychiatric conditions, with emerging evidence linking this comorbidity to accelerated aging phenotypes (Hou et al., Reference Hou, Chen, Sun, Song, Deng, Cheng, Han, Zhang, Wang, Li, Yin and Meng2023; Martin et al., Reference Martin, Asjadi, Hubbard, Kendall, Pardiñas, Jermy, Lewis, Baune, Boomsma, Hamilton, Lucae, Magnusson, Martin, McIntosh, Mehta, Mors, Mullins, Penninx and Preisig2021). Given the increasing global burden of mental disorders with poorly understood pathophysiology (GBD 2019 Mental Disorders Collaborators, 2022), research has prioritized the identification of modifiable risk factors and development of targeted interventions (Gao et al., Reference Gao, Jiang, Huang, Guo and Huang2023).
As longevity increases, geriatric syndromes and functional decline are gaining research prominence owing to their widespread epidemiological prevalence and multidimensional health impacts, particularly on affective psychopathologies (Li et al., Reference Li, Ford, Zhao, Ahluwalia, Pearson and Mokdad2009; Sarris et al., Reference Sarris, Thomson, Hargraves, Eaton, de Manincor, Veronese, Solmi, Stubbs, Yung and Firth2020; Xu, Yin, & Gong, Reference Xu, Yin and Gong2023). Current geriatric assessments lack comprehensive evaluation of biopsychosocial functioning. Addressing this gap, the World Health Organization operationalized intrinsic capacity (IC) as a multidimensional healthy aging construct (Zhou & Ma, Reference Zhou and Ma2022), formally defined as ‘the composite of biological and psychosocial reserves’(Beard et al., Reference Beard, Officer, de Carvalho, Sadana, Pot, Michel, Lloyd-Sherlock, Epping-Jordan, Peeters, Mahanani, Thiyagarajan and Chatterji2016). Contemporary research operationalizes this construct through five measurable domains: mobility, cognition, sensory function, psychological resilience, and metabolic vitality (Beard et al., Reference Beard, Si, Liu, Chenoweth and Hanewald2022; Cesari et al., Reference Cesari, Araujo, Amuthavalli Thiyagarajan, Cooper, Martin, Reginster, Vellas and Beard2018). IC has been demonstrated to predict functional decline independently of age, sex, and comorbidities, establishing it as a critical marker of aging-related health trajectories (Ramirez-Velez et al., Reference Ramírez-Vélez, Iriarte-Fernandez, Santafe, Malanda, Beard, Garcia-Hermoso and Izquierdo2023; Sánchez-Sánchez et al., Reference Sánchez-Sánchez, Lu, Gallardo-Gómez, del Pozo Cruz, de Souto Barreto, Lucia and Valenzuela2024). Previous UK Biobank evidence suggested that sleep-related impairment, including sleep pathology, may predict future depression risk (Lyall et al., Reference Lyall, Sangha, Zhu, Lyall, Ward, Strawbridge, Cullen and Smith2023). Extending this line of evidence, the present study examined multiple available IC domains and integrated metabolomic signatures and genetic susceptibility in relation to incident depression and anxiety.
Metabolomics analysis provides a unique perspective for revealing the association between IC and psychopathology by quantifying dynamic biological pathways influenced by genetics and lifestyle. A non-targeted metabolomics study has shown that IC impairment is closely associated with dysregulation of multiple classes of metabolites, including acylcarnitines, carbohydrates, and amino acids (Pan et al., Reference Pan, Li, Chhetri, Liu, Li, Liu, Shui and Ma2024). Further analysis revealed that abnormalities in the arginine-ornithine metabolic pathway are not only related to IC levels but are also significantly associated with the progression of cognitive decline. Both enrichment analysis and pathway analysis indicate that the purine metabolism pathway is highly correlated with the occurrence and development of psychological disorders, suggesting that this pathway may serve as a key link connecting IC impairment and psychopathology. Additionally, chronic inflammation, an important mechanism of aging, has been confirmed to have a causal association with the decline in individual IC (Giudici et al., Reference Giudici, de Souto Barreto, Guerville, Beard, de Carvalho, Andrieu, Rolland and Vellas2019), which biologically reflects the extensive connection between IC and functional capacity. Clinical studies have also found a clear correlation between chronic inflammatory status and mental disorders such as depression and anxiety, with elevated levels of inflammatory biomarkers in the peripheral blood of patients, and a dose–response relationship between the degree of inflammation and symptom severity (Guo et al., Reference Guo, Zhang, Hao, Wang, Zhang and Liu2023). However, current research has not clarified the specific metabolic characteristics of the relationship between IC decline and mental disorders. This critical knowledge gap highlights the need to examine the potential mediating role of metabolic pathways.
This study developed an IC-correlated metabolomic profile, subsequently evaluating its depression/anxiety risk associations, using UK Biobank’s nuclear magnetic resonance (NMR) metabolomics data. Through genome-wide polygenic risk scores (PRS), we further examined genetic moderation effects on these metabolic–psychopathology relationships.
Material and methods
Population and study design
Figure 1 depicts the study framework. The UK Biobank, a prospective cohort recruiting over 500,000 participants (Sun et al., Reference Sun, He, Sun, Han, Wang, Zhao, Li, Ma, Feng, Li, Yang and Shen2024), serves as our analytical platform. During baseline assessment, participants completed a touchscreen questionnaire and underwent a nurse-led interview, providing detailed information on lifestyle factors, physical examinations, and biological sample collection (Elliott, Peakman, & Biobank, Reference Elliott, Peakman and Biobank2008). The study received ethical clearance from the North West Multicenter Research Ethics Committee (11/NW/0382), with written informed consent secured from all participants prior to data collection.
Study design framework. IC deficit scores were computed from seven functional status measures, with subsequent metabolite profiling conducted via elastic net regression. Polygenic risk scores quantified genetic susceptibility to mood disorders. This study examined IC-associated metabolic patterns and their relationship with depression/anxiety risk, while controlling for genetic predisposition.
Figure 1. Long description
The flowchart is divided into four main sections.
1. Top-Left: A radial diagram titled Intrinsic capacity domains and factors. At the center is I C, surrounded by four colored quadrants: Psychological (red), Locomotor (orange), Sensory (blue), and Vitality (green). The outer ring lists specific measures: Exhaustion, Sleep duration, Slow walking pace, Eye problems, Hearing difficulty, Weight loss, and Low grip strength.
2. Top-Center: A box labeled Genetic susceptibility containing a D N A helix and two alleles. Below this, a red arrow points to a large central horizontal arrow containing the text 212,033 U K Biobank participants. Above the central arrow, text reads Joint effect: 1 plus 1 is greater than 2. Below the central arrow, a Mediation diagram shows circles A, B, and C with directional arrows.
3. Top-Right: Two illustrations representing Depression (a person sitting under a rain cloud) and Anxiety (a person looking distressed with exclamation marks).
4. Bottom Section: A large panel detailing the metabolic profiling of 251 metabolites. On the left are icons of a D N A strand, a microscope, and test tubes. Three horizontal dashed lines lead to analytical methods and results:
- Elastic net regression leads to I C-related metabolic signature.
- Metabolite set enrichment analysis leads to 6 enrichment groups.
- Weighted gene co-expression network analysis (W G C N A) leads to 5 W G C N A modules.
To the right of these labels are three data plots: two line graphs showing associations between metscore and depression or anxiety, a horizontal bar chart for metabolite set enrichment analysis, and two scatter plots showing Hazard Ratios for the 5 modules.
In this analysis of 503,371 individuals, participants were excluded based on the following criteria: 44,653 with incomplete IC assessments (exhaustion, sleep duration, hearing difficulties, vision problems, slow walking pace, weight loss, and low grip strength), 4,785 with consent withdrawal or preexisting depression/anxiety at baseline, 78,315 lacking genotypic data, and 163,585 lacking metabolomic profiling. The final analytical sample comprised 212,033 participants. The specific screening process is shown in Supplementary Figure S1.
Intrinsic capacity
We constructed a UK Biobank-adapted four-domain IC deficit score using seven available functional indicators mapped to four IC domains: exhaustion and sleep duration for the psychological domain, eye problems and hearing difficulty for the sensory domain, weight loss and low grip strength for the vitality domain, and slow walking pace for the locomotor domain (Supplementary Methods; Supplementary Table S1) (Beard et al., Reference Beard, Officer, de Carvalho, Sadana, Pot, Michel, Lloyd-Sherlock, Epping-Jordan, Peeters, Mahanani, Thiyagarajan and Chatterji2016; Beard, Jotheeswaran, Cesari, & Araujo, Reference Beard, Jotheeswaran, Cesari and Araujo2019; Ramírez-Vélez et al., Reference Ramírez-Vélez, Iriarte-Fernandez, Santafe, Malanda, Beard, Garcia-Hermoso and Izquierdo2023a, Reference Ramírez-Vélez, Iriarte-Fernandez, Santafé, Malanda, Beard, Garcia-Hermoso and Izquierdo2023b; Sun et al., Reference Sun, He, Sun, Han, Wang, Zhao, Li, Ma, Feng, Li, Yang and Shen2024). Cognitive capacity, one of the five WHO-defined IC domains, was not included in the primary score because comprehensive cognitive-domain data were not available for this analysis. Therefore, the exposure should be interpreted as a four-domain IC deficit score rather than a complete five-domain WHO IC measure. These indicators should be regarded as available proxies for selected aspects of the corresponding IC domains rather than complete operationalizations of each WHO-defined domain. The IC deficit score summed the seven indicators (Supplementary Table S1), with higher scores indicating poorer capacity (range 0–7). The categories ‘five factors’ (n = 982), ‘six factors’ (n = 204), and ‘seven factors’ (n = 3) were merged into a single ‘4 or more (4+) factors’ group due to the small number of responses in each. This transformation created discrete categorical variables representing the level of affected domains as 0, 1, 2, 3, or 4+.
Ascertainment of depression and anxiety
Incident depression and anxiety cases were identified through linkage to hospital admission records (both primary and secondary ICD-10 diagnoses: F32–F33 for depressive disorders, F40–F41 for anxiety disorders) and national mortality registries. Follow-up extended from baseline until the first occurrence of any of the following events: incident depression or anxiety diagnosis, mortality, loss to follow-up, or region-specific administrative censoring dates.
Metabolomics measurement
Metabolomics measurement in this study involved profiling plasma metabolites from UK Biobank participants (Supplementary Methods). The profiling was conducted by NMR spectroscopy. Briefly, a total of 251 metabolic biomarkers were assessed, including 170 metabolites and 81 composite ratio indices, covering a broad range of metabolic pathways. This standardized panel encompasses a wide range of metabolic pathways and biomolecular classes, including cholesterol metabolism, fatty acid composition, amino acids, ketone bodies, glycolysis-related metabolites, and lipid profiles across 14 lipoprotein subclasses. It reflects a broadly targeted metabolomics approach, emphasizing the predefined and accurate quantification of hundreds of known metabolic biomarkers. To address skewed distributions, metabolites with high skewness (absolute skewness >2) were log-transformed. All metabolites were converted to z-scores within each sub-study. The random forest imputation method was used to impute missing data for each metabolite, which has been previously recommended for metabolomics analysis (Wang et al., Reference Wang, Tessier, Liang, Wittenbecher, Haslam, Fernández-Duval, Heather Eliassen, Rexrode, Tobias, Li, Zeleznik, Grodstein, Martínez-González, Salas-Salvadó, Clish, Lee, Sun, Stampfer, Hu and Guasch-Ferré2023).
Genetic risk of depression and anxiety
PRS was employed to quantify baseline genetic risk for depression and anxiety (Supplementary Methods). Depression and anxiety PRS were derived from genome-wide association study summary statistics (Otowa et al., Reference Otowa, Hek, Lee, Byrne, Mirza, Nivard, Bigdeli, Aggen, Adkins, Wolen, Fanous, Keller, Castelao, Kutalik, der Auwera, Homuth, Nauck, Teumer, Milaneschi and Hettema2016; Wray et al., Reference Wray, Ripke, Mattheisen, Trzaskowski, Byrne, Abdellaoui, Adams, Agerbo, Air, Andlauer, Bacanu, Bækvad-Hansen, Beekman, Bigdeli, Binder, Blackwood, Bryois, Buttenschøn and Bybjerg-Grauholm2018). The UK Biobank’s standardized genotyping, imputation, and quality control protocols have been comprehensively documented (Bycroft et al., Reference Bycroft, Freeman, Petkova, Band, Elliott, Sharp, Motyer, Vukcevic, Delaneau, O’Connell, Cortes, Welsh, Young, Effingham, McVean, Leslie, Allen, Donnelly and Marchini2018; Sudlow et al., Reference Sudlow, Gallacher, Allen, Beral, Burton, Danesh, Downey, Elliott, Green, Landray, Liu, Matthews, Ong, Pell, Silman, Young, Sprosen, Peakman and Collins2015). PRS tertiles (low/intermediate/high risk) were defined after adjusting for sex, and 20 genetic principal components were included using clump-and-threshold pruning (Zhao et al., Reference Zhao, Yang, Sahakian, Langley, Zhang, Kuo, Li, Gan, Li, Zhao, Yu, Feng and Cheng2023). The optimally matched PRS parameters for depression and anxiety are summarized in Supplementary Table S2.
Assessment of covariates
Covariates were selected based on established determinants from prior studies, encompassing: demographic factors (age and sex); socioeconomic status (Townsend deprivation index, education, and household income); lifestyle parameters (smoking, alcohol use, and physical activity); body mass index (BMI); four individual chronic diseases (hypertension, physician-diagnosed diabetes, cardiovascular disease, and cancer), each treated as a separate covariate; and genetic factors (first 20 ancestry principal components) (Supplementary Methods) (Cabanas-Sanchez et al., Reference Cabanas-Sanchez, Esteban-Cornejo, Parra-Soto, Petermann-Rocha, Gray, Rodriguez-Artalejo, Ho, Pell, Martínez-Gómez and Celis-Morales2022; Gao et al., Reference Gao, Jiang, Huang, Guo and Huang2023).
Statistical analysis
Participant characteristics were presented as mean ± standard deviation for continuous variables, and frequency counts with percentages (%) for categorical variables. For the covariate data, we employed multiple imputations via chained equations to impute missing values. Characteristics of the non-imputed and imputed populations under investigation are presented in Supplementary Table S3.
Three Cox proportional hazards regression models were employed to estimate hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) for the associations between IC, metabolic profiles, genetic predisposition, and incident depression/anxiety (Supplementary Methods). The proportional hazards assumption was verified through Kaplan–Meier cumulative risk curves, which revealed no significant violations of the model assumptions. Furthermore, restricted cubic splines were employed to examine whether the risk of depression or anxiety exhibited a linear relationship with the increasing number of IC components, that is, the cumulative effect of IC components. The Kaplan–Meier cumulative risk curves were then generated, and comparisons were made between groups. The proportion of incident depression/anxiety cases attributable to IC was estimated using the population attributable fraction (PAF).
Elastic net regression, which integrates the strengths of both Lasso and Ridge regression (Zou & Hastie, Reference Zou and Hastie2005), was employed to examine associations between 251 metabolites and IC factors while simultaneously addressing multicollinearity and overfitting through its dual regularization mechanism. In the analysis, we implemented elastic net regression on 251 standardized metabolites, determining the optimal λ through 10-fold cross-validation with the 1-standard-error rule. This generated metabolic risk scores via a weighted linear combination of selected metabolites, where penalized regression coefficients quantified individual metabolite contributions.
Metabolite set enrichment analysis (MSEA) identified biologically annotated metabolite clusters associated with outcomes (Subramanian et al., Reference Subramanian, Tamayo, Mootha, Mukherjee, Ebert, Gillette, Paulovich, Pomeroy, Golub, Lander and Mesirov2005). Metabolite clusters identified through weighted gene co-expression network analysis (WGCNA) were further evaluated using Cox proportional hazards models to assess associations with depression and anxiety (Langfelder & Horvath, Reference Langfelder and Horvath2008).
We used additive and multiplicative scales to analyze the factors that may interact with IC (Supplementary Methods). Mediation analyses assessed metabolic profiles’ potential mediation of IC–depression/anxiety associations. Covariate definitions and subgroup analyses are detailed in Supplementary Methods. Two-tailed tests used P < 0.05 significance thresholds. Analyses utilized R v4.3.1.
Results
Basic characteristics of the study participants
The study comprised participants with a mean baseline age of 57.4 years (46.5% male, n = 113,456). Over a median follow-up of 13.7 years, 11,881 incident depression cases and 10,454 anxiety cases were identified. Participants’ baseline characteristics are shown in Table 1. Individuals who developed depression or anxiety were generally older, more likely to be female, had lower education levels, higher BMI, and a greater prevalence of comorbidities. At baseline, common comorbidities were frequent in the cohort, including hypertension in 98,885 participants (46.64%), cardiovascular disease in 62,777 participants (29.61%), diabetes in 10,142 participants (4.78%), and cancer in 16,349 participants (7.71%) (Table 1). The baseline characteristics stratified by IC deficit score and PRS are shown in Supplementary Tables S4–S6.
Characteristics of the study population at baseline
Table 1. Long description
The table presents baseline characteristics for 212,033 participants. The columns are divided into Overall (100%), Depression (No: 94%, Yes: 5.6%), and Anxiety (No: 95%, Yes: 4.9%).
* Age: Overall mean is 57.42 plus or minus 7.99. The Anxiety Yes group has the highest mean age at 58.19.
* Sex: Overall, 46.49% are Male and 53.51% are Female. In the Depression Yes group, females represent 63.38%.
* Education: 30.80% have a college education, while 69.20% have less than a college education.
* Townsend deprivation index: Overall mean is -1.65 plus or minus 2.89. The Depression Yes group has a higher mean Townsend deprivation index of -0.91.
* Income: Distributed across four tiers. The ‘Less than 18,000’ tier is higher in the Depression Yes (37.67%) and Anxiety Yes (35.39%) groups compared to the overall 23.52%.
* B M I: Overall mean is 27.38 plus or minus 4.69.
* Smoking: 54.80% never smoked, 35.39% are previous smokers, and 9.82% are current smokers.
* Drinking: 93.76% are current drinkers.
* Physical activity: 16.49% Low, 41.57% Moderate, and 41.94% High.
* Health Conditions: Overall prevalence is 46.64% for Hypertension, 29.61% for Cardiovascular disease, 4.78% for Diabetes, and 7.71% for Cancer.
* I C deficit score: Categories range from 0 to 4 plus. In the Depression Yes group, 8.23% have a score of 4 plus, compared to 2.45% overall.
a (Mean ± SD); n (%).
Associations between IC deficit score and depression/anxiety
Compared to participants with 0 IC deficit score, those with scores of 1, 2, 3, and 4+ showed a significantly increased risk of depression, which was 1.44 (95% CI: 1.38–1.50) folds in 1 score, 2.00 (1.92–2.08) folds in 2 score, 2.86 (2.72–3.00) folds in 3 score, and 3.96 (3.72–4.21) folds in 4+ score for depression (Figure 2). Similarly, individuals with IC scores of 1, 2, 3, and 4+ had 1.35 (1.30–1.40), 1.71 (1.64–1.79), 2.32 (2.20–2.45), and 3.03 (2.82–3.25) fold risks of incident anxiety, respectively, compared with those with a score of 0 (Figure 2).
Association between IC and depression/anxiety. The left panel presents results from multivariable Cox proportional hazards regression models: Model 1 included age and sex as baseline covariates; Model 2 incorporated socioeconomic status (Townsend deprivation index, education level, and annual household income), body mass index (BMI), and lifestyle factors (smoking, alcohol consumption, and physical activity); Model 3 additionally included four individual chronic diseases (hypertension, physician-diagnosed diabetes, cardiovascular disease, and cancer, each treated as a separate covariate) and genetic factors (first 20 ancestry principal components). The right panel shows restricted cubic spline analysis results, which were used to explore the nonlinear dose–response relationship between IC and depression/anxiety.
Figure 2. Long description
The left panel is a forest plot divided into two columns for Depression and Anxiety. The y-axis lists I C scores from 1 score, 2 score, 3 score, 4 plus score, and Each additional I C score. The x-axis represents Hazard Ratio H R on a scale from 0.5 to 6.0. For both conditions, H R increases significantly as I C scores rise. Three models are shown: Model 1 light blue diamond, Model 2 teal triangle, and Model 3 yellow circle. All points are marked with three asterisks indicating statistical significance. The right panel, titled Restricted cubic spline, contains two line graphs. The top graph for Depression shows a non-linear upward curve starting near H R 1.0 at I C score 0 and rising to over 6.0 at I C score 7. P-overall is less than 0.001 and P-non-linear is 0.007. The bottom graph for Anxiety shows a similar upward curve starting below H R 1.0 and rising to over 4.0 at I C score 7. P-overall is less than 0.001 and P-non-linear is 0.063. Both spline graphs include a shaded area representing the 95 percent C I.
Restricted cubic spline analyses demonstrated a nonlinear relationship between IC and the incidence of depression (P nonlinear = 0.007), contrasting with linear anxiety associations (P nonlinear = 0.063), both demonstrating significant overall trends (P overall < 0.001; Figure 2). Each unit increase in IC score elevated depression and anxiety risks by 39% (95% CI: 37–40) and 30% (95% CI: 29–32), respectively (Figure 2). Cumulative risk trajectories were visualized through Kaplan–Meier curves (Supplementary Figure S2). Furthermore, the combined PAF showed that IC score was associated with 16.32% (95% CI: 14.68–18.05) of depression cases and 10.86% (95% CI: 9.37–12.46) of anxiety cases (Supplementary Table S7). The PAF analysis of IC domains revealed that the psychological domain contributes the most to depression and anxiety, while the contribution from the locomotor domain is the least (Supplementary Table S8).
Associations of IC and metabolic signatures with risk of depression/anxiety
Elastic net regression distilled 59 IC-associated metabolites from 251 candidate biomarkers (Figure 3a,b). We further constructed the related metabolic signatures for the four domains that constitute IC. These biomarkers spanned diverse metabolic pathways. The main components of the metabolic signature were relative lipoprotein lipid concentrations (32.20%), fatty acids (16.95%), amino acids (16.95%), and lipoprotein subclasses (11.86%) (Supplementary Table S9). Positive coefficients in the IC-related metabolic signature were primarily driven by triglycerides to total lipids in IDL percentage, monounsaturated fatty acids, omega-6 fatty acids to omega-3 fatty acids ratio, free cholesterol to total lipids in very small VLDL percentage, and glycoprotein acetyls. Additionally, linoleic acid to total fatty acids percentage, cholesteryl esters in chylomicrons and extremely large VLDL percentage, saturated fatty acids to total fatty acids percentage, albumin, leucine, and DA percentage played significant roles in the negative coefficients affecting IC-related metabolic characteristics.
The metabolites constituting the metabolic signature with IC and its association with depression or anxiety risk. (a) The weights of the metabolites in signatures. (b) Associations of metabolites with IC (factors). (c) Restricted cubic splines of depression associated with metabolic signatures. (d) Associations of metabolic signatures with depression. (e) Restricted cubic spline analysis for anxiety. (f) Associations between metabolic signatures and anxiety. Chromatic scheme: Positive associations (red gradient) versus negative associations (blue gradient). Color saturation corresponds to effect strength (increased saturation = stronger association). Statistical significance markers: *p < 0.05, **p < 0.01, and ***p < 0.001 (two-tailed testing).
Figure 3. Long description
Panel A is a heatmap titled Metabolic signature coefficient. The y-axis lists 54 metabolites including Alanine, Glutamine, and various lipid fractions like X X L V L D L T G. The x-axis lists five factors: I C, Locomotor, Psychological, Sensory, and Vitality. Cells contain plus or minus signs indicating weight direction, with a color scale from blue (negative 0.50) to red (positive 0.25). Gly C A and M U F A show strong positive weights in red.
Panel B is a heatmap titled Associations between I C factors and metabolites. It uses the same axes as Panel A but displays association effects from blue (negative 0.1) to red (positive 0.2). Asterisks indicate statistical significance. Gly C A shows a strong positive association with Locomotor.
Panel C is a restricted cubic spline graph for depression. The x-axis is Metscore I C and the y-axis is H R (Hazard Ratio). A solid red line shows a non-linear increase in depression risk as the metabolic score increases, with a grey histogram showing the data distribution. P-overall is less than 0.001.
Panel D is a table showing associations between metscore and depression. It lists H R and 95 percent C I for five groups. Psychological has the highest H R at 1.79, followed by I C at 1.69.
Panel E is a restricted cubic spline for anxiety. Similar to Panel C, it shows a positive linear trend between Metscore I C and H R, though the slope is slightly less steep. P-non-linear is 0.108.
Panel F is a table for anxiety associations. Psychological shows the highest H R at 1.56, followed by I C at 1.52. Sensory is the only group not reaching statistical significance with an H R of 1.09.
Dose–response analysis indicated that increasing IC-related metabolic characteristics were associated with a linear trend in depression/anxiety risk. Fully adjusted models revealed elevated depression risk across IC domains, with HRs (95% CI) of 1.69 (1.55–1.83), 1.17 (1.14–1.19), 1.79 (1.66–1.93), 1.19 (1.07–1.31), and 1.22 (1.16, 1.28) for IC, locomotor, psychological, sensory, and vitality domains, respectively (Figure 3c,d). After full adjustment, the metabolic signatures were associated with higher risk of anxiety, with HRs (95% CI) of 1.52 (1.39–1.66), 1.13 (1.10–1.16), 1.56 (1.44–1.69), 1.09 (0.97–1.21), and 1.19 (1.13–1.26) for IC, locomotor, psychological, sensory, and vitality domains, respectively (Figure 3e,f).
Metabolite groups and modules associations for depression and anxiety
In the fully adjusted model, metabolites positively correlated with both depression and anxiety included triglycerides to total lipids in large LDL percentage, triglycerides to total lipids in medium LDL percentage, monounsaturated fatty acids to total fatty acids percentage, triglycerides in IDL, and triglycerides in large LDL. Metabolites inversely associated with both conditions comprised polyunsaturated fatty acids to total fatty acids percentage, polyunsaturated to monounsaturated fatty acids ratio, degree of unsaturation, omega-6 fatty acids to total fatty acids percentage, and cholesteryl esters to total lipids in small HDL percentage (Figure 4a,b).
Association of metabolic clusters with mood disorders. (a) Heatmap visualization of metabolite–mood disorder correlations. (b) Hazard ratios (95% CI) per standard deviation increase for top-ranked metabolites (10 positively/negatively associated). (c) Systematic evaluation of functionally annotated metabolite classes. (d) Metabolome-wide association study results organized by WGCNA-derived co-expression modules, analyzed through multivariable Cox regression. Visual keys: Chromatic intensity reflects effect magnitude (red: positive association; blue: inverse association). Statistical significance denoted as *p < 0.05, **p < 0.01, and **p < 0.001 (two-tailed tests). Note: CE: ‘cholesteryl ester’; C: ‘cholesterol’; FC: ‘free cholesterol’; MASLD: ‘metabolic dysfunction-associated steatotic liver disease’; PL: ‘phospholipid’; VLDL: ‘very low-density lipoprotein’. Particle size classifications: M: Medium-sized particles; S: Small particles; XL: Very large particles; XS: Very small particles; XXL: Exceptionally large particles.
Figure 4. Long description
Panel A is a circular heatmap showing correlations for the top 100 metabolites. At the center is a violin plot comparing H R values for anxiety in blue and depression in orange. Individual metabolites are arranged around the circle and labeled with codes like L underscore L D L underscore T G percent. Red cells indicate positive associations and blue cells indicate inverse associations. Asterisks denote statistical significance levels.
Panel B contains four tables listing the top 10 positive and negative metabolite associations for depression and anxiety. Each table includes the metabolite name, H R, and 95 percent C I. For example, L underscore L D L underscore T G percent shows a positive H R of 1.10 for depression, while P U F A percent shows a negative H R of 0.90.
Panel C is a horizontal bar chart for metabolite set enrichment analysis. The y-axis lists functional classes like Lipoprotein subclasses and Amino acids. The x-axis measures the Normalized Enrichment Score from minus 2 to 2. Blue bars represent anxiety and red bars represent depression. Lipoprotein subclasses show the strongest positive enrichment, while Amino acids show significant negative enrichment.
Panel D displays two dot plots showing associations by W G C N A modules for depression and anxiety. Five modules are color-coded: red, blue, green, orange, and purple. The x-axis represents the Hazard Ratio from 0.90 to 1.10. Module 3 in green shows a significant positive association with H R 1.07 for depression, while Module 5 in purple shows a significant inverse association with H R 0.92.
MSEA and WGCNA were applied to detect depression/anxiety-associated metabolite groups/modules. MSEA identified six biologically annotated metabolite clusters showing disorder-specific associations. Lipoprotein subclasses and triglycerides demonstrated positive correlations with both disorders, whereas amino acid and relative lipoprotein lipid concentration redistribution patterns exhibited protective associations (Figure 4c). WGCNA delineated five co-expression modules, with Module 4 demonstrating positive associations (Depression HR = 1.07, 95% CI: 1.05–1.09; Anxiety HR = 1.05, 1.03–1.07) and Module 5 showing protective effects (Depression HR = 0.92, 0.91–1.05; Anxiety HR = 0.95, 0.93–0.97) for both psychiatric outcomes (Figure 4d).
Mediation analyses
Mediation analysis revealed that metabolic signatures significantly mediated the relationship between IC domains and both depression and anxiety. Specifically, metabolic signatures acted as mediators in the associations of IC, psychological, vitality, and locomotor domains with depression (mediation-adjusted HRs [95% CI]: 1.007 [1.005–1.008], 1.010 [1.008–1.011], 1.003 [1.002–1.005], and 1.007 [1.005–1.009], respectively). Similar mediation effects were observed for anxiety, with HRs of 1.006 (1.004–1.007), 1.007 (1.005–1.009), 1.003 (1.001–1.004), and 1.006 (1.004–1.007) for the same domains (Figure 5). The IC-related metabolic signature mediated 2.39% (95% CI: 1.73%–2.94%) of the total association between IC and depression and 2.41% (95% CI: 1.79%–3.16%) of the association between IC and anxiety. Domain-specific mediation proportions were generally modest, with the largest proportions observed for the psychological metabolic signature: 3.37% (95% CI: 2.73%–3.98%) for depression and 3.17% (95% CI: 2.22%–3.99%) for anxiety. Single-metabolite mediation analysis identified triglycerides in large LDL, triglycerides in medium LDL, and the polyunsaturated fatty acids-to-monounsaturated fatty acids ratio as metabolites with the strongest mediating effects.
Mediating effects of IC-associated metabolic signatures on IC–depression/anxiety disorder relationships. Significance thresholds: *p < 0.05, **p < 0.01, and ***p < 0.001.
Figure 5. Long description
The top panel is a mediation flow diagram. At the center is a vertical orange box labeled Intrinsic capacity (I C). To the left is a blue box for Depression, and to the right is a blue box for Anxiety. Between the center and the sides are five horizontal orange boxes representing metabolic signatures: I C, Psychological, Sensory, Vitality, and Locomotor. Each signature lists Mediation effect, Direct effect, and Proportion mediated with statistical significance indicated by asterisks. Arrows indicate the flow from I C through the signatures to the mental health outcomes.
The bottom left panel is a scatter plot titled Proportion Mediated. The y-axis ranges from minus 0.005 to 0.010. Data points are divided into red (positive values) and blue (negative values). Red points include labels like Unsaturation, L_L D L_T G percent, and M_L D L_T G percent. Blue points include labels like Total_Cholines and Phosphatidylcholines. Triangles represent depression and circles represent anxiety.
The bottom right panel is a scatter plot titled Total Natural Indirect Effect. The y-axis ranges from 0.999 to 1.002. All data points are red. High-value outliers are labeled Unsaturation, L_L D L_T G percent, and P U F A forward slash M U F A. A legend at the bottom indicates that circles represent anxiety and triangles represent depression.
The joint effect of genetic risk and IC deficit score
Depression-PRS tertiles demonstrated a graded increase in depression risk, with similar dose–response relationships for anxiety-PRS and anxiety, revealing disorder-specific polygenic risk (Supplementary Figure S3). The HRs (95% CI) were 1.14 (1.09–1.20) and 1.27 (1.20–1.35) for depression in intermediate- and high-PRS groups compared to low PRS groups. The HRs (95% CI) were 1.06 (1.00–1.11) and 1.15 (1.09–1.22) for anxiety in intermediate- and high-PRS groups compared to low-PRS groups (Supplementary Figure S4A). Furthermore, we evaluated the joint association of IC and PRS with depression or anxiety (Supplementary Figure S4B). The interaction between IC and PRS on the outcome is shown in Supplementary Table S9, indicating an additive interaction but no multiplicative interaction. The highest hazard was observed in participants with high PRS and a 4+ IC deficit score; the HRs were 4.36 (95% CI: 3.70–5.13) for depression and 3.46 (95% CI: 2.87–4.17) for anxiety, compared to those with a low PRS and 0 IC score.
Subgroup analysis and sensitivity analyses
Significant interactions of baseline features with IC were found most of the time, with both additive and multiplicative interactions occurring either simultaneously or individually (P < 0.05; Supplementary Table S10). Generally, stronger effect estimates were observed in females, younger adults (aged <65 years), individuals with a college education, and those with high incomes (>£52,000) for both depression and anxiety (Supplementary Table S11). Results from the stratified analysis by drinking frequency showed that the weakest association between IC and depression and anxiety was observed among individuals who were previous drinkers (Supplementary Figure S5). Findings from multiple sensitivity analyses confirmed the robustness of the primary analysis results (Supplementary Tables S12–S15).
Discussion
In this UK Biobank analysis, we observed significant associations between IC, its metabolite profile, and elevated risks of depression and anxiety. The IC-associated metabolome encompassed lipoprotein lipid fractions, fatty acids, lipid subclasses, and amino acids. These metabolites collectively exhibited strong correlations with both psychiatric conditions. Aggregating these metabolites into an IC-related metabolic signature revealed its role in mediating the association between IC and the development of depression and anxiety. We also observed an additive interaction between IC and genetic susceptibility. These findings suggest that IC-associated depression and anxiety risk may be partially mediated through metabolomic perturbations, with effect magnitude modulated by genetic susceptibility.
To our knowledge, this study provides the first population-based cohort evidence demonstrating the longitudinal association between IC and new-onset depressive and anxiety disorders. Our findings of significant positive associations align with prior evidence reporting elevated IC prevalence in depressed populations versus nondepressed counterparts, while extending current understanding by delineating mediating metabolic pathways (Chan, Okubo, Brodie, & Lord, Reference Chan, Okubo, Brodie and Lord2020; Hoogendijk et al., Reference Hoogendijk, Afilalo, Ensrud, Kowal, Onder and Fried2019; Liu et al., Reference Liu, Yu, Wang, Li, Jiang, Kang and Liu2021). Nevertheless, the etiological mechanisms linking IC to depression and anxiety remain poorly understood. Existing literature has predominantly focused on individual IC-related components, particularly sleep patterns and muscular strength metrics, in relation to susceptibility to mental health conditions (Hou et al., Reference Hou, Chen, Sun, Song, Deng, Cheng, Han, Zhang, Wang, Li, Yin and Meng2023; Lopresti, Hood, & Drummond, Reference Lopresti, Hood and Drummond2013; Lyall et al., Reference Lyall, Sangha, Zhu, Lyall, Ward, Strawbridge, Cullen and Smith2023). For example, although Lyall et al. used UK Biobank data to show that sleep-related pathology and related factors may predict future depression risk, previous studies have rarely evaluated the combined effects of multiple available IC domains together with metabolomic profiles and genetic susceptibility. Emerging evidence indicates elevated depression/anxiety risks associated with both insomnia pathophysiology (Hou et al., Reference Hou, Chen, Sun, Song, Deng, Cheng, Han, Zhang, Wang, Li, Yin and Meng2023) and reduced muscular function as measured by grip strength (Kandola et al., Reference Kandola, Osborn, Stubbs, Choi and Hayes2020), suggesting distinct biological pathways contributing to mental health vulnerability. Our study addresses this gap by demonstrating the joint contribution of multiple IC deficits. The psychological domain had the greatest contribution to both depression and anxiety, while the locomotor domain had the least.
We first developed a metabolomic signature of IC, which was then integrated with IC deficit status in joint analyses to systematically evaluate associations with incident depression and anxiety. As hypothesized, higher metabolomic signature scores for IC were associated with increased risk of subsequent depression and anxiety. Our identification of IC-related metabolic perturbations corroborates prior work (Sha et al., Reference Sha, Wang, Wei, He, Wang, Zeng and Lei2023), while the observed depression/anxiety associations align with prospective cohort findings (Chourpiliadis et al., Reference Chourpiliadis, Zeng, Lovik, Wei, Valdimarsdóttir, Song, Hammar and Fang2024; Morgese et al., Reference Morgese, Schiavone, Maffione, Tucci and Trabace2020; Wang et al., Reference Wang, Dong, Pan, Xu, Lu, Yin, Wang, Xia, Liao and Sun2022). The mechanistic nexus between IC and metabolomic alterations remains incompletely characterized, though emerging gut microbiota research offers plausible pathways given documented IC–microbiome interactions (Mueller et al., Reference Mueller, Differding, Zhang, Maruthur, Juraschek, Miller, Appel and Yeh2021; Wang et al., Reference Wang, Wang, Lu, Chen, Yan, Yuan, Shi, Liu, Zhou, Shi, Vitiello, Han and Lu2022) and microbial regulation of metabolic profiles (Bar et al., Reference Bar, Korem, Weissbrod, Zeevi, Rothschild, Leviatan, Kosower, Lotan-Pompan, Weinberger, Le Roy, Menni, Visconti, Falchi, Spector, Adamski, Franks, Pedersen and Segal2020; Chen et al., Reference Chen, Zhernakova, Kurilshikov, Andreu-Sánchez, Wang, Augustijn, Vich Vila, Weersma, Medema, Netea, Kuipers, Wijmenga, Zhernakova and Fu2022).
Our study provides a quantitative assessment suggesting that IC-related metabolic alterations may partially mediate the association between IC and incident depression/anxiety risk. The mediation effect sizes were statistically significant but modest, suggesting that metabolic pathways may represent one component of a broader multifactorial process rather than the primary explanatory mechanism. By consolidating multiple IC-associated metabolites into a single metabolic signature, we conducted a comprehensive mediation analysis. Pathway analysis revealed that IC was associated with alterations in energy and amino acid metabolism. Mediation analysis further highlighted a potential metabolic pathway linking IC to depression/anxiety risk, involving metabolomic signatures such as relative lipoprotein lipid concentrations, fatty acids, amino acids, and lipoprotein subclasses. These associations may involve metabolic regulation, neurotransmitter homeostasis, and immune-inflammatory activation (Liu et al., Reference Liu, Fu, Yu, Li, Pan, Yang and Sun2024; Lu et al., Reference Lu, González-Bautista, Guyonnet, Martinez, Lucas, Parini, Rolland, Vellas and de Souto Barreto2023). Additionally, changes in brain morphology may represent another possible mechanism connecting lifestyle with depression/anxiety (Figee et al., Reference Figee, Riva-Posse, Choi, Bederson, Mayberg and Kopell2022; Kemp, Bernier, Lebel, & Kopala-Sibley, Reference Kemp, Bernier, Lebel and Kopala-Sibley2022). Specifically, higher lifestyle ratings have been associated with greater volumes in the orbitofrontal cortex and medial prefrontal cortex, suggesting potential relevance to cognitive control and emotional regulation (Zhang et al., Reference Zhang, Rolls, Wang, Xie, Cheng and Feng2024). Adequate sleep and grip strength may also be related to hippocampal synaptic plasticity and emotional responses to future behavioral stresses (Aime et al., Reference Aime, Calcini, Borsa, Campelo, Rusterholz, Sattin, Fellin and Adamantidis2022; Firth et al., Reference Firth, Smith, Sarris, Vancampfort, Schuch, Carvalho, Solmi, Yung, Stubbs and Firth2020).
This study further evaluated the moderating role of genetic susceptibility on IC–depression/anxiety associations by examining PRS interactions with metabolic pathways. Our findings suggest that populations with high genetic risk may experience an enhanced association between IC and mental disorders such as depression and anxiety. Individuals with high PRS for depression/anxiety may exhibit intrinsic vulnerabilities in neuroendocrine-immune regulation (Otowa et al., Reference Otowa, Hek, Lee, Byrne, Mirza, Nivard, Bigdeli, Aggen, Adkins, Wolen, Fanous, Keller, Castelao, Kutalik, der Auwera, Homuth, Nauck, Teumer, Milaneschi and Hettema2016; Wray et al., Reference Wray, Ripke, Mattheisen, Trzaskowski, Byrne, Abdellaoui, Adams, Agerbo, Air, Andlauer, Bacanu, Bækvad-Hansen, Beekman, Bigdeli, Binder, Blackwood, Bryois, Buttenschøn and Bybjerg-Grauholm2018). These genetic predispositions may make them more sensitive to IC-mediated metabolic disturbances, potentially disrupting neurotransmitter balance and synaptic plasticity, thereby amplifying the association between elevated IC levels and symptoms of depression/anxiety. This suggests that routine screening and intervention for IC may offer substantial benefits, especially in populations with high genetic susceptibility. The lack of multiplicative interaction could be partly due to the limited genetic risk explained by the SNPs included in our analysis (Zhao et al., Reference Zhao, Yang, Sahakian, Langley, Zhang, Kuo, Li, Gan, Li, Zhao, Yu, Feng and Cheng2023).
Our study has several strengths. Our findings provide novel prospective evidence elucidating IC–depression/anxiety relationships, previously understudied epidemiologically. The study’s methodological strengths include population-level sample size, extended follow-up, and innovative application of elastic net regression to develop an IC-characterizing metabolomic signature. Identification of 59 IC-associated metabolites creates translational opportunities for risk stratification biomarkers, while characterization of gene–metabolic interactions advances understanding of susceptibility modulation. Furthermore, the extensive data available from the UK Biobank allowed us to account for a wide range of confounders, including genetic predisposition to depression and anxiety. However, this study had several limitations. First, although the IC score was adapted from WHO-defined IC domains, it was based on four available domains and did not include the cognitive domain because comprehensive cognitive domain data were unavailable in the UK Biobank. This limitation means that our score cannot fully represent the complete five-domain WHO IC construct. In particular, the vitality domain was represented by weight loss and low grip strength, which capture only selected nutritional and neuromuscular aspects and do not fully cover energy metabolism, immune function, or stress-response components of the WHO vitality capacity framework. Because cognitive function is closely related to depression and anxiety, its omission may have led to under-capture of overall IC impairment and may have attenuated or otherwise incompletely represented the association between IC and depression/anxiety. As comprehensive cognitive domain data were unavailable, the exact magnitude of this potential bias could not be quantified. Although our sensitivity analysis incorporating reaction time showed consistent results, reaction time is only a limited proxy for cognitive capacity and cannot fully capture the cognitive domain. In addition, because the score was constructed by accumulating deficits across available indicators, it shares features with a frailty index approach. We therefore interpret it as a UK Biobank-adapted four-domain IC deficit score. Second, notwithstanding rigorous adjustment for recognized confounders, residual confounding may persist due to unaccounted variables not captured in the current analytical framework. Finally, the study population consisted predominantly of elderly/middle-aged White European ancestry participants, necessitating validation in multiethnic cohorts across age spectra.
Conclusions
In conclusion, a higher UK Biobank-adapted four-domain IC deficit score was prospectively associated with increased risks of incident depression and anxiety, with stronger associations among individuals with higher polygenic susceptibility. IC-related metabolomic signatures were associated with both IC status and mental health outcomes and showed modest mediation effects. These findings support the potential relevance of functional capacity and metabolic profiles for risk stratification. Further studies are needed to clarify causal mechanisms and intervention implications.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726106047.
Data availability statement
Researchers may obtain datasets through formal application procedures at the UK Biobank portal (https://www.ukbiobank.ac.uk/).
Acknowledgments
Data were analyzed under UK Biobank Application 93398. The authors would like to express their gratitude to the UK Biobank research team, the field team, and every participant.
Author contribution
Conceptualization: G.K. and Y.B. Data curation: Y.B. Funding acquisition: G.K. and Y.B. Methodology: S.S. Project administration: X.S. Software: S.S. Supervision: S.L. and Y.Z. Validation: X.S., J.G., and C.L. Writing – original draft: S.S. and X.S. Writing – review and editing: J.G., C.L., S.L., Y.Z., and Y.B.
Funding statement
The complete Funding statement should read:This work was supported by the following grants: National Natural Science Foundation of China (No. 82504406); Open Research Fund of the National Health Commission Key Laboratory of Birth Defects Prevention and the Henan Key Laboratory of Population Defects Prevention (ZD202303); Henan Medical Science and Technology Research Program (232102310069); and 2022 International Postdoctoral Exchange Fellowship Program [Talent-Introduction Program] (YJ20220181). All authors reviewed and approved the final manuscript prior to journal submission.
Competing interests
The authors declare none.
Ethical standards
All participants provided informed consent in accordance with the Declaration of Helsinki principles. Ethical approval was obtained from the NHS National Research Ethics Service (11/NW/0382). Full ethical documentation is publicly accessible through UK Biobank’s institutional portal (https://www.ukbiobank.ac.uk/learn-more-about-uk-biobank/about-us/ethics).
Open access
