Introduction
Psychiatric disorders constitute significant health and societal burdens on a global scale, consistently ranked among the most debilitating medical conditions across diverse age groups (GBD 2019 Diseases and Injuries Collaborators. (2020)). While diagnostic classification systems define disorders based on symptomatology, psychiatric comorbidity is common, with patients frequently presenting with multiple concurrent conditions. For example, a considerable proportion of patients with schizophrenia (SCZ) have comorbid depression, obsessive-compulsive disorder (OCD), and other psychiatric disorders (Gorman, Reference Gorman1996). Similarly, around 75% of patients with depression have comorbid anxiety disorders.
This comorbidity highlights the vast heterogeneity in psychiatry, as subtypes of a primary disorder with different psychiatric comorbidities may have distinct neurobiological underpinnings. Clarifying the genetic basis for these different psychiatric comorbidities can unveil biological mechanisms and promote individualized care.
Genome-wide association studies (GWASs) have achieved success in revealing the genetic basis of complex disorders (Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014; Trubetskoy et al., Reference Trubetskoy, Pardiñas, Qi, Panagiotaropoulou, Awasthi, Bigdeli, Bryois, Chen, Dennison, Hall, Lam, Watanabe, Frei, Ge, Harwood, Koopmans, Magnusson, Richards, Sidorenko, Wu, Zeng, Grove, Kim, Li, Voloudakis, Zhang, Adams, Agartz, Atkinson, Agerbo, Al Eissa, Albus, Alexander, Alizadeh, Alptekin, Als, Amin, Arolt, Arrojo, Athanasiu, Azevedo, Bacanu, Bass, Begemann, Belliveau, Bene, Benyamin, Bergen, Blasi, Bobes, Bonassi, Braun, Bressan, Bromet, Bruggeman, Buckley, Buckner, Bybjerg-Grauholm, Cahn, Cairns, Calkins, Carr, Castle, Catts, Chambert, Chan, Chaumette, Cheng, Cheung, Chong, Cohen, Consoli, Cordeiro, Costas, Curtis, Davidson, Davis, de Haan, Degenhardt, DeLisi, Demontis, Dickerson, Dikeos, Dinan, Djurovic, Duan, Ducci, Dudbridge, Eriksson, Fañanás, Faraone, Fiorentino, Forstner, Frank, Freimer, Fromer, Frustaci, Gadelha, Genovese, Gershon, Giannitelli, Giegling, Giusti-Rodríguez, Godard, Goldstein, González Peñas, González-Pinto, Gopal, Gratten, Green, Greenwood, Guillin, Gülöksüz, Gur, Gur, Gutiérrez, Hahn, Hakonarson, Haroutunian, Hartmann, Harvey, Hayward, Henskens, Herms, Hoffmann, Howrigan, Ikeda, Iyegbe, Joa, Julià, Kähler, Kam-Thong, Kamatani, Karachanak-Yankova, Kebir, Keller, Kelly, Khrunin, Kim, Klovins, Kondratiev, Konte, Kraft, Kubo, Kučinskas, Kučinskiene, Kusumawardhani, Kuzelova-Ptackova, Landi, Lazzeroni, Lee, Legge, Lehrer, Lencer, Lerer, Li, Lieberman, Light, Limborska, Liu, Lönnqvist, Loughland, Lubinski, Luykx, Lynham, Macek, Mackinnon, Magnusson, Maher, Maier, Malaspina, Mallet, Marder, Marsal, Martin, Martorell, Mattheisen, McCarley, McDonald, McGrath, Medeiros, Meier, Melegh, Melle, Mesholam-Gately, Metspalu, Michie, Milani, Milanova, Mitjans, Molden, Molina, Molto, Mondelli, Moreno, Morley, Muntané, Murphy, Myin-Germeys, Nenadić, Nestadt, Nikitina-Zake, Noto, Nuechterlein, O’Brien, O’Neill, Oh, Olincy, Ota, Pantelis, Papadimitriou, Parellada, Paunio, Pellegrino, Periyasamy, Perkins, Pfuhlmann, Pietiläinen, Pimm, Porteous, Powell, Quattrone, Quested, Radant, Rampino, Rapaport, Rautanen, Reichenberg, Roe, Roffman, Roth, Rothermundt, Rutten, Saker-Delye, Salomaa, Sanjuan, Santoro, Savitz, Schall, Scott, Seidman, Sharp, Shi, Siever, Sigurdsson, Skarabis, Slominsky, So, Sobell, Söderman, Stain, Steen, Steixner-Kumar, Stögmann, Stone, Straub, Streit, Strengman, Stroup, Subramaniam, Sugar, Suvisaari, Svrakic, Swerdlow, Szatkiewicz, Tam Ta, Takahashi, Terao, Thibaut, Toncheva, Tooney, Torretta, Tosato, Tura, Turetsky, Üçok, Vaaler, van Amelsvoort, van Winkel, Veijola, Waddington, Walter, Webb, Weiser, Williams, Witt, Wormley, Wu, Xu, Yolken, Zai, Zhou, Zhu, Zimprich, Atbaşoğlu, Ayub, Benner, Bertolino, Black, Bray, Breen, Buccola, Byerley, Chen, Cloninger, Crespo-Facorro, Donohoe, Freedman, Galletly, Gandal, Gennarelli, Hougaard, Hwu, Jablensky, McCarroll, Moran, Mors, Mortensen, Müller-Myhsok, Neil, Nordentoft, Pato, Pocklington, Roussos, Vassos, Verhage, Visscher, Yang, Posthuma, Andreassen, Kendler, Owen, Wray, Daly, Huang, Neale, Sullivan, Ripke, Walters and O’Donovan2022; Uffelmann et al., Reference Uffelmann, Huang, Munung, de Vries, Okada, Martin, Martin, Lappalainen and Posthuma2021), including psychiatric disorders. However, prior research on GWASs mainly focused on exploring the genetic basis of a single psychiatric entity, and largely ignored psychiatric comorbidity, with notable exceptions, such as the seminal cross-disorder GWASs by the Cross-Disorder Group of the Psychiatric Genomics Consortium (2013).
Our team has recently developed an innovative statistical framework, CombGWAS (Yin et al., Reference Yin, Chau, Lin, Rao, Xiang, Sham and So2021), to uncover the genetic architecture of disease combinations. In this study, we applied CombGWAS to decipher the genetic signatures distinguishing 16 psychiatric disorder combinations, such as SCZ and major depressive disorder (MDD) with and without various comorbidities. We also considered psychiatric disorders with/without comorbid obesity as disease subtypes, highlighting the intricate relationships between these conditions and their implications for cardiovascular health. Emerging evidence suggests bidirectional relationships between metabolic dysregulation (such as obesity) and psychiatric disorders (Gao et al., Reference Gao, Qin, Jiao, Hao, Zhao, Wang, Wen and Wang2024; Goldfarb et al., Reference Goldfarb, De Hert, Detraux, Di Palo, Munir, Music, Piña and Ringen2022; Hiles, Révész, Lamers, Giltay, & Penninx, Reference Hiles, Révész, Lamers, Giltay and Penninx2016; Pan et al., Reference Pan, Keum, Okereke, Sun, Kivimaki, Rubin and Hu2012).
Unlike other conventional methods, such as polygenic risk scores (Choi, Mak, & O’Reilly, Reference Choi, Mak and O’Reilly2020) or linkage disequilibrium (LD) score regression (LDSR) (Bulik-Sullivan, Loh, Finucane, Ripke, & Yang, Reference Bulik-Sullivan, Loh, Finucane, Ripke, Yang, Patterson, Daly, Price and Neale2015) that only reveal overall genetic overlap, our approach can identify specific susceptibility variants responsible for the presence or absence of a particular comorbidity. This provides deeper insights into the complex genetic underpinnings of psychiatric disorder heterogeneity. Our method can also estimate the effect size (e.g. odds ratio [OR]) of the variants contributing to the comorbidities, which cannot be achieved by other existing approaches. Furthermore, we employed additional analyses, including genetic correlation, Mendelian randomization (MR), and functional enrichment, to comprehensively characterize the shared and unique genetic architectures across psychiatric subtypes and their relationships with cardiovascular diseases (CVDs).
We summarize the key contributions and novelties of this study below:
-
1. We applied a novel statistical framework, CombGWAS, to decipher the genetic signatures that distinguish 16 psychiatric disorder combinations, including subtypes with and without specific comorbidities. We identified numerous susceptibility genes/variants across different psychiatric disease subtypes, providing insights into the genetic heterogeneity underlying these conditions.
-
2. We investigated MDD and SCZ with/without obesity as disease subtypes and revealed their genetic underpinnings. To our knowledge, this is the first GWAS-based study to uncover genetic variants underlying these comorbid conditions.
-
3. We revealed distinct genetic correlations between psychiatric disorder subtypes and CVDs, shedding light on the complex links between psychiatric and cardiometabolic traits. For example, we uncovered that MDD and SCZ with or without obesity exhibit very different, and often opposite, genetic correlations with cardiometabolic disorders.
-
4. Our MR analyses revealed that comorbid psychiatric disorders often have more profound causal effects on cardiovascular risks compared to single psychiatric disorders, but the causal effect estimates also vary across different psychiatric subtypes. Notably, obese and nonobese MDD/SCZ showed opposite causal effects on the risks of T2DM.
-
5. We provided comprehensive genetic and biological insights (through pathway, cell type, and drug enrichment analyses) that may inform the development of personalized treatment and prevention strategies for psychiatric disorder combinations.
Overall, our work represents a significant advancement in understanding the complex genetic underpinnings of psychiatric disorder heterogeneity and comorbidities, which is crucial for improving clinical management and patient outcomes in the long term.
Methods
In this study, we employed CombGWAS (Yin et al., Reference Yin, Chau, Lin, Rao, Xiang, Sham and So2021) to estimate the genetic architecture of the combinations of psychiatric disorders. CombGWAS is a statistical framework specially designed to mimic case–control GWASs of combinations of diseases (or comorbid conditions) by using summary statistics taken from case–control GWASs of the corresponding diseases only. The mathematical formulation is as follows.
Consider
$ S\sim bin\;\left(n=2,q\right) $
as a random biallelic single-nucleotide polymorphism (SNP) with
$ q $
denoting the effect allele frequency. We considered two binary psychiatric-related traits
$ {P}_1 $
and
$ {P}_2 $
, and generated a multivariate linear model on N subjects as follows:
$$ \left[\begin{array}{cc}\begin{array}{c}{P}_{11}\\ {}\begin{array}{c}{P}_{11}\\ {}\vdots \end{array}\end{array}& \begin{array}{c}{P}_{21}\\ {}\begin{array}{c}{P}_{22}\\ {}\vdots \end{array}\end{array}\\ {}{P}_{1N}& {P}_{2N}\end{array}\right]=\left[\begin{array}{cc}\begin{array}{c}1\\ {}\begin{array}{c}1\\ {}\vdots \end{array}\end{array}& \begin{array}{c}{S}_1\\ {}\begin{array}{c}{S}_2\\ {}\vdots \end{array}\end{array}\\ {}1& {S}_N\end{array}\right]\;\left[\begin{array}{cc}{\beta}_{01}& {\beta}_{02}\\ {}{\beta}_{11}& {\beta}_{12}\end{array}\right]+\epsilon $$
The left-hand side
$ P $
is an
$ N\times 2 $
phenotype matrix, and the matrices on the right-hand side are, respectively, an
$ N\times 2 $
SNP matrix,
$ 2\times 2 $
association estimates, and an
$ N\times 2 $
error matrix. The error matrix (
$ \epsilon \sim N\;\left(0,\sum \right) $
) is assumed to follow a multivariate normal distribution.
$ {\beta}_{11} $
and
$ {\beta}_{12} $
represent the regression coefficients for traits
$ {P}_1 $
and
$ {P}_2 $
for SNP
$ S $
, which are directly obtained from the summary statistics of traits
$ {P}_1 $
and
$ {P}_2 $
.
$ {\beta}_{01} $
and
$ {\beta}_{02} $
are the intercept terms estimated from the same summary statistics using the following equations:
$ {\beta}_{01}=\overline{P_1}-{\beta}_{11}\ast \overline{S}={k}_1-{\beta}_{11}\ast 2\ast {eaf}_{P_1} $
and
$ {\beta}_{02}=\overline{P_2}-{\beta}_{12}\ast \overline{S}={k}_2-{\beta}_{12}\ast 2\ast {eaf}_{P_2} $
, where
$ {k}_1 $
and
$ {k}_2 $
denote the population prevalence of traits
$ {P}_1 $
and
$ {P}_2 $
, and
$ {eaf}_{P_1} $
and
$ {eaf}_{P_2} $
denote the effect allele frequencies for the SNP in each trait’s GWAS.
The psychiatric comorbidity of two disorders (P 1 and P 2) can be defined as a function of these two traits:
$$ Comor=f\;\left({P}_1,{P}_2\right)={P}_1\times {P}_2 $$
Similarly, the ‘single’ psychiatric disorder (with the particular comorbid disorder excluded) can be defined as follows:
$$ Single=f\;\left({P}_1,{P}_2\right)={P}_1\times \left(1-{P}_2\right) $$
Our goal here is to infer the association estimates for genome-wide SNPs for the target traits (i.e. the comorbid psychiatric disorders or the single psychiatric disorder without the specific comorbid disorder).
$$ f\;\left({P}_1,{P}_2\right)=: T={\gamma}_0+{\gamma}_1S+e $$
Here,
$ e $
is assumed to follow a normal distribution with zero mean.
$ {\gamma}_0 $
and
$ {\gamma}_1 $
are the intercept and regression coefficient, respectively, for the target phenotype (comorbidity or a disorder without a comorbidity). For downstream analysis,
$ {\gamma}_1 $
serves as the primary effect size measure (equivalent to β in standard GWASs), representing the association strength with the target phenotype. According to Yin et al. (Reference Yin, Chau, Lin, Rao, Xiang, Sham and So2021), the association estimates for the combined psychiatric disorders can be estimated as follows:
$$ Comor\;\left(\hat{\gamma_0}\right)=\hat{\beta_{01}}\times \hat{\beta_{02}}+ Cov\;\left({P}_1,{P}_2\right) $$
$$ Comor\left(\hat{\gamma_1}\right)={\displaystyle \begin{array}{l}\frac{2q\left(1-q\right)}{2q\left(1-q\right)+{q}^2}\Big[\Big(\left(\hat{\beta_{01}}+\hat{\beta_{11}}\right)\times \left(\hat{\beta_{02}}+\hat{\beta_{12}}\right)-\hat{\beta_{01}}\\ {}\times \hat{\beta_{02}}\Big)\Big]+\frac{q^2}{2q\left(1-q\right)+{q}^2}\\ {}\frac{\left[\left(\left(\hat{\beta_{01}}+2\hat{\beta_{11}}\right)\times \left(\hat{\beta_{02}}+2\hat{\beta_{12}}\right)-\hat{\beta_{01}}\times \hat{\beta_{02}}\right)\right]}{2}\end{array}} $$
Here,
$ Cov\;\left({P}_1,{P}_2\right) $
denotes the covariance between the two binary traits. Along the same line, the association estimates for the single psychiatric disorder without the particular comorbid disorder can be computed as follows:
$$ Single\left(\hat{\gamma_0}\right)=\hat{\beta_{01}}\times \left(1-\hat{\beta_{02}}\right)- Cov\;\left({P}_1,{P}_2\right) $$
$$ Single\left(\hat{\gamma_1}\right)={\displaystyle \begin{array}{l}\frac{2q\left(1-q\right)}{2q\left(1-q\right)+{q}^2}\Big[\Big(\left(\hat{\beta_{01}}+\hat{\beta_{11}}\right)\times \left(1-\hat{\beta_{02}}-\hat{\beta_{12}}\right)\\ {}-\hskip2px \hat{\beta_{01}}\times \left(1-\hat{\beta_{02}}\right)\Big)\Big]+\frac{q^2}{2q\left(1-q\right)+{q}^2}\\ {}\frac{\left[\left(\left(\hat{\beta_{01}}+2\hat{\beta_{11}}\right)\times \left(1-\hat{\beta_{02}}-2\hat{\beta_{12}}\right)-\hat{\beta_{01}}\times \left(1-\hat{\beta_{02}}\right)\right)\right]}{2}\end{array}} $$
The estimates used in the above equations are presumably derived from a linear regression model. However, for binary traits, the association estimates are usually calculated from logistic regression models. Therefore, we need to transform the estimates (ORs) derived from logistic regression to those of linear regression, using the following equation:
$$ \left(\alpha k-k\right)\left(1-k\right)={\displaystyle \begin{array}{l}{\beta}_1^2\left[\hskip0.2em p\left(1-p\right)+\alpha p\left(1-p\right)\right]+{\beta}_1\Big[\alpha k\left(1-p\right)+\alpha p\\ {}-\hskip2px \alpha p k+ kp+\left(1-p\right)\left(1-k\right)\Big]\end{array}} $$
where p indicates the effect allele frequency of the SNP under study
$ S $
,
$ k $
represents the proportion of cases, β
1 represents the coefficient under a linear model, and
$ \alpha $
defines the OR of SNP
$ S $
regressed on the same binary trait. We solve the above equation for β
1. We choose the solution whose absolute value is smaller than the coefficient under a logistic model, i.e., abs(β
1)<abs(log(α)). Then we convert the coefficients back to ORs. The variances for
$ \hat{\gamma_0} $
and
$ \hat{\gamma_1} $
for the target disease combination can be estimated by the delta method, which employs a first-order Taylor expansion to estimate the variance of a function. Also, it is noteworthy that our proposed analytic framework assumes the absence of significant interactions between the comorbid trait (s) and the genetic variants under study, as explained elsewhere (Yin et al., Reference Yin, Chau, Lin, Rao, Xiang, Sham and So2021). In our previous study, we verified the validity of CombGWAS using both extensive simulations and applications to UK-Biobank data. Briefly, we reperformed a GWAS in the UK Biobank (UKBB) using the actual phenotype data of disease combinations as the outcome, and the results were highly similar to those computed from CombGWAS.
Enrichment analyses
We utilized several methods to clarify the putative functional and biological mechanisms of the identified susceptibility variants for the ‘psychiatric phenotypes/subtypes’ of interest (i.e. the comorbid psychiatric disorders, and the ‘single’ psychiatric disorders without the particular comorbid disorder). First, the MAGMA tool within FUMA (Watanabe, Taskesen, Van Bochoven, & Posthuma, Reference Watanabe, Taskesen, Van Bochoven and Posthuma2017; Watanabe, Umićević Mirkov, de Leeuw, van den Heuvel, & Posthuma, Reference Watanabe, Umićević Mirkov, de Leeuw, van den Heuvel and Posthuma2019) was used to map the identified genetic variants to their corresponding genes. This established the gene-level associations with the psychiatric phenotypes/subtypes. The definitions for independent significant SNPs and genomic risk loci in our study were also based on the criteria established by FUMA (https://fuma.ctglab.nl/tutorial#riskloci).
Second, we performed tissue specificity analysis to examine if the susceptibility genes were enriched in specific tissues. Furthermore, we carried out a cell type enrichment analysis to uncover specific cell types or cell populations that may play crucial roles in the etiology of the subtypes. We also conducted pathway analysis using the program ‘ConsensusPathDB’ (Herwig, Hardt, Lienhard, & Kamburov, Reference Herwig, Hardt, Lienhard and Kamburov2016; Kamburov, Stelzl, Lehrach, & Herwig, Reference Kamburov, Stelzl, Lehrach and Herwig2013). Moreover, we performed drug enrichment analyses on the identified susceptibility gene sets using ‘enrichr’ (Kuleshov et al., Reference Kuleshov, Jones, Rouillard, Fernandez, Duan, Wang, Koplev, Jenkins, Jagodnik, Lachmann, McDermott, Monteiro, Gundersen and Ma’ayan2016; Xie et al., Reference Xie, Bailey, Kuleshov, Clarke, Evangelista, Jenkins, Lachmann, Wojciechowicz, Kropiwnicki, Jagodnik, Jeon and Ma’ayan2021). This analysis identified potential therapeutic agents or drug targets that may be relevant to the treatment of the investigated psychiatric conditions. Notably, we also conducted tissue, pathway, cell type, and drug enrichment analyses based on subtype-specific genes.
Genetic correlation among psychiatric disease subtypes and CVDs
We employed LDSR (Bulik-Sullivan et al., Reference Bulik-Sullivan, Loh, Finucane, Ripke, Yang, Patterson, Daly, Price and Neale2015) to calculate the genetic correlation (rg) between different psychiatric entities to see if they represent genetically distinct disease subtypes with unique genetic architectures. We tested if rg ≠ 0 (assessing genetic overlap) and rg ≠ 1 (assessing subtype distinctiveness) to comprehensively characterize genetic relationships. Furthermore, we explored the genetic correlations between different psychiatric disorder subtypes and CVDs, including coronary artery disease (CAD), type 2 diabetes (T2DM), and stroke, highlighting possible links between these conditions.
MR analysis
MR is a method that leverages genetic variants as ‘instruments’ to represent the exposure for inferring causal relationships between risk factors and outcomes. We performed MR to estimate the causal effects of the studied psychiatric entities on various CVDs, testing the hypothesis that comorbidity amplifies cardiovascular risk. This analysis was included to provide clinical and translational insights into the comorbidities. A two-sample MR design was employed, using both inverse-variance weighted (MR-IVW) (Bowden, Davey Smith, Haycock, & Burgess, Reference Bowden, Davey Smith, Haycock and Burgess2016) and Egger regression (MR-Egger) (Burgess & Thompson, Reference Burgess and Thompson2017) approaches. The number of genetic variants included in the MR analysis may influence the causal estimates. To ensure the robustness of our findings, we performed MR at multiple r 2 thresholds (
$ 0.001,0.01,0.05,\mathrm{and}\;0.1 $
), taking SNP correlations into account.
Psychiatric disorders studied
We applied CombGWAS to examine seven different psychiatric disorders and obesity, namely: Alzheimer’s disease (AD), attention-deficit hyperactivity disorder (ADHD), autism spectrum disorder (ASD), anxiety disorders, insomnia, MDD, obesity, and SCZ. The datasets used are shown in Supplementary Table S1. Using CombGWAS, we estimated the effect sizes (ORs), standard errors (SEs), and corresponding p-values for each disease combination for genome-wide SNPs.
Our selection of disorders was guided by a deliberate strategy focusing on three key criteria, chosen to maximize the clinical and scientific relevance of our findings (as shown below).
We analyzed the genetic basis for 16 (or 8 pairs) psychiatric entities, with a primary focus on MDD and SCZ. The psychiatric entities include ADHD with and without comorbid ASD, MDD with and without AD, MDD with and without anxiety disorders, MDD with and without ASD, MDD with and without insomnia, MDD with and without obesity, MDD with and without SCZ, and SCZ with and without obesity.
Results
Overview
Table 1 demonstrates the independent significant SNPs (p <5e−08, r 2 < 0.6) and genomic risk loci for the eight comorbid psychiatric disorders and eight single psychiatric entities (single psychiatric disorder without a particular comorbid disorder). Our analysis revealed a total of 954 independent significant SNPs, mapped to 325 genomic risk loci, for the 8 comorbid psychiatric disorders (Figures 1 and 2, Table 1, and Supplementary Table S2). In addition, we identified 934 independent significant SNPs mapped to 328 genomic risk loci for the 8 single psychiatric entities without a comorbidity.
Number of identified independent significant SNPs and genomic risk loci for different combinations of psychiatric disorders
Manhattan plots of GWAS results for 16 psychiatric entities.
Note: ADHD, attention-deficit hyperactivity disorder; ASD, autism spectrum disorder; MDD, major depressive disorder; SCZ, schizophrenia.
Number of shared genomic risk loci, genes, and pathways across eight pairs of subtypes.
The identified susceptibility genetic variants were mapped to genes by MAGMA. We identified both subtype-specific and shared susceptibility genes, loci, and pathways across psychiatric disorder subtypes. Supplementary Table S3 provides additional details of the shared and nonshared genes associated with the psychiatric entities studied.
Tissue enrichment analysis
We conducted a tissue enrichment analysis using the FUMA tool, leveraging a precomputed list of differentially expressed genes (DEGs) across various tissues obtained from GTEx. The input consisted of significant genes identified in GWASs (via MAGMA), which were then tested for enrichment among these DEGs. We view this analysis as hypothesis-generating, as differential expression does not necessarily imply a causal role for the tissue.
Supplementary Figures S1 and S2 show the tissue enrichment analysis results for all 16 psychiatric entities studied. We observed that all the entities showed enrichment predominantly in the brain tissues. Interestingly, the basal ganglia emerged as the top-enriched tissue for all psychiatric entities involving ADHD or ASD. Abundant evidence supports its critical role in maintaining normal motor actions and cognitive functions (Damasio & Maurer, Reference Damasio and Maurer1978; Graybiel, Reference Graybiel2008; Jin & Costa, Reference Jin and Costa2015). In a study by Curtin et al. (Reference Curtin, Fleckenstein, Keeshin, Yurgelun-Todd, Renshaw, Smith and Hanson2018), ADHD was associated with a 2.4-fold increased risk of basal ganglia diseases. Notably, subtype-specific tissue enrichment was observed across different subtype pairs. For instance, significant enrichment of brain amygdala was only observed in ADHD with ASD (Supplementary Figures S1 and S2).
Another intriguing finding is that the cerebellum was the top-enriched tissue for all disease combinations involving MDD. A growing body of research has demonstrated its involvement in impaired cognitive function and emotion dysregulation (Depping, Schmitgen, Kubera, & Wolf, Reference Depping, Schmitgen, Kubera and Wolf2018; Depping et al., Reference Depping, Schmitgen, Bach, Listunova, Kienzle, Kubera, Roesch-Ely and Wolf2020; Hoche, Guell, Vangel, Sherman, & Schmahmann, Reference Hoche, Guell, Vangel, Sherman and Schmahmann2018). Compared to other psychiatric entities involving MDD, the two entities involving obesity (i.e. MDD with obesity and MDD without obesity) exhibited a lower number of enriched brain tissues, with the cerebellar hemisphere ranked as the top-enriched tissue.
Pathway enrichment analysis
We conducted pathway enrichment analysis using the web tool ConsensusPathDB (Supplementary Table S4). We observed a mixture of shared and condition-specific enriched pathways across different psychiatric entities. For instance, ‘Signaling by Nuclear Receptors’ was one of the top-enriched pathways shared between MDD with AD and MDD without AD. In a previous study, Fries, Saldana, Finnstein, and Rein (Reference Fries, Saldana, Finnstein and Rein2023) have highlighted the critical role of this pathway in coordinating an organism’s response to stress, a major risk factor for MDD. Additionally, Mandrekar-Colucci and Landreth (Reference Mandrekar-Colucci and Landreth2011) identified nuclear receptors as a promising therapeutic target for AD, as they could regulate microglial activation and mitigate the brain’s inflammatory responses.
The cardiolipin biosynthesis pathway was specifically enriched in MDD with AD. Alterations in cardiolipin content, structure, and localization have been reported to be associated with impaired neurogenesis and neuronal dysfunction (Falabella, Vernon, Hanna, Claypool, & Pitceathly, Reference Falabella, Vernon, Hanna, Claypool and Pitceathly2021), contributing to neurodegenerative diseases.
Notably, we did not find any common pathways between ADHD with and without ASD, as well as between MDD with and without ASD.
We also conducted enrichment analyses based on subtype-specific genes to elucidate subtype-specific biology (Supplementary Table S5). For example, neuronal system, nervous system development, dopaminergic synapse, and brain-derived neurotrophic factor signaling pathway are the top-enriched pathways specific to MDD with SCZ. This finding aligns well with established knowledge that both disorders are linked to disruptions in neuroplasticity (Angelucci, Brene, & Mathe, Reference Angelucci, Brene and Mathe2005; Delva & Stanwood, Reference Delva and Stanwood2021).
Cell type enrichment analysis
Cell type enrichment analyses were also performed to uncover the involved cell types for each studied disease combination (Figure 3). However, since not all cell types were available for analysis in FUMA, our results should be considered exploratory. We found that GABAergic neurons were the most enriched cell type most psychiatric entities. Fogaça and Duman (Reference Fogaça and Duman2019) reported decreased expression of GABAergic interneuron markers in the frontal cortex could result in depressive-like behaviors. Moreover, impaired GABAergic neurotransmission has been implicated in patients with depression, prompting the development of therapeutic strategies targeting this deficit (Duman, Sanacora, & Krystal, Reference Duman, Sanacora and Krystal2019). Notably, the top-enriched cell types identified for the psychiatric disease combinations were located in brain tissues. Psychiatric disorder combinations showed significant immune cell type enrichments (Supplementary Table S6), consistent with established immune dysregulation in these disorders (Gibney & Drexhage, Reference Gibney and Drexhage2013; Iakunchykova, Leonardsen, & Wang, Reference Iakunchykova, Leonardsen and Wang2024).
Cell type enrichment analysis results for each studied psychiatric disease combination.
Drug enrichment analysis
To further explore the homogeneity and heterogeneity underlying the studied psychiatric entities, we performed drug enrichment analysis on the susceptibility gene sets identified for each disease combination (Supplementary Tables S7 and S8). This analysis is considered exploratory or hypothesis-generating, as further experimental validations are required. Some enriched drugs were unique to particular psychiatric entities, while others were shared across different psychiatric entities. For example, chlorprothixene (a typical antipsychotic) was one of the top-enriched drugs specific to ADHD and ASD. Antipsychotics are often prescribed in ASD (Posey, Stigler, Erickson, & McDougle, Reference Posey, Stigler, Erickson and McDougle2008) for irritability and associated behaviors including aggression and self-injury. Antipsychotics are also sometimes prescribed in ADHD due to other comorbid conditions, especially behavior problems (Lee, Zhang, & Rose, Reference Lee, Zhang and Rose2022). Romidepsin was another top-enriched drug specific to ADHD with ASD based on subtype-specific genes. Qin et al. (Reference Qin, Ma, Wang, Hu, Matas, Wei and Yan2018) showed its potential in alleviating social deficits in mouse models. Among all disease combinations involving MDD, except for MDD with ASD, fendiline was among the top-enriched drugs. Interestingly, a previous study (So, Chau, Lau, Wong, & Zhao, Reference So, Chau, Lau, Wong and Zhao2019) suggested fendiline as a promising repurposing drug for patients with MDD.
Genetic correlation among psychiatric disorder subtypes and CVDs
We calculated genetic correlations between pairs of psychiatric entities with and without particular comorbid conditions (Table 2). The results (rg ranging from 0.1664 to 1.0271) revealed moderate- tohigh genetic correlations in most pairs, suggesting shared genetic architecture. For instance, MDD with anxiety disorder exhibited a high genetic correlation with MDD without anxiety disorder (rg = 0.9964). Similarly, a high genetic correlation was also observed between SCZ with and without obesity (rg = 0.9016). Interestingly, MDD with obesity only displayed a weak genetic correlation with MDD without obesity, implying that they may represent distinct biological subtypes of MDD (rg = 0.1663).
Genetic correlation between different psychiatric entities
Note: rg indicates the calculated genetic correlation between two disease ‘subtypes’. p (rg = 0) refers to the p-value by testing whether the rg differs from zero; p (rg = 1) refers to the p-value by testing whether the rg differs from 1.
To further investigate the genetic overlap between psychiatric entities and CVDs, we analyzed their genetic correlations by LDSR (Table 3 and Supplementary Table S9). We prioritize subtype pairs with distinct genetic architectures. As shown in Table 3, most ‘pairs’ of psychiatric subtypes (i.e. a disorder with and without a comorbidity) exhibited significantly different genetic correlations with CVDs, indicating distinct genetic relationships.
Genetic correlation between psychiatric entities and cardiovascular diseases
Note: rg indicates the calculated genetic correlation between two psychiatric disorder subtypes and cardiovascular diseases. Results for MDD + AD and MDD + ANX are not included due to space limits. Bold values indicate a p-value of ≤ 0.05, while italic values represent a p-value between 0.05 and 0.1. We included full results in Supplementary Table S9a.
Interestingly, some psychiatric subtypes displayed opposite genetic correlations with the same CVD. For example, while SCZ with obesity exhibited a positive genetic correlation with T2DM (rg = 0.2481), SCZ without obesity demonstrated a negative genetic correlation with T2DM (rg = −0.1539). Similarly, SCZ with and without obesity displayed opposing genetic correlations with CAD. Notably, the genetic correlations between ‘unsubtyped’ (originally defined) psychiatric disorders and CVDs generally lie between those for ‘subtyped’ (comorbid or only single) disorders. These findings highlight the complex interactions between genetic factors associated with psychiatric entities and CVDs, indicating potential distinct genetic signatures for different psychiatric subtypes.
Results of MR analysis
MR analysis was conducted to investigate whether a certain psychiatric disease combination is causally linked to a significantly higher or lower risk of cardiovascular outcomes. Most results were consistent across different
$ {r}^2 $
cutoffs. For simplicity, we primarily report analysis results at
$ {r}^2=0.001,\mathrm{p}=5E-08 $
, which are standard settings in TwoSampleMR. Table 4 summarizes the results (Supplementary Table S10 for full results). It is noteworthy that with a binary exposure, the causal estimate (OR) reflects the average effect on the outcome associated with a 2.72-fold increase in the prevalence of the exposure (e.g. an increase in exposure prevalence from 1 to 2.72%) (Burgess & Labrecque, Reference Burgess and Labrecque2018).
MR analysis results for cardiovascular outcomes
Note: MR analysis results were primarily based on MR-IVW.
a The causal analysis results were derived from MR-Egger instead of MR-IVW due to the presence of imbalanced horizontal pleiotropy, as indicated by a significant Egger intercept. Since the exposure is binary, the OR represents the average effect to the outcome if the exposure prevalence (prevalence) is increased by 2.72 times. Bold values indicate a p-value of ≤ 0.05, while italic values represent a p-value between 0.05 and 0.1.
Many studied psychiatric disorder combinations exhibited a significant causal relationship with increased risks of CVDs. Importantly, the magnitude of effect sizes varies across different psychiatric disorder subtypes. For example, both MDD with and without anxiety were found to be causally linked to significantly increased CAD risks. The OR of MDD with anxiety on CAD is
$ 2.106\;\left(\mathrm{MR}-\mathrm{IVW},p=2.35\mathrm{E}-05\right) $
, while that of MDD without anxiety on CAD is
$ 1.108\;\left(\mathrm{MR}-\mathrm{Egger},p=3.78\mathrm{E}-05\right) $
.
Additionally, we found that comorbid psychiatric disorders often had stronger causal effects on cardiovascular outcomes than single disorders. For example, MDD with insomnia was causally linked to a significantly increased risk of CAD with an OR of
$ 3.721\;\left(\mathrm{MR}-\mathrm{IVW},p=1.72\mathrm{E}-06\right) $
, while MDD without insomnia had a much lower OR of
$ 1.054\;\left(\mathrm{MR}-\mathrm{IVW},p=2.71\mathrm{E}-03\right) $
.
Both obese MDD and obese SCZ, but not nonobese MDD or SCZ, were found to be causally linked to increased T2DM risk, highlighting obesity’s role in amplifying CVD liability. Specifically, obese MDD and obese SCZ were linked to T2DM risk with ORs of
$ 5.418\;\left(\mathrm{MR}-\mathrm{IVW},p=3.58\mathrm{E}-06\right) $
and
$ 1.364\;\left(\mathrm{MR}-\mathrm{Egger},p=2.81\mathrm{E}-03\right) $
, respectively. In contrast, nonobese MDD (
$ \mathrm{MR}-\mathrm{Egger},\mathrm{OR}=0.292,p=9.87\mathrm{E}-04 $
) and nonobese SCZ (
$ \mathrm{MR}-\mathrm{Egger},\mathrm{OR}=0.948,p=2.58\mathrm{E}-02 $
) were associated with significantly decreased T2DM risk.
The above results underscore the complex causal relationships between specific psychiatric disease combinations and various cardiovascular outcomes.
Discussion
Overview
This study represents a conceptual and methodological advance in psychiatric genetics, moving beyond the identification of genetic risk factors of individual disorders to the direct delineation of clinical heterogeneity. While prior research established that psychiatric disorders share a polygenic architecture, the exact genetic basis of comorbid disorders has not been investigated systematically. Our work demonstrates that a specific comorbidity can fundamentally alter a disorder’s genetic signature and its relationship with other health outcomes.
A key innovation of this approach is its ability to reveal clinically vital genetic relationships that are undetectable with standard methods. For instance, our analysis shows that the genetic risk for T2DM is positively correlated with SCZ with comorbid obesity, yet negatively correlated with SCZ without obesity. This novel finding provides proof of concept for a genetically informed nosology, suggesting that ‘metabolically normal’ and ‘metabolically abnormal’ SCZ may be distinct biological subtypes. The same also applies to MDD. This distinction has immediate implications for patient stratification and the clinical management of cardiovascular risk. By demonstrating this principle across multiple disorder pairs, we also revealed distinct loci, divergent health relationships, and unique pathway enrichments for various psychiatric disorder combinations. These findings are largely unreported in previous studies. This work provides a roadmap toward precision psychiatry, laying the foundation for subtype-specific biomarkers and personalized therapies.
Unique genetic underpinnings of psychiatric disorder subtypes
A key finding of this study is the demonstration of distinct genetic underpinnings of psychiatric disorder subtypes. The proportion of subtype-specific loci/SNPs varies across psychiatric entity pairs, with the smallest proportion exceeding 50%. We observed a slightly different pattern for subtype-specific genes, with proportions ranging between 8.9 and 97.5%, but most subtypes showed over 50% unique genes. These genes provide insights into the subtype-specific biological mechanisms and pathophysiology. For instance, KDM4A and SEMA6D emerged as top susceptibility genes only in ADHD without ASD, indicating the unique genetic underpinnings for ADHD subtypes. In a recent study, Guo et al. (Reference Guo, Hong, Xiong, Yan, Xie and Guan2024) revealed KDM4A as a negative regulator of engram formation for memory separation. These subtype-specific susceptibility genes could be utilized in the differentiation between subtypes and inform targeted treatment strategies.
Furthermore, we performed a comprehensive analysis of enriched pathways, cell types, and drugs based on the GWAS results. These enrichment analyses shed light on the potential biological mechanisms and pathways involved in the development and manifestation of specific subtypes. We also revealed some interesting patterns. For example, although we observed high proportions of subtype-specific genes for MDD with insomnia and MDD without insomnia, the proportion of subtype-specific pathways was lower. These findings suggest that a high proportion of subtype-specific genes does not always translate to a similarly high proportion of unique pathways. For example, it can possibly be due to different combinations of genes corresponding to different pathways.
Genetic correlations among disease subtypes and CVDs
The majority of disease subtypes exhibited high (rg ≥ 0.8) or moderate (0.4 ≤ rg < 0.8) correlation with one another. Among the eight pairs of subtypes examined, only one pair, that is, MDD with obesity versus MDD without obesity, demonstrates a weak correlation. Despite their high genetic correlations, most subtypes exhibited distinct genetic correlations with the same CVD. Some pairs even displayed entirely opposing genetic correlations, particularly those subtype pairs that involved obesity. These findings suggest that the psychiatric disorder combinations we studied may represent biologically distinct subtypes with both shared and unique genetic signatures.
Evidence indicates that normal weight may protect against CVDs (Powell-Wiley et al., Reference Powell-Wiley, Poirier, Burke, Després, Gordon-Larsen, Lavie and Sanders2021), potentially counteracting the elevated risks of CVD conferred by psychiatric disorders. This aligns with the observed opposing or weak genetic correlations between nonobese psychiatric disorders and CVD.
Varying causal effects of different psychiatric disorder subtypes on cardiovascular outcomes
While many psychiatric disorder combinations showed increased CVD risks, comorbid disorders typically demonstrated stronger causal effects than single disorders. This highlights the importance of considering the cumulative impact of comorbid disorders on cardiovascular health.
Of note, obese and nonobese MDD/SCZ showed opposite causal effects on T2DM risk. Obesity in MDD or SCZ enhances CVD risk, informing targeted interventions (e.g. weight management). These findings underscore the need for more comprehensive and personalized screening, prevention, and treatment strategies that address both psychiatric and cardiovascular conditions in individuals with comorbidities.
Integrated genetic insights from multiple levels of analysis
In this study, we employed a comprehensive series of complementary analyses to elucidate both shared and distinct genetic architectures across psychiatric disorder subtypes. Our multilevel approach, encompassing GWAS, gene, pathway, cell type, drug enrichment analyses, genetic correlation, and MR, provided an in-depth understanding by revealing both convergent and divergent biological mechanisms underlying psychiatric disorder combinations.
For instance, several lines of evidence converged to highlight specific biological mechanisms for subtypes of ADHD. Both gene-based analyses (implicating KDM4A) (Guo et al., Reference Guo, Hong, Xiong, Yan, Xie and Guan2024) and pathway analyses (implicating, e.g. synaptic vesicle cycle) (Wang et al., Reference Wang, Liang, Jin, Liu, Luo, Zhang, Wang, Dong and Zhang2025) independently pointed toward memory-related mechanisms in ADHD without ASD. This convergence across different analytical levels strongly suggests that memory formation and related processes may represent a distinct etiological pathway for this subtype.
Notably, our multimethod approach also revealed instructive divergences between analytical findings, which are equally crucial for a comprehensive understanding. For example, while genetic correlation analysis indicated a high degree of genetic similarity between MDD with AD and MDD without AD (
$ rg=1.0255,\mathrm{SE}=0.0265 $
), the MR analysis revealed significantly different causal effects on diabetes mellitus (
$ \mathrm{OR}=0.712,p=2.14\mathrm{E}-05 $
for MDD with AD versus
$ \mathrm{OR}=1.224,p=4.23\mathrm{E}-03 $
for MDD without AD). This example illustrates that conditions sharing substantial underlying genetic risk factors (high rg) may nonetheless exert differential downstream causal effects on other health outcomes. It should be noted that even if rg is high, the actual effect sizes (beta) of SNPs for traits 1 and 2 can still differ, for example, when the LD structure and allele frequencies differ across the two cohorts, or when heritabilities differ between the two traits.
Furthermore, the contrasting causal effects observed in the MR analysis for obese versus nonobese MDD/SCZ on T2DM risk (e.g. obese MDD increasing T2DM risk while nonobese MDD decreasing it) provide another example of how the same psychiatric disorder may have substantially different clinical consequences depending on the comorbidity status, which can be fully appreciated by integrating genetic correlation and MR findings.
Limitations
This study has several limitations. First, the GWAS summary statistics used were primarily derived from European samples. As such, some results may be population-specific and may not be generalizable to non-European groups. This limitation may be mitigated with the availability of summary statistics from other ethnic groups. While our analysis provides valuable insights into the genetic signatures underlying different psychiatric disease subtypes, further exploration is required to validate these findings. Replication studies using independent cohorts and individuals with comorbid conditions, as well as more diverse populations, would enhance the robustness and generalizability of the observed genetic associations. Besides, our study only included a limited number of psychiatric disease combinations. Also, the biological mechanisms underlying the psychiatric disorder subtypes will require further experimental studies.
Another limitation is that substantial differences in heritabilities and sample sizes between the trait pairs may influence the results. Large disparities between traits could disproportionately bias the contribution toward traits with higher heritabilities or sample sizes. For example, given two traits (A and B), if A is much more heritable and the sample size is larger, the genetic architecture of A + B (comorbidity) may be biased toward that of trait A itself. This concern may be particularly relevant for highly heritable disorders with large available sample sizes, such as SCZ.
Conclusions
To conclude, our study successfully identified the genetic basis and risk genes for eight pairs of psychiatric disease subtypes. Secondary analyses, including pathway, cell type, and drug enrichment analyses, provided further biological insights into the pathophysiology of these disorder subtypes. Genetic correlation and MR analyses shed light on the links of these subtypes with various CVDs, which may have important clinical implications. Furthermore, the identified nonshared susceptibility genes may provide a way to differentiate the different ‘subtypes’.
Future research should incorporate data from diverse ethnic groups, conduct replication studies, and perform functional investigations to enhance the robustness and generalizability of the observed genetic associations. By elucidating the genetic basis of a wide variety of psychiatric disorder subtypes, our study paves the way for the development of personalized treatment and preventive strategies for individuals affected by these complex disorders.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291725101396.
Data availability statement
The GWAS summary statistics of the studied disorders were downloaded from publicly available databases.
Acknowledgments
This work was supported partially by a Young Collaborative Research Grant (C4003-23Y), a National Natural Science Foundation China (NFSC) Young Scientist grant (31900495), a National Natural Science Foundation China grant (81971706), and the Lo Kwee Seong Biomedical Research Fund from The Chinese University of Hong Kong and the KIZ-CUHK Joint Laboratory of Bioresources and Molecular Research of Common Diseases, Kunming Institute of Zoology, and The Chinese University of Hong Kong, China.
Competing interests
The authors declare none.
Open access
