Longitudinal association of homocysteine with depressive and anxiety symptoms among urban adults: healthy aging in neighborhoods of diversity across the life span study

Abstract

Longitudinal associations of homocysteine (HCY) with depressive symptoms scores among urban adults remain under-studied, especially across sex, race and levels of anxiety. We examined longitudinal associations of homocysteine (HCY) with depressive symptoms scores among urban adults, before and after stratifying by sex, race and anxiety level, using data from 1460 Healthy Aging in Neighborhoods of Diversity across the Lifespan Study (HANDLS) participants aged 30–64 y at v₁ (2004–2009), followed across 3 visits up to 2017. In addition to LnHcy_v1, we used group-based trajectory models predicting z-transformed likelihood of greater LnHcy with age (Hcy_traj). Total and domain-specific depression symptoms were scored using Center for Epidemiologic Studies Depression (CES-D) scale. Mixed-effects linear regression models and Cox proportional hazards models were utilized. A positive association was found between baseline LnHcy_v1 and CES-D total scores in reduced socio-demographic- adjusted Model 1 (β (standard error [SE]) = + 2.337 (0.902), P = 0.010), a relationship slightly attenuated in fully adjusted Model 2 (Model 1 adjusting for lifestyle and health factors) with a β (SE) = + 1.825 (0.883), P = 0.039. Individuals with lower anxiety levels experienced faster CES-D domain 2 score annualized increase over time (interpersonal problems) with higher LnHcy_v1 (β (SE) = 0.041 (0.018), P = 0.024). Hcy_traj was linked to incident elevated depressive symptoms (CES-D total score ≥16) overall (fully adjusted model: HR = 1.09, 95% CI: 1.03–1.14, P = 0.001), particularly among women and those living in poverty. Baseline and “high trajectory” of LnHcy were positively associated with depressive symptoms and elevated depressive symptom incidence, in a sex-, race-, poverty status- and anxiety-level specific manner.

Introduction

According to the World Health Organization and United Nations Development Program projections, the number of those over 60 would increase to 2.1 billion by 2050, from 1.4 billion in 2030 [1, 2]. Declines in social, cognitive, and physical function associated with aging has been related to worse quality of life [3]. An earlier study discovered a reciprocal association between quality of life in older age and mental health outcomes, such as anxiety and depression [4]. Depression, while being underdiagnosed frequently, is a major contributor to the worldwide sickness burden and has been linked to higher rates of morbidity and mortaltiy in the elderly [2, 5]. Hu et al.‘s latest 48-study meta-analysis estimated that 28.4% of older adults had depression, with a 95% confidence interval (CI) ranging from 24.8% to 32.0% [1]. Personal characteristics associated with geriatric depression include female sex, aging, being single or divorced, having less education, being unemployed, low income, no health insurance, smoking, having grown up through traumatic experiences, low self-esteem, social deprivation, being alone or lonely, bereavement, chronic conditions, cognitive impairment, poor health, and history of depression [2].

Studies suggest that stressful life experiences and long-term chronic stress may be linked to geriatric depression [6–8]. Physical symptoms are more common in late life onset depression than in early onset depression, and vascular disease may contribute to, exacerbate, or prolong depressive symptoms in old age [5]. A meta-analysis of 61 prospective cohort studies found that the presence of late-life depression may raise the risk of cardiovascular disease-related death by >30% [5]. Investigating connection between stressors and symptoms of depression can help identify strategies for averting late-life depression and its detrimental impact on health. Homocysteine (HCY), which plays a crucial role in the biochemical balance within the central nervous system, shows strong associations with cardiovascular (i.e., stroke and coronary artery disease), neurological (i.e., dementia), and psychiatric disease (depression and anxiety) when elevated [9–22]. A meta-analysis of 46-studies revealed that elevated HCY is associated with depression [23]. Several studies identified positive relationships between elevated HCY levels and anxiety disorders, including PTSD [13, 17, 18, 24]. However, despite such evidence, findings concerning HCY’s associations with depressive symptoms, in the presence or absence of anxiety disorders remains minimal.

In this study, urban middle-aged adults participating in the Health Aging in Neighborhoods of Diversity across the Life Span (HANDLS) investigation had their long-term associations between HCY and depression examined. We hypothesized that, differently depending on race and sex, higher HCY would be associated with higher symptoms of anxiety and depression. We also investigated whether anxiety levels affected the relationship over time between HCY and depressive symptoms. Both anxiety and depressive symptoms were compared to the long-term levels and change in HCY with age as a secondary analysis.

Methods

Database

Initiated in 2004 by the National Institute on Aging (NIA) Intramural Research Program (IRP), the HANDLS study is a prospective cohort study conducted to address research topics linked to health disparities in age-associated disorders. Using an area probability sampling strategy, middle-aged African American and White individuals of both sexes (baseline age: 30-64 years) were chosen from thirteen Baltimore city neighborhoods with widely varying household incomes in order to create the sample of urban adults for the HANDLS study [25]. Additionally, to increase participation rates and retention among non-traditional research participants, the HANDLS project makes use of mobile medical research vehicles (MRVs) and innovative research methods. The HANDLS project was authorized by the National Institutes of Health’s Institutional Review Board, and participants gave written informed consent [25].

Between 2004 and 2009, baseline (Visit 1) data on HANDLS participants were gathered in two stages: (i) an in-home interview including health status, health care utilization, psychosocial factors, diet, neighborhood characteristics, and demographics comprised the first phase of the study and (ii) the subsequent stage encompassed the following aspects: medical history, physical examination, dietary recall, cognitive assessment, psychophysiological evaluations (e.g., heart rate variability, arterial thickness, carotid ultrasonography, muscle strength, bone density assessments), and laboratory assessments (e.g., blood chemistries, hematology, biomarkers of oxidative stress, biomaterials for genetic studies) all carried out in MRVs. Following the initial visit, HANDLS participants were contacted every five years for Visits 2 and 3, which were conducted between 2009 and 2013 and 2013 and 2017, respectively. HANDLS data documentation are available at https://handls.nih.gov/06Coll-w00dataDocR.cgi. While some HANDLS study visits involved distinct forms of evaluations, several assessments were repeated throughout time. This analysis harnessed data from Visits 1, 2, and 3, including data on blood HCY concentrations and depressive symptom scores.

Measures

Homocysteine (HCY)

Plasma HCY was measured by the Alinity i analyzer Aeon Technologies, LLC (Frostburg, MD). First, the serum quality was measured by hemolysis, lipemia, and icteria detection. All serum samples passed the quality check threshold determined by Aeon Technologies. HCY was quantified using the Alinity i Homocysteine assay on the Alinity i analyzer [26]. This assay is a one-step immunoassay that uses chemiluminescent microparticle immunoassay (CMIA) technology. The analytical measuring range (AMR) of the Alinity i Homocysteine assay is 1.00 to 50.00 mmol/L (0.14 to 6.76 mg/mL). Samples were run in 12 batches and a serum sample (Cat # 200-0162; Stem Cell Technologies) was run in each batch as a control. The homocysteine values for this serum control ranged from 4.55 to 6.18, with an inter-assay coefficient of variation of 8.38%.

Using a STATA plugin (traj and trajplot) modified from a well-established SAS approach [27, 28], group-based trajectory modeling (GBTM) was carried out for Log_e transformed HCY, or LnHcy measured at visits 1, 2, and 3. This allowed for the identification of adult groups with comparable developmental trajectories in terms of level and trends throughout time. This group-based method uses maximum likelihood and a multinomial modeling strategy to estimate model parameters. The quasi-Newton method was selected as the likelihood optimization technique. The traj command uses maximum likelihood estimation to preserve data and mitigate biases from listwise elimination, assuming missingness at random.

Group-based trajectories with age were displayed with 95% confidence intervals (CI) for each group trajectory. We defined a censored normal distribution for LnHcy, with intercept (0), linear (1), or quadratic (2) orders for each group trajectory, as appropriate. The most parsimonious model with best the fit was determined using the BIC. If linearity was established using the trajplot, up to three groups were taken into consideration, and the linear model was selected. Group membership probability was the main metric of interest obtained from the GBTM. If there was an identified trajectory group with less than 10% prevalence in the three-group model, then two groups were selected. Given prior evidence, the trajectory group with elevated and/or increasing HCY over time was selected as the main exposure of interest [9–22], and the exposure was operationalized using group membership probability.

Depressive symptoms

Using self-reported information from the Center for Epidemiological Studies Depression (CES-D) questionnaire, a depressed symptoms score was computed [29]. Several samples of older adults have demonstrated the adequate psychometric properties of the CES-D questionnaire (e.g. [30]). With 20 items and item scores ranging from “0” to “3,” the CES-D total score range anywhere from “0” and “60.” The frequency and intensity of depressed symptomatology during the previous week were the main aspects of the CES-D questions. Participants in the HANDLS study were asked to select whether they experienced an item most or all of the time (score = 3), seldom or a substantial portion of the time (score = 2), rarely or never (score = 0), or some or a little of the time (score = 1). Some questions required reverse coding. Additionally, we looked at domain-specific CES-D scores, such as those for (1) depressive affect (such as feeling depressed); (2) interpersonal problems (such as experiencing social anxiety); (3) somatic complaints (such as difficulty sleeping or eating); and (4) positive affect (such as thinking positively) [29]. By summing the scores for depressed symptoms for each item under each domain, we were able to determine the raw CES-D sub-scores for each domain. Information about the items used to calculate each domain-specific CES-D sub-score is described elsewhere [29]. A binary version of CES-D total score was also computed using a cutoff of 16, commonly used in other studies [31], for a sensitivity analysis on elevated depressive symptoms (EDS) incidence.

Anxiety symptoms

Self-report information from the Psychiatric Diagnostic Screening Questionnaire Generalized Anxiety Disorder subscale (PDSQga) was used to compute an anxiety symptom score. There were strong psychometric properties ascribed to the PDSQga among adults [32, 33]. There are a total of 10 items with a scoring of “No”, “Yes”, “Don’t Know”, “Not Applicable”. A total PDSQga score ≥ 6 indicated substantial anxiety symptoms. Information about the items used to calculate anxiety symptoms is described in Supplementary Method 1.

For a secondary analysis whereby anxiety symptoms were the outcome of interest, GBTM was used to obtain two groups for anxiety scores between visits 1 and 3 (High vs. Low), using a zero-inflated Poisson (zip) distribution. This binary trajectory outcome was then converted into a Log_e(odds of “High” anxiety) and used as main continuous outcome of interest in a series of multiple linear regression models [27, 28]. Anxiety disorder was also elicited at all 3 visits. This variable was used for validation only by cross-tabulating it with the main anxiety disorder variable (above vs. below median) and examining mean Log_e(odds of “High” anxiety) across baseline anxiety disorder.

Covariates

The 2010 Healthy Eating Index [HEI-2010] and health (body mass index [BMI; weight/height2 in kg.m⁻², continuous]) characteristics were among factors taken into account when examining the hypothesized relationships between HCY and depressive symptoms. Other key confounders included demographics (sex (male, female), age (years), race (White, African American), poverty status (<125% federal poverty line, ≥125% federal poverty line), education (less than high school, high school, more than high school), lifestyle (current cigarette smoking [Yes, No]) and use of drugs (Yes, No [using any of marijuana, opiates, and cocaine]). The age at baseline (Visit 1) was examined as a continuous variable, and the follow-up time durations were computed using the age at Visits 2 and 3. The Department of Health and Human Services’ poverty standards were used to operationalize the state of poverty based on household income and total household size [34]. The HEI-2010 [34] based its overall diet quality recommendations on food and macronutrient standards for the American population from the Dietary Guidelines for Americans. We also described the relationship between HCY and several downstream health-related factors including self-rated health and comorbidities without including those into the main models given their potential mediating effect between HCY and depressive symptoms. Comorbidities were characterized as follows: self-reported history of any of several cardiovascular diseases (no, yes), including atrial fibrillation, angina, coronary artery disease, congestive heart failure, and myocardial infarction; diabetes (non-diabetic, pre-diabetic, diabetic); dyslipidemia ([or statin use] no, yes); and hypertension (no, yes). Three categories for self-rated health were established: very good/excellent, good, and poor/average.

Effect modifiers

The main effect modifiers in this study were sex (male vs. female) and race (African American vs. White). In addition, baseline anxiety symptoms (above vs. below median of total score) was also considered as an important effect modifier in the association between HCY and depressive symptoms over time.

Statistical methods

Stata version 18 (StataCorp, College Station, TX) was used to conduct descriptive, bivariate, and multivariable analyses. While counts and percentages were employed to describe categorical data, measures of central tendency (mean, median) and dispersion (standard deviation, standard error, interquartile range) were used to characterize continuous variables. Testing for multicollinearity among the variables included in mixed-effects models using correlation matrices was one of the model-building procedures. The mixed command in Stata removes missing data from analysis by removing observations with missing outcomes, but can cause data loss and biases if not random. Nevertheless, the model includes observations with at least one outcome available, assuming missingness at random. Given that each covariate had less than 5% missing data on average, we ensured sample sizes were constant between different adjusted models by performing multiple imputations (5 imputations, 10 iterations) using the chained equations methodology in order to reduce missing data caused by the addition of covariates into different models. Only the potentially confounding covariates were imputed, while outcome and exposures were not. Stata commands to this end included mi impute, mi passive and mi estimate among others. During this estimation method, all covariates were entered simultaneously, and continuous covariates were centered on their means, just like in earlier research [35, 36]. First, using the largest sample after excluding HANDLS subjects with missing CES-D data, baseline sociodemographic, lifestyle, and health characteristics, CES-D test scores (at baseline and change over time), as well as LnHcy_v1 and z-transformed probability of higher LnHcy with age, were described before and after stratifying according to Hcy_v1 tertiles. Differences across tertiles were tested using a series of bivariate linear regression and multinomial logistic regression models to compare means and odds of belonging to specific categories across these tertiles. Specifically, the uppermost two tertiles of LnHcy_v1 were compared to the lowest tertile. Second, different sets of covariates were accounted for when building a series of mixed-effects linear regression models for baseline HCY as a predictor of CES-D test scores (at baseline and change over time) and z-transformed probability of higher HCY trajectory as a predictor of CES-D test scores (at baseline and change over time), (Supplementary method 2). Time spent on study, measured in years, between visits 1 and 3 was the time variable employed. Age, sex, race, poverty status, inverse mills ratio (IMR), time spent on studies, and its interaction with variables including LnHcy_v1 and Hcy_traj alternative exposures were all taken into account while adjusting Model 1. Model 2 included time on study and its interaction with LnHcy_v1 or Hcy_traj and other covariates such as age, sex, race, poverty status, education, literacy, smoking, drug use, and the 2010-HEI, BMI, and IMR. For Models 1 and 2, the interaction effects of LnHcy_v1 or Hcy_traj with sex and race were assessed as a sensitivity analysis. Moreover, stratified analyses were carried out independently for White and African American HANDLS participants, males, and women, as was done in previous studies examining blood biomarkers in relation to depressive symptoms [31, 37–39]. Therefore, we applied Models 1-2 to two exposures (LnHcy_v1 and Hcy_traj), five CES-D test scores (one total score and four domain-specific scores), two stratifying variables (sex, race), and up to two repeats (impact on baseline CES-D test scores and effect on change in CES-D test scores). Using a two-stage Heckman selection technique, we corrected for sample selectivity resulting from missing data in all models [26]. We estimated those mixed-effects linear regression models adjusted for the IMR in addition to the aforementioned covariates after using a probit regression model to predict an indicator of selection with sex, age at Visit 1, race, and poverty status. This model produced an IMR, or a function of the probability of being selected given these characteristics [40]. Predictive margins from the main mixed-effects linear regression models (Model 1, CES-D total score) were obtained and plotted across time for illustration purposes.

As a secondary outcome, the global anxiety score was treated similarly, but with a maximum of two visits (visits 1 and 3). In contrast to the CES-D score outcome, Log_e odds(“high anxiety trajectory”) was the main outcome of interest which was entered into a linear regression model. Prior to multiple testing correction, the type I error rate for the main and interaction effects was predetermined to be 0.05 and 0.10, respectively [41]. We used the familywise Bonferroni correction approach [42] to adjust for outcome multiplicity (i.e., five CES-D test scores and one global anxiety test score), especially for Model 1. Following that, Model 2 was regarded as a sensitivity model that incorporated variables that could be mediating or confounding. In line with earlier research [43], we therefore changed the significance thresholds for the main effects to p < 0.010 (0.05/5) and the two-way interaction terms to 0.10/5 = 0.020.

Four sensitivity analyses were also performed. First, Ln(odds of elevated anxiety) was the outcome in a linear regression model with primary predictor being baseline Log_e(Hcy) [LnHcy]. A reduced and a fully adjusted model was conducted. Second, mixed-effects linear regression models were carried out whereby the main outcomes were CES-D total and domain-specific scores and the predictor was annualized change in LnHcy between visits 1 and 2. The models were incrementally adjusted as in earlier models. Moreover, a third sensitivity analysis was conducted whereby a mixed-effects linear regression model with time-dependent LnHcy as the outcome was conducted in the overall sample examining its association with various exposures, including baseline CES-D total score, ordinal anxiety score, probability of higher anxiety score and anxiety disorder at baseline. Finally, Cox proportional hazards models were conducted on the time-dependent binary version of the CES-D total score with a cutoff of 16. Time on study was used as the time variable. Analyses were also conducted on multiple imputed data. Specifically, the fully adjusted models were presented, examining associations of baseline and trajectory exposures of Hcy with incidence of EDS, after adjustment for the same vector of covariates included in the mixed-effects linear regression models. Stratification by sex, race, poverty status and anxiety score level was carried out, while also testing for interaction between each exposure and effect modifier separately in the fully adjusted models. The full Stata script will be made available on github at: https://github.com/baydounm/HANDLS_Hcy_Depression_Anxiety.

Results

As shown in Fig. 1, of 3,720 participants recruited initially in the HANDLS study, 1,460 participants had complete data on Hcy at visit 1, all of whom had complete data on CES-D scores at any of visits 1, 2 or 3, which made the final analytic sample for the main outcome of depressive symptoms over time. Based on Fig. 2, three separate groups were identified using group-based trajectory modeling for LnHcy levels, indicating a positive linear trend with age and an initial level that can be labelled as “Low”, “Medium” and “High”. The probability of belonging to the “High” group was on average 0.12. Importantly, 1 SD in Hcyt_raj in the selected sample was equivalent to a 0.28 increased probability of membership in the “High” group (data shown on https://github.com/baydounm/HANDLS_Hcy_Depression_Anxiety). Figure 3 shows the findings for the anxiety score using GBTM, which yielded two groups, namely “High” (46.5%) vs. “Low” (53.5%).

Fig. 1. Study Flowchart – HANDLS (2004-2017).

Notes: CES-D Center for Epidemiological Studies Depression, HANDLS healthy aging in neighborhoods of diversity across the lifespan study.

Fig. 2. Group-based trajectories for homocysteine – HANDLS (2004-2017).

Notes: HANDLS healthy aging in neighborhoods of diversity across the lifespan study; A = Table display of intercept and linear terms for the three trajectories in homocysteine identified using group-based trajectories; B = A graphical display of the three groups identified using group-based trajectory modeling is shown, whereby Ln_Hcy represents homocysteine total score (Log_e transformed) and Age (years) represents the time variable.

Fig. 3. Group-based trajectories for anxiety score – HANDLS (2004–2017).

Notes: HANDLS healthy aging in neighborhoods of diversity across the lifespan study; A = Table display of intercept and linear terms for the three trajectories in anxiety score, ordered (ANXIETY) identified using group-based trajectories; B = A graphical display of the two groups identified using group-based trajectory modeling is shown, whereby ANXIETY is the ordered anxiety score (range:0–10) and Age (years) represents the time variable.

Table 1 describes socio-demographic, lifestyle and health characteristics at baseline as well as HCY and CES-D total scores (at baseline and change over time), overall, and across tertiles of baseline HCY. The means ( ± standard error [19]) for baseline LnHcy and CES-D total scores were estimated at 2.15 ( ± 0.01) and 14.01 ( ± 0.29), respectively. Similarly, the means ( ± SEM) for probability of high LnHcy and between-visit change in CES-D total score (annualized empirical bayes estimator) were estimated at 0.12 ± 0.01 and −0.12 ± 0.00, respectively. Means and proportions of several characteristics differed across baseline HCY tertiles. Mean age increased across the tertiles of Hcy as did the prevalence proportions of hypertension, diabetes, and elevated cholesterol. In contrast, we observed lower HEI-2010 scores, proportions of female subjects, and proportions of cigarette smokers across higher tertiles of HCY. Non-linear J-shaped associations between Hcy tertiles and CES-D total and sub-domain scores were detected. Similarly, ordered baseline anxiety score exhibited a non-linear inverted U-shaped association with HCY tertiles and an inverse relationship with the Log-odds of “high anxiety” probability.

Table 1.

Summary statistics for socio-demographic, lifestyle and health characteristics at baseline (visit 1: 2004–2009); homocysteine, depressive symptoms and anxiety scores (at baseline and change over time), overall, and according to tertiles of baseline homocysteine (N = 1460): HANDLS 2004–2017.

	% or Mean ± SEM	Hcy at visit 1 tertiles
	N = 1460	n = 485	n = 489	n = 486
		1st	2nd	3rd
Homocysteine:
Hcy at visit 1	9.18 ± 0.14	6.29 ± 0.04	8.42 ± 0.03***	12.82 ± 0.35***
LnHcy at visit 1	2.15 ± 0.01	1.83 ± 0.01	2.13 ± 0.00***	2.49 ± 0.01***
Hcytraj: probability of “High” LnHcy	0.12 ± 0.01	0.02 ± 0.00	0.05 ± 0.01*	0.29 ± 0.02***
Hcychange: annualized change between visits 1 and 2	0.25 ± 0.04	0.30 ± 0.02	0.21 ± 0.03	0.25 ± 0.13
Socio-demographic:
Sex, %:
Male	42.4	24.1	43.0***	60.5***
Female	58.0	76.0	57.5	40.0
Age (years):
Continuous	47.92 ± 0.24	45.73 ± 0.42	48.30 ± 0.42***	49.73 ± 0.39***
Race, %:
White	43.2	44.3	44.0	42.0
African American	57.0	56.0	56.4	58.4
Poverty status, %:
< 125% federal poverty line	37.0	37.3	36.4	37.0
≥ 125% federal poverty line	63.1	63.0	64.0	63.0
Education:
Less than high school	6.2	6.1	6.0	7.0
High school	57.0	56.0	56.0	59.0
More than high school	37.2	38.3	38.5	35.0
Lifestyle:
Cigarette smoking:
Yes	43.5	40.0	44.0	20.1
No	57.0	60.2	56.5	80.0
Drug use:
Yes	18.2	15.1	19.5	47.2
No	82.0	85.0	81.0	53.0
HEI-2010 score:	43.11 ± 0.33	44.34 ± 0.63	43.34 ± 0.59	41.66 ± 0.50**
HEALTH:
Body mass index (kg/m2):	29.89 ± 0.19	30.08 ± 0.34	29.99 ± 0.34	29.61 ± 0.33
Self-rated health:
Poor/Average	21.0	19.2	19.0	25.0
Good	39.1	39.2	39.5	39.0
Very good/Excellent	40.0	42.0	42.0	36.3
Hypertension:
Yes	40.1	33.0	42.0**	46.0*
No	60.0	67.2	58.3	54.0
Diabetes:
None	68.1	72.2	69.0	63.0
Pre-diabetes	18.0	15.0	18.1	21.0**
Diabetes	14.0	13.0	13.2	16.0
High Cholesterol:
Yes	24.0	21.0	23.2	27.3
No	76.2	79.1	77.0	73.0
Cardiovascular disease:
Yes	14.5	16.0	14.0	14.2
No	86.0	84.4	86.4	86.0
Depressive symptoms: a
Visit 1
CES-D total score (n = 1445)	14.01 ± 0.29	14.34 ± 0.51	13.20 ± 0.49	14.50 ± 0.53*
CES-D domain 1 score (n = 1449)	4.34 ± 0.13	4.51 ± 0.22	4.03 ± 0.21	4.49 ± 0.23*
CES-D domain 2 score (n = 1449)	0.91 ± 0.03	0.95 ± 0.06	0.88 ± 0.06	0.89 ± 0.06*
CES-D domain 3 score (n = 1449)	6.42 ± 0.12	6.56 ± 0.20	6.09 ± 0.20	6.61 ± 0.21*
CES-D domain 4 score (n = 1449)	9.66 ± 0.07	9.69 ± 0.12	9.80 ± 0.12	9.50 ± 0.12*
CES-D total score ≥ 16, %	36.6	36.9	35.3	37.7
Visit 1 to Visit 3
Anxiety:
Anxiety total score, ordered (n = 1181)	2.78 ± 0.09	2.78 ± 0.15	2.94 ± 0.17*	2.62 ± 0.16*
Z-scored probability “High anxiety” trajectory (n = 1383)	−0.05 ± 0.03	−0.03 ± 0.05	−0.02 ± 0.05	−0.11 ± 0.05
“High anxiety” Trajectory probability (n = 1383)	0.44 ± 0.01	0.45 ± 0.02	0.46 ± 0.02	0.41 ± 0.02
Log-odds of “High anxiety” trajectory (n = 1224)	−1.10 ± 0.21	−0.62 ± 0.36	−0.98 ± 0.38	−1.73 ± 0.37*
Anxiety total score, below and above median (n = 1181), %
Below median	52.0	49.0	51.0	55.4
Above median	48.3	51.3	49.2	45.0
Anxiety Disorder (n = 1346), %
Yes	11.0	11.5	12.0	9.3
No	89.2	89.0	88.4	91.0
Depressive symptoms annualized rate of change, visits 1 through 3: a
CES-D total, empirical bayes estimator	−0.12 ± 0.00	−0.12 ± 0.01	−0.11 ± 0.00	−0.13 ± 0.01*
CES-D domain 1, empirical bayes estimator	−0.07 ± 0.00	−0.07 ± 0.00	−0.07 ± 0.00	−0.07 ± 0.00
CES-D domain 2, empirical bayes estimator	−0.01 ± 0.00	−0.01 ± 0.00	−0.01 ± 0.00	−0.01 ± 0.00
CES-D domain 3, empirical bayes estimator	−0.08 ± 0.00	−0.08 ± 0.00	−0.07 ± 0.00	−0.08 ± 0.00
CES-D domain 4, empirical bayes estimator	−0.05 ± 0.00	−0.05 ± 0.00	−0.06 ± 0.00*	−0.04 ± 0.00

Statistically significant association: *p < 0.05, **p < 0.01, ***p < 0.001, using t-tests from bivariate linear and multinomial logit regression models with Hcy tertile as the only predictor, comparing the two upper tertiles of Hcyv1 with lowest Hcyv1 tertile.

CES-D Center for Epidemiological Studies Depression, Hcy homocysteine, HEI healthy eating index, N sample size, SEM standard error of the mean.

^aDepressive symptoms scores include the CES-D total score, the CES-D domain 1 score [depressive affect], the CES-D domain 2 score [interpersonal problems], the CES-D domain 3 score [somatic complaints] and the CES-D domain 4 score [positive affect].

The associations of LnHcy_v1 with depressive symptoms over time (total and domain-specific scores), both baseline and annualized change are presented in Table 2, overall, and across sex and race. Overall, and after adjusting for multiple testing, a positive cross-sectional relationship was observed baseline LnHcy and CES-D total scores in Model 1 (β (standard error [SE]) = + 2.337 (0.902), P = 0.010), a relationship that was slightly attenuated in Model 2 but remained statistically significant [Model 2 (β (SE) = 1.825 (0.883), P = 0.039). The finding from Model 1 is illustrated in Fig. 4 as predictive margins of CES-D total scores across time based on this model. Among women, LnHcy_v1 was only associated with reduced scores on the positive affect sub-domain (Domain #4) in both the reduced (Model 1: β (SE) = −0.86 (0.30), P = 0.004) and the fully adjusted models (Model 2: β (SE) = −0.68 (0.30), P = 0.025). Importantly, when we stratified the analyses by levels of anxiety (below vs. above median), heterogeneity across these groups were detected. Most notably, among individuals below median anxiety, LnHcy was associated with a faster increase in CES-D domain 2 score over time (“interpersonal problems”: β (SE) = 0.043 (0.018), P = 0.018) for Model 1; and β (SE) = 0.041 (0.018), P = 0.024) for Model 2). In contrast, among individuals above median anxiety, LnHcy showed a strong positive cross-sectional association with CES-D total score (β (SE) = 4.02 (1.46), P = 0.006) for Model 1; and β (SE) = 3.59 (1.45), P = 0.013) for Model 2).

Table 2.

Parameter estimates from mixed effects linear regression models that regressed depressive symptoms total and domain-specific scores (baseline and between-visit change) on Log_e-transformed homocysteine (HCY) values.

	X = Ln(Homocysteine) at v1 (LnHcy)
	Model 1a		Model 2b
Overallc (N = 1460, K = 2.8):	β (SE)	P value	β (SE)	P value
CES-D total score:
LnHcy	+2.337 (0.902)	0.010	+1.825 (0.883)	0.039
LnHcy × Time	−0.115 (0.108)	0.286	−0.114 (0.109)	0.294
CES-D domain 1 score:
LnHcy	+0.951 (0.394)	0.016	+0.773 (0.389)	0.047
LnHcy × Time	−0.067 (0.047)	0.156	−0.066 (0.048)	0.166
CES-D domain 2 score:
LnHcy	+0.085 (0.106)	0.422	+0.051 (0.105)	0.626
LnHcy × Time	−0.003 (0.015)	0.839	−0.002 (0.015)	0.895
CES-D domain 3 score:
LnHcy	+0.784 (0.356)	0.028	+0.570 (0.348)	0.102
LnHcy × Time	−0.010 (0.047)	0.831	−0.014 (0.048)	0.770
CES-D domain 4 score:
LnHcy	−0.481 (0.210)	0.022	−0.381 (0.208)	0.067
LnHcy × Time	+0.036 (0.032)	0.265	+0.033 (0.032)	0.302
MEN (n = 619, K = 2.7):
CES-D total score:
LnHcy	+1.587 (1.195)	0.184	+1.401 (1.170)	0.231
LnHcy × Time	−0.191 (0.148)	0.198	−0.182 (0.148)	0.220
CES-D domain 1 score:
LnHcy	+1.026 (0.506)	0.043	+0.943 (0.482)	0.050
LnHcy × Time	−0.142 (0.062)	0.021	−0.132 (0.062)	0.033
CES-D domain 2 score:
LnHcy	+0.033 (0.162)	0.839	+0.010 (0.160)	0.948
LnHcy × Time	−0.011 (0.022)	0.636	−0.010 (0.023)	0.672
CES-D domain 3 score:
LnHcy	+0.536 (0.485)	0.269	+0.472 (0.475)	0.321
LnHcy × Time	−0.038 (0.067)	0.568	−0.036 (0.067)	0.592
CES-D domain 4 score:
LnHcy	−0.040 (0.285)	0.890	−0.001 (0.283)	0.998
LnHcy × Time	+0.017 (0.046)	0.719	+0.016 (0.046)	0.732
WOMEN (n = 841, K = 2.8):
CES-D total score:
LnHcy	+2.932 (1.325)	0.027	+1.818 (1.300)	0.162
LnHcy × Time	+0.003 (0.156)	0.986	+0.003 (0.159)	0.985
CES-D domain 1 score:
LnHcy	+0.925 (0.586)	0.114	+0.480 (0.559)	0.390
LnHcy × Time	+0.001 (0.069)	0.984	+0.004 (0.071)	0.957
CES-D domain 2 score:
LnHcy	+0.163 (0.141)	0.246	+0.099 (0.142)	0.486
LnHcy × Time	+0.003 (0.020)	0.884	+0.006 (0.020)	0.775
CES-D domain 3 score:
LnHcy	+0.892 (0.513)	0.082	+0.446 (0.501)	0.373
LnHcy × Time	+0.037 (0.066)	0.574	+0.029 (0.067)	0.666
CES-D domain 4 score:
LnHcy	−0.863 (0.303)	0.004	−0.680 (0.303)	0.025
LnHcy × Time	+0.040 (0.045)	0.369	+0.036 (0.046)	0.430
WHITE (N = 630, K = 2.8):
CES-D total score:
LnHcy	+1.807 (1.479)	0.222	+1.362 (1.458)	0.350
LnHcy × Time	−0.205 (0.186)	0.270	−0.229 (0.189)	0.225
CES-D domain 1 score:
LnHcy	+0.535 (0.616)	0.384	+0.403 (0.613)	0.511
LnHcy × Time	−0.094 (0.081)	0.249	−0.103 (0.082)	0.209
CES-D domain 2 score:
LnHcy	+0.087 (0.165)	0.600	+0.055 (0.165)	0.741
LnHcy × Time	−0.005 (0.024)	0.849	−0.003 (0.024)	0.915
CES-D domain 3 score:
LnHcy	+0.599 (0.567)	0.291	+0.407 (0.554)	0.462
LnHcy × Time	−0.055 (0.077)	0.469	−0.070 (0.078)	0.369
CES-D domain 4 score:
LnHcy	−0.487 (0.353)	0.167	−0.385 (0.351)	0.272
LnHcy × Time	+0.049 (0.053)	0.352	+0.052 (0.054)	0.336
African American (n = 830, K = 2.7):
CES-D total score:
LnHcy	+2.593 (1.150)	0.024	+2.064 (1.114)	0.064
LnHcy × Time	−0.030 (0.135)	0.827	−0.023 (0.136)	0.868
CES-D domain 1 score:
LnHcy	+1.224 (0.504)	0.015	+1.013 (0.473)	0.032
LnHcy × Time	−0.046 (0.058)	0.431	−0.039 (0.058)	0.503
CES-D domain 2 score:
LnHcy	+0.110 (0.139)	0.428	+0.070 (0.138)	0.610
LnHcy × Time	−0.004 (0.019)	0.839	−0.003 (0.019)	0.880
CES-D domain 3 score:
LnHcy	+0.821 (0.458)	0.073	+0.612 (0.444)	0.168
LnHcy × Time	+0.024 (0.060)	0.691	+0.023 (0.061)	0.706
CES-D domain 4 score:
LnHcy	−0.447 (0.259)	0.085	−0.353 (0.257)	0.169
LnHcy × Time	+0.014 (0.041)	0.724	+0.009 (0.041)	0.820
Below median anxiety score (n = 610, K = 2.8):
CES-D total score:
LnHcy	+1.288 (1.022)	0.208	+0.918 (1.008)	0.362
LnHcy × Time	+0.055 (0.141)	0.695	+0.038 (0.142)	0.789
CES-D domain 1 score:
LnHcy	+0.052 (0.398)	0.895	−0.024 (0.422)	0.954
LnHcy × Time	+0.029 (0.060)	0.631	+0.011 (0.059)	0.848
CES-D domain 2 score:
LnHcy	−0.039 (0.121)	0.748	−0.042 (0.122)	0.731
LnHcy × Time	+0.043 (0.018)	0.018	+0.041 (0.018)	0.024
CES-D domain 3 score:
LnHcy	+0.817 (0.444)	0.066	+0.599 (0.450)	0.183
LnHcy × Time	+0.021 (0.064)	0.741	+0.018 (0.064)	0.782
CES-D domain 4 score:
LnHcy	−0.390 (0.251)	0.120	−0.323 (0.251)	0.198
LnHcy × Time	+0.040 (0.044)	0.361	+0.040 (0.044)	0.357
Above median anxiety score (n = 571, K = 2.8):
CES-D total score:
LnHcy	+4.015 (1.461)	0.006	+3.587 (1.451)	0.013
LnHcy × Time	−0.178 (0.194)	0.361	−0.195 (0.197)	0.321
CES-D domain 1 score:
LnHcy	+1.834 (0.666)	0.006	+1.679 (0.647)	0.009
LnHcy × Time	−0.095 (0.085)	0.264	−0.099 (0.086)	0.248
CES-D domain 2 score:
LnHcy	+0.400 (0.183)d	0.029	+0.362 (0.183)d	0.047
LnHcy × Time	−0.044 (0.027)d	0.101	−0.043 (0.027)d	0.11
CES-D domain 3 score:
LnHcy	+1.120 (0.549)	0.041	+0.927 (0.543)	0.088
LnHcy × Time	−0.029 (0.083)	0.729	−0.039 (0.083)	0.638
CES-D domain 4 score:
LnHcy	−0.694 (0.352)	0.049	−0.628 (0.362)	0.082
LnHcy × Time	+0.012 (0.054)	0.828	+0.015 (0.056)	0.791

Results are presented for models fit to the overall sample of HANDLS 2004–2017 participants (N = 1460) and by stratifying variables.

CES-D Center for Epidemiological Studies Depression, Hcy Homocysteine, K Mean number of visits per subject, N Sample size, SE Standard error.

^aModel 1 is adjusted for age, sex, race, poverty status, inverse mills ratio as well as time on study between visits 1 and 2 (in years) and its interaction with homocysteine and covariates.

^bModel 2 is adjusted for age, sex, race, poverty status, education, smoking, drug use, 2010 healthy eating index, body mass index, inverse mills ratio as well as time on study between visits 1 and 2 (in years) and its interaction with homocysteine and covariates.

^cDepressive symptoms scores include the CES-D total score, the CES-D domain 1 score [depressive affect], the CES-D domain 2 score [interpersonal problems], the CES-D domain 3 score [somatic complaints] and the CES-D domain 4 score [positive affect].

^dP < 0.05 for null hypothesis that LnHcy × Z = 0 or LnHcy × Time × Z = 0 in unstratified model with Z being sex, race or anxiety status.

Fig. 4. Predictive margins of CES-D total score from Model 1 of LnHcy vs. CES-D total score in the overall sample – HANDLS (2004–2017).

Notes: CES-D Center for Epidemiologic Studies-Depression, HANDLS healthy aging in neighborhoods of diversity across the lifespan study. Based on Model 1 mixed-effects linear regression model with outcome being CES-D total score and main exposure LnHcy at baseline visit centered at 2.15. Model adjusted for baseline age, sex, race, poverty status and inverse mills ratio. Three trajectories are shown for LnHcy levels of 1.15, 2.15 and 3.15 to illustrate difference in baseline CES-D total score per unit increase in the exposure (β ± SE: +2.337 ± 0.902, P = 0.010).

The probability of belonging to the “High” LnHcy trajectory was z-scored (Hcyt_raj) and entered into another set of mixed-effects linear regression models with outcome being CES-D total and sub-domain scores (baseline and between-visit change), overall, and by stratifying variables (Table 3). Overall, a “High” LnHcy trajectory probability (z-scored), or Hcy_traj, was directly associated with baseline CES-D total score for Model 1 (β (SE) = 0.74 (0.29), P = 0.009), an association that was slightly attenuated in Model 2 (β (SE) = 0.64 (0.28), P = 0.021). This association was mainly driven by the relationship between Hcy_traj and the interpersonal problems (Domain 2 score) and somatic complaints (Domain 3 score) sub-domains. Hcy_traj was also linked to a higher score on the depressed affect sub-domain (Domain 1 score) among men at baseline, an association that remained statistically significant in Model 2. Among individuals with an above median anxiety score, Hcy_traj was inversely related to the annualized rate of increase in the “interpersonal problems” sub-domain (Domain 2 score) in the reduced and fully adjusted models.

Table 3.

Parameter estimates from mixed effects linear regression models that regressed depressive symptoms total and domain-specific scores (baseline and between-visit change) on homocysteine trajectory (Hcy_traj).

	Hcytraj: probability of “High” LnHcy, z-scored
	Model 1a		Model 2b
	β (SE)	P	β (SE)	P
Overallc (N = 1457, K = 2.8):
CES-D total score:
Hcy traj	+0.742 (0.285)	0.009	+0.640 (0.278)	0.021
Hcy traj × Time	−0.021 (0.034)	0.535	−0.022 (0.034)	0.518
CES-D domain 1 score:
Hcy traj	+0.258 (0.125)	0.039	+0.224 (0.123)	0.068
Hcy traj × Time	−0.010 (0.015)	0.492	−0.010 (0.015)	0.504
CES-D domain 2 score:
Hcy traj	+0.075 (0.033)	0.025	+0.069 (0.033)	0.038
Hcy traj × Time	−0.006 (0.005)	0.161	−0.006 (0.005)	0.164
CES-D domain 3 score:
Hcy traj	+0.311 (0.112)	0.006	+0.265 (0.110)	0.016
Hcy traj × Time	−0.001 (0.015)	0.948	−0.001 (0.015)	0.924
CES-D domain 4 score:
Hcy traj	−0.077 (0.066)	0.246	−0.059 (0.066)	0.371
Hcy traj × Time	+0.001 (0.010)	0.900	+0.002 (0.010)	0.848
Men (n = 618, K = 2.7):
CES-D total score:
Hcy traj	+0.773 (0.340)	0.023	+0.698 (0.335)	0.037
Hcy traj × Time	−0.003 (0.041)	0.939	−0.001 (0.042)	0.980
CES-D domain 1 score:
Hcy traj	+0.397 (0.144)	0.006	+0.370 (0.144)	0.010
Hcy traj × Time	−0.016 (0.017)	0.360	−0.014 (0.017)	0.405
CES-D domain 2 score:
Hcy traj	+0.086 (0.046)	0.062	+0.077 (0.046)	0.096
Hcy traj × Time	−0.007 (0.006)	0.281	−0.006 (0.006)	0.302
CES-D domain 3 score:
Hcy traj	+0.272 (0.138)	0.049	+0.248 (0.136)	0.070
Hcy traj × Time	+0.004 (0.019)	0.825	+0.005 (0.019)	0.793
CES-D domain 4 score:
Hcy traj	−0.025 (0.082)	0.755	−0.007 (0.081)	0.935
Hcy traj × Time	−0.013 (0.013)	0.295	−0.014 (0.013)	0.289
Women (n = 839, K = 2.8):
CES-D total score:
Hcy traj	+0.673 (0.469)	0.151	+0.409 (0.456)	0.370
Hcy traj × Time	−0.021 (0.055)	0.698	−0.021 (0.056)	0.703
CES-D domain 1 score:
Hcy traj	+0.085 (0.207)	0.682	−0.020 (0.203)	0.921
Hcy traj × Time	+0.001 (0.025)	0.971	+0.001 (0.025)	0.974
CES-D domain 2 score:
Hcy traj	+0.074 (0.050)	0.139	+0.060 (0.050)	0.227
Hcy traj × Time	−0.006 (0.007)	0.366	−0.006 (0.007)	0.392
CES-D domain 3 score:
Hcy traj	+0.319 (0.181)	0.078	+0.213 (0.176)	0.225
Hcy traj × Time	+0.002 (0.023)	0.944	+0.001 (0.024)	0.953
CES-D domain 4 score:
Hcy traj	−0.147 (0.108)	0.171	−0.102 (0.107)	0.339
Hcy traj × Time	+0.016 (0.016)	0.323	+0.015 (0.016)	0.349
White (n = 627, K = 2.8):
CES-D total score:
Hcy traj	+0.736 (0.493)	0.136	+0.660 (0.489)	0.177
Hcy traj × Time	+0.051 (0.061)	0.402	+0.042 (0.062)	0.495
CES-D domain 1 score:
Hcy traj	+0.261 (0.206)	0.207	+0.231 (0.213)	0.278
Hcy traj × Time	+0.009 (0.027)	0.745	+0.008 (0.027)	0.775
CES-D domain 2 score:
Hcy traj	+0.091 (0.055)	0.101	+0.085 (0.055)	0.125
Hcy traj × Time	+0.006 (0.008)	0.430	+0.006 (0.008)	0.421
CES-D domain 3 score:
Hcy traj	+0.296 (0.190)	0.119	+0.263 (0.185)	0.155
Hcy traj × Time	+0.020 (0.025)	0.419	+0.016 (0.026)	0.539
CES-D domain 4 score:
Hcy traj	−0.022 (0.119)	0.856	−0.006 (0.118)	0.959
Hcy traj × Time	−0.021 (0.017)	0.227	−0.019 (0.018)	0.272
African American (n = 830, K = 2.7):
CES-D total score:
Hcy traj	+0.676 (0.352)	0.054	+0.557 (0.341)	0.102
Hcy traj × Time	−0.038 (0.041)	0.357	−0.036 (0.041)	0.378
CES-D domain 1 score:
Hcy traj	+0.241 (0.154)	0.119	+0.196 (0.145)	0.176
Hcy traj × Time	−0.015 (0.018)	0.382	−0.014 (0.018)	0.431
CES-D domain 2 score:
Hcy traj	+0.069 (0.042)	0.102	+0.060 (0.042)	0.155
Hcy traj × Time	−0.013 (0.006)	0.029	−0.012 (0.006)	0.033
CES-D domain 3 score:
Hcy traj	+0.278 (0.140)	0.047	+0.229 (0.136)	0.093
Hcy traj × Time	−0.005 (0.018)	0.796	−0.004 (0.018)	0.811
CES-D domain 4 score:
Hcy traj	−0.095 (0.079)	0.233	+0.071 (0.079)	0.368
Hcy traj × Time	+0.007 (0.012)	0.553	+0.007 (0.012)	0.588
Below median anxiety score (n = 608, K = 2.8):
CES-D total score:
Hcy traj	+0.412 (0.309)	0.182	+0.343 (0.322)	0.286
Hcy traj × Time	+0.005 (0.045)	0.912	−0.007 (0.045)	0.880
CES-D domain 1 score:
Hcy traj	+0.047 (0.128)	0.714	+0.032 (0.126)	0.802
Hcy traj × Time	+0.008 (0.019)	0.687	+0.005 (0.019)	0.783
CES-D domain 2 score:
Hcy traj	+0.039 (0.039)	0.316	+0.039 (0.039)	0.309
Hcy traj × Time	+0.009 (0.006)	0.126	+0.008 (0.006)	0.152
CES-D domain 3 score:
Hcy traj	+0.277 (0.142)	0.052	+0.226 (0.140)	0.106
Hcy traj × Time	−0.003 (0.020)	0.872	-0.005 (0.020)	0.818
CES-D domain 4 score:
Hcy traj	−0.013 (0.081)	0.872	−0.002 (0.084)	0.979
Hcy traj × Time	+0.011 (0.014)	0.431	+0.014 (0.015)	0.338
Above median anxiety score (n = 570, K = 2.8):
CES-D total score:
Hcy traj	+0.293 (0.451)	0.516	+0.285 (0.446)	0.522
Hcy traj × Time	−0.017 (0.059)	0.774	−0.021 (0.059)	0.718
CES-D domain 1 score:
Hcy traj	+0.074 (0.206)	0.720	+0.073 (0.198)	0.712
Hcy traj × Time	−0.013 (0.026)	0.619	−0.015 (0.026)	0.567
CES-D domain 2 score:
Hcy traj	+0.096 (0.056)	0.088	+0.093 (0.056)	0.096
Hcy traj × Time	−0.023 (0.008)d	0.005	−0.022 (0.008)d	0.006
CES-D domain 3 score:
Hcy traj	+0.157 (0.169)	0.353	+0.154 (0.166)	0.353
Hcy traj × Time	+0.000 (0.025)	0.994	−0.002 (0.025)	0.922
CES-D domain 4 score:
Hcy traj	+0.040 (0.109)	0.716	+0.044 (0.111)	0.694
Hcy traj × Time	−0.020 (0.016)	0.209	−0.020 (0.017)	0.231

Results are presented for models fit to the overall sample of HANDLS 2004–2017 participants (N = 1460) and by stratifying variables.

CES-D Center for Epidemiological Studies Depression, Hcy _traj z-transformed probability of belonging to a group with higher homocysteine over time according to group-based trajectory modeling, K mean number of visits per subject, N sample size, SE standard error.

^aModel 1 is adjusted for age, sex, race, poverty status, inverse mills ratio as well as time on study in years between visits 1 and 3 and its interaction with homocysteine trajectory and covariates.

^bModel 2 is adjusted for age, sex, race, poverty status, education, literacy, smoking, drug use, 2010 healthy eating index, body mass index, inverse mills ratio as well as time on study in years between visits 1 and 3 and its interaction with homocysteine trajectory and covariates.

^cDepressive symptoms scores include the CES-D total score, the CES-D domain 1 score [depressive affect], the CES-D domain 2 score [interpersonal problems], the CES-D domain 3 score [somatic complaints] and the CES-D domain 4 score [positive affect].

^dP < 0.05 for null hypothesis that LnHcy × Z = 0 or LnHcy × Time × Z = 0 in unstratified model with Z being sex, race or anxiety status.

Four sensitivity analyses were also conducted. In the first sensitivity analysis, LnHcy at visit 1 was studied in relation to the Log_e odds of high anxiety (Table S1). No association was detected in both the reduced and fully adjusted models. In a second sensitivity analysis (Table S2), annualized rate of change in Hcy between visits 1 and 2 was studied in relation to CES-D total and sub-scores in the overall sample. After adjustment for multiple testing, annualized change in LnHcy between the first two visits was not associated with neither baseline nor annualized change in CES-D scores, in both reduced and fully adjusted models. Third, a mixed-effects linear regression model with LnHcy as the outcome was conducted in the overall sample examining its association with various exposures, including baseline CES-D total score, ordinal anxiety score, probability of higher anxiety score and anxiety disorder at baseline (Table S3). Most notably, baseline CES-D total score was associated with LnHcy_v1 in both the reduced and fully adjusted models. In contrast, a higher probability of elevated anxiety over time was associated with faster increase in LnHcy over time, a finding only detected in the reduced model. In the fourth sensitivity analysis (Table S4), the two main Hcy exposures of interest were tested as predictors for incidence of EDS over time, using a series of fully adjusted Cox proportional hazards models, while stratifying and testing for potential effect modification by sex, race, poverty status and anxiety score levels. Baseline Hcy was not associated with incident EDS overall. In fact, there was strong heterogeneity in this association across race, poverty status and anxiety levels. Most notably, baseline Hcy was inversely associated with incident EDS among White participants and participants living above poverty while being positively associated with this outcome among African American adults, adults with low level of anxiety at baseline, and adults living below poverty (P < 0.05 for 2-way interaction terms between baseline Hcy exposure and those 3 effect modifiers). In contrast, Hcy_traj was positively associated with incident EDS overall (HR = 1.09, 95% CI: 1.03–1.14, P = 0.001), more strongly so among women and individuals living below poverty (P < 0.05 for 2-way interaction between Hcy_traj and those 2 effect modifiers). For Hcy_traj, the strongest association with incident EDS was noted in the “Below poverty” stratum: HR = 1.22, 95% CI: 1.13–1.18, P < 0.001.

Discussion

In this prospective cohort study of 1460 urban adults, 30 to 64 years of age at baseline, mean LnHcy_v1 and CES-D total scores were estimated at 2.15 and 14.01, respectively. Overall, a positive cross-sectional relationship was observed LnHcy_v1 and CES-D total scores in Model 1 (β ([SE]) = + 2.337 (0.902), P = 0.010), a relationship that was slightly attenuated in Model 2 [(β (SE) = + 1.825 (0.883), P = 0.039). However, heterogeneity was detected across anxiety levels, with individuals below median anxiety experiencing faster increases in CES-D domain 2 scores with increased LnHcy_v1 (interpersonal problems: β [SE] = 0.041 [0.018], P = 0.024). A “High” LnHcy trajectory probability (z-scored), or Hcy_traj, was directly associated with baseline CES-D total score for Model 1 (β (SE) = + 0.74 (0.29), P = 0.009), an association that was slightly attenuated in Model 2 (β (SE) = + 0.64 (0.28), P = 0.021). This association was mainly driven by the relationship between Hcy_traj and the interpersonal problems and somatic complaints sub-domains. Among individuals with an above median anxiety score, Hcy_traj was inversely related to the annualized rate of increase in the “interpersonal problems” sub-domain (Domain 2 score) in the reduced and fully adjusted models. Hcy_traj was positively associated with incident elevated depressive symptoms (CES-D total score ≥ 16) overall (fully adjusted model: HR = 1.09, 95% CI: 1.03–1.14, P = 0.001), though more strongly so among women and individuals living below poverty.

Though research in this area of work is limited, prior studies suggested that higher levels of HCY were related to depressive episodes, symptoms, and risk for depression and anxiety disorders, though it remains unclear if anxiety partially explains these relationships [13, 17–22]. Differences in sample compositions across age, sex, nationality, and prior medical history of psychiatric disorders may be reasons for these discrepancies. Furthermore, studies have varied on their assessments of depression, depressive symptoms, and anxiety, as well as that of other conditions such as PTSD. The growing body of work in this area warrants further inquiry into how these relationships develop over time and to what extent the heterogeneity across anxiety levels observed in the present study highlight more nuanced influences per depression- or anxiety-related disorders.

The cross-sectional and longitudinal evidence of the positive association of HCY and symptoms of depression are consistent with the findings from other studies [15, 16]. Most notably, a systematic review and meta-analysis comparing homocysteine levels in healthy subjects and those with depression found that individuals with higher homocysteine levels had a higher risk of depression [15]. This was found in studies using various depression diagnostic tools, including DSM-IV, GDS, ZDRS, and BDI-II [15]. The review and meta-analysis also found that participants with hyperhomocysteinaemia had a higher chance of depression [15]. The findings suggest that future research should focus on the tools used for depression assessment to better understand the relationship between homocysteine levels and depression [15]. A cross-sectional study along with a systematic review, and meta-analysis were conducted on 3752 men aged 70 years or older [16]. The results showed that the odds ratio of prevalent depression increased 4% with every unit increase of total homocysteine (tHcy) [16]. The methylenetetrahydrofolate reductase (MTHFR) C677T TT genotype was 0.19 mg/L higher among participants with the MTHFR C677T TT genotype compared to the CC genotype [16]. The meta-analysis showed that older adults with high tHcy had an increased risk of depression, and TT carriers were 22% more likely than CC carriers to have current depression or a history of depression [16]. The triangular association between the MTHFR genotype, tHcy, and depression implies that higher concentrations of tHcy increase the risk of depression [16]. Confirmatory data from sufficiently powered randomized trials of homocysteine-lowering therapy are now required to test the causal relationship between tHcy and depression [16].

Nevertheless, a study conducted in the Longitudinal Aging Study Amsterdam (LASA) involving 1352 men and women aged ⩾65 years, assessed depressive symptoms six times from 1995/1996 to 2011/2012 [14]. Multiple linear regression and mixed models were used to assess the associations of vitamin B-12 as well as Hcy with depressive symptoms over 16 years [14]. The results showed that vitamin B-12 was not cross-sectionally or prospectively associated with depressive symptoms, and no association was found with incident depression [14]. For Hcy, no associations were found, except for a lower risk of depression in younger participants [14]. The study concluded that further research is needed to understand the influence of Hcy metabolism on mental health [14]. Our present study detected a similar inverse association between baseline Hcy and incident EDS among White adults and individuals living over poverty with no detectable association overall. In contrast, there was a positive association detected among African American adults and individuals living below poverty. Therefore, the inverse association detected in the previous study may be specific to middle-aged adults of higher SES, independently of other socio-demographic, lifestyle and health-related factors.

Although the relationship of HCY to depression was first noted in 1970 [9], the exact mechanisms explaining the association remain unclear. High levels of homocysteine can result from a variety of reasons including a dietary deficiency of vitamins B_6, B_12, and folate, genetic variation of the enzymes essential for the metabolism of homocysteine such as MTHFR and cystathionine beta-synthase, enzymes, gastric atrophy, inflammatory bowel disease and methionine loading [9, 16]. Elevated HCY may result in deficiency of methyl transfer reactions, resulting in deficits to neurotransmitter synthesis, and is associated with reduced cortical and hippocampal volumes in the brain. These changes may explain how HCY levels mediate symptoms of depression [44]. The effects of HCY on depression have been reported to be independent of dementia [45]. However, the interaction of HCY and depression can influence cognitive performance [46]. Evaluated HCY and depression appear to be associated with white matter damage, gray matter atrophy, β-amyloid deposition, and cardiovascular disease risk [46, 47]. Yet not all studies have found the association between HCY and depression which could reflect the many causes for elevated HCY, including genetic predisposition and brain disease, and the several types of depression in samples of different ethnicities and ages [9, 48].

A growing interest in Plasma HCY and its sequelae has spurred a set of evidence linking HCY to adverse outcomes for brain health. Much of this interest has grown out of observational studies linking high plasma HCY and allelic variants of the MTHFR gene to psychiatric abnormalities [9, 49]. HCY production is controlled through one-carbon metabolism [50, 51]. Methionine synthase, a B12-dependent enzyme, is a key enzyme that governs the utilization of HCY [50, 51]. Vitamin B12 deficiency or a mutation in the enzyme slow the conversion of HCY to methionine, leading to the accumulation of HCY [50, 51]. An inborn error of metabolism in the MTHFR gene, which is responsible for the production of the other substrate of the methionine synthase reaction, 5-methyl THF, can also cause HCY accumulation [52]. Animal studies have shown that loss-of-function mutations in MTHFR decrease levels of monoamine neurotransmitters in key regions of the brain, such as glutamate in the amygdala and γ-aminobutyric acid in the thalamus [53]. Hyperhomocysteinemia has also been linked to the production of homocysteic acid, a compound that is known to be neurotoxic to dopaminergic regions of the brain and that functions as an N-methyl-D-aspartate (NMDA) receptor agonist [54]. Indeed, it appears that the downstream neurotoxic effects of HCY on dopaminergic circuits and the changes in monoamine neurotransmitter concentrations may help explain the observed associations from our study, which we find are also consistent across several other epidemiological studies [9, 55].

Our study has several strengths including its being the first study conducted among ethnically and socio-economically diverse urban middle-aged adults to examine these intricate research questions. Second, the study was well-powered to examine associations across sex and race, as well as across anxiety levels. Furthermore, analyses included advanced techniques such as multiple imputation coupled with mixed-effects linear regression models and 2-stage Heckman selection. There was also a wealth of data available to control for potential confounders in all our models. A few limitations are worth considering when interpreting study results. First, subsamples of the initial HANDLS participants were used to assess potential associations between HCY and depressive symptoms, which may have resulted in selection bias. Second, as a large number of exposure, outcome, and covariate assessment factors were self-reported, measurement errors are probably present and could result in skewed measures of association. Third, our ability to compare our findings with published literature may have been hampered by various criteria used to categorize HCY as well as different means of measuring it in serum. Fourth, there’s a chance that the amount of time spent in follow-up between HANDLS Visits 1, 2, and 3 was insufficient to see any clinically significant improvements in depression symptoms. Similarly, fluctuation of depressive symptoms over time cannot be fully captured by the total CES-D score. Future research should therefore investigate proposed correlations over extended follow-up periods. Fifth, even after adjusting for a number of factors, residual confounding is likely because HANDLS is an observational study, which means we were unable to demonstrate causal associations. Sixth, while analyzing HCY in connection to depressive symptoms, the impact of interaction effects by sex and race might have been understated. Lastly, the HANDLS study’s sampling design limits the generalizability of the study’s findings to middle-aged and older persons living in metropolitan areas across the United States.

Conclusion

Among urban middle-aged adults, baseline and “high trajectory” of LnHcy were positively associated with depressive symptoms and elevated depressive symptom incidence, in a sex-, race-, poverty status- and anxiety-level specific manner. More research is needed using longitudinal designs with larger sample sizes and longer follow-up times, in order to replicate and confirm our present study findings.

Author contributions

Michael F. Georgescu: study concept, plan of analysis, assistance with data management and statistical analysis, literature search and review, write-up of parts of the manuscript, revision of the manuscript. May A. Beydoun: study concept, plan of analysis, data management and statistical analysis, literature search and review, write-up of parts of the manuscript, revision of the manuscript. Christian A. Maino Vieytes: study concept, plan of analysis, literature search and review, write-up of parts of the manuscript, revision of the manuscript. Marie T. Fanelli-Kuczmarski: study concept, plan of analysis, literature search and review, write-up of parts of the manuscript, revision of the manuscript. Jason Ashe: plan of analysis, literature search and review, write-up of parts of the manuscript, revision of the manuscript. Hind A. Beydoun: plan of analysis, literature search and review, write-up of parts of the manuscript, revision of the manuscript. Sharmin Hossain: plan of analysis, literature search and review, write-up of parts of the manuscript, revision of the manuscript. Nicole Noren Hooten: study concept, literature search and review, write-up of parts of the manuscript, review of the manuscript. Michele K. Evans: data acquisition, plan of analysis, write-up of parts of the manuscript, revision of the manuscript. Alan B. Zonderman: data acquisition, plan of analysis, write-up of parts of the manuscript, revision of the manuscript.

Funding

This research was supported entirely by the Intramural Research Program of the NIH, National Institute on Aging (Z01-AG000513). Open access funding provided by the National Institutes of Health.

Data availability

The study protocol (09-AG-N248) received approval from the National Institute on Environmental Health Sciences’ Institutional Review Board (IRB) of the National Institutes of Health (NIH). Upon request, data can be made available to researchers with approved proposals, after they have agreed to confidentiality as required by our IRB. Policies are publicized on: https://handls.nih.gov. Data access request can be sent to principal investigators (PI) or the study manager, Jennifer Norbeck at norbeckje@mail.nih.gov. These data are owned by the National Institute on Aging at the NIH. The PIs have made those data restricted to the public for two main reasons: “(1) The study collects medical, psychological, cognitive, and psychosocial information on racial and poverty differences that could be misconstrued or willfully manipulated to promote racial discrimination; and (2) Although the sample is fairly large, there are sufficient identifiers that the PIs cannot guarantee absolute confidentiality for every participant as we have stated in acquiring our confidentiality certificate.” [56]

Competing interests

The authors declare no competing interests.

Ethics approval and consent

The HANDLS project was authorized by the National Institutes of Health’s Institutional Review Board, and participants gave written informed consent [25]. All methods were performed in accordance with the relevant guidelines and regulations.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-024-03111-7.