Genome-wide interaction association analysis identifies interactive effects of childhood maltreatment and kynurenine pathway on depression

Abstract

Childhood maltreatment stands out as a pivotal risk factor for depression, with gene-by-environment interaction serving as a crucial mechanism. Here we perform genome-wide interaction analyzes of childhood maltreatment in the UK Biobank, integrating methylation evidence through colocalization analysis and identifying associated brain structure abnormalities from childhood to adulthood. A genome-wide significant genomic region interacting with childhood maltreatment is identified at 8p11.21 (IDO2 rs7846217, P = 2.02e-08), implicating the tryptophan-kynurenine pathway. Colocalization analysis reveals that IDO2 rs11777027, rs2340953 and rs28631334 are associated with depression in individuals exposed to childhood maltreatment and colocalize with methylation signals in both blood and brain for IDO2. These interactions affect cortical thickness of the left supramarginal gyrus in children (P = 9.72e-04) and adults (P = 1.34e-04), as well as cortical volume in the right angular gyrus in children (P = 1.02e-04). Furthermore, the interactions significantly predict new-onset depression at a 2-year follow-up in children. Stunted increase in cortical thickness of the left middle-anterior cingulate gyrus and sulcus significantly mediates the interaction between childhood maltreatment and IDO2 on childhood depression. These interactions also moderate antidepressant treatment efficacy at 4-6 weeks.

Fig. 1. Study frame.

MDD = major depressive disorder, HC = healthy control, G-by-E or G × E = gene(G)-by-environment(E), ABCD = the Adolescent Brain Cognitive Development Study, ROI = region of interest.

Fig. 2. Genome-wide by environment interaction results of cumulative childhood maltreatment experiences in UK Biobank.

We conducted a genome-wide by environment interaction analysis with cumulative childhood maltreatment (CM) experiences to search for genetic variants associations with major depressive disorder (MDD) using the SaddlePoint Approximation implementation of Gene-by-Environment (G-by-E) analysis (SPAGE) approach in UK Biobank (N_cases = 26,664, N_controls = 85,715). a Shown is G-by-E interaction P values versus chromosomal position. We identified one locus at genome-wide significance (two-tailed genome-wide corrected threshold at P < 5×10 − 8; red dashed horizontal line). The arrow indicates the index variant IDO2 rs7846217. The orange dashed line indicates genome-wide suggestive significant threshold P < 1e-05. Source data are provided through Figshare (10.6084/m9.figshare.27124452). b Shown is odds ratios (ORs) of different types of CM on risk for MDD using logistic regression, stratified by IDO2 rs7846217 genotypes (three genotypes: CC, CT, and TT). The center of each dot represents the OR, with error bars indicating the 95% confidence intervals (CIs). For CM (yes/no, indicating binary CM experience), the ORs reflect the risk for MDD in individuals exposed to CM refer to those unexposed. For cumulative CM (Cum-CM), physical abuse (PhyAbu), emotional abuse (EmoAbu), sexual abuse (SexAbu), physical neglect (PhyNeg), and emotional neglect (EmoNeg), the ORs represent the odds of MDD risk per 1 standard deviation (SD) increase in the respective CM severity scale. Two-tailed P values for the G-by-E interaction effects are shown below the x-axis. Source data are provided as a Source Data file. c Shown is odds ratios (ORs) of IDO2 rs7846217 TT genotype refer to TC/CC genotype on risk for MDD, stratified by CM exposure. The center of each dot represents the OR, with error bars indicating the 95% CIs. Two-tailed P values are shown below the x-axis.

Fig. 3. Colocalization of genomic signal for depression with methylation levels in individuals exposed to childhood maltreatment.

a Shown is colocalization results of blood/brain methylation quantitative trait loci (meQTL) effect and major depressive disorder (MDD) genomic summary data obtained from individuals exposed to childhood maltreatment (N_cases = 17,225, N_controls = 45,372). Blue circle indicates meQTL effects, and red triangle indicates genetic effects on MDD; three different blue/red colors represent 3 colocalized variants rs11777027, rs2340953 and rs28631334, respectively. The center of each dot represents the beta, with error bars indicating the 95% confidence intervals (CIs). Source data are provided as a Source Data file. b Shown is scatter plot for the colocalization pair of blood meQTL cg25981315 and MDD genetic signal within CM exposure group rs2340953, with the highest colocalization posterior probability of 16.8% from eCAVIAR. c Shown is locus zoom around the index variant rs7846217 generated from LocusZoom.org. Three colocalized variants rs2340953, rs11777027 and rs28631334 are labeled, showing linkage disequilibrium (LD) with rs7846217. The LD r² are 0.94, 0.96 and 0.76, respectively, with 1000 Genome as reference. d Shown is protein-protein interaction (PPI) network for IDO2 quired from STRING. The tryptophan-kynurenine catabolic process is the top rank pathway in PPI and are labeled using red dots.

Fig. 4. Brian structures affected by interaction effects between IDO2 and childhood maltreatment.

a–b Shown are brain structural magnetic resonance imaging (sMRI) indicators affected by gene-by environment (G-by-E) interaction of IDO2 rs7846217 and cumulative childhood maltreatment (CM) experience in adults using UK Biobank (a, N = 20,901) and in children using ABCD cohort (b, N = 5,335). Each dot represents an individual brain region, and different colors represent cortical area (CA), cortical thickness (CT), cortical volume (CV), sulcal depth (SD), respectively. Linear regression models with adjustments made for age, sex and top ten genomic principal components (PCs) were performed, and the -logP for G-by-E are illustrated here. The red dashed line indicates the Bonferroni-corrected P threshold for morphometric measures with P = 3.38e-04 (calculated by 0.05/148) for (a), and P = 3.31e-04 (calculated by 0.05/151) for (b). The sMRI indicators passed threshold are labeled. GPIS-LCT = mean thickness of G-pariet-inf-Supramar in the left hemisphere, GPIA-RCV = cortical volume of G-pariet-inf-Angular for right hemisphere. Source data are provided as a Source Data file. c Shown is a diagram representing cross-validation. The significant indicator identified in adults was validated in children, but not vice versa. d–f Shown are effects of cumulative CM experience on GPIS-LCT in adults (d, N = 20,901) and children (e, N = 5,335), and on GPIA-RCV in children (f, N = 5,335), stratified by rs7846217 genotypes. Linear regression models with adjustments made for age, sex and top ten PCs were performed. The center of each dot represents the beta coefficient, with error bars indicating the 95% confidence intervals (CIs). P values across subgroups of rs7846217 genotypes are labeled, and two-tailed P < 0.05 was considered significant. Source data are provided as a Source Data file.

Fig. 5. Interaction effects of IDO2 and childhood maltreatment on children’s depression and antidepressant efficacy.

Source data are provided as a Source Data file. a Shown is odds ratios (ORs) of childhood maltreatment (CM) experience (yes vs. no) on risk for new onset of child-report major depressive disorder (MDD) at 2-year follow-up using logistic regression, stratified by IDO2 genotypes. Children not diagnosed as MDD at baseline were included into analysis (N = 4,855). The center of each dot represents the OR, with error bars indicating the 95% confidence intervals (CIs). Two-tailed P < 0.05 for the gene-by-environment (G-by-E) interaction is considered as significant. b–c Shown are betas of CM on 2-year follow-up brain structural magnetic resonance imaging (sMRI) partitioned by rs28631334 and rs2340953. Linear regressions with G-by-E for sMRI at 2-year follow-up were performed. sMRI indicators used standardized values. Five sMRI indicators passed multi-correction threshold (P < 3.31e-04) are shown. The center of each dot represents the beta, with error bars indicating the 95% CIs. P-LCA: cortical area for left hemisphere cortical precuneus, OP-LCV: cortical volume for left hemisphere cortical occipital pole, MOTSLS-RSD: sulcal depth for right hemisphere cortical medial occipito-temporal sulcus and lingual sulcus, MACGS-LCT: cortical thickness for left hemisphere cortical middle-anterior part of the cingulate gyrus and sulcus, CS-RSD: sulcal depth for right hemisphere cortical central sulcus. d CM increases risk for MDD at 2-year follow-up by affecting MACGS-LCT, this path was moderated by rs28631334 genotype. e Shown is box-plot of Hamilton’s Depression Scale (HAMD) change rate at midterm-treatment refer to baseline (N = 142) across groups by CM experience and rs2340953 (TCCC/CM: n = 5, TCCC/noCM: n = 20, TT/CM: n = 30, TT/noCM: n = 87; t = 1.99, P = 0.048).The center line represents the median in each group, the black diamond represents the mean in each group, whiskers extend from the box to the minimum and maximum data points within 1.5 times the interquartile range (IQR), box limits indicate the 25^th and 75^th quartiles of the data, with the box representing the IQR.