Introduction

Depression is a complex and pervasive mental health disorder that affects millions of people worldwide1. It is characterized by a range of behavioral symptoms, including persistent feelings of sadness, hopelessness, worthlessness, and a lack of interest or pleasure in daily activities. These symptoms are often accompanied by concentration difficulties, changes in appetite, sleep disturbances—either insomnia or hypersomnia—and suicide attempts2,3,4. Additionally, cognitive impairments, which occur in 85–94% of clinical cases5, affect various domains such as attention, executive functions, memory, and processing speed6,7. The diverse nature of depression extends beyond these aspects, as researchers have explored its connection with neurobiology. Subcortical brain structures, such as the amygdala and hippocampus, play a crucial role in regulating emotions, memory, and cognitive functions, establishing their significance in the neurobiology of depression.

Neuroimaging studies have progressively revealed volumetric changes within subcortical structures in depression. For instance, the amygdala, pivotal in emotional processing, often exhibits alterations in volume under depressive conditions8,9,10. These changes may contribute to increased emotional reactivity and the heightened emotional experiences commonly observed in individuals with depression11. Similarly, the hippocampus, crucial for memory formation and stress regulation, frequently shows a reduction in volume among those with depression12,13,14. This hippocampal atrophy is associated with cognitive deficits and difficulties in stress response regulation in individuals experiencing depression15. The observed alterations in the nucleus accumbens, a key region involved in reward processing, suggest a broader impact of depression on neural circuits related to emotional and reward-related processes16. Additionally, other subcortical regions, such as the thalamus, caudate, and putamen, also exhibit structural changes in depression17,18,19. These structural alterations highlight the influence of depression on the physical dimensions of critical brain regions, providing insight into the neurobiological underpinnings of the disorder.

Recent advances in genetic research, particularly through genome-wide association study (GWAS), have provided invaluable insights into the genetic factors underlying depression and subcortical volumetric traits. In a meta-GWAS involving over 1.3 million individuals, 243 independent risk loci for depression were identified20. Concurrently, ongoing investigations into the genetic architecture of subcortical volumes have contributed significantly to our understanding of the complex interactions between genetic factors and these traits21,22,23. Given the high heritability and polygenic nature observed in both depression24 and subcortical volumetric traits21, it is likely that the neuroimaging findings could be attributed to a shared genetic basis influencing these conditions. Several studies exploring this genetic overlap, using approaches such as linkage disequilibrium score regression (LDSC) and polygenic risk score (PRS), have uncovered weak or absent genome-wide genetic correlations25,26,27. However, these approaches might underestimate genetic overlap, particularly in scenarios involving a balanced mixture of genetic variants with both concordant and discordant effects on the phenotypes. Recently, Frei et al. introduced the bivariate causal mixture model (MiXeR)28, a method that evaluates genetic overlap by considering all variants, regardless of the direction of effect. Furthermore, conditional and conjunctional false discovery rate (cond/conjFDR) analyses, which utilize an empirical Bayesian statistical framework, enhance the power of both GWASs in discovering novel risk loci and identifying shared genetic loci for the two traits29,30,31. These innovative approaches have been successfully applied in various studies, illuminating the genetic relationships between complex traits32,33.

In this work, we aim to identify the shared genetic basis between depression and subcortical volumes. Specifically, the MiXeR method is employed to investigate the shared polygenic architecture. Following this, the cond/conjFDR approaches are employed to identify both novel loci for each trait and the genetic loci shared between these traits. Furthermore, functional annotation and gene mapping of the shared loci are conducted. The expression patterns of these shared genes are then examined across various developmental stages, and their associations with depression and subcortical volumes, as well as their potential impacts on cognitive abilities and behavioral symptoms, are explored. By integrating advanced statistical models with recent large-cohort GWASs, the power to discover shared genomic loci in depression and subcortical brain structures is increased, thereby better revealing the shared genetic basis between depression and subcortical volumes. A systematic flow of the study design is shown in Supplementary Fig. 1.

Results

Quantification of genetic overlap with MiXeR

Using univariate MiXeR34, we estimated the heritability, polygenicity, and discoverability for depression and each subcortical volumetric phenotype. All 14 subcortical phenotypes showed a good model fit, as indicated by positive Akaike information criterion (AIC) values. As shown in Fig. 1a and Supplementary Data 1, univariate MiXeR analysis revealed that depression exhibited lower single nucleotide polymorphism (SNP) heritability (0.051), higher polygenicity (9400 trait-influencing variants), and lower discoverability (8.3E−6) compared with all subcortical volumetric phenotypes. In contrast, these volumetric phenotypes demonstrated SNP heritability ranging from 0.181 to 0.295, polygenicity ranging from 939 to 3207 trait-influencing variants, and discoverability ranging from 1.2E−4 to 3.5E−4. Additionally, due to the lower discoverability, the estimated sample size required to reach 90% heritability was more than ten times larger for depression than for subcortical volumetric phenotypes (65.8 million vs. 2.5–6.6 million, Supplementary Fig. 2, Supplementary Data 1).

Fig. 1: Genetic overlap between depression and subcortical volumes using MiXeR.
figure 1

a Heritability, polygenicity, and discoverability for each phenotype estimated by univariate MiXeR. b Venn diagrams showing MiXeR results of the estimated number of variants shared between depression and subcortical volumes. The numbers of shared and trait-specific variants are shown in thousands with standard errors. The size of the circles reflects the degree of polygenicity. c The points indicate the proportions of shared variants with concordant effect directions. Source data are provided as a Source Data file. L left, R right.

Full size image

Applying bivariate MiXeR28, we found considerable quantities of shared variants in each pair of depression-volumetric phenotypes. Of the 14 subcortical structures, eight demonstrated a good model fit for MiXeR analyses with depression (both AICbest vs. min and AICbest vs. max were positive; Supplementary Data 2). As shown in Fig. 1b and Supplementary Data 2, subcortical volumetric traits shared 23.4% to 83.3% of their trait-influencing variants with depression. Conversely, depression shared 4.1% to 17.9% of its trait-influencing variants with subcortical structures. The Dice coefficient, which is the ratio of shared variants to the total number of variants, ranged from 7.4% to 28.5%, indicating varying degrees of genetic overlap between depression and the various subcortical traits. Additionally, the allelic effects of the variants shared between depression and subcortical volumes exhibited divergent directions, with the proportion of shared variants with concordant effects ranging from 19.7% to 73.6% (Fig. 1c, Supplementary Data 2). Within the shared component, the bilateral caudate, bilateral putamen, and right pallidum demonstrated positive correlations with depression, with the largest coefficient observed in the left caudate (rho = 0.634). In contrast, other subcortical structures displayed negative correlations with depression, with the largest coefficient observed in the left hippocampus (rho = -0.784), as shown in Supplementary Data 2. Details of the MiXeR results are provided in Supplementary Figs. 3–16.

Identification of novel and shared genetic loci (cond/conjFDR)

To improve the discovery of genetic variants associated with depression and subcortical volumes, we applied condFDR analysis29,30,31 to examine the associations of depression with subcortical volumes, and vice versa. By conditioning on subcortical volumetric phenotypes, we identified 94 distinct loci associated with depression at condFDR <0.01, including 7 novel loci (Supplementary Data 3, Supplementary Fig. 17). The candidate SNPs in these novel loci were mapped to 14 genes (Supplementary Data 4). Enrichment analysis was conducted on these genes to explore potential biological pathways, however, due to the small number of genes, no significant results were obtained. Additionally, when conditioning on depression, a total of 410 loci associated with the volume of the thalamus (left N = 22, right N = 28), caudate (left N = 43, right N = 41), putamen (left N = 44, right N = 43), pallidum (left N = 20, right N = 35), hippocampus (left N = 25, right N = 33), amygdala (left N = 19, right N = 12), and accumbens (left N = 25, right N = 20) were identified (Supplementary Data 5, Supplementary Fig. 18). Combining loci across all subcortical volumetric traits resulted in a total of 233 distinct loci, among which 13 were newly identified (Supplementary Data 6). The candidate SNPs in these novel loci were mapped to 57 genes (Supplementary Data 7), and subsequent enrichment analysis revealed associations with several gene ontology (GO) biological process terms, including cerebral cortex GABAergic interneuron development (FDR q = 4.0E−2) and pyrimidine nucleoside biosynthetic process (FDR q = 4.0E−2).

The conditional quantile‒quantile (Q‒Q) plots demonstrated substantial genetic overlap for each depression-volume pair (Supplementary Figs. 19–32). Thus, conjFDR analysis was subsequently performed to identify genomic loci jointly associated with both traits. As shown in Fig. 2a, b and Supplementary Data 8, conjFDR unveiled 77 loci shared between depression and subcortical volumetric phenotypes: the thalamus (left N = 9, right N = 13), caudate (left N = 8, right N = 4), putamen (left N = 3, right N = 6), pallidum (left N = 2, right N = 6), hippocampus (left N = 3, right N = 9), amygdala (left N = 4, right N = 5), and accumbens (left N = 2, right N = 3). Among them, the most significant locus was chr18: 50358109-51055069 for depression and bilateral putamen volume (lead SNP: rs4632195, left: conjFDR = 1.0E−8; right: conjFDR = 4.3E−9; Fig. 2c). Of the 77 shared loci, a total of 51 genomic loci were jointly associated with depression and at least two subcortical volumetric phenotypes, resulting in 44 distinct shared loci across all depression-volume pairs. Within these shared loci, 7 were novel for depression, 17 were novel for subcortical volumetric phenotypes, and 2 were novel for both (Supplementary Data 9).

Fig. 2: Results of loci shared between depression and subcortical volumes.
figure 2

a Manhattan plot showing the associations between depression and subcortical volumes identified by conjFDR. Each point represents a SNP and the point with a larger size and black border indicates a lead SNP. The y-axis represents the -log10-transformed conjFDR values for each SNP, and the x-axis reflects the chromosomal position. The dashed line represents the threshold of conjFDR = 0.05, and the color of the dots represents different phenotypes. b Brain map showing the number of loci of each subcortical structure shared with depression. A color transition from white to red indicates an increasing number of shared loci. c LocusZoom plots of a genomic locus (chr18: 50358109-51055069, lead SNP: rs4632195) showing the regional SNP associations (two-sided -log10(p)) for depression and bilateral putamen volumes from the original GWAS summary data d. MRI slice plot of bilateral putamen volumes. Source data are provided as a Source Data file. Chr chromosome, L left, R right.

Full size image

The primary locations of the candidate SNPs in the distinct shared loci were within intronic regions (58.7%) and non-coding RNA intronic regions (25.4%), as illustrated in Fig. 3a and detailed in Supplementary Data 10. Additionally, 18 SNPs with non-synonymous mutations in exons were identified. Approximately 3.7% of the candidate SNPs were indicated as pathogenic, with combined annotation-dependent depletion (CADD) scores exceeding 12.37. About 10.3% of the candidate SNPs exhibited a higher likelihood of regulatory functionality, as indicated by a RegulomeDB level of less than 3. Moreover, 90.1% of the candidate SNPs were located in regions with a predominantly open chromatin configuration, characterized by a chromatin state score of less than 8. Among the 44 lead SNPs, five (rs11047209, rs1976052, rs2246556, rs4593875, rs62618693) had CADD larger than 12.37, with rs62618693, a non-synonymous mutation in QSER1, having the highest CADD value of 31 (Supplementary Data 9). Furthermore, the candidate SNPs in 44 distinct shared loci were mapped to 269 protein-coding genes (Supplementary Data 11), and enrichment analysis revealed 42 GO biological process terms (FDR q < 0.05), mainly including the regulation of neuron projection development (FDR q = 3.4E−3), regulation of axonogenesis (FDR q = 1.1E−2), and central nervous system development (FDR q = 2.8E−2); see Supplementary Data 12 and Fig. 3b for details.

Fig. 3: Results of functional annotation and gene expression analyses for the shared loci.
figure 3

a Distribution of functional consequences of candidate SNPs in the genomic loci shared between depression and 14 subcortical volumes. b The bubble plot shows the enriched GO terms of biological processes for protein-coding genes. The x-axis indicates -log10-transformed values of FDR-adjusted two-sided p-values, and the y-axis shows each GO term. The color of the dots indicates the magnitude of the -log10(q) values, and the size of the dots represents the number of genes enriched in a given term. c Developmental trajectories of average gene expression in whole brain postmortem-bulk RNA data (PsychEncode) for all genes mapped by shared candidate SNPs (conjFDR < 0.05) between depression and seven subcortical volumetric phenotypes (left and right subcortical phenotypes are combined through averaging). The shaded region indicates 95% confidence intervals, and the line indicates the usual timing of birth. The x-axis represents the nine developmental windows, and the y-axis represents the log2-transformed mean expression value. d The plot shows all significant expression-trait associations (two-sided q < 0.05, FDR corrected) identified by S-MultiXcan for both depression and at least one subcortical volume. The x-axis provides the names of 55 genes and the y-axis lists the names of depression and 14 subcortical structures. The magnitudes of the -log10-transformed q values are coded as different colors from white to red. Source data are provided as a Source Data file. L left, ncRNA non-coding RNA, PCW postconceptional week, PY postnatal year, R right.

Full size image

To quantify the differences and similarities in how shared lead SNPs identified by conjFDR influence different subcortical volumetric phenotypes, we utilized Cochran’s Q test to detect heterogeneity in 44 shared lead SNPs across 14 subcortical volumetric phenotypes, and for each volumetric phenotype pair. We found that 30 shared lead SNPs showed heterogeneity across all subcortical phenotypes (p < 0.05), whereas 14 lead SNPs showed no evidence of effect heterogeneity (Supplementary Data 13). Additionally, the heterogeneity for each pair of subcortical phenotypes is shown in Supplementary Data 14 and Supplementary Fig. 33. Specifically, we observed no heterogeneity in shared lead SNPs between bilateral subcortical structures, except for the bilateral accumbens, which exhibited heterogeneity in one locus. Among the other subcortical phenotype pairs, the pair with the highest number of heterogeneous SNPs was the left caudate and the right hippocampus, with 20 out of 44 SNPs showing heterogeneity. In contrast, the pair with the lowest number of heterogeneous SNPs was the left hippocampus and the left amygdala, with only 2 out of 44 SNPs showing heterogeneity.

Developmental trajectories of shared genes

Using spatiotemporal brain gene expression data from PsychENCODE35, we investigated the developmental expression trajectories of genes shared between depression and each subcortical structure. As shown in Fig. 3c, genes shared between depression and the accumbens, pallidum and thalamus showed similar trajectories with peak expression during the mid-gestation period (windows 3-4, post-conceptional weeks [PCW] 16-22), followed by a slight decline in expression, and then entering a plateau phase after window 6 (postnatal years [PY] 0.5–2.6). The expression of genes in the hippocampus, amygdala, and caudate displayed a peak expression level during the early prenatal period (window 1, PCW 8-9), followed by a continuous decline until reaching the lowest point in PY 0.5–2.6. Regarding the putamen, its expression peaked at PCW 8-9 and then decreased steadily.

Expression-trait associations

For each depression-volume pair, we investigated whether the expression of shared genes in the brain was associated with both depression and the corresponding subcortical volumes using the S-MultiXcan tool36. Out of these shared genes, we found 137 significant associations between gene expression (involving 71 genes) and volumetric phenotypes. Among these 71 genes, 55 genes were also found to be associated with depression (Supplementary Data 15, 16, Fig. 3d). For example, the expression levels of LRRC37A2, SPPL2C, ARL17A, FMNL1, PLEKHM1, ARHGAP27, and CRHR1 in the brain were associated with the volumes of the bilateral thalamus, bilateral pallidum, left accumbens, and right hippocampus, as well as with depression. Moreover, the expression of NPEPPS, GINM1, PRMT6, EFCAB13, NUP43, LATS1, ULBP2, ACYP1, and AREL1 showed correlations with the bilateral thalamic volume and depression. The MSH6 was correlated with the bilateral hippocampal volume and depression. The most significant expression-depression association was found for CELSR3, which was also linked to the volume of the right pallidum.

Cognitive and behavioral associations

To investigate whether the genes shared between depression and subcortical volumetric phenotypes exerted further influence on cognitive abilities or behavioral symptoms associated with depression, GWAS summary statistics for 8 cognitive abilities and 15 behavioral symptoms were acquired (Supplementary Data 17, 18). The associations between the expression of the 55 shared genes (i.e., those associated with depression and at least one subcortical volume) and cognitive abilities, as well as behavioral symptoms, were also analyzed using S-MultiXcan. Out of these 55 shared genes, we found 30 significant associations between gene expression and behavioral symptoms, as well as 59 significant associations with cognition, involving a total of 30 genes (Fig. 4a, Supplementary Data 19). Notably, the expression of five genes in the brain, including ARL17A, CRHR1, GPX1, SPPL2C, and LRRC37A2, had significant associations with more than five cognitive or behavioral traits, indicating their widespread influence on cognition and behavior. The most significant expression-cognition association was found for LRRC37A2 with reaction time (RT), which was also correlated with depression and six subcortical structures.

Fig. 4: Associations of the expression of shared genes in the brain with cognitive abilities and behavioral symptoms.
figure 4

a Bubble plot showing significant expression-trait associations (two-sided q < 0.05, FDR corrected). The x-axis shows the shared genes, and the y-axis shows cognitive abilities and behavioral symptoms. The size and color of the bubbles demonstrate the significance level. The gray bubbles represent non-significant associations. b The bar plots show the significance level (uncorrected two-sided p-values) of the expression of LRRC37A2 in five subcortical structures with the VNR and RT. c The bar plots show the significance level (uncorrected two-sided p-values) of PLEKHM1 expression in five subcortical structures associated with feelings of worthlessness during the worst period of depression and VNR. Source data are provided as a Source Data file. ConcWorst difficulty concentrating during worst depression, DSST digit symbol substitution task, Inadequacy recent feelings of inadequacy, Interest recent lack of interest or pleasure in doing things, Movement recent changes in speed/amount of moving or speaking, PM pairs matching, RT reaction time, SleepIssues trouble falling or staying asleep, or sleeping too much, SleepMuch sleeping too much, SleepTrouble trouble falling asleep, TiredRecent recent feelings of tiredness or low energy, TiredWorst feelings of tiredness during the worst episode of depression, TMT-B trail making test B, VNR verbal–numerical reasoning, WorthlessWorst feelings of worthlessness during worst period of depression.

Full size image

Additionally, we observed that the gene expression levels of the same gene across different subcortical regions were associated with distinct cognitive abilities or behavioral symptoms (Fig. 4b, c). For instance, the expression of LRRC37A2 in the amygdala exhibited the strongest association with verbal–numerical reasoning (VNR), whereas its expression in the accumbens did not correlate with VNR. Conversely, the expression of LRRC37A2 in the accumbens was predominantly associated with RT (Fig. 4b). Furthermore, the expression of PLEKHM1 in the hippocampus was most significantly associated with feelings of worthlessness during the worst period of depression, while its expression in the caudate was primarily associated with VNR (Fig. 4c).

Control analyses

Despite previous studies finding weak or no genome-wide genetic correlation between depression and subcortical volumes25,26, we employed LDSC analyses37,38 to investigate the correlation using updated GWAS summary statistics, in order to confirm or refute the relationship. We found that the absolute genetic correlations ranged from 0.002 for left pallidum volume to 0.093 for right caudate volume, with nominal statistical significance only observed for bilateral caudate volume (left p = 0.015, right p = 0.004; see details in Supplementary Data 20).

To validate the shared loci identified by conjFDR analysis, further analysis was undertaken using local analysis of [co]variant association (LAVA)39. In general, LAVA analyses supported the findings of the conjFDR analysis. Among the 77 shared loci, LAVA identified 56 genetic loci showing nominally significant local heritability in both depression and volumetric phenotypes. Within these 56 loci, 33 exhibited a significant local genetic correlation (FDR q < 0.05), with 20 displaying a positive correlation (ranging from 0.63 to 1.00) and 13 showing a negative correlation (ranging from -0.54 to -1.00), as shown in Supplementary Data 8. The most significant positive local genetic correlation was observed at the same locus (chr18: 50358109-51055069, lead SNP: rs4632195) between depression and the left putamen (r = 0.64, p = 6.0E−4, conjFDR = 1.0E−8), as well as the right putamen (r = 0.63, p = 4.5E−4, conjFDR = 4.3E−9). The most significant negative local genetic correlation was observed between depression and right thalamic volume at chr4: 3247007-3285389 (lead SNP: rs3135170, r = −1.00, p = 1.8E-3, conjFDR = 0.044).

Discussion

Here, we explored the shared genetic architecture between depression and subcortical volumetric phenotypes. Bivariate MiXeR analysis revealed that subcortical volumetric traits shared a substantial proportion of trait-influencing variants with depression, with the left amygdala showing the highest Dice coefficient of 28.5%. The identification of shared loci, encompassing 44 distinct loci associated with both depression (including 7 novel loci) and subcortical volumes (including 17 novel loci), through conjFDR analysis enriches our understanding of their intricate genetic connections. Functional annotation, gene mapping, and gene expression analyses provide vital information about the molecular mechanisms behind the identified shared loci. Unraveling these mechanisms will deepen our comprehension of how genetic factors contribute to the co-occurrence of depression and alterations in subcortical brain volumes.

In previous studies utilizing LDSC25,26, it was demonstrated that there are weak or absent genome-wide genetic correlations between depression and subcortical brain volumes, even when applying a lenient statistical threshold (i.e., uncorrected p < 0.05). For example, one study found no significant genetic correlation between depression and subcortical volumes25, and another study found that depression was weakly positively correlated with hippocampal volume26. In the present study, we validated these findings using updated GWAS summary statistics, revealing nominal significance exclusively in the genetic correlation between depression and bilateral caudate volume (Supplementary Data 20). Notably, genome-wide genetic correlations can only capture consistent allelic effect directions among shared variants and do not consider mixed patterns of directional effects. In contrast, MiXeR analysis indicated that all subcortical volumes shared a significant proportion of variants (ranging from 23.4% to 83.3%) with depression. These shared variants exhibit divergent directions, as evidenced by the proportion of shared variants with concordant effects ranging from 19.7% to 73.6%. This balanced mixture of directional effects aligns with the nonsignificant genome-wide genetic correlation between depression and subcortical volumetric phenotypes, suggesting the potential involvement of both agonistic and antagonistic overlapping molecular genetic mechanisms in its pathophysiology.

The condFDR analysis, focused on uncovering novel genetic variants associated with depression and subcortical volumetric phenotypes, yielded compelling results. Under the stringent condFDR threshold (< 0.01), conditioning on subcortical volumetric phenotypes revealed 94 distinct loci linked to depression, including 7 novel loci. Conversely, conditioning on depression identified 233 distinct loci associated with subcortical volumetric phenotypes, including 13 novel loci. Among the lead SNPs of these 13 novel loci, rs636926 is a nonsynonymous exonic variant within TMEM171 (Supplementary Data 6). This variant results in a change in the corresponding amino acid, potentially directly impacting the structure or function of the protein. Subsequent enrichment analysis linked 57 mapped genes from these 13 novel loci predominantly to the development of GABAergic interneurons in the cerebral cortex and pyrimidine nucleoside biosynthetic process. GABAergic interneurons are inhibitory neurons in the nervous system that play a vital role in the functional maturation of the cortex and hippocampal neural circuitry and activity40,41. Several studies have indicated that alterations in GABAergic interneurons contribute to clinical features in neuropsychiatric disorders such as autism spectrum disorder42 and schizophrenia43. Dysregulation of pyrimidine nucleoside biosynthetic process may influence neurogenesis, synaptic plasticity, and neurotransmitter signaling within subcortical structures, potentially leading to changes in the brain’s structural architecture. By exploring the biological functions of these genes, biological processes or pathways related to the functioning of subcortical brain regions can be revealed, providing valuable insights for further investigation.

The Q‒Q plots demonstrate a significant genetic overlap for each pair of depression and subcortical volume, as illustrated in Supplementary Figs. 19–32. Additionally, the conjFDR analysis identified 77 loci shared between depression and subcortical volumetric phenotypes. Within the shared loci, there are both concordant effects (42 out of 77 SNPs) and discordant effects (35 out of 77 SNPs) in the top lead SNPs (Supplementary Data 8), providing additional insights into the lack of a genome-wide genetic correlation revealed by the LDSC25. Notably, there were 44 distinct shared loci across all depression-volume pairs, indicating unique associations that contribute to the shared genetic architecture between depression and subcortical brain structures. The most significant shared locus, chr18: 50358109-51055069 (lead SNP: rs4632195), between depression and bilateral putamen volumes (left: conjFDR = 1.0E−8; right: conjFDR = 4.3E−9; Supplementary Data 8), showed positive local genetic correlations (left: rho = 0.644, right: rho = 0.627). This locus is located within the intronic region of the DCC gene, which encodes a netrin receptor involved in axon guidance and migration. Previous studies have linked SNPs within DCC to putamen volume44 and suggested correlations with depression45,46 and schizophrenia47. Moreover, we observed that rs62618693 (CADD = 31), a non-synonymous mutation within the exons of the QSER1, was jointly associated with bilateral amygdala volume and depression, and the locus (chr11: 32623621-32956492) with rs62618693 as the lead SNP was a novel locus for subcortical volume.

Among the 44 distinct shared loci, two are novel for both depression and subcortical volumes. One novel locus (chr2: 48024876-48024876, lead SNP: rs3136329, conjFDR = 3.5E−2, Supplementary Data 9) is located in FBXO11, which is ubiquitously expressed in the brain (reads per kilobase of transcript per million mapped reads [RPKM] 21.6) and has been found to be associated with schizophrenia48. Another novel locus (chr10: 98011878-98162702, lead SNP: rs11188695, conjFDR = 1.6E−2, Supplementary Data 9) can be mapped to TLL2 by expression quantitative trait loci (eQTL) mapping. TLL2 encodes an astacin-like zinc-dependent metalloprotease and is associated with stress sensitivity49. Additionally, the lead SNPs in other shared loci implicate genes like CDK14, NRD1, DYRK1A, KLF11, ARHGEF10, TWF1P1, and LRRC71 in the genetic relationship between depression and subcortical volumetric phenotypes (Supplementary Data 11). Furthermore, the candidate SNPs in the loci shared between depression and subcortical volumetric traits were mapped to 269 protein-coding genes. Enrichment analysis of these genes elucidated their involvement in essential biological processes, such as the regulation of neuron projection development, regulation of axonogenesis, and central nervous system development. These findings provide valuable insights into potential molecular mechanisms and pathways underlying the connection between depression and subcortical brain structures.

By employing S-MultiXcan to explore expression-trait associations, our study revealed 55 shared genes that not only exhibited correlations with volumetric phenotypes but also demonstrated associations with depression. For example, the expression of LRRC37A2, SPPL2C, ARL17A, FMNL1, PLEKHM1, ARHGAP27, and CRHR1 in the brain is linked to the bilateral thalamus, bilateral pallidum, left accumbens, right hippocampus and depression, suggesting shared genetic underpinnings between subcortical brain characteristics and depression. Moreover, we found 89 significant associations between the expression of shared genes and cognitive abilities as well as behavioral symptoms, involving a total of 30 genes. Among these, five genes—ARL17A, CRHR1, GPX1, SPPL2C, and LRRC37A2—had significant associations with more than five cognitive or behavioral traits. Our study also revealed diverse implications for these genes. For instance, PLEKHM1 variants are associated with neuroticism50 and global cortical volume51. LRRC37A2 and ARHGAP27 have been linked to Parkinson’s disease52,53. Additionally, CRHR1 encodes a G protein-coupled receptor implicated in the regulation of the hypothalamic-pituitary-adrenal axis, which is associated with stress-related psychopathology54,55,56. CRHR1 has been identified as a genetic risk factor for affective disorders and is linked to metabolic activity in the anterior hippocampus and amygdala57. ARL17A is involved in intracellular protein and vesicle-mediated transport and is a risk gene for progressive supranuclear palsy58. GPX1, a critical component of the antioxidant system involved in the major antioxidant molecule glutathione, may affect cognition and emotional behavior through mechanisms involving oxidative stress-induced neuronal damage, synaptic dysfunction, and neurotransmitter dysregulation, and is also related to the risk of depression59,60. These complex relationships provide a comprehensive understanding of the interactions among gene expression, subcortical volumetric phenotypes, depression, and cognitive abilities as well as behavioral symptoms, offering valuable perspectives into potential contributing mechanisms.

Our findings should be considered for the following limitations. First, the analysis was restricted to individuals of European ancestry due to the absence of adequately powered GWAS on other ancestries. Future research in large non-European populations is crucial to enhance the external validity and generalizability of our results. Second, our results are confined to common variants on autosomal chromosomes, while rare variants also play an important role in depression61,62. Therefore, future studies on rare variants are warranted. Third, the absence of subtype-specific data restricts our ability to thoroughly investigate the shared genetic mechanisms underlying various depression subtypes and subcortical brain structures, and our findings may not fully elucidate the associations between shared genes and the diverse clinical symptoms observed across different depression subtypes. This emphasizes the need for future research efforts to incorporate subtype-specific data to enhance our understanding of this condition. Fourth, both the MiXeR and condFDR/conjFDR approaches are limited to bivariate analyses, preventing the determination of whether the identified genetic overlap is influenced by additional factors. Finally, the observation of negative AIC values in the bivariate MiXeR analysis indicates that these results should be interpreted with caution. Future studies with larger sample sizes and more comprehensive data may be necessary to validate our findings.

In conclusion, our study employed innovative methods, including MiXeR and cond/conjFDR analyses, to explore the genetic overlap between depression and subcortical brain volumes. By unraveling the shared genetic architecture, we identified several loci associated with both depression and specific volumetric traits, and uncovered novel loci that provide insights into the molecular mechanisms underlying this complex relationship. The functional annotation highlighted pathways related to neuron projection development, regulation of axonogenesis, and central nervous system development. These findings enhance our understanding of the shared genetic basis of depression and subcortical brain structures, contributing valuable knowledge to the intersection of psychiatric genetics and neurobiology.

Methods

GWAS data

The GWAS summary statistics for depression were derived from a recent meta-analysis involving 371,184 individuals diagnosed with depression and 978,703 controls20. Participants from the UK Biobank were excluded to avoid sample overlap, and samples from 23andMe were omitted due to access restrictions, resulting in a GWAS subsample consisting of 166,773 cases and 507,679 controls. Additionally, summary statistics for 14 subcortical volumetric phenotypes, encompassing bilateral volumes of the thalamus, caudate, putamen, pallidum, hippocampus, amygdala and accumbens, were acquired from the UK Biobank with a sample size of 33,224 individuals23. Notably, both the depression and volumetric trait datasets consisted of individuals of European ancestry. For further details regarding the GWAS summary statistics, please refer to the Supplementary Methods.

Univariate and bivariate MiXeR analyses

The MiXeR (v1.3, https://github.com/precimed/mixer) approach was used to assess the polygenic overlap between depression and subcortical volumetric traits28,34. Univariate MiXeR analyses were first conducted to estimate heritability, polygenicity, and discoverability for each phenotype. Heritability was calculated as the cumulative effect of all trait-influencing variants, which are common variants exhibiting a non-zero additive effect on the trait. Polygenicity was defined as the number of trait-influencing variants accounting for 90% of the heritability for each phenotype. Traits with higher polygenicity have a larger number of genetic variants affecting them than those with lower polygenicity. Discoverability, measured as the variance of additive genetic effects, indicated that traits with higher discoverability exhibit more detectable genetic signals. MiXeR’s power analysis was further conducted to determine the proportions of phenotypic variance explained by causal SNPs reaching genome-wide significance at current sample sizes and to estimate the sample sizes necessary to explain larger portions of SNP heritability. Traits that are less genetically discoverable require larger sample sizes to explain a given proportion of SNP heritability.

Bivariate MiXeR analyses28 were then performed to model the estimated additive genetic effects as a mixture of four Gaussian components. These components represent variants that do not influence either trait, variants affecting only one of the traits, and variants influencing both traits. MiXeR simultaneously estimated the proportion of shared trait-influencing variants with concordant effects (i.e., the same effect directions) on both traits. To evaluate the polygenic overlap, MiXeR calculated a Dice coefficient, which is the ratio of shared variants to the total number of variants, providing a measure of the similarity in trait-influencing variants between depression and volume traits. The results for MiXeR are presented in a Venn diagram illustrating unique and shared polygenic components across the traits.

The AIC was employed to assess model fitting performance. In the univariate MiXeR analysis, positive AIC values indicate that the GWAS summary statistics have sufficient power to justify the use of the MiXeR model compared to the LDSC model. In the bivariate MiXeR, a positive AIC value indicates adequate discrimination between the best model fit and the model assuming minimal or maximum possible polygenic overlap28. Notably, SNPs within the major histocompatibility complex (MHC) region (chr6: 25000000-35000000) were excluded from the MiXeR analyses to avoid potential bias from extreme complex LD in this region. For a detailed description of MiXeR, please see the Supplementary Methods.

Cond/conjFDR analyses

CondFDR and conjFDR analyses were employed by “pleiofdr” software (https://github.com/precimed/pleiofdr) to identify novel loci for each trait and specific shared loci for each depression-volume pair, respectively29,30,31. To enhance the discovery of genetic variants for each individual trait, the test statistics of the primary phenotype were re-ranked by the condFDR statistical framework based on the strength of their association with a secondary phenotype. Additionally, conditional Q‒Q plots were generated to visually depict cross-trait enrichment patterns. These plots present p-value distributions across all SNPs for the primary phenotype and within SNP strata determined by their association significance with the secondary trait (p < 0.1, p < 0.01, and p < 0.001). In these plots, enrichment is observed as an increased degree of leftward deflection from the expected line as the association significance increases in the secondary phenotype. To ensure the model fit, both the MHC and 8p23.1 (chr8: 7200000-12500000) regions were excluded from the conditional Q‒Q plot63,64. Subsequently, conjFDR analysis was performed to identify genomic loci jointly associated with both traits, with the conjFDR value defined as the maximum of the two mutual condFDR values. The conjFDR approach estimates a posterior probability, indicating the likelihood that a SNP is null for either one or both phenotypes, given that the p-values for both phenotypes are lower than the observed p-values. The significance thresholds were set at 0.01 for condFDR and 0.05 for conjFDR, as recommended in previous studies33,65,66,67. For more details, please see the Supplementary Methods.

Genomic locus definition

For the cond/conjFDR results, independent genomic loci were defined using the “sumstats.py” script (https://github.com/precimed/python_convert), which implements the same logic as in functional mapping and annotation (FUMA, https://fuma.ctglab.nl/)68. Specifically, independent significant SNPs were those that achieved a condFDR <0.01 or conjFDR <0.05 in their respective analyses and were also required to be independent from each other with an r2 < 0.6. Among these significant independent SNPs, lead SNPs were defined as those that were independent of each other with an r2 < 0.1. Candidate SNPs were defined as all SNPs with an r2 ≥ 0.6 with one of the independent significant SNPs, which determined the borders of each locus. If two loci were within 250 kb of each other, they were merged into a single genomic locus. All LD information was calculated from the 1000 Genomes European reference panel. Subsequently, independent loci across all depression-volume pairs were combined into distinct loci. In cases where the genetic positions of two or more loci from different depression-volume pairs overlapped, they were merged into one distinct locus with the union of their genomic boundaries. The lead SNP with the minimum FDR value was designated the final lead SNP in the merged locus.

To determine the novelty of independent loci identified by cond/conjFDR, several steps were taken. First, we verified whether any candidate SNPs within the identified loci had been reported to be significantly associated with depression or subcortical volumetric traits in the NHGRI-EBI GWAS Catalog69. Next, the identified loci were compared with significant loci (p < 1.0E−6) reported in the original GWAS20,23 and findings from previous cond/conjFDR studies33,70,71,72,73,74,75,76,77,78,79,80,81 or other GWASs44,73,82,83 (Supplementary Data 21, 22). A locus was considered novel if all candidate SNPs had not been previously reported in the NHGRI-EBI GWAS Catalog and if it extended beyond 500 kb from the borders of any reported loci in the original GWAS summary, previous cond/conjFDR studies, or other GWAS studies. Notably, if the start and end positions of the locus were not provided, the locus border was defined as 1000 kb away from the lead SNP.

Functional annotation

The FUMA online platform was used to perform functional annotation for candidate SNPs within specific genomic loci. These SNPs were annotated using CADD scores, RegulomeDB scores, and 15-core chromatin states. CADD scores, derived from over 60 genomic features, predict the likelihood of a variant being pathogenic, with scores above 12.37 suggesting a deleterious effect84. RegulomeDB scores assess the regulatory functionality of SNPs, inferred from the eQTL and chromatin markers—a lower score indicates a greater likelihood of regulatory functionality85. The 15-core chromatin state classification, based on ChromHMM predictions and utilizing multiple chromatin markers across 127 epigenomes, denotes various functional states of the genome, including promoters, enhancers, and transcriptional activity regions. Categories 1–7 indicate open chromatin states, which are associated with transcriptional regulation and gene expression86,87. Moreover, candidate SNPs were mapped to protein-coding genes using any of three strategies: (1) positional mapping within a 10-kb window, (2) eQTL mapping by associating cis-eQTL SNPs with genes whose expression varies with allelic variation at the SNP level (GTEx v8 brain), and (3) chromatin interaction mapping to establish connections through three-dimensional DNA‒DNA interactions between the genomic regions of SNPs and proximal or distal genes. Furthermore, an assessment of the enrichment of GO biological processes for all mapped genes was conducted using g:Profiler (https://biit.cs.ut.ee/gprofiler)88, and the Benjamini‒Hochberg FDR method was applied to correct for multiple comparisons, with a significance threshold of q < 0.05.

Heterogeneity of shared loci across subcortical volumetric phenotypes

To assess the differences and similarities in how shared genetic variants affect different subcortical volumetric phenotypes, the effect sizes and standard errors of 44 lead SNPs for 14 subcortical volumetric phenotypes were extracted from the original GWAS summary statistics. Cochran’s Q test, a statistical method designed to detect heterogeneity, was then applied to these statistical values. Specifically, two types of heterogeneity tests were performed: one assessing all 14 subcortical phenotypes together and the other conducting pairwise comparisons between each pair of the 14 subcortical phenotypes. A p-value of less than 0.05 from Cochran’s Q test was considered to indicate significant heterogeneity among the subcortical volumetric phenotypes for a given SNP.

Spatiotemporal gene expression trajectory analysis

Spatiotemporal brain expression trajectory analysis was performed to explore how the expression of genes shared between depression and subcortical volumetric phenotypes varies across different developmental stages, aiming to elucidate the potential roles these genes play in the pathophysiology of depression and associated structural changes in the brain. Processed mRNA-seq data from PsychENCODE35, comprising 607 tissue samples from 41 postmortem brains ranging from 8 PCW to 40 PY, were utilized to identify the expression patterns of these shared genes across lifespan windows. Following the methodology of the PsychENCODE study35, we divided the data into nine developmental windows: 8-9 PCW, 12-13 PCW, 16-17 PCW, 19-22 PCW, 35 PCW to 4 months, 0.5-2.6 PY, 2.8-10.7 PY, 13-19 PY, and 21-40 PY. Gene expression levels, represented in RPKM, were log2-transformed and centered to the mean expression level for each sample89. The expression values for each volumetric phenotype were then calculated as the mean values across all shared genes mapped by their shared loci with depression. Finally, smoothed LOESS curves were plotted to visualize the expression trajectories.

Expression-trait association analysis

To evaluate whether the brain’s expression levels of shared genes, identified through conjFDR analyses of each depression-volume pair, are significantly correlated with these traits (i.e., depression and corresponding volumetric phenotypes), we employed the “S-MultiXcan” tool (https://github.com/hakyimlab/MetaXcan)36. This method integrates GWAS summary data with pre-computed eQTL prediction models of all the brain tissues available from the GTEx v8, including the amygdala, anterior cingulate cortex (BA24), caudate, cerebellar hemisphere, cerebellum, cortex, frontal cortex (BA9), hippocampus, hypothalamus, nucleus accumbens, putamen, cervical spinal cord C1, and substantia nigra. Genes lacking significant brain eQTL associations were excluded from the analysis, and a significant expression-trait association was determined with an FDR corrected q < 0.05.

Cognitive and behavioral associations

To explore whether the expression of genes shared between depression and subcortical brain regions influences specific cognitive or behavioral functions, we utilized GWAS summary data for 8 cognitive abilities90 and 15 behavioral symptoms. The cognitive abilities included digit symbol substitution task (DSST), VNR, pairs matching (PM), RT, matrix pattern recognition (Matrix), tower rearranging (Tower), trail-making test B (TMT-B), and a general factor of intelligence (g-factor), as determined by a genomic structural equation model. These cognitive ability domains cover reasoning ability, executive function, processing speed, and memory, offering a comprehensive view of neurocognitive performance (for more details, see Supplementary Data 17). Additionally, the GWAS summary for behavioral symptoms included appetite change, sleep disturbance, and difficulty concentrating, among others (for more details, see Supplementary Data 18). Using the “S-MultiXcan” tool, we performed association analyses between the expression of shared genes—those whose expression was linked to both depression and at least one subcortical volume—and these cognitive abilities and behavioral symptoms, with significance determined by an FDR-corrected q < 0.05.

Control analyses

Genome-wide genetic correlation

The LDSC analysis was performed using “ldsc” software (v1.0.1, https://github.com/bulik/ldsc)37,38 to calculate genome-wide genetic correlations for each depression-volume pair based on the summary statistics. All analyses were exclusively performed on HapMap3 variants, excluding the MHC and 8p23.1 regions, and utilized pre-calculated LD scores from the 1000 Genomes European data.

Local genetic correlation

LAVA (v0.1.0, https://github.com/josefin-werme/LAVA)39 was employed to validate the shared loci identified by conjFDR analysis. As the detection of valid and interpretable local genetic correlations depends on the presence of sufficient local genetic signals, univariate tests in the LAVA approach served as a filtering step for bivariate local genetic correlation analyses. Bivariate local genetic correlation analyses were then exclusively performed for pairs of depression and volumetric traits with a nominally significant univariate local genetic signal.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.