Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Nov 1.
Published in final edited form as: Psychoneuroendocrinology. 2021 Sep 20;133:105420. doi: 10.1016/j.psyneuen.2021.105420

Plasma metabolomic profiles associated with chronic distress in women

Katherine H Shutta a,*, Raji Balasubramanian a,*, Tianyi Huang b,c, Shaili C Jha d, Oana A Zeleznik b,c, Candyce H Kroenke e, Lesley F Tinker f, Jordan W Smoller g,h, Ramon Casanova i, Shelley S Tworoger d,j, JoAnn E Manson b,c,d,k, Clary B Clish l, Kathryn M Rexrode c,m, Susan E Hankinson a,**, Laura D Kubzansky n,**
PMCID: PMC8547060  NIHMSID: NIHMS1744267  PMID: 34597898

Abstract

Several forms of chronic distress including anxiety and depression are associated with adverse cardiometabolic outcomes. Metabolic alterations may underlie these associations. Whether these forms of distress are associated with metabolic alterations even after accounting for comorbid conditions and other factors remains unclear. Using an agnostic approach, this study examines a broad range of metabolites in relation to chronic distress among women.

For this cross-sectional study of chronic distress and 577 plasma metabolites, data are from different substudies within the Women’s Health Initiative (WHI) and Nurses’ Health Studies (NHSI, NHSII). Chronic distress was characterized by depressive symptoms and other depression indicators in the WHI and NHSII substudies, and by combined indicators of anxiety and depressive symptoms in the NHSI substudy. We used a two-phase discovery-validation framework, with WHI (N=1,317) and NHSII (N=218) substudies in the discovery phase (identifying metabolites associated with distress) and NHSI (N=558) substudy in the validation phase. A differential network analysis provided a systems-level assessment of metabolomic alterations under chronic distress. Analyses adjusted for potential confounders and mediators (demographics, comorbidities, medications, lifestyle factors). In the discovery phase, 46 metabolites were significantly associated with depression measures. In validation, six of these metabolites demonstrated significant associations with chronic distress after adjustment for potential confounders. Among women with high distress, we found lower gamma-aminobutyric acid (GABA), threonine, biliverdin, and serotonin and higher C16:0 ceramide and 3-methylxanthine. Our findings suggest chronic distress is associated with metabolomic alterations and provide specific targets for future study of biological pathways in chronic diseases.

Keywords: chronic distress, depression, anxiety, women’s health, metabolomics, network analysis

1. Introduction

Chronic diseases of aging, including cardiometabolic disease, are leading causes of morbidity and mortality. Evidence increasingly links various manifestations of chronic distress to higher risk of chronic disease (Cohen et al. 2007; Gupta et al. 2016), while the molecular processes driving these associations are not fully understood. Metabolite profiling technologies quantify small molecules (metabolites) that are products of cellular activities, reflecting the combined effects of the environment and the genome (Goodacre et al. 2004; Roberts et al. 2012). Identifying metabolites altered by chronic distress may shed light on biological underpinnings of the relationship between distress and chronic disease. Relevant research has focused primarily on anxiety and depression; thus, we use chronic distress as an umbrella term encompassing both of these characterized by either symptoms or diagnoses.

Cross-sectional studies in humans have evaluated metabolomic profiles with depression, anxiety, and related manifestations of chronic distress. Several small investigations using candidate approaches (<100 cases with distress) found alterations in lipid (Karabatsiakis et al. 2015; Liu et al. 2015; Liu et al. 2016) and tryptophan-serotonin (Karabatsiakis et al. 2015; Liu et al. 2015) metabolism and the glutamine-glutamate pathway (Hashimoto et al. 2016) with high distress. Altmaier et al. examined metabolomic alterations in relation to type D personality, which is characterized by negative affect and social inhibition (Altmaier et al. 2013). This agnostic study considered 668 metabolites measured in 1,502 participants, finding significantly lower levels of the tryptophan metabolite kynurenine among those with type D personality compared to controls (Altmaier et al. 2013). A recent pooled analysis of 5,283 cases and 10,145 controls evaluating 230 lipids and other metabolites in relation to depression identified associations with several plasma lipids, fatty acids, and low molecular weight metabolites (Bot et al. 2020). Though large, the study lacked information on several important covariates such as comorbidities and lifestyle factors. In our own recent study of 45 candidate metabolites, we found depression was associated with higher levels of glutamate and PC 36:1/38:3, and lower levels of tryptophan and GABA-to-glutamate and GABA-to-glutamine ratios (Huang et al. 2020).

Here, we used an agnostic approach to evaluate associations of 577 metabolites with chronic distress. We utilized a robust discovery-validation study design involving three independent, well-characterized datasets that include a range of covariates. Of note, we used two of these datasets for our prior study of association between depression and candidate metabolites as described above (Huang et al. 2020). This agnostic study differs from the prior work and adds to the literature in several important ways. First, we used an agnostic approach, investigating all 577 measured metabolites. Second, we characterized chronic distress broadly by considering both depression and anxiety-related indicators. Third, this work includes a newly measured validation dataset of 279 cases with chronic distress and 279 healthy controls, nested in the Nurses’ Health Study, that has not been previously published; these data were used as an independent validation of the significant findings in discovery datasets. Finally, in this validation dataset, we employed a differential network analysis for a systems-level view of metabolomic alterations under chronic distress. Identifying a broad set of validated metabolites associated with chronic distress can point to new directions in the study of molecular processes linking psychological distress to the development of chronic disease.

2. Material and methods

2.1. Study protocol and population

Participants were from sub-studies within the Women’s Health Initiative hormone therapy trials (WHI–HT), the Nurses’ Health Study (NHS) and the Nurses’ Health Study II (NHSII) (Figure 1). Sub-studies included the Mind-Body Study (MBS) (Huang et al. 2019) nested within NHSII, a coronary heart disease (CHD) case-control study nested within WHI-HT (Paynter et al. 2018), and a nested case-control study of chronic distress within NHS. NHSII and WHI-HT substudies were used in the discovery stage to identify metabolites associated with chronic distress; in the validation stage, we assessed significance of these metabolites using data from the NHS sub-study (Figure 1). All participants provided written informed consent. This study was approved by the Institutional Review Board of UMass-Amherst.

Figure 1:

Figure 1:

Open in a new tab

Participant selection into the discovery (WHI-HT, NHSII) and the validation (NHS) cohorts and two -stage discovery-validation framework for identifying metabolites associated with chronic distress.

* Statistical significance is assessed via a likelihood ratio test of the minimal model with the three exposures (“tripleexposure model (minimal)”) vs. the minimal model with age and race only (“baseline model”).

**Minimal conditional logistic regression model implicitly adjusts for matching factors by the use of conditional logistic regression. No other covariates are included.

1 pair removed

due to missing

metabolomics

2.1.1. Discovery.

The NHSII includes 116,429 female US registered nurses aged 25-42 years old at study enrollment in 1989. The MBS sub-study within NHSII consists of 233 postmenopausal women sampled from participants who reported childhood trauma or sexual abuse on a 2001 questionnaire (Huang et al. 2019). An 8-hr fasting blood sample, used for plasma metabolomic profiling, was collected from 2013-2014. Storage information for the blood samples is detailed in the Supplement. Within one month of blood collection, participants also completed a questionnaire assessing psychosocial factors, including various forms of distress.

The WHI includes 161,808 postmenopausal women aged 50-79 recruited at 40 clinical centers across the US from 1993 to 1998. A subset participated in two hormone therapy trials (WHI-HT, N=27,347). Our study participants are from an ancillary CHD case/control study including 1362 participants selected from the placebo and active hormone arms of the WHI-HT trials (Paynter et al. 2018; Balasubramanian et al. 2020). Women provided a 12-hr fasting blood sample and completed questionnaires assessing demographic, medical, and psychosocial characteristics at entry into WHI (1993-1998). Storage information for the blood samples is detailed in the Supplement.

2.1.2. Validation.

The NHS consists of 121,700 US registered female nurses aged 30-55 years at study enrollment in 1976 (Belanger et al. 1978). The NHS includes two blood collections. Women who provided blood samples at the first blood collection (1988–1990) were invited to give a second blood sample (2000-2002) as previously described (Tworoger et al. 2013). Samples were stored using the same protocols as in NHSII, as detailed in the Supplement. Participants eligible for the current study provided blood samples in the second collection, completed questionnaires with depression-related (1992 through 2004) and anxiety-related (1988, 2004) measures, and were free of cancer or cardiovascular disease at the time of blood collection (N=13,194). We selected a subset of these women for a matched case-control analysis described below.

2.2. Outcome and covariate assessment

Chronic distress was assessed using questionnaire-based symptom measures of anxiety and depression (generally used for screening) and self-reported use of antidepressants, physician diagnosis of depression, and history of depression. Participants missing information on distress assessments were excluded (Figure 1).

2.2.1. Discovery.

In NHSII, chronic distress was assessed at or near the time of blood draw via three depression indicators. Depressive symptoms were assessed in 2013/2014 using the validated 10-item Center for Epidemiologic Studies Depression Scale (CES-D-10) (Bjorgvinsson et al. 2013). Regular antidepressant use within the last two years was self-reported at the time of blood draw. Physician-diagnosed depression was self-reported on biennial NHSII questionnaires beginning in 2003. In the WHI-HT, chronic distress was assessed via three depression indicators at participants’ baseline visit. Depressive symptoms were measured by the validated 6-item CES-D questionnaire (CES-D-6). Current antidepressant use was assessed by an inventory of drug labels presented by participants. Participants self-reported a history of depression by indicating they had been (1) depressed for ≥ two weeks in the past year, and (2) depressed for much of the time for ≥ two years in their lifetimes. Measures of anxiety were not available in NHSII or WHI-HT.

2.2.2. Validation.

In NHS, cases and controls were defined using repeated assessments of depression (symptoms, reports of physician diagnosis, antidepressant use) and anxiety (symptoms) measured repeatedly prior to and around the time of blood draw. Cases consisted of women with past and/or current depression and anxiety who reported experiencing severe distress levels. Controls consisted of women who reported neither depression nor anxiety (past or current). 280 cases and 280 controls were selected, with 1:1 matching on age, race/ethnicity, menopausal status, fasting status, and day/time of blood draw. Plasma metabolomic profiles were obtained for 279 matched pairs.

2.2.3. Covariate assessment.

For discovery, covariates included age, race/ethnicity, BMI, smoking, use of statins or other lipid-lowering drugs, hormone therapy, history of hypertension or diabetes, diet, and physical activity. In the validation analysis, age, race/ethnicity, fasting status, menopausal status, and day and time of blood draw were used as matching variables. Other covariates included the covariates used in the discovery phase with the exception of hormone therapy. See Supplement for additional details on case and control selection, exposure, and covariate definitions.

2.3. Metabolomics assay

Plasma metabolomics profiling was conducted at the Broad Institute of the Massachusetts Institute of Technology and Harvard using liquid chromatography tandem mass spectrometry (LC-MS)-based metabolomics platform as described in detail previously (Paynter et al. 2018). Three complementary LC-MS methods were used to yield data on 487 identified metabolites in the NHSII and 417 metabolites in the WHI-HT, totaling 577 unique metabolites measured across discovery datasets (327 metabolites measured in both). In the NHS validation dataset, 480 metabolites were measured. The median (10th - 90th percentile) of the percent coefficient of variation (CV) of metabolites measured across datasets and identified with an HMDB ID were 10.4 (6.3-24.6) n=473, 9.5 (5.4-34.2) n=530 and 10.3 (6.0-20.8) n=530, respectively. See Supplement for additional details.

2.4. Statistical Analysis

Metabolite relative abundances were log-transformed and standardized to have mean zero and standard deviation one. We excluded metabolites missing in > 50% of the participants (N=22 in WHI-HT; N=1 in NHSII); in the NHS, all discovery metabolites available for analysis met the criterion (Supplementary Data 2). See Supplement for details of metabolite level imputation.

In the study design stage, power calculations were performed to ensure appropriate sample size. Discovery analyses were powered to detect an odds ratio of 1.65 with univariate power 0.97 and multivariate power 0.95 at Bonferroni-corrected significance level 0.05/400 (assuming approximately 400 metabolites would be detected). Validation analyses were proposed with power to detect an odds ratio of 1.75 with univariate power 0.91 and multivariate power 0.86 at Bonferroni-corrected significance level 0.05/20 (assuming approximately 20 metabolites would be carried forward from discovery to validation).

An assumption that metabolite levels are normally distributed is applied in the linear analyses in the discovery phase. We used the log transformation to transform distributions to be approximately symmetric. Due to the large number of metabolites, individual tests for normality were not conducted. The validation analysis does not make a distributional assumption on the metabolite levels.

2.4.1. Discovery.

Each study evaluated associations of metabolites with depression measures in metabolite-by-metabolite linear models. Metabolite level was the outcome and the three depression indicators (depressive symptoms, history of depression, and antidepressant use) at or near the time of blood draw were included simultaneously as predictors, adjusting for age and race. Significance of each association was assessed by a likelihood ratio test (LRT) of the model with the three depression indicators vs. a minimal model adjusting for age and race only. A second, fully-adjusted model included covariates described above. Adjustment for multiple testing was as proposed by Benjamini and Hochberg (Benjamini 2006). Significant associations satisfied a minimal model p < 0.05 and a false discovery rate (FDR)-adjusted p < 0.20 in WHI-HT and < 0.30 in NHSII. We assessed significance of each depression indicator using a Wald test. Finally, we conducted an over-representation analysis (ORA) to identify metabolite classes enriched for associations with depression (see Supplement).

2.4.2. Validation.

Metabolites meeting the significance threshold in discovery analyses were evaluated in metabolite-by-metabolite conditional logistic regression models. Log-transformed, standardized metabolite relative abundance was the primary independent variable and chronic distress case/control status was the outcome. Covariate adjustment was implemented as in discovery analyses. Significant metabolites satisfied a threshold of p < 0.05 and FDR-adjusted p < 0.05 in the minimal model. A differential network between distressed cases and controls was constructed using the DINGO algorithm (Ha et al. 2015). Metabolites were annotated with network-level measures, including hub(Kleinberg 1999), betweenness, and closeness centrality (Freeman 1978). Betweenness and closeness centrality interpret edge weights as distances; we set edge weights to be the inverse of the absolute partial correlation between metabolites. As a secondary analysis, we performed a pathway analysis to identify KEGG pathways enriched for association with chronic distress in the validation dataset (see Supplement).

2.4.3. Sensitivity analyses.

Several sensitivity analyses were conducted to assess (i) the role of antidepressant use in associations of distress with validated metabolites; (ii) the role of smoking, which can affect a broad range of metabolites; (iii) the extent to which covariates attenuated effect sizes in the validation study. See Supplement for additional details on these sensitivity analyses.

3. Results

Characteristics of participants at or near the time of blood draw are shown in Table 1. Women in the WHI-HT and NHS cohorts were approximately 4-7 years older at blood draw on average than women in NHSII. WHI-HT women were more racially diverse than NHS or NHSII women. When depressed status was defined as one or more of the following: a CES-D score ≥ 10, antidepressant use, or history of depression, the prevalence of depressed status was higher in NHSII (44.0%) versus WHI-HT (25.3%).

Table 1:

Basic characteristics of the WHI-HT, NHSII and NHS datasets, assessed at or near time of blood collection.

Discovery Validation
WHI-HT NHSII NHS
Recruitment: 1993-1998
Blood collection: at study entry		 Recruitment: 1989
Blood collection: 2013-2014		 Recruitment: 1976
Blood collection: 2000-2002
Not Depressed Depressed 1 Not Depressed Depressed 2 Not Distressed Distressed 3
N 984 333 122 96 279 279
Age in years, mean(sd) 67.1 (6.8) 65.4 (7.1) 60.6 (4.2) 60.5 (3.8) 64.3 (6.6) 64.3 (6.6)
Non-white, % 14.6% 19.8% 2.5% 7.3% 2.5% 0.7%
Distress measures
CES-D-6 1.4 (1.3) 5.4 (3.1) - - - -
CES-D-10 - - 3.7 (2.7) 8.7 (5.4) 3.7 (2.5) 15.6 (4.3)
Antidepressant use,% 0 25.8 0 56.3 1.4 47.7
History of depression,% 0 45.0 0 75.0 0 55.2
Medical factors
Type 2 diabetes, % 13.3 20.4 3.3 8.3 2.9 13.3
Hypertension, % 41.2 47.2 29.5 37.5 38.7 53.8
Aspirin use, % 29.4 29.4 32.0 39.6 50.54 51.25
Statin use, % 13.5 13.8 23.8 30.2 17.2 27.6
Other lipid-lowering drug, % 2.4 3.3 3.3 5.2 1.8 5.4
Hormone therapy, % 6.6 4.8 23.0 24.0 62.2 62.3
Body mass index, kg/m4 29.1 (5.8)1 30.0 (6.2)1 26.1 (6.1) 27.7 (5.9) 25.7 (4.5) 28.5 (7.6)
Lifestyle factors
Smoking status (%):
• Never/past smoking 86.5 79.0 96.7 96.9 94.3 91.8
• Current smoking 11.4 19.2 3.3 3.1 4.7 7.9
• Missing 2.1 1.8 0 0 1.0 0.3
Recreational physical activity in MET-hrs/week, mean(sd) 9.8 (11.9) 8.2 (10.3) 31.2 (28.4) 24.7 (24.7) 25.9 (28.5) 16.9 (22.1)
Diet quality score5 mean(sd) 66.7 (10.7) 64.9 (11.0) 68.4 (13.4) 69.30 (12.6) 57.5 (13.1) 53.6 (12.5)
Caffeine intake, mg/day mean(sd) 176.9 (147.6) 182.0 (155.1) 170.8 (141.6) 187.8 (151.9) 124.1 (115.1) 144.9 (137.9)
Alcohol intake, g/day mean(sd) 4.3 (9.4) 3.6 (13.2) 9.1 (12.0) 5.6 (9.4) 7.3 (12.0) 5.4 (9.9)
Open in a new tab
1

Defined as one or more of CES-D-6 >= 5, antidepressant use, or history of depressed mood

2

Defined as one or more of CES-D-10 >= 10, antidepressant use, or history of physician-diagnosed depression

3

Defined as detailed in Methods

4

BMI missing for 2 participants

5

Measured by HEI-2005 in WHI and AHEI-2010 in NHSII and NHS. AHEI-2010 missing for 2 participants in NHS.

3.1. Discovery of novel metabolite associations with depression status

In WHI-HT (n=1,317), 417 metabolites were evaluated for associations with depression status. In minimally-adjusted models, 21 metabolites reached statistical significance (p < 0.05; FDR-adjusted p < 0.20) as determined by the global LRT of model with all three depression indicators (“triple-exposure model”) (Figure 2, Supplementary Table S1, Supplementary Data 1). In fully-adjusted models, 15 of these remained significant (p < 0.05); N6-acetylysine, C20:2 CE, C24:0 LPC, C36:1 PE and 3-methylxanthine were marginally significant (p < 0.10) (Supplementary Data 1). The association of cotinine with depression status was attenuated in models adjusting for smoking status (p=0.64). In NHSII (n=218), 488 metabolites were evaluated and 26 reached statistical significance (p < 0.05 and FDR-adjusted p < 0.30) according to the global LRT (Figure 2, Supplementary Table S1, Supplementary Data 1). In fully-adjusted models, 21 remained significant (p < 0.05); four (proline, C38:3 PC, N4-acetylcytidine and docosapentanoate) were marginally significant (p≤0.10) (Supplementary Data 1). The association of the remaining metabolite (DMGV) with depression status was attenuated in the fully adjusted models (p=0.14). See Supplement for additional details of these analyses.

Figure 2:

Figure 2:

Open in a new tab

(a) Plot of 46 candidate metabolites identified in the discovery stage as associated with depression in WHI-HT and NHSII. Individual dots are colored to reflect the direction and magnitude of association with each depression exposure (CES-D score, history of depression, antidepressant use) and sized in proportion to the negative log of p-value associated with each coefficient. Effects are from a minimally adjusted model adjusting for age and race. (b) Discovery (WHI-HT, NHSII) and validation (NHS) results for the 11 validated metabolites*, by class. Effects shown are from minimally adjusted models adjusting for age and race explicitly via inclusion in the linear model in discovery analyses and for age, race, menopausal status, fasting status, and date and time of blood draw implicitly via conditional logistic regression in validation analyses.

Of the 577 metabolites measured in at least one of the discovery substudies, 375 were measured in both. 45 metabolites met the threshold for statistical significance in one of the two datasets and one metabolite (serotonin) met the threshold in both (Supplementary Table 1). See Supplement for details about CVs, percent missing, and reproducibility of these 46 metabolites.

In the ORA, 4 of 43 metabolite classes (FDR-adjusted p < 0.05) were significantly overrepresented among discovery metabolites. These classes included amino acids (p = 4.1x10−4, FDR-adjusted p = 0.01), bilirubins (p = 6.5x10−4, FDR-adjusted p = 0.01), eicosanoids (p = 6.9x10−4, FDR-adjusted p = 0.01), and purines and pyrimidines (p = 0.003, FDR-adjusted p = 0.04) (Supplementary Table S2).

3.2. Validation of metabolite associations with chronic distress (cases versus controls)

Of 46 discovery metabolites, 43 were available for validation in NHS (threosphingosine, C20:2 CE, and 5-hydroxytryptophan were not available). Eleven of these 43 met the validation threshold with directions of associations consistent across discovery and validation datasets, with raw and FDR-adjusted p-values < 0.05 in the minimally adjusted model (Table 2, Figure 2, Supplementary Data 1). CVs and percent missing values of these validated metabolites were within acceptable ranges (Supplementary Data 2). Six metabolites remained significantly associated with chronic distress after adjusting for potential confounders (Table 2). In fully-adjusted models, higher levels of serotonin, biliverdin, GABA, and threonine were associated with lower odds of chronic distress. Higher levels of C16:0 ceramide and 3-methylxanthine were associated with higher odds of chronic distress.

Table 2:

Validated metabolites sorted by metabolite class. Results indicate the significance of metabolites differences and association (95% confidence interval*) of standardized, log-transformed metabolite levels by case-control status. p-values and FDR-adjusted p-values are from minimal models**. Bold indicates metabolite was significant after adjusting for potential confounders.

Metabolite Category p FDRp Minimal Model OR Full Model OR
DMGV Amino Acids 2.0*10−4 0.002 1.46 (1.2,1.79) 1.16 (0.9,1.48)
GABA Amino Acids 0.012 0.041 0.79 (0.66,0.95) 0.74 (0.59,0.93)
threonine Amino Acids 0.015 0.047 0.79 (0.66,0.96) 0.73 (0.59,0.92)
biliverdin Bilirubins 3.5*10−5 0.001 0.66 (0.54,0.8) 0.65 (0.51,0.82)
C16:0 Ceramide (d18:1) Ceramides 0.001 0.005 1.41 (1.16,1.72) 1.23 (1,1.53)
serotonin Indoles and Indole Derivatives 1.3*10−7 5*10−6 0.58 (0.47,0.71) 0.65 (0.52,0.83)
pseudouridine Nucleoside and nucleotide analogues 0.011 0.041 1.29 (1.06,1.57) 1.05 (0.83,1.32)
3-methylxanthine Purines and Pyrimidines 0.003 0.024 1.3 (1.09,1.56) 1.24 (1.01,1.53)
N4-acetylcytidine Purines and Pyrimidines 0.005 0.024 1.32 (1.09,1.6) 0.97 (0.76,1.23)
cotinine Pyridines and derivatives 0.005 0.024 1.32 (1.09,1.6) 1.49 (0.98,2.27)
hydroxycotinine Pyridines and derivatives 0.008 0.035 1.3 (1.07,1.59) 1.39 (0.94,2.05)
Open in a new tab
*

OR (95% CI) from conditional logistic regression models performed within the NHS validation cohort, with confidence intervals constructed by exponentiating the endpoints of 95% Wald-type confidence intervals for the log odds ratio

**

Minimal models adjusted for matching factors via the use of conditional logistic regression. Matching factors include age, race/ethnicity, menopausal status, fasting status, and day and time of blood draw. Full models additionally adjusted for medical and lifestyle factors.

In our NHS validation dataset, we reproduced four of the six findings we previously reported in our examination of candidate metabolites and depression (conducted using the same datasets used for discovery in the present study using an agnostic approach). (Huang et al. 2020). In minimally-adjusted models, PC 38:3 was positively associated (OR=1.21, 95% CI:[1.02,1.44]) and three metabolites were inversely associated with distress, including tryptophan (OR=0.75, 95% CI:[0.62,0.90]), GABA-to-glutamate (OR: 0.75, 95% CI: [0.61,0.91]), and GABA-to-glutamine (OR:0.77, 95%CI: [0.65,0.92]) ratios. While the other two metabolites that reached statistical significance in our candidate study, glutamate nor PC 36:1 were not associated with distress in our validation dataset, directions of association were concordant (glutamate: OR=1.14, 95% CI: [0.94,1.39]; PC 36:1: OR=1.10, 95% CI: 0.93, 1.29]). See Supplement for details on the ratio analyses.

3.3. Differential metabolomics networks in distressed cases versus non-distressed controls

Using the NHS validation data, we estimated a differential network comparing cases with controls. Cotinine and hydroxycotinine, metabolites involved in nicotine metabolism, dominated this network, suggesting systems-level metabolomic changes due to smoking were most visible (Supplementary Figure 2, Supplementary Table S3).

Given findings that highly distressed individuals are more likely to smoke (Cohen et al. 2015), to facilitate identifying metabolic alterations linked to distress beyond those that occur because of smoking, we repeated the analysis among non-current smokers (N=494). Among non-current smokers, using 68 metabolites meeting criteria for inclusion into the DINGO algorithm we estimate a differential network comparing cases with controls. The differential network included 50 metabolites and 97 differential edges meeting the threshold for statistical significance (FDR-adjusted p < 0.05 on 100 bootstrap samples) (Figure 3). Cortisol and 21-deoxycortisol are evident as key players. Supplementary Table S4 shows the annotation of each metabolite with network-level measures, providing insights into each metabolite’s role within the network.

Figure 3:

Figure 3:

Open in a new tab

Differential network reflecting statistically significant differences (FDRp<0.05) in partial correlations between distressed cases and controls in the validation (NHS) dataset, excluding current smokers. Red (blue) edges denote partial correlations that were higher (lower) in distressed cases versus controls. Red nodes correspond to validated metabolites; blue nodes correspond to metabolites from the 46 discovery metabolites that were not significantly associated with distress in the validation dataset; green nodes correspond to metabolites significantly associated with distress in the validation dataset which were not found in the discovery phase.

3.4. Sensitivity analyses

Sensitivity analyses conducted to distinguish effects of antidepressant use from chronic distress suggested not all observed metabolomic changes can be attributed solely to antidepressant use, with numerous validated metabolites remaining significantly associated with distress after accounting for antidepressant use in various ways (Supplementary Figure S5). Further sensitivity analyses examining metabolomic changes among nonsmoking women demonstrated that most findings were not attributable to smoking (Supplementary Figure S1), and separately that while estimate for validated metabolites were often attenuated by covariate adjustment, associations remained evident in most cases. (Supplementary Figure S6). See Supplement for further details.

4. Discussion

We validated 11 metabolites significantly associated with chronic distress among women in models adjusted for matching factors through a discovery-validation study involving three independent datasets in three cohorts. Validated metabolites included amino acids (DMGV, GABA, threonine), biliverdin, C16:0 ceramide, serotonin, pseudouridine, purine and pyrimidine derivatives (3-methylxanthine, N4-acetylcytidine) and pyridine derivatives (cotinine, hydroxycotinine). In fully-adjusted models, six metabolites remained significant: serotonin, biliverdin, GABA, threonine, C16:0 Ceramide, and 3-methylxanthine. Three (biliverdin, GABA, and threonine) were also significantly associated with distress among women not taking antidepressants relative to non-distressed women. The validated metabolites are active in a range of biochemical pathways, including tryptophan, porphyrin/chlorophyll, alanine/aspartate/glutamate metabolism, and glycine/serine/threonine metabolism (Supplementary Table S5). Several have also been associated with other aspects of chronic distress and/or cardiometabolic disorders.

In the ORA of the discovery metabolites, four metabolite classes were significantly overrepresented. Four of the validated six metabolites belong to these classes: GABA and threonine (amino acids), biliverdin (bilirubins), and 3-methylxanthine (purines and pyrimidines). GABA is an inhibitory neurotransmitter involved in blocking neuronal signals, resulting in decreased nervous system activity. Similar to our study, prior work has found GABA inversely associated with depressive symptoms or chronic distress (Daniele et al. 2012; Shabel et al. 2014; Ma et al. 2016; Gabbay et al. 2017). While one recent investigation found increased GABA abundance among cases with major depressive disorder (MDD), sample size was limited (50 cases with MDD and 50 controls) and analyses did not adjust for important confounders such as comorbidities (Pan et al. 2018). Threonine, an amino acid involved in protein biosynthesis, was inversely associated with chronic distress in our study. A study of treatment-resistant depression in both males and females identified decreased levels of serum threonine among N=9 patients who did not respond to a 5-week antidepressant treatment vs. N=16 responders; although this direction is consistent with our results, the small sample size limits comparison (Maes et al. 1998). While a urine metabolomics study found higher threonine levels in 32 participants with depression and anxiety symptoms versus 32 controls, findings may not be comparable as the sample was small and the analysis was unadjusted for important potential confounders (Chen et al. 2018; Pan et al. 2018).

Biliverdin, a bile pigment and product of heme catabolism (Walter Boron 2016) belonging to the class of bilirubins, was inversely associated with chronic distress in our study, consistent other work finding heme oxygenase 1 expression (which drives the degradation of heme to biliverdin) inversely associated with more severe depressive symptoms (Robaczewska et al. 2016). Biliverdin is a strong antioxidant, and a negative association with chronic distress is consistent with prior work linking higher psychosocial distress with elevated oxidative stress (Black et al. 2017). Notably, other work has found inverse associations of biliverdin with cardiometabolic diseases (Kunutsor et al. 2015; Abbasi et al. 2015; Wang and Bautista 2015).

3-methylxanthine is a purine derivative involved in caffeine metabolism (Sanchez 2017) belonging to the class of purines and pyrimidines. While no prior population-based studies have found associations between 3-methylxanthine and depression and related disorders, we found 3-methylxanthine was positively associated with chronic distress even after adjustment for dietary caffeine (Table 2, Supplementary Figure S6). This marker may represent a novel factor associated with distress, although our results may be limited by inaccuracies in self-reported dietary caffeine intake.

Validated metabolites not contained in ORA-significant metabolite classes include serotonin and C16:0 ceramide. In our study, serotonin levels were lower in women with versus without chronic distress. Serotonin is a well-studied neurotransmitter, with numerous studies implicating reduced activity of serotonin pathways and serotonin imbalance in the pathogenesis of depression (Cowen and Browning 2015). Selective Serotonin Reuptake Inhibitors (SSRIs), a type of antidepressant, block serotonin uptake in the brain, increasing the extracellular availability of serotonin. Prior work has observed an inverse relationship between plasma serotonin and antidepressant use in patients with MDD (Holck et al. 2019), and decreases in plasma serotonin at 4 and 8 weeks after SSRI treatment (Gupta et al. 2016). Decreased plasma serotonin has also been observed in animal studies of chronic antidepressant treatment (Siesser et al. 2013). Levels of the sphingolipid C16:0 ceramide were positively associated with chronic distress in our study, consistent with findings from population-based studies (Gracia-Garcia et al. 2011; Dinoff et al. 2017). Sphingolipids are a diverse class of lipids commonly found in brain cell membranes (Merrill et al. 2005; Muller et al. 2015), and have been studied as potential targets for treating depression and other psychiatric disorders (Schneider et al. 2017). Additional work links aberrations within sphingolipid metabolism to coronary artery disease and metabolic disorders (Laaksonen et al. 2016; Dinoff et al. 2017).

Our novel differential network analysis provides important insight into systems-level metabolomic differences between women with and without chronic distress. Cotinine and hydroxycotinine, two metabolites involved in nicotine metabolism, were key nodes in the differential network. Tobacco use is generally higher among individuals with chronic distress for several reasons, including self-medication (Moran 2012). Additionally, cotinine has been proposed as a potential therapeutic for some forms of distress as well as Parkinson’s disease, due to its effects on the biology of the brain (Barreto et al. 2014; Moran 2012). The differential network estimated among non-smokers differed substantially from that constructed using the full sample: cotinine did not appear and hydroxycotinine played a minimal role, while cortisol and 21-deoxycortisol were implicated as key nodes (Supplementary Figure S2, Supplementary Table 4).

These six metabolites have also been evaluated in animal models, although the direction of effects is not always consistent with our findings. For example, GABA has been implicated in several previous animal models of various forms of distress. One study of chronic mild stress (CMS) in rats demonstrated decreased levels of GABA in the hippocampus and prefrontal cortex in CMS rats vs. healthy controls (Hemanth Kumar et al. 2012), and a study using a mouse model of depression similarly identified decreased GABAergic activity in the prefrontal cortex in mice with a depression-like phenotype vs. healthy controls (Veeraiah et al. 2014). However, conflicting results regarding the direction of the relationship between GABA and distress have been reported in other studies using animal models of distress; for example, one study found increased GABA in the hippocampus in mice subjected to chronic mild unpredictable stress vs. controls (Sekar et al. 2019). Sekar et al. also note that is difficult to compare across these studies due to differences across various animal models and tissues studied.

Other experimental animal studies have found an association of chronic distress with threonine, although the direction of effects is not always consistent with our findings of an inverse association. For example, threonine was elevated in rats exposed to chronic mild unpredictable stress vs. healthy controls (Ni et al. 2008); however, we note that this study was conducted on a very small sample of male mice (6 cases and 6 controls) and examined levels of threonine in several regions of the brain, while our work was conducted with a large sample of human females and considered levels of threonine in plasma. Interestingly, in a macaque model of naturally-occurring depression (NOD), both plasma and CSF threonine levels were elevated in 12 female macaques with NOD vs. 10 female controls (Xu et al. 2015; Deng et al. 2019).

In another experimental animal study, biliverdin reductase was elevated nearly two-fold in mice bred to be more anxious vs. mice with normal anxiety levels; biliverdin reductase catalyzes the degradation of biliverdin to bilirubin, a process which corroborates the decreased levels of biliverdin that we observed in women with chronic distress (Szego et al. 2010). Serotonin’s role in the pathophysiology of depression is implicated in numerous animal studies observing reduced levels of tryptophan, a serotonin pre-cursor, in the plasma of stressed animals compared to controls (Depke et al. 2008; Li et al. 2010; Xiong et al. 2016; Puurunen et al. 2016); plasma serotonin and tryptophan were both inversely associated with chronic distress in our NHS validation data. C16:0 ceramide is a sphingolipid; while this metabolite specifically has not been assessed in experimental studies, sphingolipid metabolism aberrations have been observed in animal models of common mental disorders like depression and anxiety (Gulbins et al. 2013; Gulbins et al. 2016). To our knowledge, 3-methylxanthine has not been examined in animal models of distress.

We note that, in keeping with experimental animal studies more generally, these animal studies generally relied on small sample sizes and used a broad range of proteomic or metabolomic platforms and tissues, making cross-study comparisons challenging. Moreover, it is unclear how well animal models of stress are representative of the experience of chronic distress in humans either with regard to the psychological or biological manifestations and sequelae (Steimer 2011).

A recent meta-analysis (Bot et al. 2020) integrated NMR metabolomics from 5,283 cases with depressive symptoms and 10,145 controls across nine Dutch cohorts and some comparison with our study is possible. Adjusting for sex, age, smoking, fasting status, and lipid-modifying medication, differential levels of various lipids and amino acids were observed in depressed cases relative to controls in the meta-analysis. Of these, isoleucine, tyrosine, palmitoleic acid and oleic acid, and several TAGs and DAGs were available in our study, but none met our discovery threshold. In minimally adjusted analyses, isoleucine and five TAGs (C54:1, C54:2, C54:3, C52:1, C58:9) had nominal significance (p < 0.05) in the NHS validation dataset but after adjustment for covariates, associations were attenuated (p > 0.35). We note that one-to-one mapping of metabolites between the two studies is limited by platform differences.

Strengths of our study include the use of multiple richly characterized cohorts enabling adjustment for a broad range of covariates along with the use of a well-characterized metabolomics platform measuring a large set of metabolites with robust CVs and low missingness. Our study was limited by the cross-sectional nature of the metabolomic profiling; temporal changes in metabolite profiles could not be investigated. In a previous investigation on the stability of metabolite profiles conducted in NHS, 90% of metabolites measured demonstrated within-individual reproducibility (Spearman correlation or intra-class correlation coefficient > 0.4 over 1-2 years)(Townsend et al. 2013). Several discovery metabolites were most strongly associated with antidepressant use; as antidepressant use is highly concordant with severity of distress, we are unable to disentangle antidepressant effects from effects of severe distress. As our study population was likely comprised of higher-functioning women, the most severe cases of chronic distress may not be represented. Due to data availability by cohort, chronic distress was assessed using indicators of depression in the discovery datasets and indicators of depression and anxiety in the validation dataset. Prior evidence suggests associations between metabolites and anxiety or depression are similar(Kessler et al. 2005), and thus we expect differences in assessment of distress between datasets did not dramatically influence findings. Between-cohort differences in participant characteristics, sample size, and data capture protocols may have hindered our ability to detect consistent signals between studies; these differences also argue that our validated findings are robust.

Using an agnostic approach, we identified six metabolites associated with chronic distress across diverse cohorts of women after adjusting for participant demographic, lifestyle, and medical characteristics. We identified systems-levels metabolomic differences between women with and without chronic distress by coupling a novel differential network analysis with our conventional analyses. Our findings suggest that a shared component of metabolic dysregulation may underlie various indicators of chronic distress. As chronic distress has been consistently linked to increased risk of cognitive and cardiometabolic disorders, our findings point to specific, testable hypotheses regarding metabolites that may be involved in the etiology of these and other chronic diseases of aging.

Supplementary Material

1
2
3
4

Highlights.

  • Metabolomic profiling found a metabolite signature of chronic distress in women

  • Our findings suggest metabolic dysregulation occurs in chronic distress

  • Metabolomic networks differ between women with distress and healthy controls

Acknowledgments

Metabolomic analysis in the Women’s Health Initiative (WHI) was funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services through contract HHSN268201300008C. The WHI program is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services through contracts HHSN268201600018C, HHSN268201600001C, HHSN268201600002C, HHSN268201600003C, and HHSN268201600004C.

We thank the participants and the staff of the Nurses’ Health Study (NHS) and the Nurses’ Health Study II (NHSII) for their valuable contributions. The NHS is supported by the National Institutes of Health (UM1CA186107, R01CA49449). The NHS II is also supported by the National Institutes of Health (U01CA176726, R01CA67262).

Research reported in this publication is supported by the National Institutes of Health under award number R01AG05160001A1.

This content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The authors assume full responsibility for analyses and interpretation of these data.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Declarations of interest: none

Code availability statement

The code used to conduct these analyses is available upon request from the corresponding author.

Human research participants statement

All participants in the WHI, NHS, and NHSII cohorts provided written informed consent. This study was approved by the Institutional Review Board of UMass-Amherst.

Competing interest statement

The authors declare no competing interests.

Data availability statement

Women’s Health Initiative data: Metabolite data analyzed in this study have been deposited in and are available from the dbGaP database under dbGaP accession no. phs001334.v1.p3.22. All other WHI data described in the manuscript, code book, and analytic code are available by request/research proposal at Women’s Health Initiative (www.whi.org).

Nurses’ Health Study and Nurses’ Health Study II data: NHS/NHSII data analyzed in this study are available by request/research proposal from the Nurses Health Studies (https://www.nurseshealthstudy.org/researchers).

REFERENCES

  1. Abbasi A, Deetman PE, Corpeleijn E, Gansevoort RT, Gans RO, Hillege HL, van der Harst P, Stolk RP, Navis G, Alizadeh BZ, and Bakker SJ. 2015. ‘Bilirubin as a potential causal factor in type 2 diabetes risk: a Mendelian randomization study’, Diabetes, 64: 1459–69. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Altmaier E, Emeny RT, Krumsiek J, Lacruz ME, Lukaschek K, Hafner S, Kastenmuller G, Romisch-Margl W, Prehn C, Mohney RP, Evans AM, Milburn MV, Illig T, Adamski J, Theis F, Suhre K, and Ladwig KH. 2013. ‘Metabolomic profiles in individuals with negative affectivity and social inhibition: a population-based study of Type D personality’, Psychoneuroendocrinology, 38: 1299–309. [DOI] [PubMed] [Google Scholar]
  3. Balasubramanian R, Paynter NP, Giulianini F, Manson JE, Zhao Y, Chen JC, Vitolins MZ, Albert CA, Clish C, and Rexrode KM. 2020. ‘Metabolomic profiles associated with all-cause mortality in the Women’s Health Initiative’, Int J Epidemiol, 49: 289–300. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Barreto GE, Iarkov A, and Moran VE. 2014. ‘Beneficial effects of nicotine, cotinine and its metabolites as potential agents for Parkinson’s disease’, Front Aging Neurosci, 6: 340. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Belanger CF, Hennekens CH, Rosner B, and Speizer FE. 1978. ‘The nurses’ health study’, Am J Nurs, 78: 1039–40. [PubMed] [Google Scholar]
  6. Benjamini Y; Krieger AM;Yekutieli D . 2006. ‘Adaptive Linear Step-up Procedures that Control the False Discovery Rate’, Biometrika, 93: 491–507. [Google Scholar]
  7. Bjorgvinsson T, Kertz SJ, Bigda-Peyton JS, McCoy KL, and Aderka IM. 2013. ‘Psychometric properties of the CES-D-10 in a psychiatric sample’, Assessment, 20: 429–36. [DOI] [PubMed] [Google Scholar]
  8. Black CN, Bot M, Scheffer PG, and Penninx BW. 2017. ‘Oxidative stress in major depressive and anxiety disorders, and the association with antidepressant use; results from a large adult cohort’, Psychol Med, 47: 936–48. [DOI] [PubMed] [Google Scholar]
  9. Bot M, Milaneschi Y, Al-Shehri T, Amin N, Garmaeva S, Onderwater GLJ, Pool R, Thesing CS, Vijfhuizen LS, Vogelzangs N, Arts ICW, Demirkan A, van Duijn C, van Greevenbroek M, van der Kallen CJH, Kohler S, Ligthart L, van den Maagdenberg Amjm, Mook-Kanamori DO, de Mutsert R, Tiemeier H, Schram MT, Stehouwer CDA, Terwindt GM, Willems van Dijk K, Fu J, Zhernakova A, Beekman M, Slagboom PE, Boomsma DI, Penninx Bwjh, and Bbmri-Nl Metabolomics Consortium. 2020. ‘Metabolomics Profile in Depression: A Pooled Analysis of 230 Metabolic Markers in 5283 Cases With Depression and 10,145 Controls’, Biol Psychiatry, 87: 409–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Chen JJ, Bai SJ, Li WW, Zhou CJ, Zheng P, Fang L, Wang HY, Liu YY, and Xie P. 2018. ‘Urinary biomarker panel for diagnosing patients with depression and anxiety disorders’, Transl Psychiatry, 8: 192. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Cohen BE, Edmondson D, and Kronish IM. 2015. ‘State of the Art Review: Depression, Stress, Anxiety, and Cardiovascular Disease’, Am J Hypertens, 28: 1295–302. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Cohen S, Janicki-Deverts D, and Miller GE. 2007. ‘Psychological stress and disease’, JAMA, 298: 1685–7. [DOI] [PubMed] [Google Scholar]
  13. Cowen PJ, and Browning M. 2015. ‘What has serotonin to do with depression?’, World Psychiatry, 14: 158–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Daniele S, Da Pozzo E, Abelli M, Panighini A, Pini S, Gesi C, Lari L, Cardini A, Cassano GB, and Martini C. 2012. ‘Platelet uptake of GABA and glutamate in patients with bipolar disorder’, Bipolar Disord, 14: 301–8. [DOI] [PubMed] [Google Scholar]
  15. Deng FL, Pan JX, Zheng P, Xia JJ, Yin BM, Liang WW, Li YF, Wu J, Xu F, Wu QY, Qu CH, Li W, Wang HY, and Xie P. 2019. ‘Metabonomics reveals peripheral and central short-chain fatty acid and amino acid dysfunction in a naturally occurring depressive model of macaques’, Neuropsychiatr Dis Treat, 15: 1077–88. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Depke M, Fusch G, Domanska G, Geffers R, Volker U, Schuett C, and Kiank C. 2008. ‘Hypermetabolic syndrome as a consequence of repeated psychological stress in mice’, Endocrinology, 149: 2714–23. [DOI] [PubMed] [Google Scholar]
  17. Dinoff A, Saleem M, Herrmann N, Mielke MM, Oh PI, Venkata SLV, Haughey NJ, and Lanctot KL. 2017. ‘Plasma sphingolipids and depressive symptoms in coronary artery disease’, Brain Behav, 7: e00836. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Freeman LC 1978. ‘Centrality in social networks conceptual clarification’, Social networks, 1: 215–39. [Google Scholar]
  19. Gabbay V, Bradley KA, Mao X, Ostrover R, Kang G, and Shungu DC. 2017. ‘Anterior cingulate cortex gamma-aminobutyric acid deficits in youth with depression’, Transl Psychiatry, 7: e1216. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Goodacre R, Vaidyanathan S, Dunn WB, Harrigan GG, and Kell DB. 2004. ‘Metabolomics by numbers: acquiring and understanding global metabolite data’, Trends Biotechnol, 22: 245–52. [DOI] [PubMed] [Google Scholar]
  21. Gracia-Garcia P, Rao V, Haughey NJ, Bandaru VV, Smith G, Rosenberg PB, Lobo A, Lyketsos CG, and Mielke MM. 2011. ‘Elevated plasma ceramides in depression’, J Neuropsychiatry Clin Neurosci, 23: 215–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Gulbins A, Grassme H, Hoehn R, Wilker B, Soddemann M, Kohnen M, Edwards MJ, Kornhuber J, and Gulbins E. 2016. ‘Regulation of Neuronal Stem Cell Proliferation in the Hippocampus by Endothelial Ceramide’, Cell Physiol Biochem, 39: 790–801. [DOI] [PubMed] [Google Scholar]
  23. Gulbins E, Palmada M, Reichel M, Luth A, Bohmer C, Amato D, Muller CP, Tischbirek CH, Groemer TW, Tabatabai G, Becker KA, Tripal P, Staedtler S, Ackermann TF, van Brederode J, Alzheimer C, Weller M, Lang UE, Kleuser B, Grassme H, and Kornhuber J. 2013. ‘Acid sphingomyelinase-ceramide system mediates effects of antidepressant drugs’, Nat Med, 19: 934–8. [DOI] [PubMed] [Google Scholar]
  24. Gupta M, Neavin D, Liu D, Biernacka J, Hall-Flavin D, Bobo WV, Frye MA, Skime M, Jenkins GD, Batzler A, Kalari K, Matson W, Bhasin SS, Zhu H, Mushiroda T, Nakamura Y, Kubo M, Wang L, Kaddurah-Daouk R, and Weinshilboum RM. 2016. ‘TSPAN5, ERICH3 and selective serotonin reuptake inhibitors in major depressive disorder: pharmacometabolomics-informed pharmacogenomics’, Mol Psychiatry, 21: 1717–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Ha MJ, Baladandayuthapani V, and Do KA. 2015. ‘DINGO: differential network analysis in genomics’, Bioinformatics, 31: 3413–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Hashimoto K, Bruno D, Nierenberg J, Marmar CR, Zetterberg H, Blennow K, and Pomara N. 2016. ‘Abnormality in glutamine-glutamate cycle in the cerebrospinal fluid of cognitively intact elderly individuals with major depressive disorder: a 3-year follow-up study’, Transl Psychiatry, 6: e744. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Hemanth Kumar BS, Sushanta Kumar Mishra, Poonam Rana, Sadhana Singh, and Subash Khushu. 2012. ‘Neurodegenerative evidences during early onset of depression in CMS rats as detected by proton magnetic resonance spectroscopy at 7T’, Behavioural Brain Research, 232: 53–59. [DOI] [PubMed] [Google Scholar]
  28. Holck A, Wolkowitz OM, Mellon SH, Reus VI, Nelson JC, Westrin A, and Lindqvist D. 2019. ‘Plasma serotonin levels are associated with antidepressant response to SSRIs’, J Affect Disord, 250: 65–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Huang T, Balasubramanian R, Yao Y, Clish CB, Shadyab AH, Liu B, Tworoger SS, Rexrode KM, Manson JE, Kubzansky LD, and Hankinson SE. 2020. ‘Associations of depression status with plasma levels of candidate lipid and amino acid metabolites: a meta-analysis of individual data from three independent samples of US postmenopausal women’, Mol Psychiatry. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Huang T, Trudel-Fitzgerald C, Poole EM, Sawyer S, Kubzansky LD, Hankinson SE, Okereke OI, and Tworoger SS. 2019. ‘The Mind-Body Study: study design and reproducibility and interrelationships of psychosocial factors in the Nurses’ Health Study II’, Cancer Causes Control, 30: 779–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Karabatsiakis A, Hamuni G, Wilker S, Kolassa S, Renu D, Kadereit S, Schauer M, Hennessy T, and Kolassa IT. 2015. ‘Metabolite profiling in posttraumatic stress disorder’, J Mol Psychiatry, 3: 2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Kessler RC, Berglund P, Demler O, Jin R, Merikangas KR, and Walters EE. 2005. ‘Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the National Comorbidity Survey Replication’, Arch Gen Psychiatry, 62: 593–602. [DOI] [PubMed] [Google Scholar]
  33. Kleinberg Jon M. 1999. ‘Authoritative sources in a hyperlinked environment’, Journal of the ACM (JACM), 46: 604–32. [Google Scholar]
  34. Kunutsor SK, Bakker SJ, Gansevoort RT, Chowdhury R, and Dullaart RP. 2015. ‘Circulating total bilirubin and risk of incident cardiovascular disease in the general population’, Arterioscler Thromb Vasc Biol, 35: 716–24. [DOI] [PubMed] [Google Scholar]
  35. Laaksonen R, Ekroos K, Sysi-Aho M, Hilvo M, Vihervaara T, Kauhanen D, Suoniemi M, Hurme R, Marz W, Scharnagl H, Stojakovic T, Vlachopoulou E, Lokki ML, Nieminen MS, Klingenberg R, Matter CM, Hornemann T, Juni P, Rodondi N, Raber L, Windecker S, Gencer B, Pedersen ER, Tell GS, Nygard O, Mach F, Sinisalo J, and Luscher TF. 2016. ‘Plasma ceramides predict cardiovascular death in patients with stable coronary artery disease and acute coronary syndromes beyond LDL-cholesterol’, Eur Heart J, 37: 1967–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Li ZY, Zheng XY, Gao XX, Zhou YZ, Sun HF, Zhang LZ, Guo XQ, Du GH, and Qin XM. 2010. ‘Study of plasma metabolic profiling and biomarkers of chronic unpredictable mild stress rats based on gas chromatography/mass spectrometry’, Rapid Commun Mass Spectrom, 24: 3539–46. [DOI] [PubMed] [Google Scholar]
  37. Liu X, Zheng P, Zhao X, Zhang Y, Hu C, Li J, Zhao J, Zhou J, Xie P, and Xu G. 2015. ‘Discovery and validation of plasma biomarkers for major depressive disorder classification based on liquid chromatography-mass spectrometry’, J Proteome Res, 14: 2322–30. [DOI] [PubMed] [Google Scholar]
  38. Liu Y, Yieh L, Yang T, Drinkenburg W, Peeters P, Steckler T, Narayan VA, Wittenberg G, and Ye J. 2016. ‘Metabolomic biosignature differentiates melancholic depressive patients from healthy controls’, BMC Genomics, 17: 669. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Ma K, Xu A, Cui S, Sun MR, Xue YC, and Wang JH. 2016. ‘Impaired GABA synthesis, uptake and release are associated with depression-like behaviors induced by chronic mild stress’, Transl Psychiatry, 6: e910. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Maes M, Verkerk R, Vandoolaeghe E, Lin A, and Scharpe S. 1998. ‘Serum levels of excitatory amino acids, serine, glycine, histidine, threonine, taurine, alanine and arginine in treatment-resistant depression: modulation by treatment with antidepressants and prediction of clinical responsivity’, Acta Psychiatr Scand, 97: 302–8. [DOI] [PubMed] [Google Scholar]
  41. Merrill AH Jr., Sullards MC, Allegood JC, Kelly S, and Wang E. 2005. ‘Sphingolipidomics: high-throughput, structure-specific, and quantitative analysis of sphingolipids by liquid chromatography tandem mass spectrometry’, Methods, 36: 207–24. [DOI] [PubMed] [Google Scholar]
  42. Moran VE 2012. ‘Cotinine: Beyond that Expected, More than a Biomarker of Tobacco Consumption’, Front Pharmacol, 3: 173. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Muller CP, Reichel M, Muhle C, Rhein C, Gulbins E, and Kornhuber J. 2015. ‘Brain membrane lipids in major depression and anxiety disorders’, Biochim Biophys Acta, 1851: 1052–65. [DOI] [PubMed] [Google Scholar]
  44. Ni Y, Su M, Lin J, Wang X, Qiu Y, Zhao A, Chen T, and Jia W. 2008. ‘Metabolic profiling reveals disorder of amino acid metabolism in four brain regions from a rat model of chronic unpredictable mild stress’, FEBS Lett, 582: 2627–36. [DOI] [PubMed] [Google Scholar]
  45. Pan JX, Xia JJ, Deng FL, Liang WW, Wu J, Yin BM, Dong MX, Chen JJ, Ye F, Wang HY, Zheng P, and Xie P. 2018. ‘Diagnosis of major depressive disorder based on changes in multiple plasma neurotransmitters: a targeted metabolomics study’, Transl Psychiatry, 8: 130. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Paynter NP, Balasubramanian R, Giulianini F, Wang DD, Tinker LF, Gopal S, Deik AA, Bullock K, Pierce KA, Scott J, Martinez-Gonzalez MA, Estruch R, Manson JE, Cook NR, Albert CM, Clish CB, and Rexrode KM. 2018. ‘Metabolic Predictors of Incident Coronary Heart Disease in Women’, Circulation, 137: 841–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Puurunen J, Tiira K, Lehtonen M, Hanhineva K, and Lohi H. 2016. ‘Non-targeted metabolite profiling reveals changes in oxidative stress, tryptophan and lipid metabolisms in fearful dogs’, Behav Brain Funct, 12: 7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Robaczewska J, Kedziora-Kornatowska K, Kucharski R, Nowak M, Muszalik M, Kornatowski M, and Kedziora J. 2016. ‘Decreased expression of heme oxygenase is associated with depressive symptoms and may contribute to depressive and hypertensive comorbidity’, Redox Rep, 21: 209–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Roberts LD, Souza AL, Gerszten RE, and Clish CB. 2012. ‘Targeted metabolomics’, Curr Protoc Mol Biol, Chapter 30: Unit 30 2 1–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Sanchez JM 2017. ‘Methylxanthine Content in Commonly Consumed Foods in Spain and Determination of Its Intake during Consumption’, Foods, 6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Schneider M, Levant B, Reichel M, Gulbins E, Kornhuber J, and Muller CP. 2017. ‘Lipids in psychiatric disorders and preventive medicine’, Neurosci Biobehav Rev, 76: 336–62. [DOI] [PubMed] [Google Scholar]
  52. Sekar S, Grandjean J, Garnell JF, Willems R, Duytschaever H, Seramani S, Su H, Ver Donck L, and Bhakoo KK. 2019. ‘Neuro-metabolite profiles of rodent models of psychiatric dysfunctions characterised by MR spectroscopy’, Neuropharmacology, 146: 109–16. [DOI] [PubMed] [Google Scholar]
  53. Shabel SJ, Proulx CD, Piriz J, and Malinow R. 2014. ‘Mood regulation. GABA/glutamate co-release controls habenula output and is modified by antidepressant treatment’, Science, 345: 1494–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Siesser WB, Sachs BD, Ramsey AJ, Sotnikova TD, Beaulieu JM, Zhang X, Caron MG, and Gainetdinov RR. 2013. ‘Chronic SSRI treatment exacerbates serotonin deficiency in humanized Tph2 mutant mice’, ACS Chem Neurosci, 4: 84–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Steimer T 2011. ‘Animal models of anxiety disorders in rats and mice: some conceptual issues’, Dialogues Clin Neurosci, 13: 495–506. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Szego EM, Janaky T, Szabo Z, Csorba A, Kompagne H, Muller G, Levay G, Simor A, Juhasz G, and Kekesi KA. 2010. ‘A mouse model of anxiety molecularly characterized by altered protein networks in the brain proteome’, Eur Neuropsychopharmacol, 20: 96–111. [DOI] [PubMed] [Google Scholar]
  57. Townsend MK, Clish CB, Kraft P, Wu C, Souza AL, Deik AA, Tworoger SS, and Wolpin BM. 2013. ‘Reproducibility of metabolomic profiles among men and women in 2 large cohort studies’, Clin Chem, 59: 1657–67. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Tworoger SS, Eliassen AH, Zhang X, Qian J, Sluss PM, Rosner BA, and Hankinson SE. 2013. ‘A 20-year prospective study of plasma prolactin as a risk marker of breast cancer development’, Cancer Res, 73: 4810–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Veeraiah P, Noronha JM, Maitra S, Bagga P, Khandelwal N, Chakravarty S, Kumar A, and Patel AB. 2014. ‘Dysfunctional glutamatergic and gamma-aminobutyric acidergic activities in prefrontal cortex of mice in social defeat model of depression’, Biol Psychiatry, 76: 231–8. [DOI] [PubMed] [Google Scholar]
  60. Boron Walter, Boulpaep Emile. 2016. Medical Physiology (Elsevier; ). [Google Scholar]
  61. Wang L, and Bautista LE. 2015. ‘Serum bilirubin and the risk of hypertension’, Int J Epidemiol, 44: 142–52. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Xiong Z, Yang J, Huang Y, Zhang K, Bo Y, Lu X, Su G, Ma J, Yang J, Zhao L, and Wu C. 2016. ‘Serum metabonomics study of anti-depressive effect of Xiao-Chai-Hu-Tang on rat model of chronic unpredictable mild stress’, J Chromatogr B Analyt Technol Biomed Life Sci, 1029-1030: 28–35. [DOI] [PubMed] [Google Scholar]
  63. Xu Fan, Wu Qingyuan, Xie Liang, Gong Wei, Zhang Jianguo, Zheng Peng, Zhou Qinmin, Ji Yongjia, Wang Tao, Li Xin, Fang Liang, Li Qi, Yang Deyu, Li Juan, Melgiri Narayan D., Shively Carol, and Xie Peng. 2015. ‘Macaques Exhibit a Naturally-Occurring Depression Similar to Humans’, Scientific Reports, 5: 9220. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1744267-supplement-1.xlsx (43KB, xlsx)
2
NIHMS1744267-supplement-2.xlsx (18.6KB, xlsx)
3
NIHMS1744267-supplement-3.docx (2.8MB, docx)
4
NIHMS1744267-supplement-4.xlsx (27.2KB, xlsx)

Data Availability Statement

Women’s Health Initiative data: Metabolite data analyzed in this study have been deposited in and are available from the dbGaP database under dbGaP accession no. phs001334.v1.p3.22. All other WHI data described in the manuscript, code book, and analytic code are available by request/research proposal at Women’s Health Initiative (www.whi.org).

Nurses’ Health Study and Nurses’ Health Study II data: NHS/NHSII data analyzed in this study are available by request/research proposal from the Nurses Health Studies (https://www.nurseshealthstudy.org/researchers).

RESOURCES