Genome-wide interaction study of a proxy for stress-sensitivity and its prediction of major depressive disorder

Individual response to stress is correlated with neuroticism and is an important predictor of both neuroticism and the onset of major depressive disorder (MDD). Identification of the genetics underpinning individual differences in response to negative events (stress-sensitivity) may improve our understanding of the molecular pathways involved, and its association with stress-related illnesses. We sought to generate a proxy for stress-sensitivity through modelling the interaction between SNP allele and MDD status on neuroticism score in order to identify genetic variants that contribute to the higher neuroticism seen in individuals with a lifetime diagnosis of depression compared to unaffected individuals. Meta-analysis of genome-wide interaction studies (GWIS) in UK Biobank (N = 23,092) and Generation Scotland: Scottish Family Health Study (N = 7,155) identified no genome-wide significance SNP interactions. However, gene-based tests identified a genome-wide significant gene, ZNF366, a negative regulator of glucocorticoid receptor function implicated in alcohol dependence (p = 1.48x10-7; Bonferroni-corrected significance threshold p < 2.79x10-6). Using summary statistics from the stress-sensitivity term of the GWIS, SNP heritability for stress-sensitivity was estimated at 5.0%. In models fitting polygenic risk scores of both MDD and neuroticism derived from independent GWAS, we show that polygenic risk scores derived from the UK Biobank stress-sensitivity GWIS significantly improved the prediction of MDD in Generation Scotland. This study may improve interpretation of larger genome-wide association studies of MDD and other stress-related illnesses, and the understanding of the etiological mechanisms underpinning stress-sensitivity.


Introduction
Using unrelated individuals from two large population-based samples, UK Biobank (UKB; N = 23,092) and Generation Scotland: Scottish Family Health Study (GS:SFHS; N = 7,155), we sought to identify genes involved in stress-sensitivity by performing GWIS for the interaction between MDD status and SNP allele on neuroticism score. We identified a gene significantly associated with stress-sensitivity and show that a PRS derived from the interaction term of the GWIS, significantly predicts liability to depression independently of the PRS for MDD and/or neuroticism.

UK Biobank (UKB) participants
UKB is a major national health resource that aims to improve the prevention, diagnosis and treatment of a wide range of illnesses. It recruited more than 500,000 participants aged from middle to older age who visited 22 assessment centres across the UK between 2006 and 2010. Data were collected on background and lifestyle, cognitive and physical assessments, sociodemographic factors and medical history. The scientific rationale, study design, ethical approval, survey methods, and limitations are reported elsewhere [51,52]. UKB received ethical approval from the NHS National Research Ethics Service North West (Research Ethics Committee Reference Number: 11/NW/0382). All participants provided informed consent. The present study was conducted on genome-wide genotyping data available from the initial release of UKB data (released 2015). Details of sample processing specific to UKB project are available at http://biobank.ctsu.ox.ac.uk/crystal/refer.cgi?id=155583 and the Axiom array at http://media.affymetrix.com/support/downloads/manuals/axiom_2_assay_auto_workflow_ user_guide.pdf. UKB genotyping and the stringent QC protocol applied to UKB data before it was released can be found at http://biobank.ctsu.ox.ac.uk/crystal/refer.cgi?id=155580. SNPs genotyped on GS:SFHS were extracted from the imputed UKB genotype data [53] (imputed by UKB using a merged panel of the UK10K haplotype reference panel and the 1000 Genomes Phase 3 reference panel) with quality > 0.9 was hard-called using PLINK v1.9 [54]. Individuals were removed based on UKB genomic analysis exclusion (UKB Data Dictionary item #22010), non-white British ancestry (#22006: genetic ethnic grouping; from those individuals who selfidentified as British, principal component analysis was used to remove outliers), high genotype missingness (#22005), genetic relatedness (#22012; no pair of individuals have a KING-estimated kinship coefficient > 0.0442), QC failure in UK BiLEVE study (#22050 and #22051: UK BiLEVE Affymetrix and UK BiLEVE genotype quality controls for samples) and gender mismatch (#22001: genetic sex). Further, from the initial release of UKB data and using PLINK pihat < 0.05, individuals who were also participants of GS:SFHS and their relatives were excluded to remove any overlap of individuals between discovery and target samples. A dataset of 109,283 individuals with 557,813 SNPs remained for further analysis, aged 40-79 (57,328 female, 51,954 male; mean age = 57.1 years, s.d. = 7.99), of which 109,282 had data available for neuroticism score and 23,092 had data available on MDD status (n cases = 7,834, n controls = 15,258, n female = 11,510, n male = 11,582; mean age = 57.7 years, s.d. = 8.04). Thus, the final dataset comprised 23,092 unrelated individuals. environmental covariates and linked to routine health records [55,56]. All components of GS: SFHS obtained ethical approval from the Tayside Committee on Medical Research Ethics on behalf of the NHS (Research Ethics Committee Reference Number: 05/S1401/89) and participants provided written consent. The protocol for recruitment is described in detail in previous publications [57,58]. GS:SFHS genotyping and quality control is detailed elsewhere [59]. Briefly, individuals with more than 2% missing genotypes and sex discrepancies were removed, as well as population outliers. SNPs with genotype missingness > 2%, minor allele frequency < 1% and a Hardy-Weinberg Equilibrium test p < 1x10 −6 were exclude. Finally, individuals were removed based on relatedness (pi-hat < 0.05), maximizing retention of case individuals, using PLINK v1. 9

Phenotype assessment
Neuroticism score (EPQN). Participants in both UKB and GS:SFHS cohorts were assessed for neuroticism using 12 questions from the Eysenck Personality Questionnaire-Revised Short Form's Neuroticism Scale (EPQN) [60][61][62][63]. Neuroticism can be scored by adding up the number of "Yes" responses on EPQN. This short scale has a reliability of more than 0.8 [64]. EPQN distributions were found to be sufficiently "normal" after assessment for skewness and kurtosis to be analysed using linear regression (both coefficients were between -1 and 1).
MDD diagnoses. In UKB, the MDD phenotype was derived following the definitions from Smith et al. [63] Current and previous depressive symptoms were assessed by items relating to the lifetime experience of minor and major depression [60], items from the Patient Health Questionnaire [65] and items on help-seeking for mental health [63]. Using a touchscreen questionnaire, participants were defined as probable cases if they i) answered "Yes" to the question "Ever depressed for a whole week" (UKB field: 4598), plus at least 2 weeks duration (UKB field: 4609), or ii) did report having seen a GP or psychiatrist for nerves, anxiety, tension or depression (UKB fields: 2090 and 2010) and reported symptoms (UKB field: 4631) with at least 2 weeks duration (UKB field: 5375). In our unrelated sample, 7,834 participants were diagnosed with MDD (with single, moderate or recurrent episodes) and 15,258 were controls (N = 23,092).
In GS:SFHS, participants took in-person clinical visits where they were screened for a history of psychiatric and emotional disorders (i.e., psychiatric, mood state/psychological distress, personality and cognitive assessment) by trained researchers using the Structured Clinical Interview for DSM-IV Non-Patient Version (SCID) [66], which is internationally validated to identify episodes of depression. Those participants that were positive in the initial screening continue through clinical interview and were administered the mood sections of the SCID. The SCID elicited the presence or absence of a lifetime history of MDD, age of onset and number of episodes. Participants fulfilling the criteria for at least one major depressive episode within the last month were defined as current MDD cases. Participants who were screened positive for Bipolar I Disorder were excluded. Those participants who were negative during the initial screening or did not fulfilled criteria for MDD were assigned as controls. Further details regarding the diagnostic assessment are reported elsewhere [56,57]. All interviewers were trained for the administration of the SCID. Inter-rater reliability for the presence or absence of a lifetime diagnosis of major depressive disorder was good (Kappa = 0.86, p < 0.001, 95%CI 0.7 to 1.0). In our unrelated GWIS sample (N = 7,155), 2,010 had a lifetime diagnosis of MDD and 5,145 were controls.

Statistical methods
GWIS and derivation of a genetic stress-sensitivity effect. The effect size of an stresssensitivity effect (β SS ) was derived by performing a GWIS for the effect of the MDD status and SNP allele on EPQN (dependent variable) in both UKB and GS:SFHS cohorts using PLINK 1.90 (PLINK-command-gxe; fitting MDD diagnosis as a binary "group" effect) [54]. PLINKcommand-gxe estimates the difference in allelic association with a quantitative trait (EPQN) between two groups (MDD cases vs. controls) producing effect estimates on each group and a test of significance for the interaction between SNP allele and MDD status. The interaction p value reflects the difference between the regression coefficient of the allelic effect in a linear model for EPQN in MDD cases (β A ) and the same regression coefficient in a linear model for EPQN in controls (β B ). The stress-sensitivity interaction effect was defined as the difference in allele effect between MDD cases and control groups.
Considering one SNP, the effect it confers to EPQN can be modelled by MDD status (control = 0, MDD case = 1) as follows: This is equivalent to modelling the effect on MDD cases as follows: Or, it can be modelled as a whole as: Where COV stands for covariates, β 2 stands for β 1 −β 0 , and β 2c stands for β 1c −β 0c . Thus, the interaction effect (β SS ) can be estimated as the difference in allelic effect on EPQN between MDD cases (β A ) and controls (β B ) as follows, b SS is therefore defined as the effect size reflecting the genetic stress-sensitivity effect on MDD cases compared to controls (S1 Fig). Stress-sensitivity GWIS, main additive effect GWASs, meta-analysis and gene-set analysis. For GWIS and subsequent analyses, sample specific covariates were applied as follows: UKB. All phenotypes were adjusted for centre, array and batch as random effects prior to analyses. Analyses were adjusted for age, sex and 15 informative principal components (PCs; UKB Data Dictionary items #22009.01 to #22009.15) as fixed effects to take account of possible population stratification. GS:SFHS. All the analyses were adjusted for age, sex and 20 PCs.
GWAS for MDD and neuroticism, using logistic and linear models of additive allelic effects respectively, were conducted on the same sample sets for comparison and generation of matched PRS using PRSice-2 [67].
Results from the GWIS of UKB and GS:SFHS were combined in a sample size weighted meta-analysis performed using METAL [68]. While the use of standard error weighting is more common, the different diagnostic scheme and MDD prevalence between the two cohorts (GS:SFHS; 12.2%, UKB: 25.8%) [57, 63] may indicate systematic differences in the measurement of MDD. Generalized gene-based analysis of the meta-analysis was performed using MAGMA [69] implemented through FUMA [70] (http://fuma.ctglab.nl). Briefly, SNP summary statistics were mapped to 17,931 protein-coding genes. Individual SNP p values from a gene were combined into a gene test-statistic using a SNP-wise model and a known approximation of the sampling distribution used to obtain a gene-based p value. Genome-wide significance was defined at p = 0.05/17,931 = 2.79x10 -6 .
LD Score regression. The summary statistics from the meta-analysis were used to examine the genetic overlap between the polygenic architecture of stress-sensitivity, MDD and neuroticism. LD score regression was used to derive the genetic correlations (r G ) between these traits [71,72] using meta-analysed GWAS and GWIS summary statistics. SNP-based heritability was also estimated using LD score regression, using the summary statistics from single-SNP analyses.
Polygenic profiling. PRS were produced using PRSice-2 [67], permuted 10,000 times and standardized to a mean of 0 and a standard deviation of 1. Using GWIS summary statistics, we created PRS for stress-sensitivity (PRS SS ) by weighting the sum of the reference alleles in an individual by the stress-sensitivity effect (β SS ). Additional PRS were generated weighting by MDD main additive effects (PRS D ) and neuroticism main additive effects (PRS N ) using GWAS summary statistics from GS:SFHS or UKB. In addition, PRS D and PRS N were also generated using summary statistics from the most recent Psychiatric Genetic Consortium (PGC) MDD meta-analysis [42] (excluding GS:SFHS, and UKB individuals when required; N = 155,866 & 138,884) and the Genetics of Personality Consortium (GPC) neuroticism metaanalysis [24,77] (N = 63,661). Generalized linear models were implemented in R 3.1.3 [78]. The direct effect of PRS SS (model 1), PRS D (model 2) and PRS N (model 3) on MDD risk were assessed in independent logistic regression models on GS:SFHS (target cohort) using GWAS and GWIS statistics from UKB (the largest cohort) as the discovery sample to weight PRS. Multiple regression models fitting both PRS D and PRS N (model 4) and fitting each of them separately with PRS SS (models 5 and 6) were also calculated. Finally, full additive multiple regression models fitting PRS weighted by all three effects (full model) was assessed using both PRS SS , PRS D and PRS N at their best-fit in independent models . Further, results were also assessed using PRS D and PRS N weighted by PGC2 MDD [42] and GPC neuroticism [77] summary statistics. Further detail is given in 'Polygenic Profiling' in S1 Supporting Information. All models were adjusted by sex, age and 20 PCs. A null model was estimated from the direct effects of all covariates on MDD. 10,000 permutations were used to assess significance of each PRS. The predictive improvement of combining the effects of multiple PRS over a single PRS alone was tested for significance using the likelihood-ratio test.
Cross-validation was performed using UKB as target sample and GS:SFHS as discovery sample. Additional analyses using PRS D and PRS N weighted by PGC2 MDD [42] and GPC neuroticism [77] summary statistics were also tested. MDD status on UKB was adjusted by centre, array and genotyping batch as random effects and scaled (between 0 and 1) prior to analysis, giving a quasi-binomial distribution of MDD status on UKB. Models implemented on UKB (quasi-binomial regression) were adjusted by sex, age and 15 PCs. Nagelkerke's R 2 coefficients were estimated to quantify the proportion of MDD liability explained at the observed scale by each model and converted into R 2 coefficients at the liability scale (prevalence: 12.2% in GS:SFHS [57] and 25.8% in UKB [63]) using Hong Lee's transformation [79] available from GEAR: GEnetic Analysis Repository [80].
Using stress-sensitivity to stratify depression. GS:SFHS MDD cases (n cases = 2,016; n female = 1,345, n male = 671) have data available on MDD course (single or recurrent), age of onset (n = 1,964) and episode count (n = 2,016), as well as on neuroticism (n = 2,010). In addition, a subset were evaluated by Mood Disorder Questionnaire [81] (MDQ; n = 1,022) and Schizotypal Personality Questionnaire [82] (SPQ; n = 1,093). The reduced sample number of MDQ and SPQ reflects the later addition of these questionnaires to the study and does not reflect a particular subgroup of GS:SFHS.
Difference in PRS SS and PRS D between MDD cases and controls on GS:SFHS were tested using a Student's two sample t-test (two tailed). Cases of MDD on GS:SFHS with data available on each trait analyzed were stratified by quintiles based on PRS SS and PRS D (5x5 groups). Post hoc, the effects on each trait of quintiles based on PRS SS and its interaction effect with quintiles based on PRS D were assessed using linear regression models adjusting by sex and age in an attempt to identify a characteristic subtype of MDD patients with differential stress-sensitivity levels. The same analysis was reproduced using PRSs as continuous variables.

Results
We confirmed the elevated neuroticism score in MDD cases in our samples. Individuals with a diagnosis of MDD had significantly higher EPQN scores compared to healthy controls (all p < 1.9.x10 -279 ) in both GS:SFHS (mean controls = 3.16; mean cases = 6.42) and UKB (mean controls = 2.79; mean cases = 5.64). Neuroticism levels differ significantly between males and females. To control for this and any age/polygenic effects, which may account for differences in the prevalence of MDD, we created a matched set of cases and controls. The difference in neuroticism levels between cases and controls remained significant after matching the controls for PGC PRS D , sex and age. (GS:SFHS: mean controls = 3.51; UKB: mean controls = 2.97; all p < 2.7x10 -158 ; S1 Table).

Polygenic risk scores for stress-sensitivity predict MDD liability
PRS were used to investigate whether common variants affecting stress-sensitivity predict MDD risk. We generated PRS (PRS SS ) for stress-sensitivity based on the summary statistics from the GWIS. After 10,000 permutations, PRS SS significantly predicted MDD risk in GS: SFHS using weights from the larger UKB summary data (Empirical-p = 0.04; p = 5.2x10 -3 ; β = Position of the SNP respect to closest gene transcripts within 100kb (including UTRs) from 5 prime (5') or 3prime (3'). LD score regression was performed to obtain genetic correlations between stress-sensitivity, MDD and neuroticism. As previously shown, there was a significant genetic correlation between MDD and neuroticism (r G = 0.637, s.e. = 0.0704, p = 1.39x10 -19 ). However, we found no evidence for a genetic correlation between stress-sensitivity and MDD (rG = -0.099, s.e. = 0.182, p = 0.585) or between stress-sensitivity and neuroticism (r G = 0.114, s.e. = 0.107, p = 0.285).  Table). On the liability scale, the MDD variance explained in GS:SFHS by PRS SS was modest (R 2 = 0.195%). This was less than predicted by PRS weighted by the genetic main effects of MDD or neuroticism (PRS D : R 2 = 0.368%; PRS N : R 2 = 0.459%; Table 2 and S6 Table). However, this association was not cross-validated in UKB using summary data from the smaller GS:SFHS GWIS (Empirical-p = 0.68; p = 0.23; β = 0.004, s.e. = 0.003; best-fit p threshold = 0.005; PRS SS R 2 = 0.013%; S6 Table), likely due to lack of power as a result of the small discovery sample size. PRS D (R 2 = 0.204%) and PRS N (R 2 = 0.166%) derived from GS:SFHS significantly predicted MDD in UKB (Table 2 and S6  Table).
Due to the known genetic correlations between MDD, neuroticism and stressful life events [21], models jointly fitting the effects of multiple PRS were analysed. Multiple regression analyses in GS:SFHS showed that, compared to PRS D effects alone, the stress-sensitivity effect derived from the UKB GWIS effects significantly explains an additional 0.195% (a predictive improvement of 53.1%, p = 5. In models fitting PRS D and PRS N , the variances explained were non-additive, demonstrating the partial overlap between MDD risk prediction from PRS D and PRS N main additive effects. This is consistent with the known genetic correlation between these two traits. An overlap was not seen between the variance explained by PRS SS effect and the variance explained by PRS D and/or PRS N . Multiple regression analyses fitting PRS D and PRS N derived from worldwide consortiums (Fig 3) showed that the increased sample size from GWAS used to derive PRS D resulted in an increment of MDD variance explained in GS:SFHS by PRS D (from 0.368% to 1.378%). However, there was no change in the proportion of the variance explained by the PRS SS in the full model (PRS SS p = 3.5x10 -3 ). These results suggest that PRS SS explains a proportion of MDD risk not accounted for by PRS D or PRS N at current sample sizes. However, these findings were not cross-validated in UKB using PRS SS derived from GS:SFHS GWIS, likely due to lack of power as a result of the small discovery sample size (S6 Fig). Using stress-sensitivity to stratify MDD in GS:SFHS MDD cases show significantly higher PRS SS (p = 2x10 -3 ) and PRS D (p = 1.8x10 -4 ) than controls. Association between MDD-related traits and stress-sensitivity risk quintiles was assessed Stress-sensitivity proxy and depression on MDD cases in order to identify a subgroup of MDD patients, perhaps defining a characteristic aetiological subtype of MDD. However, stratification analysis failed, and no quintile based on PRS SS nor its interaction with quintiles based on PRS D showed statistically significant effects on any trait analyzed. Individuals with high PRS SS were not significantly different from other cases for sex, MDD course, age of onset or episode count, nor neuroticism, mood disorder or schizotypal personality scores (p > 0.05; S7 Table). Results remained non significant when PRSs were fitted as continuous variables (p > 0.05).

Discussion
The existence of genetic variants affecting an individual's risk of depression in response to stress has been predicted previously [46,49,50] and is consistent with the departure from a simple additive genetic model seen in twin-studies of recurrent depressive disorder [83].
Through international research efforts such as the PGC and UK Biobank, there are everincreasing sample sizes available for understanding the genetics of MDD. These resources are beginning, and will continue to, identify genome-wide significant loci [42, 84,85]. However, the lack of environmental data and/or their reliability, makes the study of genetic individual's response to their negative effects, and their contribution to the onset of MDD and other stressrelated disorders, difficult. As a way to address this limitation, we generated a proxy for stresssensitivity through modelling the interaction between SNP allele and MDD status on neuroticism score in a GWIS approach. Thus, we sought to identify the genetic underpinnings of individual's sensitivity to stress response (stress-sensitivity) through those variants that contribute to higher neuroticism levels only in individuals with a lifetime diagnosis of MDD but not in healthy controls. We performed a GWIS to identify loci showing differential effects on neuroticism scores in individuals with and without MDD (so called stress-sensitivity proxy). No SNPs reached genome-wide significance, but 14 SNPs from 8 loci reached suggestive significance levels (see S4 Table for prior evidence of associated phenotypes). Enrichment analysis showed no evidence for enrichment of specific pathways or tissues. The top two loci, PTP4A1-PHF3-EYS and ZNF366 have been previously associated with alcohol dependence [86][87][88][89][90], alcohol intake (dbGaP: phs000342) and glucocorticoid receptor function [91][92][93]. The most significant SNP in this study, rs319924, is an intronic variant in EYS that is a potential eQTL for LGSN [76], a gene previously associated with male-specific depression [94]. This is of particular interest given previous studies linking alcohol consumption, stress and the risk of depression [95][96][97][98][99][100]. However, findings should be interpreted with caution, as these loci did not reach genome-wide significance at current sample size. Evidence of an eQTL effect was predicted for a lead SNP in LATS2, a positive regulator of histone methyltransferase activity [101] a process important in anxiety-related behaviours [102]. The prior association of the top two loci in this study with alcohol related-phenotypes suggests that genes involved in the sensitivity to stress may mediate the effects of stress on alcohol consumption. Some PHF3 paralogs have been shown to be linked with depression and modulate stress response [103,104].
Gene-based analysis identified a genome-wide significant association between ZNF366 and stress-sensitivity. ZNF366 (also known as DC-SCRIPT) is a corepressor of transcription found in nuclear receptor complexes including the glucocorticoid receptor. ZNF366 represses glucocorticoid receptor-mediated transcription in monocyte-derived dendritic cells [91]; and may act through histone deacetylases to modulate immune response [92]. There is evidence from a large-scale mRNA display study that PHF3, in the region underlying the most significant peak in the single SNP analysis, may also interact, directly or indirectly, with the glucocorticoid receptor (IntAct database [93]) but this has not been confirmed. These results reinforce the hypothesis that our proxy for stress-sensitivity truly reflects the genetic architecture of sensitivity to respond to stress.
We estimated a significant lower bound on common SNP-based heritability for stress-sensitivity of 5%. Whilst the known genetic overlap between MDD and neuroticism was detectable, the lack of genetic correlation with stress-sensitivity, reinforced by results from multiple regression analyses, indicated a lack of significant overlap in the genetics factors underpinning stress-sensitivity and MDD or neuroticism. This analysis may be limited by our sample size, although using the largest available meta-analyses of MDD and neuroticism [42,77] did not decrease the proportion of liability explained by the PRS SS . We note, that as such meta-analyses increase in size it is likely, as with the effects of smoking in schizophrenia [105,106], that the indirect genetic effects of the environment on the risk of depression will be detected by GWAS. However, through studies such as ours, or similar, the mechanism for the effect of the risk alleles may be clarified.
Further, we show that such genetic information in stress-sensitivity could significantly improve the proportion of liability to MDD predicted by PRS based only on additive genetic effects on MDD identified by large GWAS. The summary results from the GWIS were used to derive a PRS reflecting the genetic difference in stress-sensitivity. This variable significantly predicted liability to MDD in GS:SFHS (p = 5.2x10 -3 , Empirical-p = 0.04 after 10,000 permutations), although this finding could not be replicated in UKB (Empirical-p = 0.68), likely due to lack of power. This is consistent with the expectation that the larger the discovery sample (i.e. UKB), the greater the accuracy of the weighting and the more predictive the PRS [107]. Multiple regression models in GS:SFHS suggest that inclusion of PRS weighted by stress-sensitivity significantly improves MDD prediction over use of either MDD and/or neuroticism weighted PRS alone (improvement in full model p = 8.5x10 -3 ). However, we were unable to identify a subgroup of MDD cases with higher PRS ss . The polygenic interaction approach used in our study may, therefore, improve the interpretation of both positive and negative findings from GWAS studies (i.e. pathways and mechanisms involved, lack of replication, or negative findings in variants mediating environmental effects). Added to paralleling recent developments in GWAS analyses, it may maximize our power to detect gene-by-environment effects in this heterogeneous disorder.
Future studies will be required to further investigate the effects of adverse life events in individuals with high or low polygenic risk scores for stress-sensitivity. However, the methodology presented allows addressing the genetic response to negative outcomes via proxy in the absence of prospective environmental data.
Here we identify an independent set of risk variants for an individual's response to negative outcomes and show that incorporating information across many loci provides clear and replicable evidence for a genetic effect of stress-sensitivity on MDD risk; identifying a potential genetic link with alcohol intake. These results require further study, but may inform treatment of comorbid alcohol dependency and depression.