Socioeconomic Status Is Associated With Reward Processing, Interleukin 1β, Striatal Connectivity, and Antidepressant Outcomes in Individuals With Major Depressive Disorder: A CAN-BIND-1 Report

Abstract

Background

Major depressive disorder (MDD) is a common condition with heterogeneous risk factors. Socioeconomic status (SES) is one such risk factor, which is negatively linked to MDD treatment outcomes and symptom severity. SES is associated with altered resting-state functional connectivity (RSFC) in reward-processing circuitry and elevated proinflammatory cytokine levels in individuals without depression. However, how the negative consequences of low SES exacerbate MDD psychopathology is poorly understood.

Methods

Data on SES (household income and education), depression severity, self-reported reward processing, serum proinflammatory cytokine levels, and neuroimaging for 323 adult participants (211 patients with MDD receiving open-label escitalopram, 112 control participants without depression; 63.4% female) were obtained from the CAN-BIND-1 (Canadian Biomarker Integration in Depression Study-1) dataset. General linear models assessed the effects of MDD diagnosis and SES on self-reported reward processing and proinflammatory cytokine levels. Whole-brain seed-to-voxel RSFC analyses were performed for the dorsal and ventral striatum leveraging 249 participants (150 patients with MDD, 99 control participants; 62.2% female). We also assessed the impact of SES on response to open-label escitalopram.

Results

Participants with MDD from households with lower incomes displayed lower goal pursuit behaviors, decreased interleukin 1β levels, and slower improvement to escitalopram relative to those from households with higher incomes. Using a lenient z > 2.3 threshold, MDD household income correlated with striatal RSFC with the dorsolateral prefrontal and posterior cingulate cortices.

Conclusions

Our results elucidate the role of SES and its negative consequences in altering reward processing and antidepressant treatment efficacy in MDD, corroborating previous literature suggesting that SES significantly impacts health outcomes. Better characterizing the relationship between SES and MDD psychopathology may inform future treatment approaches and intervention development.

Plain Language Summary

In this study, we explored how socioeconomic status (SES) influences depression symptoms and treatment. Using brain scans, blood tests, and clinical assessments from more than 300 adults, we found that people with depression from lower SES backgrounds showed reduced reward drive, changes in brain networks related to reward, and slower improvement with antidepressant treatment compared with people from higher SES backgrounds. These findings suggest that SES plays a key role in cognition, brain function, and recovery from depression. Understanding this link could help guide more effective and tailored treatments for people with depression.

Plain Language Summary

In this study, we explored how socioeconomic status (SES) influences depression symptoms and treatment. Using brain scans, blood tests, and clinical assessments from more than 300 adults, we found that people with depression from lower SES backgrounds showed reduced reward drive, changes in brain networks related to reward, and slower improvement with antidepressant treatment compared with people from higher SES backgrounds. These findings suggest that SES plays a key role in cognition, brain function, and recovery from depression. Understanding this link could help guide more effective and tailored treatments for people with depression.

Major depressive disorder (MDD) is a heterogeneous disorder affecting approximately 3.2% of adults worldwide (1,2) and carries significant economic and personal burden (3,4). First-line treatments for MDD include pharmaco- and psychotherapy, and only one-third of patients achieve symptom remission with their first antidepressant treatment (5). This may be because MDD has multifactorial causes and heterogeneous risk factors, and the interaction between genetic and environmental factors may influence individual MDD susceptibility and treatment prognosis (6). However, the association between environmental risk factors and MDD pathophysiology is poorly understood.

Socioeconomic status (SES) is one environmental factor thought to influence MDD severity and prognosis due to disparities in access to financial, educational, social, and health resources (4). SES is the combination of social and economic factors that determine an individual’s social position relative to others (7). SES is a complex construct that can be quantified using objective measures (e.g., household income) or subjective self-reported scales (8). Individuals from lower SES backgrounds are at increased risk for MDD and respond less favorably to treatment (9, 10, 11, 12). Individuals from low SES backgrounds also have higher levels of serum proinflammatory biomarkers, which have in turn been associated with increased depressive symptom severity and prevalence (13, 14, 15, 16, 17). Consequently, characterizing the influence of SES on MDD pathophysiology is likely essential to improving treatment outcomes for individuals with MDD.

SES may contribute to MDD pathophysiology via exacerbating chronic stress (CS) due to disparities in economic or social resources. Low SES is correlated with numerous contributors to stress burden, including worse physical health, inadequate access to health care, and insufficient nutrient intake (18, 19, 20). Individuals from low SES backgrounds have higher levels of stress biomarkers, including cortisol (21,22). Elevated CS has been linked to increased psychological distress and impaired reward processing, as evidenced by CS being a strong predictor for experiencing anhedonia (23, 24, 25). CS is also correlated with increased neuroinflammation, indexed by elevated serum proinflammatory biomarkers including interleukin 6 (IL-6), IL-1β, and tumor necrosis factor α (TNF-α) (26). These markers are strongly associated with risk for developing MDD and increased symptom severity (27, 28, 29). Furthermore, levels of these proinflammatory cytokines have been linked to alterations in striatal resting-state functional connectivity (RSFC) and reward processing in individuals without depression (control participants), as well as poorer antidepressant treatment response in MDD (30, 31, 32, 33). These findings imply that SES and its relationship to neuroinflammation may contribute to altered neural structure and function and specific symptoms observed in individuals with MDD.

Evidence from neuroimaging studies suggests that the consequences of low SES impact cognition and brain function. Individuals without depression from lower SES backgrounds tend to display altered executive and reward functioning compared with individuals from higher SES backgrounds (34,35). Compared with people from high SES backgrounds, people from low SES backgrounds display altered connectivity in regions associated with reward anticipation and goal-directed action, including hyperactivity in the caudate nucleus and hypoactivity in the anterior cingulate cortex, consistent with the finding that low SES is associated with greater responses to both rewards and losses (36,37). Overall, growing evidence indicates that SES is related to neural function and cognition.

Few studies exist on the relationships between SES, cognition, and brain function in MDD, and those that do exist focus primarily on SES during childhood. For example, one study found that school-age children from a lower SES background exhibited reduced RSFC from the hippocampus and amygdala and increased depression severity (38). Similarly, another study identified a relationship between childhood SES and increased hippocampal connectivity in adults with MDD (39). More research is required to identify the role of SES in mediating brain structure, function, and cognition in adults with MDD.

Despite SES being identified as a predictor of antidepressant response and studies showing that childhood SES influences brain structure and function later in life, these relationships have not yet been established for adults with MDD and current SES (12,38, 39, 40). Previous work also suggests that RSFC in reward-related regions in MDD, such as RSFC from the nucleus accumbens (NAc) and dorsal striatum (DS), is associated with SES in individuals without depression (36,37,41,42). Therefore, our primary aim was to characterize the impact of SES on reward processing, proinflammatory cytokines, striatal RSFC, and escitalopram response in adults with MDD. We hypothesized that individuals diagnosed with MDD from a lower SES background would be most negatively impacted in reward processing, proinflammatory cytokine serum concentration, and striatal RSFC compared with individuals without MDD from a similar socioeconomic background and individuals diagnosed with MDD from a higher socioeconomic background. Furthermore, we predicted that SES would be a negative predictor of antidepressant treatment response to an 8-week course of the first-line pharmacotherapy, escitalopram.

Methods and Materials

Recruitment

Participants were recruited between 2013 and 2017 at the 6 sites as part of the CAN-BIND-1 (Canadian Biomarker Integration in Depression Study-1) (NCT01655706), and the protocol is detailed elsewhere (43) and in the Supplement. Briefly, MDD participants were 18 to 60 years old, met DSM-IV-TR criteria for MDD (44), and were at least moderately depressed (Montgomery–Åsberg Depression Rating Scale [MADRS] score ≥ 24) (45). Control participants were 18 to 60 years old with no psychiatric or other medical diagnoses. All participants provided written informed consent to participate, and this study was approved by each site’s research ethics board.

Treatment

All participants completed a baseline visit involving neuroimaging, molecular, and clinical/demographic assessments, detailed below. Participants with MDD then received open-label escitalopram at a flexible dose (10–20 mg) for 8 weeks. Assessments were repeated at 2, 8, and 16 weeks. Control participants completed the same study visits but did not receive treatment.

Clinical Measures

The MADRS was the primary clinical outcome for the study. Reward responsivity was measured using the Behavioral Inhibition Scale/Behavioral Activation Scale (BIS/BAS), which consists of 4 subscales measuring approach and avoidance behavior (46). The 3 BAS subscales measure 1) reward responsiveness, or the pleasure experienced when receiving a reward; 2) drive, or the willingness to engage in behaviors to pursue a reward; and 3) fun seeking, or the tendency to seek out novel rewards. The BIS subscale measures a person’s motivation to prevent aversive outcomes.

Demographic and SES Measures

Measures of SES included self-reported years of education, household income, and employment status. Education was measured as a continuous variable. Participants who chose to disclose their household income (180/211 participants with MDD, 91/112 control participants) reported their responses on a Likert scale consisting of 8 income brackets (Table 1). For all analyses, responses to this scale were modeled as continuous. Participants also reported employment status coded as a categorical variable with 8 options and were asked to self-describe their ethnicity using 14 alphabetical options aligned with the Canadian Census. Participants could endorse multiple ethnicities. Participants disclosed their sex assigned at birth, coded as a binary variable, and their current height and weight were measured to calculate body mass index (BMI). Missing demographic data were not imputed.

Table 1.

Descriptive Statistics for the CAN-BIND-1 Sample

	MDD Group		Control Group		Test Statistic	p Value
	n	Mean (SD) or n (%)	n	Mean (SD) or n (%)
Age	211	35.303 (12.649)	112	33.036 (10.737)	U = 10,901.500	.252
Education	209	16.890 (2.138)	112	18.357 (2.262)	U = 7365.500	<.001
Body Mass Index	208	26.546 (6.395)	109	24.348 (4.729)	U = 9046.500	.003
Depression Severity	211	29.877 (5.606)	112	0.830 (1.681)	U = 0.000	<.001
BIS	210	23.219 (3.467)	111	17.932 (3.435)	U = 3174.500	<.001
BAS Reward Responsiveness	210	14.310 (2.642)	111	16.09 (1.847)	U = 7146.000	<.001
BAS Drive	210	8.965 (2.437)	111	10.865 (1.979)	U = 6190.00	<.001
BAS Fun Seeking	210	9.810 (2.598)	111	11.162 (2.061)	U = 7926.000	<.001
Sex, Female	–	133 (63.0%)	–	71 (63.4%)	χ21 = 0.004	.949
Ethnicity
Aboriginal	–	2 (0.9%)	–	1 (0.9%)	χ21 = 0.002	.961
Arab	–	2 (0.9%)	–	1 (0.9%)	χ21 = 0.002	.961
Black	–	11 (5.2%)	–	2 (1.8%)	χ21 = 2.225	.136
Chinese	–	14 (6.6%)	–	10 (8.9%)	χ21 = 0.560	.455
East Asian	–	2 (0.9%)	–	1 (0.9%)	χ21 = 0.002	.961
Filipino	–	4 (1.9%)	–	2 (1.8%)	χ21 = 0.0005	.944
Japanese	–	1 (0.5%)	–	0 (0.0%)	χ21 = 0.532	.466
Jewish	–	2 (0.9%)	–	0 (0.0%)	χ21 = 1.068	.301
Korean	–	2 (0.9%)	–	0 (0.0%)	χ21 = 1.068	.301
Latin American/Hispanic	–	14 (6.6%)	–	3 (2.7%)	χ21 = 2.297	.130
Self-described	–	12 (5.7%)	–	5 (4.5%)	–	–
South Asian	–	7 (3.3%)	–	11 (9.8%)	χ21 = 5.881	.015
Southeast Asian	–	3 (1.4%)	–	2 (1.8%)	χ21 = 0.064	.801
West Asian	–	1 (0.5%)	–	0 (0.0%)	χ21 = 0.532	.466
White	–	154 (73.0%)	–	76 (67.9%)	χ21 = 0.939	.333
Employment Status
Working now	–	104 (49.3%)	–	73 (65.2%)	χ21 = 7.458	.006
Disabled, permanent/temporarily	–	17 (8.1%)	–	0 (0.0%)	χ21 = 9.525	.002
Temporarily laid off, temporary leave	–	13 (6.2%)	–	0 (0.0%)	χ21 = 7.190	.007
Keeping house	–	3 (1.4%)	–	1 (0.9%)	χ21 = 0.167	.682
Looking for work	–	28 (13.3%)	–	4 (3.6%)	χ21 = 7.711	.005
Student	–	27 (12.8%)	–	30 (26.8%)	χ21 = 9.853	.002
Retired	–	2 (0.9%)	–	0 (0.0%)	χ21 = 1.068	.301
Other	–	16 (7.6%)	–	2 (1.8%)	–	–
Household Income
<$10,000	–	21 (11.7%)	–	4 (4.4%)	χ21 = 4.172	.041
$10,000–$24,999	–	25 (13.9%)	–	12 (13.2%)	χ21 = 0.093	.761
$25,000–$49,999	–	43 (23.9%)	–	16 (17.6%)	χ21 = 1.820	.177
$50,000–$74,999	–	37 (20.6%)	–	25 (27.5%)	χ21 = 1.080	.299
$75,000–$99,999	–	21 (11.7%)	–	14 (15.4%)	χ21 = 0.491	.483
$100,000–$149,999	–	21 (11.7%)	–	13 (14.3%)	χ21 = 0.213	.645
$150,000–$199,999	–	7 (3.9%)	–	3 (3.3%)	χ21 = 0.100	.752
>$200,000	–	5 (2.8%)	–	4 (4.4%)	χ21 = 0.390	.532

Group differences were not assessed for self-described ethnicity or employment status.

BAS, Behavioral Activation Scale; BIS, Behavioral Inhibition Scale; CAN-BIND-1, Canadian Biomarker Integration in Depression Study-1; MDD, major depressive disorder.

Biological Sample Collection and Processing

Sample collecting and measurement of proinflammatory cytokines has been previously published and are detailed in the Supplement (47). Briefly, venous whole-blood samples were acquired at baseline and at treatment weeks 2, 8, and 16 from all participants (43). For our secondary analysis, we obtained baseline measurements of proinflammatory cytokines linked to SES and MDD: IL-6, IL-1β, and TNF-α (26).

Neuroimaging Acquisition

All participants underwent structural and functional neuroimaging; the protocol, acquisition parameters, and quality control (QC) pipelines have been published previously and are provided in the Supplement (48). For this analysis, we used the T1-weighted scan (1 mm³) and 10-minute resting-state scan (eyes-open; TR = 2000 ms, 4 mm³, flip angle = 75°, matrix dimension = 64 × 64 pixels). QC was considered during study design, and automated and manual QC procedures were applied during and following data collection. For example, a “human phantom” traveled to each site at which neuroimaging data were being collected to quantify site differences (48).

Neuroimaging Preprocessing and First-Level Analysis

The preprocessing and QC protocols have been detailed elsewhere (48,49). Whenever possible, participants were requested to return to the site for rescanning in the case of any images deemed unusable for further processing due to insufficient full-brain coverage or too many imaging artifacts. The functional magnetic resonance imaging (fMRI) data were preprocessed using the Optimization of Preprocessing Pipelines for NeuroImaging pipeline, as described in the Supplement (50,51). Briefly, this pipeline included rigid-body motion correction, censoring, slice time correction, spatial smoothing (full width at half maximum = 6 mm), skull stripping, tissue segmentation, physiological and motion denoising, low pass filtering (0.10 Hz), and spatial normalization to the structural Montreal Neurological Institute (MNI) 152 template (4 mm³). We had clean fMRI data for 150 participants with MDD and 99 control participants.

The NAc and DS were selected as regions of interest (ROIs) for first-level whole-brain seed-to-voxel–based correlation analysis due to their involvement in reward-related cognition (52,53). These seeds were extracted using an existing subcortical parcellation (54). For the purposes of this study, we selected the NAc from scale I and the anterior caudate nucleus from scale II of this atlas. To extract whole-brain seed-to-voxel–based RSFC at the first level, we performed 2 fixed-effects general linear models (GLMs) in FSL FEAT version 6.00, one for each ROI demeaned time series as the explanatory variable (55).

Statistical Analysis

First, we performed GLMs to test the effect of SES on reward processing in patients with MDD compared with control participants. We performed 4 GLMs to investigate this effect with different subscales of the BIS/BAS. Age, sex, recruitment site, diagnosis, household income, and years of education were modeled as regressors for each GLM. Diagnosis was modeled as a single dummy regressor. We included two-way interactions modeling diagnosis × household income and diagnosis × education to investigate differential effects of SES on reward processing by diagnosis. GLMs were corrected for multiple comparisons using the false discovery rate (FDR) (56). Post hoc, we also investigated significant interactions using diagnosis-specific GLMs. We also tested the impact of BMI, employment status, and ethnicity on significant effects. To avoid binning participants and maintain the statistical testing assumptions, we investigated the impact of the 7 (50%) ethnicities with the greatest sample size, using the largest group as the comparator. Our analyses centered primarily on education and household income, with employment status being modeled as a covariate in post hoc analyses. Income, education, and employment status are overlapping constructs and are all predictive of health outcomes across the lifespan (57,58). However, employment status consists of multiple factors, including social status and an individual’s material resources (59). As such, many studies regarding the relationship between SES and health outcomes have focused on income and education (22,36).

Next, we tested whether socioeconomic indicators associated with reward processing were differentially correlated with 3 proinflammatory cytokines—IL-1β, IL-6, and TNF-α—in participants with MDD and control participants. We performed 3 GLMs with the serum cytokine level as the dependent variable and age, sex, recruitment site, diagnosis, education, and household income as regressors. As in the previous analyses, we included two-way interaction terms to investigate differential effects of SES on cytokine level by MDD status. We examined regressors for GLMs that had a significant omnibus test after FDR correction (56). We also performed 2 mediation analyses in the MDD sample to investigate whether significant proinflammatory cytokines mediated the relationship between SES and reward processing. Mediation models were evaluated with and without site, age, sex, education, and BMI as covariates. In addition, 95% CIs were generated using bootstrapping (1000 iterations).

We proceeded to investigate whether different striatal RSFC correlated with SES in participants with MDD compared with control participants. We performed 2 mixed-effects GLMs per ROI (NAc and DS) (55). We modeled this as a continuous covariate interaction model with 12 predictors, 2 modeling diagnostic group, 2 modeling household income or education (MDD and control household income/education were modeled as separate regressors but mean centered using all data), age, sex, and dummy variables representing recruitment site. Two linear contrasts determined whether the slope of MDD household income or education on striatal connectivity differed significantly from the slope of household income or education for control participants. Clusters were corrected for multiple comparisons using Gaussian random field theory (60). Because the standard height threshold of z > 3.1 yielded nonsignificant findings, we performed exploratory analyses using a more lenient z > 2.3 and cluster p < .025 two-tailed [adjusted for the 2 striatal ROIs (45,46,48, 49, 50, 51,61)]. First-level parameter estimates were extracted for significant clusters and retained for post hoc analyses.

To test associations with depression severity, we performed a GLM predicting baseline MADRS severity, with age, sex, site, education, and household income as covariates. Furthermore, we used repeated-measures analysis of variance to test the effect of SES on escitalopram-induced symptom improvement from baseline to treatment week 8, with regressors modeling age, sex, education, and household income. Post hoc GLMs predicting severity across time (weeks 2, 4, 6, and 8), with regressors modeling age, sex, education, household income, and baseline severity, were used to further characterize significant within-subjects effects (FDR corrected). Our sample was adequately powered to assess different RSFC between groups, with previous rs-fMRI studies with comparable sample sizes (62). According to results reported by Zhong et al. (63) performing multivariate whole-brain RSFC analyses at α = 0.05 and power = 0.8, we would need approximately 40 participants per group to detect a significant difference in RSFC across groups, which our sample exceeds.

Results

Household Income Correlates With Behavioral Approach in MDD

Individuals with MDD (n = 211) and control participants (n = 112) did not differ significantly on age, sex, ethnicity, or household income (Table 1). However, the MDD group had significantly fewer years of education and higher BMI and were less likely to be currently working or a student and more likely to be on permanent or temporary disability, temporarily laid off, on sick or maternity leave, or looking for work relative to the control group.

First, we tested for differences in the relationship between reward processing and SES in participants with MDD and control participants. Omnibus tests indicated that all 4 GLMs for BIS/BAS subscales were significant (p_FDR < .001) (Tables S1–S5). All models had a significant main effect of diagnosis, such that participants with MDD had higher behavioral avoidance and lower behavioral approach scores (p_FDR < .001). One interaction was significant: group × household income with the BAS Reward Drive scale (estimate = 0.491, SE = 0.174; 95% CI, 0.148 to 0.833; t₂₅₆ = 2.820; p_FDR = .020) (Figure 1A). A group × education interaction with BAS Fun Seeking trended to significance (estimate = 0.365, SE = 0.146; 95% CI, 0.077 to 0.653; t₂₅₆ = 2.498; p = .013; p_FDR = .052). Post hoc testing revealed that BAS Reward Drive scores were significantly associated with household income in the MDD group (estimate = 0.288, SE = 0.107; 95% CI, 0.077 to 0.499; t₁₆₉ = 2.693; p_FDR = .031) and nonsignificant in control participants (estimate = −0.195, SE = 0.158; 95% CI, −0.508 to 0.119; t₈₀ = −1.237; p_FDR = .220).

Figure 1.

Household income is associated with the Behavioral Activation Scale (BAS) reward drive and interleukin 1β (IL-1β) in individuals diagnosed with major depressive disorder (MDD). (A) Household income was positively associated with BAS reward drive in individuals with MDD, but not in participants without depression. (B) IL-1β was positively associated with BAS reward drive in individuals with MDD, but not in participants without depression. Violin plots visualize the distribution of raw BAS reward drive subscale scores and transformed IL-1β levels for each income bin, with a least squares line indicating the trend of effect before accounting for covariates.

We also tested whether these two-way interactions remained after considering ethnicity, employment status, or BMI. For group × household income on BAS Reward Drive scores, the interaction remained significant after considering ethnicity (estimate = 0.459, SE = 0.188; 95% CI, 0.089 to 0.828; t₂₂₃ = 2.447; p = .015), employment status (estimate = 0.473, SE = 0.173; 95% CI, 0.133 to 0.813; t₂₃₄ = 2.738; p = .007), or BMI (estimate = 0.486, SE = 0.176; 95% CI, 0.139 to 0.832; t₂₅₀ = 2.762; p = .006). Similarly, the group × education interaction on BAS Fun Seeking effect still trended to significance after covarying with ethnicity (estimate = 0.423, SE = 0.153; 95% CI, 0.121 to 0.724; t₂₂₃ = 2.764; p = .006), employment status (estimate = 0.381, SE = 0.151; 95% CI, 0.083 to 0.680; t₂₃₄ = 2.520; p = .012), or BMI (estimate = 0.374, SE = 0.152; 95% CI, 0.075 to 0.673; t₂₅₀ = 2.462; p = .015). In sum, the results indicate that SES differentially impacted approach behaviors in participants with MDD relative to control participants, and these findings were robust when considering additional demographic factors.

Household Income Correlates With IL-1β in MDD

Next, we tested whether serum proinflammatory cytokines were differentially related to SES in participants with MDD and control participants. Only the omnibus test assessing IL-1β was significant (IL-1β: F_12,239 = 5.504, p_FDR < .001, adjusted R² = 0.162; IL-6: F_12,239 = 1.636, p_FDR = .082, adjusted R² = 0.030; TNF-α: F_12,239 = 1.905, p_FDR = .052, adjusted R² = 0.042) (Tables S6 and S7). The IL-1β model had a significant group × household income interaction (estimate = 0.199, SE = 0.090; 95% CI, 0.022 to 0.376; t₂₃₉ = 2.212; p = .028) (Figure 1B), which similarly trended when considering BMI (estimate = 0.176, SE = 0.089; 95% CI, −0.000 to 0.352; t₂₃₄ = 1.966; p = .051) and remained significant when considering employment status (estimate = 0.192, SE = 0.092; 95% CI, 0.012 to 0.373; t₂₁₇ = 2.102; p = .037) or ethnicity (estimate = 0.205, SE = 0.099; 95% CI, 0.010 to 0.400; t₂₀₇ = 2.077; p = .039). In post hoc GLMs, we found that BMI in the MDD group was significantly associated with IL-1β (estimate = 0.036, SE = 0.143; 95% CI, 0.009 to 0.062; t₁₈₄ = 2.625; p = .009) but not education (estimate = 0.006, SE = 0.025; 95% CI, −0.044 to 0.056; t₁₉₇ = 0.230; p = .818) or household income (estimate = 0.024, SE = 0.021; 95% CI, −0.019 to 0.066; t₁₆₈ = 1.110; p = .268). Post hoc tests indicated that household income in the MDD group was positively associated with IL-1β (estimate = 0.136, SE = 0.049; 95% CI, 0.039 to 0.233; t₁₅₈ = 2.761; p_FDR = .013) and nonsignificant in control participants (estimate = 0.106, SE = 0.093; 95% CI, −0.078 to 0.290; t₇₄ = 1.145; p_FDR = .256). Mediation analyses indicated that the relationship between household income and reward drive was not significantly mediated by IL-1β (see the Supplement). The results indicate a positive association between household income and IL-1β in MDD.

DS RSFC Correlates With Household Income in MDD

Next, we identified striatal RSFC that correlated differentially with SES by diagnosis. A total of 150 participants with MDD and 99 control participants were included (MDD: mean age [SD] = 34.44 [12.37] years, 94 female; control: mean age [SD] = 33.03 [10.87] years; 64 female) (Table S8). We did not find any significant clusters for NAc RSFC or for education for either ROI. However, in exploratory analyses using the more lenient cluster threshold of z > 2.3, we found a significant diagnosis × household income interaction in DS RSFC with the left premotor and dorsolateral prefrontal cortex (DLPFC) (140 mm³, cluster peak MNI: −28, +16, +30; cluster p = .017; Brodmann areas 6, 8, 9) and posterior cingulate cortex (PCC), retrosplenial cortex, and precuneus (220 mm³, cluster peak MNI: −16, −56, +10; cluster p = .001; Brodmann areas 30, 31) (Figure 2A). This interaction remained significant after incorporating education, BMI, employment status, or ethnicity (all models p < .001). In participants with MDD, household income was positively associated with DS-DLPFC (estimate = 0.047, SE = 0.018; 95% CI, 0.082 to 0.230; t₁₂₂ = 2.601; p_FDR = .010) (Figure 2B) and DS-PCC RSFC (estimate = 0.055, SE = 0.017; 95% CI, 0.088 to 0.287; t₁₂₂ = 3.287; p_FDR = .003) (Figure 2C). The opposite was true for control participants (DLPFC: estimate = −0.082, SE = 0.027; 95% CI, −0.136 to 0.028; t₇₂ = −3.049; p_FDR = .004; PCC: estimate = −0.097, SE = 0.030; 95% CI, −0.156 to 0.038; t₇₂ = −3.278; p_FDR = .003).

Figure 2.

Dorsal striatum (DS) resting-state functional connectivity (RSFC) is differentially associated with household income in patients with major depressive disorder (MDD) and control participants. (A) A significant diagnosis × household interaction was observed in DS RSFC with the posterior cingulate cortex (DS-PCC) (220 mm³, cluster peak Montreal Neurological Institute [MNI]: −16, −56, +10; cluster p = .001; Brodmann areas 30, 31; yellow-red), and dorsolateral prefrontal cortex (DS-DLPFC) (140 mm³, cluster peak MNI: −28, +16, +30; cluster p = .017; Brodmann areas 6, 8, 9; yellow-red). The blue cluster on the left visualizes the region of interest used for the seed-to-voxel–based analysis. (B, C) Post hoc general linear models for each clinical group indicated that DS RSFC was significantly positively correlated with self-reported household income in the MDD group and anticorrelated in the control group. Violin plots depict the z-transformed first-level parameter estimates for DS RSFC with either the DLPFC or PCC for each income bin, with a least squares line indicating the effect before accounting for covariates.

Household Income Correlates With Early Antidepressant Improvement

We tested whether SES was associated with baseline severity and symptom improvement. Regarding the relationship between SES and baseline symptom severity (omnibus test: F_9,170 = 2.190, p = .025, adjusted R² = 0.056), neither education (estimate = −0.247, SE = 0.190; 95% CI, −0.623 to 0.128; t₁₇₀ = −1.299; p = .196) nor household income (estimate = −0.035, SE = 0.241; 95% CI, −0.511 to 0.442; t₁₇₀ = −0.144; p = .886) was significant (Figure 3A and Table S9). However, there was a significant time × household income effect on improvement (Greenhouse-Geisser F_{3.194,472.461} = 3.737, p = .010) (Figure 3B; Tables S10 and S11). Individuals with MDD who reported a lower household income had significantly higher MADRS scores at treatment weeks 2 (estimate = −1.219, SE = 0.320; 95% CI, −1.851 to 0.587; t₁₅₃ = −3.810; p_FDR = .001), 4 (estimate = −0.859, SE = 0.363; 95% CI, −1.575 to 0.142; t₁₅₀ = −2.367; p_FDR = .038), and 6 (estimate = −0.833, SE = 0.383; 95% CI, −1.589 to 0.076; t₁₄₄ = −2.175; p_FDR = .042) but not week 8 (estimate = −0.426, SE = 0.442; 95% CI, −1.299 to 0.447; t₁₄₅ = −0.965; p_FDR = .336). Lastly, we tested whether incorporating significant biobehavioral correlates of household income impacted the relationship with improvement. The effect remained significant after including baseline BAS Reward Drive scores, IL-1β, and DS RSFC (Greenhouse-Geisser F_{2.998,311.745} = 2.644; p = .049), indicating that socioeconomic indicators may drive trajectories of antidepressant response independent of its relationship to pretreatment reward processing, proinflammatory cytokines, or striatal RSFC.

Figure 3.

Household income is associated with early improvements in response to open-label escitalopram. (A) Socioeconomic indicators were not significantly associated with baseline depression severity, as measured by the Montgomery–Åsberg Depression Rating Scale (MADRS). (B) A repeated-measures analysis of variance revealed a significant time × household income interaction, such that individuals who reported a higher socioeconomic status experienced greater reductions in depression severity at treatment weeks 2, 4, and 6. To visualize the effect more simply, we binned income groups above or below the study median (median income bin: $50,000–$74,999) and calculated the mean MADRS severity at each time point for individuals who were below the median (<$50,000), at the median ($50,000–$74,999), or above the median (>$74,999). Error bars depict the standard error.

Discussion

Independent studies have found robust correlations of SES with depressive symptoms such as anhedonia, antidepressant treatment outcomes, and proinflammatory biomarkers. To our knowledge, this study was the first to examine the relationship between these factors using clinical, behavioral, and functional neuroimaging measures in a deeply phenotyped MDD sample. Our results indicated that household income but not education was correlated with reward drive, IL-1β, and DS RSFC in patients with MDD. Furthermore, individuals with higher household incomes reported earlier escitalopram improvement. Consistent with our hypotheses, these results suggest that SES influences biological markers and treatment outcomes in MDD.

Household income was associated with reward drive in MDD. Previous research with control participants confirms that low SES is correlated with elevated reward sensitivity (64), which may be adaptive because it enables disadvantaged individuals to take immediate advantage of opportunities in unpredictable environments (65,66). Our finding suggests that an MDD diagnosis moderates the relationship between reward processing and SES, consistent with prior work (67, 68, 69). Furthermore, DS-PCC RSFC was differentially correlated with household income by diagnosis. DS activity during reward tasks is correlated with reward drive (70). Default mode network (DMN) function, which implicates the PCC and engages during reward processing (68, 69, 70), is positively correlated with childhood SES in control participants (71, 72, 73, 74). Furthermore, DMN connectivity is consistently atypical in MDD, but the direction of the effect is mixed (75). Our results indicate that SES correlates with DMN connectivity in MDD, potentially influencing these mixed findings.

DS-DLPFC RSFC was correlated with household income in MDD. The DLPFC plays a significant role in executive function, including working memory, decision making, and task initiation, domains in which individuals with MDD often display deficits (76,77). Frontostriatal hypoconnectivity is frequently reported in MDD and is correlated with deficits in reward processing (78, 79, 80, 81) and poorer treatment outcomes (82). Furthermore, DLPFC volume in control participants is anticorrelated with SES (83) and mediates the relationship between SES and reward-related cognition (84). Our results suggest that SES and MDD may interact to play a compounding role on frontostriatal connectivity, as individuals with MDD from lower SES backgrounds displayed lower frontostriatal connectivity than their higher SES counterparts. These findings add to a growing body of work that shows that socioeconomic disparities may impact frontostriatal connections implicated in treatment outcomes.

In contrast to previous work, IL-1β was positively correlated with household income in MDD, although this was rendered nonsignificant after accounting for BMI. Previous work indicates that individuals with MDD display abnormally high serum levels of proinflammatory cytokines and that SES is anticorrelated with serum proinflammatory levels in control participants (69,85). Consequently, we anticipated that individuals with MDD from lower SES backgrounds would display the highest levels of serum proinflammatory cytokines, the opposite of what we observed. Similar to our findings, some studies of pregnant people have found a positive correlation between IL-1β and SES, potentially related to desensitization of proinflammatory pathways following CS related to social or economic disadvantage (86,87). Previous work indicates that SES (88, 89, 90, 91) and MDD (92) are associated with higher BMI, which in turn is associated with increased IL-1β (93, 94, 95). Our findings support the notion that diagnosis is associated with higher BMI, but we did not observe a significant association with SES in MDD. Regardless, covarying for BMI may partially explain variance in the relationship between SES and neuroinflammation in individuals with MDD because we found a strong association between BMI and cytokine function consistent with prior literature (47). Additionally, we observed significant sex differences, which have been observed previously in control participants and people with MDD (13,96,97). Taken together, these findings highlight the complex and potentially interacting roles of SES, sex, and dysregulated immune activity in contributing to MDD. Future work should aim to elucidate the longitudinal relationship between SES and cytokine function, ideally in a sex-stratified analysis, in contributing to MDD pathophysiology.

Individuals with depression with higher annual household incomes responded more quickly to escitalopram. Early response to antidepressants is a consistent predictor of later remission (98,99) and has been suggested as a criterion for early medication switches or augmentation therapy (100). However, treatment switching is associated with increased economic costs and greater health care use (101, 102, 103). This highlights the disparities in mental health outcomes faced by those from lower SES backgrounds and suggests that socioeconomic factors, if unaccounted for during model training, could confound the prediction accuracy of decision trees or other heuristics used for early medication switches or augmentation therapy. Consistent with this notion, the accuracy of fMRI-based models for predicting behavior is impacted by a complex relationship with sociodemographic factors (104). Our results highlight the importance of considering SES factors in characterizing trajectories or biomarkers of treatment response.

We note several limitations of this study. First, while household income measured on a Likert scale is a valid measure of SES, income-to-needs ratio is more robust because it accounts for household size and cost of living (105). We were unable to quantify this ratio, as household size or income as a continuous variable was not collected. Second, the experience of individuals from lower SES backgrounds is not monolithic. Consistent with prior studies, we observed divergent results across our measures of SES, reinforcing the nature of SES as a complex construct (36,57,58). Health care outcomes can vary depending on individual life factors, the age at which individuals experience low SES, or the way SES is defined (8). Similarly, other sociodemographic factors were not acquired that could provide nuance, including gender and self-described ethnicity. While we used available ethnicity data, we acknowledge that the data may not fully represent the ethnic, immigrant, or cultural groups present in Canada and are often misused to represent other socioeconomic disparities and experiences such as discrimination (106, 107, 108). Furthermore, the sample was disproportionately White, which has limitations when used as a statistical reference group (109,110). Third, we resorted to a lenient cluster-forming threshold, which potentially increases the risk of type I error (56). This threshold has been previously used to quantify differences in RSFC (111, 112, 113, 114, 115, 116), including in MDD (61). Fourth, data regarding reward processing were collected using the BIS/BAS questionnaire. While this survey is a validated measure of reward sensitivity, future work should incorporate task-based measures of reward-oriented behavior (46,117). Fifth, since the fMRI data were collected, there have been several updates to fMRI technology, including the ability to record data with greater spatial resolution (118). Lastly, this secondary analysis was not preregistered, which serves to minimize bias.

Conclusions

To our knowledge, the current study is the first to explore the relationships between SES, neuroinflammation, and reward-related RSFC in individuals with MDD treated with escitalopram. This study showed that SES influences goal pursuit, inflammatory cytokines, and DS RSFC in patients with MDD. Individuals who reported a higher household income exhibited earlier symptom improvements in response to escitalopram, which may have implications for clinical decisions to continue or adjust medications. This work contributes to an existing body of literature suggesting that low SES contributes to negative outcomes in MDD (11,119, 120, 121). As such, better characterizing how SES interacts with an individual’s well-being and their neurobiology, and by extension identifying how these factors can be directly targeted, is vital for ensuring the best possible outcomes for people affected by MDD.