Assessing and refining a metabolite-based distress score for use in different populations within three US cohorts

Abstract

Introduction:

Psychological distress is associated with metabolic alterations that contribute to adverse cardiometabolic outcomes across diverse populations. We previously developed a metabolite-based distress score (MDS) in predominantly White women. Because men and Black individuals suffer high rates of cardiometabolic disease, we evaluated whether the MDS is similarly associated with depression in these populations, and derived a multi-ethnic MDS (MDS-ME) among Black individuals with replications in independent samples.

Methods:

Data were from three U.S. cohorts, the Multi-Ethnic Study of Atherosclerosis (MESA), Women’s Health Initiative (WHI), and Nurses’ Health Study (NHS), evaluated in 4 stages. Stages 1-3 included 2,477 White and 1,625 Black MESA participants with depression status data and plasma metabolomics assessed at baseline. Stage 1: We calculated the MDS and examined its association with depression status. Stage 2: We conducted an agnostic analysis among a 70% random subset of Black participants to identify additional relevant metabolites. Stage 3: We derived an MDS-ME in Black participants. Stage 4: We evaluated the MDS-ME using independent samples from all three cohorts.

Results:

The previously validated 20-metabolite MDS showed stronger associations with prevalent depression in White (OR: 1.93, 95% CI: 1.71, 2.19) versus Black (OR: 1.13, 95% CI: 0.97, 1.33) MESA participants. Associations were generally similar in men and women. The MDS-ME comprised 33 metabolites, and relative to the MDS was more strongly associated with depression in Black participants (OR: 1.45; 95% CI: 1.04, 2.03), with associations also evident in White participants (OR: 1.81; 95% CI: 1.60, 2.04). The MDS-ME was similarly associated with depression in Black women in WHI and NHS.

Conclusion:

Results suggest that incorporating metabolites identified in Black participants enhances the utility and generalizability of the MDS across more diverse populations.

Introduction

Multiple forms of psychological distress, such as depression, are increasingly recognized as key risk factors for cardiometabolic disease and as contributors to health disparities (Djuric et al., 2008; Talala et al., 2011; Turner, 2013; Vaccarino et al., 2025). The ways in which psychological distress is experienced, expressed, and biologically embedded may vary across population groups. Cultural norms, social contexts, and gender roles can influence how distress is perceived and reported, as well as how distress may engage downstream biologic pathways (Assari, 2017; Lincoln et al., 2011; Pamplin and Bates, 2021). Such norm- and context-related variation may differ across racial groups and between men and women, potentially contributing to heterogeneity in the biological correlates of psychological distress. However, the pathways linking psychological distress to cardiometabolic risk, including distress-associated metabolic profiles that may differ by race and sex, remain poorly understood.

Metabolic alterations are important biological correlates of psychological distress and may provide insight into pathways through which distress promotes cardiometabolic disease risk. Metabolomic studies have shown that perturbations in amino acid, lipid, and energy metabolism are associated with depressive symptoms and chronic stress, suggesting that metabolic pathways may serve as key mediators leading to adverse physical health outcomes (Bot et al., 2020; Caspani et al., 2021; Huang et al., 2021; Miao et al., 2023; Shutta et al., 2021; Zhu et al., 2022). In recent work in predominantly White women, we identified and validated a metabolite-based distress score (MDS) consisting of 20 plasma metabolites that showed robust differences across levels of psychological distress captured by a composite measure of depression and anxiety (Balasubramanian et al., 2023; Shutta et al., 2021). Further, the MDS was associated with increased risk of incident diabetes and cardiovascular disease among predominantly White men and women, independent of conventional cardiometabolic risk factors (Balasubramanian et al., 2023; Huang et al., 2024). However, this MDS has not yet been evaluated in more diverse populations, particularly among Black individuals or men, populations that suffer high rates of cardiometabolic disease.

Interestingly, a growing body of evidence highlighted considerable differences in plasma metabolomic profiles between Black and White individuals, which have been partly attributed to behavioral and environmental factors (Butler et al., 2023; Hu et al., 2022; McGee et al., 2024). Notable metabolomic differences were also observed between men and women, with patterns that varied dynamically with age (Bell et al., 2021; Costanzo et al., 2022). It is possible that the MDS previously developed in primarily White, post-menopausal women may not fully reflect distress-related metabolic profiles in Black individuals and men, which may be partly attributable to differences in socioenvironmental exposures, lifestyles, stress physiology, and/or genetic or epigenetic regulation.

Therefore, the objectives of the current study are (1) to evaluate the association of our previously developed MDS with depression in White and Black men and women and (2) as appropriate, to evaluate if additional distress-related metabolites might better capture a metabolite profile of distress in these groups by deriving a multi-ethnic MDS (MDS-ME). To address these objectives, we conducted this work across multiple U.S. cohorts, including the Multi-Ethnic Study of Atherosclerosis (MESA), the Women’s Health Initiative (WHI) and the Nurses’ Health Study (NHS). The study was conducted in four stages to evaluate, identify, revise, and replicate the metabolomic signature of distress in Black participants, and secondarily, to assess potential sex differences (Figure 1). This multi-cohort, multi-stage approach allows us to build upon our prior work, assess the consistency of observed associations, identify cohort-specific differences, and strengthen the robustness and reproducibility of the results. Understanding racial and sex differences in the metabolic manifestations of distress is critical for advancing mechanistic insight into health disparities.

Figure 1.

Flow chart for the analytic samples. Note that our analysis was restricted to Black and White participants in the Multi-Ethnic Study of Atherosclerosis (MESA), as they were the ancestry groups for which replication was possible in the Women’s Health Initiative (WHI and the Nurses’ Health Study (NHS). Participants from other ethnic groups in MESA (Hispanics and Chinese) were not included in this study.

Methods

Study population

Data were from three US cohorts (MESA, WHI, and NHS). MESA is a prospective observational cohort that recruited 6,814 participants free of cardiovascular disease from six US communities in 2000-2004 (baseline age: 45-84 years) (Bild et al., 2002). The WHI consists of 161,808 postmenopausal women who were recruited from 40 clinical centers across the US in 1993-1998 (baseline age: 50-79 years) (1998). The NHS is an ongoing, longitudinal cohort of 121,700 female registered US nurses who completed biennial follow-up surveys since enrollment in 1976 (baseline age: 30-55 years) (Colditz and Hankinson, 2005). In all cohorts, race was self-reported by participants.

In our four-stage study (Figure 1), Stages 1-3 use data from 2,477 White and 1,625 Black MESA participants with complete data on plasma metabolomics and depression-related measures. The analysis was further restricted to a 70% training subset of Black participants in Stages 2-3 to derive the MDS-ME. In Stage 4, to provide additional independent assessment of the MDS-ME in MESA, we utilized the remaining withheld 30% subset of MESA Black participants, as well as additional Black and White women from WHI and NHS for whom at least a subset of the newly identified metabolites in the MDS-ME were available. In WHI, our analytic sample included 149 Black women and 1,107 White women who participated in the Observational Study (WHI-OS) and were included in a previous nested case-control study of plasma metabolomics and coronary heart disease (Paynter et al., 2018). The NHS analysis included 214 Black women and 1856 White women who were included in a prior analysis of racial differences in metabolomic profiles (McGee et al., 2024). Of note, while the original 20-metabolite MDS was developed in samples of primarily White women from WHI and NHS, the replication samples included in the current study were independent, with no overlap with our prior analyses.

Depression assessment

In MESA, following previous work in this area, depression status at baseline was defined by the presence of one or both of the following: (1) current antidepressant use or (2) a score ≥16 on the 20-item Center for Epidemiologic Studies Depression Scale (CES-D), an established cutoff for probable depression (Radloff, 1977). In WHI and NHS, depression status was ascertained according to use of antidepressants or obtaining a score above the specified cut point on a validated depressive symptom measure, with a third element used in each dataset to capture self-reported depression history. Specifically, in WHI, depression was defined as meeting any of the following: (1) current antidepressant use, (2) CES-D score ≥16, or (3) self-reported history of depressed mood. In NHS, using depression measures obtained on surveys closest to blood collection (range 2-6 years after blood collection), we defined depression as: (1) current antidepressant use, (2) scoring ≤52 on the 5-item Mental Health Index (MHI-5), an established cut point previously used with this measure,(Berwick et al., 1991) or (3) self-reported history of receiving a clinician’s diagnosis of depression.

Metabolomic measurement

Plasma metabolomics profiling was performed at the Broad Institute of the Massachusetts Institute of Technology and Harvard using liquid chromatography tandem mass spectrometry (LC-MS) platforms for all three samples. In MESA, data were generated from four platforms: (1) the C8-positive platform primarily detects moderately hydrophobic metabolites including lipids and sterol derivatives, (2) the C18-negative platform is optimized for compounds of intermediate polarity, such as fatty acids, bile acids and complex lipids, (3) the HILIC-positive platform measures polar, water-soluble metabolites, including amino acids, carnitines, and biogenic amines, and (4) the HILIC-negative platform captures polar and acidic metabolites such as organic acids and carbohydrates. Together, these platforms capture a broad range of identified metabolites (n=764 metabolites). Details on profiling methods have been described previously (Clish, 2015; Paynter et al., 2018).

In WHI, metabolomics data were available for the same four platforms; however, because these data were generated several years prior to the MESA profiling, they featured a smaller number of annotated metabolites identified at the time (n=472 metabolites) (Hu et al., 2022; Paynter et al., 2018). In NHS, metabolomics data were available from only two platforms (HILIC-positive and C8-positive) and also had limited coverage (n=334) (McGee et al., 2024).

Covariates.

We examined two sets of covariates to provide estimates adjusted for key confounders (Model 1) and for a wider range of medical and biobehavioral factors (Model 2). In MESA, covariates were assessed at baseline, concurrent with blood collection. Model 1 included age (years), sex, study site, education (high school or less, some college, college graduate, graduate school), employment status (homemaker, employed, unemployed, retired), and household income (<$20,000, $20,000-$49,999, $50,000-$99,999, ≥$100,000). Model 2 further adjusted for BMI (continuous), smoking status (current, former, never), physical activity (MET hrs/week, continuous), diet quality scores (continuous), systolic blood pressure (continuous), HDL cholesterol (continuous), total cholesterol (continuous), diabetes status (yes, no), and use of antihypertensive and lipid-lowering medications (yes, no). We attempted to adjust for comparable sets of covariates in WHI and NHS; however, a small number of covariates could not be included because of data limitations. Details on covariate assessment are provided in Supplemental Table 1.

Statistical analysis

Stage 1: Evaluation of the original MDS.

Using data from MESA White and Black participants, we calculated the MDS based on 20 metabolites previously identified (Supplemental Table 2).¹⁴ Each metabolite was log-transformed and z-scored separately among White and Black participants before being weighted by its corresponding coefficient to generate the 20-metabolite MDS, which was then z-scored within each racial group. Using logistic regression, depression status was regressed on the standardized MDS, such that the odds ratio (OR) represented the odds of probable depression per 1-SD increase in the MDS. All analyses were conducted separately in White and Black participants, and within each racial group we further stratified by sex to evaluate potential differences between men and women. Statistical interactions by sex were assessed using a likelihood ratio test comparing models with and without the cross-product term. Two multivariable-adjusted models were considered in these analyses as described above.

Stage 2: Identification of additional distress-related metabolites in Black individuals.

We randomly divided the MESA Black participants into two subsets (70% for discovery and 30% for replication). We conducted an agnostic analysis assessing associations between depression and 764 metabolites in the 70% discovery subset (n=1,140) to identify additional metabolites potentially associated with depression among Black participants beyond the 20 metabolites included in the original MDS. Following a similar strategy used for developing the original MDS, we used logistic regression to estimate the odds ratio (OR) and 95% CI of probable depression for each metabolite (standardized to z-scores), applying the same two multivariable-adjusted models described above. Because our primary objective was to identify informative metabolites that may enhance the original MDS in Black participants rather than to make definitive causal inferences, we used a liberal inclusion threshold (p<0.05) to select candidate metabolites, prioritizing sensitivity over strict control of false positives.

Stage 3: Development of the MDS-ME in MESA Black participants.

Metabolites nominally associated with depression in the previous step (p<0.05) were considered as candidates for the MDS-ME. Further, to build the MDS-ME based on our prior work, all 20 metabolites included in the original MDS were also included as candidates regardless of whether their associations reached the nominal significance level among Black participants. To select metabolites, candidate metabolite z-scores were entered into a LASSO logistic regression model as explanatory variables, with depression status as the response variable (Friedman et al., 2010; Tibshirani, 1996). Model tuning was performed using 10-fold cross-validation to determine the optimal penalty parameter that minimized the cross-validated deviance. Metabolites with nonzero coefficients in the optimal model were retained. The MDS-ME was calculated as the weighted sum of these selected metabolites using their LASSO-derived coefficients.

Stage 4: Evaluation of the MDS-ME with depression in three independent samples.

To evaluate reproducibility and generalizability, we evaluated the association of the MDS-ME with depression in the independent 30% replication set of MESA Black participants (n=485). We also evaluated the performance of the MDS-ME among MESA White participants (n=2,477) and assessed potential sex differences in White and Black men and women. To assess the generalizability of the MDS-ME in other cohorts, we evaluated the MDS-ME in independent samples from the WHI (149 Black and 1,107 White women) and NHS (214 Black and 1856 White women). Of note, only a subset of metabolites identified for the MDS-ME in MESA were available in WHI and NHS, with slightly different subsets available in each cohort. We recreated a reduced MDS-ME in MESA by including only metabolite subsets that were available in WHI and NHS, and compared them with the full MDS-ME in MESA. We adjusted for comparable sets of covariates in WHI and NHS as possible; however, a small number of covariates could not be included because of data limitations (Supplemental Table 1). We implemented random-effects meta-analysis of the fully-adjusted estimates of the associations of MDS-ME with depression among Black participants, with inference based on the Hartung-Knapp-Sidik-Jonkman method to assess potential heterogeneity across cohorts (IntHout et al., 2014). All analyses were conducted in RStudio (version 4.3.5).

Results

Sample characteristics

In MESA, the mean age (SD) was 62.6 (10.2) years in 2,477 White participants and 62.2 (10.1) years in 1,625 Black participants (Table 1). Despite slightly higher depressive symptoms (both continuous and binary CESD measures) in Black compared with White participants, antidepressant use was more common in White participants. As a result, probable depression was more prevalent in White (19.9%) than in Black participants (14.3%). Compared with White participants, Black participants tended to report lower educational attainment and household income, unemployment or retirement, current smoking, higher physical activity and poorer diet quality. On average, Black versus White participants exhibited higher systolic blood pressure and greater use of antihypertensive medications, lower total cholesterol, and less use of lipid-lowering medications. BMI was higher and diabetes was more prevalent among Black participants. Similar differences were observed between White and Black women in WHI and NHS (Supplemental Table 3).

Table 1.

Sample characteristics in White and Black participants from MESA

	White	Black	P1
N	2477	1625
CESD	6.9 (6.9)	7.3 (7.3)	0.10
CESD≥16, %	10.1	11.7	0.13
Antidepressant use, %	12.6	3.8	<0.001
Depression status, %	19.9	14.3	<0.001
Sociodemographic factors
Age, yrs	62.6 (10.2)	62.2 (10.1)	0.21
Men, %	47.6	46.2	0.38
Study site, %			<0.001
Baltimore County, MD	20.1	30.3
Chicago, IL	21.0	15.9
Forsyth County, NC	21.7	26.7
New York, NY	8.2	17.7
Los Angeles County, CA	5.2	9.4
St Paul, MN	23.9	0.0
Education, %			<0.001
High school or less	21.9	31.0
Some college	28.3	35.3
College graduate	22.3	18.0
Graduate school	27.5	15.7
Household income, %			<0.001
<$20,000	10.7	20.1
$20,000-$49,999	31.2	37.8
$50,000-$99,999	32.1	34.3
>$100,000	26.0	7.8
Employment status, %			<0.001
Homemaker	11.0	6.8
Employed	51.4	44.9
Unemployed	1.4	2.8
Retired	36.2	45.6
Clinical factors
Systolic blood pressure, mmHg	123.4 (20.4)	131.8 (21.8)	<0.001
HDL cholesterol, mg/dL	52.4 (15.8)	52.2 (15.1)	0.74
Total cholesterol, mg/dL	196.0 (35.4)	189.3 (36.2)	<0.001
Diabetes status, %	6.1	17.1	<0.001
Antihypertensive medications, %	33.4	50.0	<0.001
Lipid-lowering medications, %	18.4	16.7	0.20
Lifestyle factors
BMI, kg/m2	27.7 (5.1)	30.1 (5.9)	<0.001
Current smokers, %	11.4	18.1	<0.001
Physical activity, MET-hrs/week	94.6 (89.5)	109.0 (115.0)	<0.001
Alternative Healthy Eating Index	53.5 (10.5)	51.2 (9.8)	<0.001

¹P-values were calculated by ANOVA for continuous variables and chi-square tests or Fisher’s exact tests for categorical variables.

Stage 1: Evaluation of the original MDS

The 20-metabolite MDS, previously developed from independent samples of primarily White women, was strongly associated with depression status in MESA White participants (Table 2). After adjusting for age, sex, study site, education, household income and employment status, each 1-SD increment in the MDS was associated with 86% higher odds of depression (95% CI: 1.67, 2.09). The association became somewhat stronger with further adjustment of clinical and lifestyle factors (OR: 2.00; 95% CI: 1.77, 2.25). When examining associations stratified by sex, no appreciable differences were observed between White women (n=1,298) and men (n=1,179) in either model (p-interaction>0.88; Table 2). In contrast, the association between the original MDS and depression status was substantially weaker among Black participants (comparable OR: 1.16; 95% CI: 1.00, 1.35) after adjusting for sociodemographic factors and remained essentially the same after additionally adjusting for clinical and lifestyle factors. When considering associations in Black men (n=750) and women (n=875) separately, the association appeared stronger in men (OR: 1.35; 95% CI: 1.03, 1.78) but was null in women (OR: 1.06; 95% CI: 0.86, 1.30; p-interaction=0.08).

Table 2.

Associations between the original 20-metabolite MDS (per 1-SD increment) and depression status in White and Black participants from MESA

		Model 11		Model 22
	n	OR (95% CI)	p-value	OR (95% CI)	p-value
White participants
Overall	2477	1.83 (1.64, 2.05)	6.0E-26	1.93 (1.71, 2.19)	2.5E-26
Women	1298	1.81 (1.58, 2.09)	1.6E-16	1.94 (1.67, 2.27)	6.0E-17
Men	1179	1.83 (1.52, 2.22)	3.5E-10	1.91 (1.56, 2.35)	4.6E-10
p-int by sex		0.90		0.88
Black participants
Overall	1625	1.16 (1.00, 1.35)	0.06	1.13 (0.97, 1.33)	0.12
Women	875	1.06 (0.87, 1.29)	0.55	1.06 (0.86, 1.30)	0.60
Men	750	1.37 (1.06, 1.76)	0.01	1.35 (1.03, 1.78)	0.03
p-int by sex		0.04		0.08

¹Ajusted for age, sex, study site, education, income, and employment status

²Model 1 + adjusted for BMI, smoking status, physical activity, diet quality, systolic blood pressure, HDL cholesterol, total cholesterol, diabetes status, antihypertensive medications, and lipid-lowering medications

Stage 2: Identification of additional distress-related metabolites in Black individuals

Given the weaker association between the original MDS and depression among Black participants and especially in Black women, we performed an agnostic analysis by examining the associations of individual 764 metabolites with depression in a subset of MESA Black participants (n=1,140; Supplemental Table 4). In Model 1, 57 metabolites were nominally associated with depression, while in Model 2 (fully-adjusted) 46 metabolites were nominally associated of which 41 were identified in Model 1. Based on Model 2 (Figure 2), the three metabolites most strongly associated with lower odds of depression were serotonin (OR: 0.64, 95% CI: 0.53, 0.77), hydroxytryptophan (OR: 0.67, 95% CI: 0.57, 0.80), and adenosine (OR: 0.73, 95% CI: 0.62, 0.87). Other metabolites inversely associated with depression were primarily various carnitine species. Metabolites positively associated with depression mainly included PCs, LPCs, LPEs, long-chain saturated fatty acids (arachidic and nonadecylic acid) and several metabolites related to niacin metabolism (trigonelline, N1-methyl-2-pyridone-5-carboxamide, N1-methyl-4-pyridone-3-carboxamide). Notably, only 3 metabolites overlapped with the previously developed 20-metabolite MDS, including serotonin, thiamine and LPE 18:0/0:0. Further, while γ-aminobutyric acid (GABA) was included in the original MDS but did not reach nominal significance in our sample of Black individuals, the GABA derivative, 4-acetamidobutanoic acid, was identified.

Figure 2.

Metabolites (n=46) associated with depression status among Black MESA participants (n=1,140). Estimates represent the odds ratios for depression per one-unit increase in each metabolite z-score, adjusted for age, sex, study site, education, income, employment status, BMI, smoking status, physical activity, diet quality, systolic blood pressure, HDL cholesterol, total cholesterol, diabetes status, antihypertensive medications, and lipid-lowering medications.

Stage 3: Development of the MDS-ME in MESA Black participants

LASSO selection was conducted based on 63 metabolites, including 46 nominally associated in Stage 2 in MESA Black participants and 20 previously identified in White participants (with 3 overlapping metabolites between the two sets). Of these, 33 metabolites had non-zero coefficients in LASSO selection, including 23 metabolites associated with depression in Black participants and 10 present in the original MDS (Table 3). The MDS-ME based on these 33 metabolites was correlated with a Pearson correlation of 0.33 (p<0.001) with the original MDS in this discovery sample of Black participants.

Table 3.

Metabolites selected by LASSO from 63 candidates¹

Metabolites	HMDB ID	Coefficients for the selected metabolites (n=33)	Metabolites available in WHI (n=23)	Metabolites available in NHS (n=17)
Newly identified in Black participants
Hydroxytryptophan	HMDB0000472	−0.118185703	Yes	Yes
Adenosine	HMDB0000050	−0.133749853	Yes	Yes
Trigonelline	HMDB0000875	0.068687941	Yes	Yes
Inosine	HMDB0000195	−0.084331067	No	No
CAR 12:2	HMDB0241240	−0.069096016	No	No
LPC 18:1/0:0	HMDB0002815	.
DG 34:0	HMDB0007100	0.031482078	Yes	No
DG 38:5	HMDB0007199	0.060904641	Yes	No
Tauromuricholic acid	HMDB0000932	0.226951482	No	No
CAR 14:3	HMDB0241394	.
Methionine sulfoxide	HMDB0002005	0.034465313	Yes	No
LPE 20:4/0:0	HMDB0011517	0.111292059	Yes	Yes
LPC 20:4/0:0	HMDB0010395	.
Arachidic acid	HMDB0002212	.
PC 34:1	HMDB0007972	.
TG 51:4	HMDB0052411	−0.109898503	No	No
ATP	HMDB0000538	.
LPE 16:0/0:0	HMDB0011503	.
CAR 14:2	HMDB0013331	.
N1-methyl-2-pyridone-5-carboxamide	HMDB0004193	0.082816157	Yes	Yes
CAR 16:3	HMDB0241493	.
S-Adenosyl-L-homocysteine	HMDB0000939	0.167287077	Yes	No
Nonadecylic acid	HMDB0000772	0.175189304	No	No
Caffeine	HMDB0001847	0.091264702	Yes	Yes
TG 40:0	HMDB0099325	0.101370556	Yes	No
LPE 18:1/0:0	HMDB0011506	.
CAR 12:3	HMDB0241271	−0.071808343	No	No
DG 38:4	HMDB0007170	.
CAR 10:1	HMDB0013205	.
4-Acetamidobutanoic acid	HMDB0003681	.
N-Acetyl-D-tryptophan	HMDB0013713	.
CAR 10:0;OH	HMDB0061636	−0.165006348	No	No
LPE 22:6/0:0	HMDB0011526	0.053165491	Yes	Yes
Undecanedioic acid	HMDB0000888	−0.21615763	No	No
p-Hydroxyphenylacetic acid	HMDB0000020	0.077748506	No	No
N1-Methyl-4-pyridone-3-carboxamide	HMDB0004194	.
LPC 0:0/22:5	NA	.
Linoleoyl ethanolamide	HMDB0012252	−0.086326041	Yes	Yes
Xanthosine	HMDB0000299	.
LPC 22:6/0:0	HMDB0010404	.
CAR 16:2	HMDB0013334	.
Malonic acid	HMDB0000691	−0.158687583	No	No
PC P-34:1	HMDB0011210	.
Previously identified in White women
3-Methylxanthine	HMDB0001886	.
Biliverdin	HMDB0001008	.
Cotinine	HMDB0001046	0.153840971	Yes	Yes
GABA	HMDB0000112	−0.071403319	Yes	Yes
N4-Acetylcytidine	HMDB0005923	0.016805711	Yes	Yes
Pseudouridine	HMDB0000767	.
Threonine	HMDB0000167	0.005407185	Yes	Yes
Creatinine	HMDB0000562	−0.089912137	Yes	Yes
Glutamine	HMDB0000641	.
N2,N2-Dimethylguanosine	HMDB0004824	.
Tryptophan	HMDB0000929	−0.054883174	Yes	Yes
C16:0 Ceramide (d18:1)	HMDB0004949	.
PC 34:3	HMDB0008006	.
PC 36:5 plasmalogen	HMDB0011220	.
PC 38:3	HMDB0008047	−0.020767236	Yes	No
Hippurate	HMDB0000714	−0.075685314	Yes	Yes
Arachidonate	HMDB0001043	.
Identified in both Black and White participants
Serotonin	HMDB0000259	−0.138403955	Yes	Yes
Thiamine	HMDB0000235	0.027853993	Yes	Yes
LPE 18:0/0:0	HMDB0011130	.

¹The 63 candidate metabolites included 43 newly identified metabolites in the MESA discovery sample of 1,170 Black participants plus 20 metabolites previously identified in White participants

Stage 4: Evaluation of the MDS-ME with depression in three independent samples

In the independent withheld replication sample of 485 MESA Black participants, the MDS-ME also had a Pearson correlation of 0.33 with the original MDS. After full adjustment in Model 2, the association between MDS-ME and depression was statistically significant (OR per 1-SD increment in the MDS-ME: 1.45, 95% CI: 1.04, 2.03), with no evidence for sex differences (p-interaction>0.78; Table 4). The MDS-ME also performed well among MESA White participants (n=2,477; fully-adjusted OR: 1.81; 95% CI: 1.60, 2.04), and the associations were similar in White men and women (p-interaction>0.49).

Table 4.

Associations between the multi-ethnic MDS (per 1-SD increment) and depression status in White and Black participants from MESA

		Model 11		Model 22
	n	OR (95% CI)	p-value	OR (95% CI)	p-value
MESA
Black participants (30% replication set)
Overall	485	1.47 (1.09, 1.99)	0.01	1.45 (1.04, 2.03)	0.03
Women	264	1.43 (0.98, 2.13)	0.07	1.46 (0.95, 2.27)	0.09
Men	221	1.45 (0.89, 2.50)	0.15	1.46 (0.79, 2.81)	0.24
p-int by sex		0.94		0.78
White participants
Overall	2477	1.83 (1.63, 2.05)	3.69E-25	1.81 (1.60, 2.04)	2.67E-22
Women	1298	1.76 (1.52, 2.03)	1.50E-14	1.75 (1.51, 2.04)	2.02E-13
Men	1179	1.98 (1.64, 2.41)	3.78E-12	1.94 (1.59, 2.39)	2.39E-10
p-int by sex		0.49		0.58

¹Adjusted for age, sex, study site, education, income, and employment status

²Model 1 + adjusted for BMI, smoking status, physical activity, diet quality, systolic blood pressure, HDL cholesterol, total cholesterol, diabetes status, antihypertensive medications, and lipid-lowering medications

In WHI, 23 out of the 33 metabolites selected by the LASSO in the MDS-ME developed in MESA were available; In NHS, 17 of 33 metabolites were available (Table 3). Among 485 MESA Black participants, the 33-metabolite MDS-ME was strongly correlated with the reduced 23-metabolite MDS-ME available in WHI (r=0.74) and with the reduced 17-metabolite MDS-ME available in NHS (r=0.64); the corresponding correlations were r=0.74 and r=0.65 among White MESA participants. These correlation levels suggested that using these reduced MDS-ME as proxies for the replication analyses is reasonable. In WHI, the reduced MDS-ME was associated with depression with a similar magnitude in White (n=1,107; OR: 1.21; 95% CI: 1.04, 1.40) and Black (n=149; OR: 1.34; 95% CI: 0.85, 2.21) women (Table 5). In NHS, the reduced MDS-ME was also associated with depression similarly in White (n=1,856; OR: 1.11; 95% CI: 0.95, 1.30) and Black (n=214; OR: 1.20; 95% CI: 0.69, 2.06) women. Notably, though the associations did not reach statistical significance among the two small samples of Black participants in WHI and NHS, we found no evidence for significant between-study heterogeneity based on the meta-analysis of fully-adjusted estimates in Black participants using the full MDS-ME in MESA, the 23-metabolite MDS-ME in WHI and the 17-metabolite MDS-ME in NHS (pooled OR: 1.36; 95% CI: 1.07, 1.74; I²=0%; p for heterogeneity=0.84; Figure 3).

Table 5.

Associations between the reduced, multi-ethnic MDS (per 1-SD increment) and depression status in White and Black participants from WHI, NHS and MESA

		Model 11		Model 22
	n	OR (95% CI)	p-value	OR (95% CI)	p-value
Reduced 23-metabolite MDS3
WHI
Black women	149	1.29 (0.89, 1.94)	0.19	1.34 (0.85, 2.21)	0.23
White women	1107	1.24 (1.08, 1.44)	0.003	1.21 (1.04, 1.40)	0.01
MESA Black participants (30% replication set)
Overall	485	1.29 (0.96, 1.73)	0.09	1.31 (0.94, 1.84)	0.11
Women	264	1.16 (0.79, 1.71)	0.44	1.23 (0.80, 1.92)	0.34
Men	221	1.46 (0.90, 2.43)	0.13	1.32 (0.73, 2.39)	0.37
p-int by sex		0.52		0.83
MESA White participants
Overall	2477	2.04 (1.82, 2.30)	6.08E-33	2.07 (1.83, 2.35)	2.23E-30
Women	1298	1.91 (1.65, 2.22)	5.20E-18	1.96 (1.68, 2.29)	4.53E-17
Men	1179	2.29 (1.89, 2.81)	2.65E-16	2.31 (1.88, 2.88)	1.12E-14
p-int by sex		0.25		0.26
Reduced 17-metabolite MDS4
NHS
Black women	214	1.45 (0.90, 2.32)	0.12	1.20 (0.69, 2.06)	0.51
White women	1856	1.15 (1.00, 1.33)	0.048	1.11 (0.95, 1.30)	0.19
MESA Black participants (30% replication set)
Overall	485	1.17 (0.87, 1.56)	0.29	1.21 (0.88, 1.67)	0.25
Women	264	1.03 (0.70, 1.50)	0.88	1.12 (0.73, 1.72)	0.59
Men	221	1.40 (0.86, 2.34)	0.18	1.31 (0.72, 2.46)	0.38
p-int by sex		0.47		0.79
MESA White participants
Overall	2477	2.10 (1.87, 2.36)	7.66E-36	2.13 (1.89, 2.42)	5.91E-34
Women	1298	1.98 (1.71, 2.31)	2.39E-19	2.03 (1.74, 2.38)	1.41E-18
Men	1179	2.32 (1.92, 2.81)	4.07E-18	2.34 (1.92, 2.87)	1.23E-16
p-int by sex		0.30		0.30

¹Adjusted for age, sex (MESA only), study site (MESA only), education, income, employment status (MESA only), menopausal status (NHS only)

²Model 1 + adjusted for BMI, smoking status, physical activity, diet quality, systolic blood pressure, HDL cholesterol (MESA and WHI only), total cholesterol (MESA and WHI only), diabetes status (MESA and WHI only), antihypertensive medications (MESA and WHI only), and lipid-lowering medications (MESA and WHI only)

³The multi-ethnic MDS in WHI was based on 23 of the 33 metabolites selected in MESA

⁴The multi-ethnic MDS in NHS was based on 17 of the 33 metabolites selected in MESA.

Figure 3.

Meta-analysis of the association between the multi-ethnic MDS (MDS-ME) and depression among Black participants from MESA, WHI and NHS. The MDS-ME was calculated based on the full 33 metabolites in MESA, 23 metabolites in WHI and 17 metabolites in NHS.

In sensitivity analyses, as expected, the reduced MDS-ME showed weaker associations with depression than the full MDS-ME among MESA Black participants; however, the point estimates were remarkably similar to those observed in WHI and NHS Black women (Table 5). Moreover, both reduced MDS-ME demonstrated consistently strong positive associations with depression among MESA White participants.

Discussion

In this study, we found that the MDS, originally developed in several independent samples of White women, was strongly associated with phenotypic measures of depression in both White men and women in MESA. In contrast, associations were weaker in Black participants, particularly Black women. We conducted an agnostic analysis to identify additional plasma metabolites associated with psychological distress in Black participants and revised the MDS by incorporating newly identified metabolites. The MDS-ME demonstrated stronger associations than the MDS with depression in Black men and women in MESA, as well as comparable associations in external samples of Black women from WHI and NHS. Nonetheless, the magnitude of the association between the MDS-ME and depression remained weaker in Black participants compared with that in White participants in the MESA sample.

Of note, the MDS-ME, developed among Black men and women in combination, was as strongly linked to depression in White men and women as the original MDS developed specifically among White women. These findings have several important implications. When examining metabolomic processes and their relationship to distress (or other psychosocial factors), it is important to consider potential racial differences and recognize that the relevant phenotypes may be heterogeneous across populations. Also important to recognize is that more heterogeneous samples may provide greater insight into how metabolomic processes are associated with higher order factors. Similar themes have emerged in psychiatric genetics, where multi-ancestry studies of major depression have identified novel loci and shown that newer polygenic score methods that integrate ancestry-diverse GWAS data improve cross-population prediction (2025; Hoggart et al., 2024^; Meng et al., 2024^; Ruan et al., 2022). Likewise, studies of plasma metabolomics have documented substantial Black-White differences, particularly in lipids and amino acids (Butler et al., 2023; Hu et al., 2022; McGee et al., 2024). In women, a metabolomic pattern capturing such Black-White differences was associated with incident coronary heart disease risk independent of self-reported race and traditional risk factors (Hu et al., 2022), underscoring the value of more heterogeneous samples for clarifying how distress-related metabolomic signatures may contribute to health disparities.

Among Black participants, the three metabolites that were most strongly associated with depression status and remained significant after correction for multiple comparisons were serotonin, hydroxytryptophan, and adenosine. Hydroxytryptophan is a natural intermediate product on the pathway that converts tryptophan to serotonin, a key neurotransmitter involved in stress responses, mood regulation, and sleep (Maffei, 2020). Because hydroxytryptophan directly influences serotonin synthesis, altered serotonin signaling due to changes in hydroxytryptophan levels or receptor functions have been implicated in depressive symptoms (Correia and Vale, 2022; Richard et al., 2009). Adenosine is known to have a predominantly inhibitory effect on the central nervous system. Reduced adenosine levels have been associated with elevated chronic inflammation, dysregulated dopamine and glutamate activity, and exaggerated stress responses, all of which may increase vulnerability to depression (Haskó and Cronstein, 2013; Pasquini et al., 2022; Yue et al., 2025). This is consistent with our observation of an inverse association between adenosine and depression. Interestingly, we also observed a positive association of caffeine with depression. Caffeine primarily acts as an adenosine receptor antagonist in increasing alertness, enhancing concentration, and increasing wakefulness (the opposite of symptoms found in depression) (Ribeiro and Sebastião, 2010). Given that caffeine metabolism may vary across individuals and may be influenced by broader contextual factors (Rybak et al., 2015), further investigation is required to fully elucidate the complex biological consequences of the balance between adenosine, caffeine, and adenosine receptor across populations.

Other metabolites associated with depression among Black participants included positive associations for several lipid species and inverse associations for multiple carnitines. Changes in lipid profiles, such as increased levels of PCs, LPCs, and LPEs, may be linked to depression by disrupting cell-membrane integrity, neurotransmitter signaling, inflammation, and mitochondrial energy metabolism, which can impair neuronal function and stress regulation (Ge et al., 2025; Pinto et al., 2022). Conversely, carnitines tend to show the opposite metabolic patterns by facilitating fatty-acid transport into mitochondria for increased mitochondrial efficiency, enhancing neuroplasticity, and exerting anti-inflammatory effects, with lower levels indicating impaired energy metabolism and diminished adaptive neural responses commonly seen in depression (Ait Tayeb et al., 2024; Bigio et al., 2024; Milaneschi et al., 2022). Thus, while altered lipids often signal metabolic stress and inflammation, lower carnitine levels suggest mitochondrial underperformance as a complementary biological pathway in depression. In addition, thiamine (vitamin B1) and several metabolites related to niacin (vitamin B3) metabolism were positively associated with depression status. It is possible that increased niacin, which is synthesized from tryptophan, may reduce tryptophan bioavailability for serotonin production, thereby altering serotonin signaling and increasing depression risk (Davidson et al., 2022; Tang et al., 2025). However, some evidence suggests that vitamin B deficiency may also contribute to depressive symptoms (Han et al., 2025; Mikkelsen et al., 2017). Given the conflicting evidence, the role of niacin-related metabolites in depression warrants further study.

Of note, many of the metabolites discussed above based on findings in Black participants had not shown the most robust signals with depression in our previous study of White women. In fact, only three metabolites (serotonin, thiamine and LPE 18:0/0:0) previously included in the MDS developed in White women were significantly associated with depression in MESA Black participants; several others, including GABA, did not reach nominal significance. Such differences in the distress-related metabolic profiles between White and Black participants may reflect underlying variation in genetic architecture, metabolic regulation, or inflammatory responses across populations. Socio-environmental factors, such as unemployment, discrimination, diet, and neighborhood context, may also influence metabolic profiles in ways that intersect with depression risk (Cardel et al., 2021; Li et al., 2019; Vrany et al., 2021). Additionally, unequal access to treatment and differences in comorbid conditions could influence both metabolic pathways and depression. Collectively, these factors likely interact, leading to different metabolomic signatures of depression that emerge as important in diverse groups.

Despite our efforts to refine the MDS for Black participants, the strength of its association with depression among Black participants with MDS was still weaker than associations among White participants using either the original MDS or the MDS-ME. Methodologically, a more limited sample size for Black participants may have reduced power for metabolite discovery and replication, potentially attenuating observed associations. Biologically, our findings may suggest that depression, or more generally psychological distress, is a more heterogenous phenotype in Black participants, influenced by a broad range of social and environmental risk factors (Assari, 2017; Assari and Moazen-Zadeh, 2016; Vyas et al., 2020; Wang and Narcisse, 2025). If future studies confirm reliable differences, this could suggest caution in generalizing biomarker-based screening tools developed in homogeneous samples. At the population level, these findings highlight the importance of considering how structural and social determinants of health can affect metabolic processes associated with psychological distress in diverse groups.

This study has several key strengths. First, it focused on two populations at high risk for cardiometabolic disease, Black individuals and men, to identify metabolomic signatures associated with psychological distress. Second, a rigorous approach was applied to revise the MDS and replicate the associations both internally and externally in independent samples. Third, a common, extensively validated metabolomic profiling method was used across datasets, facilitating direct comparison. Finally, key covariates were available across different samples, allowing for comprehensive and comparable adjustment in all analyses.

Several limitations should also be noted. Our study was cross-sectional by design, and the identified metabolomic profiles may reflect both metabolic precursors and consequences of depression. Longitudinal studies and causal approaches (e.g., Mendelian randomization) are needed to disentangle these potentially bidirectional relationships. While our study represents the first and largest investigation on metabolomic profiles of psychological distress specifically for Black individuals, the sample size for Black participants was still modest, particularly compared with that for White participants. This may have limited our ability to identify certain metabolites with smaller effect sizes. Despite consistency in metabolomic profiling and covariate assessment, there were variations in the definition of depression and population characteristics (e.g., age and sex distributions) across study samples. Also, our analysis was limited to metabolites consistently annotated across independent cohorts. Because data processing pipelines and annotation approaches have advanced significantly over time, older cohorts contain more unannotated features that, if identified, could have strengthened validation or further improved the MDS-ME. Furthermore, the influence of long-term and variable blood sample storage on plasma metabolites is not well understood (Townsend et al., 2016; Yin et al., 2015). These could compromise replication efforts and reduce the generalizability of the findings to contemporary populations. Future studies with robust replication are also needed for other racial and ethnic groups (e.g., Asians, Hispanics) to develop a comprehensive understanding of the metabolic correlates of psychological distress in diverse populations.

In summary, the MDS originally developed in White women showed similar associations with depression in White men but weaker associations in Black individuals. Incorporating additional metabolites associated with depression in Black individuals yielded MDS-ME that showed more consistent positive associations with depression status across three independent replication samples of Black individuals, highlighting a broader array of metabolites that may underlie depression than was previously identified. However, the association between the MDS-ME and depression in Black individuals remained weaker than that between the original MDS and depression in White individuals, suggesting additional metabolites may emerge with larger and more diverse samples. Future studies should extend investigations of distress-related metabolic profiles to larger and more diverse populations and assess how such profiles contribute to cardiometabolic and other health outcomes and if they provide insight into health disparities across diverse groups.