Introduction
Bipolar disorder (BD) is a complex genetic disorder with high heritability (Smoller & Finn, Reference Smoller and Finn2003). Clinically, each patient shows unique (manic or depressive) episode symptoms with relapsing and remitting patterns (Nierenberg et al., Reference Nierenberg, Agustini, Kohler-Forsberg, Cusin, Katz, Sylvia and Berk2023), which suggests a heterogeneity in its genetic basis. A significant portion of patients also demonstrate overlapping clinical features with other psychiatric diagnoses, such as schizophrenia, anxiety disorders, attention-deficit hyperactivity disorder (ADHD), and eating disorders (Mantere et al., Reference Mantere, Isometsä, Ketokivi, Kiviruusu, Suominen, Valtonen and Leppämäki2010). This phenomenon has generated controversial issues regarding diagnostic boundaries and comorbidities. From a genomic perspective, it might reflect a pleiotropy among BD and other psychiatric illnesses (Lee, Feng, & Smoller, Reference Lee, Feng and Smoller2021; Stearns, Reference Stearns2010).
Several studies have sought genetically valid clinical subtypes of BD. In addition to bipolar 1 (BD-I) v. bipolar 2 (BD-II) subtyping, various clinical characteristics have been investigated in genome-wide studies (Coombes et al., Reference Coombes, Markota, Mann, Colby, Stahl, Talati and Biernacka2020; Faraone, Su, & Tsuang, Reference Faraone, Su and Tsuang2004; Kerner, Lambert, & Muthén, Reference Kerner, Lambert and Muthén2011; Labbe et al., Reference Labbe, Bureau, Moreau, Roy, Chagnon, Maziade and Merette2012; Maciukiewicz et al., Reference Maciukiewicz, Czerski, Leszczynska-Rodziewicz, Kapelski, Szczepankiewicz, Dmitrzak-Weglarz and Karlowski2012, Reference Maciukiewicz, Dmitrzak-Weglarz, Pawlak, Leszczynska-Rodziewicz, Zaremba, Skibinska and Hauser2014; Meier et al., Reference Meier, Mattheisen, Vassos, Strohmaier, Treutlein, Josef and Szelinger2012; Monahan et al., Reference Monahan, Stump, Coryell, Harezlak, Marcoulides, Liu and Nurnberger2015; Ruderfer et al., Reference Ruderfer, Fanous, Ripke, McQuillin, Amdur, Gejman and Kendler2014). These studies have suggested different genetic basis for various subtypes. However, there is a lack of consistency in applied subtypes, which limits the comparison between studies. In a previous study (Baek et al., Reference Baek, Ha, Kim, Cho, Yang, Choi and Hong2019), the authors extracted six phenotype dimensions of BD from comprehensive lifetime clinical characteristics as possible homogenous phenotypes for genomic studies. Application of valid phenotypes would be an essential step to clarify the genetic architecture of complex genetic disorders.
In addition to genetic heterogeneity within single disorders, a strong possibility of pleiotropy, sharing common genomic risks among different illnesses, has been suggested in psychiatric disorders, including BD. Recent large-scale genome-wide association studies (GWAS) have consistently revealed genetic correlations among them (Frei et al., Reference Frei, Holland, Smeland, Shadrin, Fan, Maeland and Dale2019; Kim et al., Reference Kim, Gonçalves, Husain, Müller, Mulsant, Zai and Kloiber2023; Lee et al., Reference Lee, Ripke, Neale, Faraone, Purcell, Perlis and Wray2013). Given overlapping clinical features between BD and other psychiatric illnesses, certain subtypes of BD might share genes with other psychiatric disorders beyond just BD. To test this hypothesis, research designs incorporating intra-disease heterogeneity and inter-disease commonality into a single genomic analysis need to be applied.
In the present study, we attempted to test whether phenotypically heterogeneous subtypes of BD might have distinct polygenic risks that are shared with other psychiatric illnesses. We applied lifetime phenotype dimensions identified in our previous study (Baek et al., Reference Baek, Ha, Kim, Cho, Yang, Choi and Hong2019) as BD subtypes. To investigate whether each phenotype dimension had a different association pattern with polygenic risks of major psychiatric illnesses, polygenic risk scores (PRSs) of six psychiatric illnesses from East Asian (EA) and other available populations were analyzed.
Methods
Study participants
A total of 467 patients with DSM-defined bipolar I (BD-I) or II disorder (BD-II) were recruited from Samsung Medical Center and Seoul National University Bundang Hospital. Of these participants included in the study, a total of 307 were included in our previous study, and 160 were additionally recruited. Basic demographic characteristics showed no significant difference between those who were included in our previous study and those who were later recruited in terms of basic demographic characteristics. Table 1 displays the basic demographic characteristics of the study participants.
Basic sociodemographic characteristics of study participants (n = 467)
s.d., standard deviation.
Detailed recruitment and evaluation processes were described elsewhere (Baek et al., Reference Baek, Ha, Kim, Cho, Yang, Choi and Hong2019). In brief, we recruited those with BD-I or BD-II who were clinically in stable conditions defined by the Clinical Global Impression of Severity scale score ⩽3. Written informed consent was obtained from all subjects after a complete explanation of the study. This study was also approved by the Institutional Review Boards at Samsung Medical Center and Seoul National University Bundang Hospital.
Symptom dimension generation
The detailed process was described in our previous manuscript (Baek et al., Reference Baek, Ha, Kim, Cho, Yang, Choi and Hong2019). In brief, clinical symptoms were assessed on a lifetime basis mainly through direct patient interviews using the Korean version of the Diagnostic Interview for Genetic Studies (DIGS) (Joo et al., Reference Joo, Joo, Hong, Hwang, Maeng, Han and Kim2004) or the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID) (Han et al., Reference Han, Ahn, Siong, Cho, Kim, Bae and Park2000), using the DIGS at Samsung Medical Center or the SCID at Seoul National University Bundang Hospital. We selected 41 variables of lifetime characteristics of BD (including clinical course, seasonality, and chronotypes), symptom characteristics during acute episodes, and psychiatric comorbid conditions (six variables related to clinical courses, 29 variables covering lifetime symptoms of mood episodes, and 6 specific comorbid conditions) from the clinical data of 307 patients with BD. After conducting multiple imputations, iterative principal component analysis and varimax rotation were performed. Six factor models were selected based on scree plot acceleration factor rules, Velicer's minimum average partial test, and face validity evaluations. Variables with rotated factor loading <0.2 were excluded. As a result, a total of 37 phenotypes were included in phenotype dimension construction.
The six dimensions generated in our study were cyclicity, depression, atypical vegetative symptoms, elation, psychotic/irritable mania, and comorbidity dimensions. Detailed phenotypes included in each dimension are listed in online Supplementary Table S1.
In addition to phenotype dimension data from the original 307 participants used in phenotype dimension calculation, phenotype dimension scores of additional 160 participants were calculated using the same model. There was no significant difference in each dimension score between the initial samples used to develop the dimension model and those who were later included in the analysis.
Genotyping and imputation
The Korea Biobank Array (Moon et al., Reference Moon, Kim, Han, Hwang, Shin, Park and Kim2019) were used for genotyping DNA samples. We performed sample-level quality control (QC) and removed the indels before imputation. For the sample-level QC, the principal components of genetic ancestry were calculated using PLINK version 1.912 (Purcell et al., Reference Purcell, Neale, Todd-Brown, Thomas, Ferreira, Bender and Sham2007) and used as covariates in the GWAS analysis. Two samples were removed after sample-level QC. Phasing and imputation were performed with Eagle v2.4 and Minimac 4 using the Korean Imputation Service Phase 1 reference panel with this set of shared single nucleotide variants (SNVs) (Hwang, Choi, Won, Kim, & Kim, Reference Hwang, Choi, Won, Kim and Kim2022). After imputation, variants with an imputation quality of R 2 <0.80 or call rate <98% were removed. Finally, 9 556 467 SNVs were used for data analysis.
PRS construction
We constructed PRS using the PRS-CS method (Ge, Chen, Ni, Feng, & Smoller, Reference Ge, Chen, Ni, Feng and Smoller2019). We calculated PRSs for BD (PRS-BD) and five mental illnesses (i.e. schizophrenia [SCZ], PRS-SCZ; major depressive disorder [MDD], PRS-MDD; obsessive-compulsive disorder [OCD], PRS-OCD; anxiety disorder [ANX], PRS-ANX; attention-deficit hyperactivity disorder [ADHD], PRS-ADHD) showing significant genetic correlations with BD in prior studies. PRSs for BD-I (PRS-BD-I) and BD-II (PRS-BD-II) were also separately calculated. As reference data, we used (1) the most recent and the largest GWAS and (2) GWAS with ancestrally appropriate populations. As there is limited availability of East Asian (EA) GWAS data, we could use summary data of EA GWASs for SCZ, BD, and MDD (PRS-SCZ-EA, PRS-BD-EA, and PRS-MDD-EA, respectively). For EA GWAS data of BD, we excluded Korean BD samples in order to avoid possible sample overlaps. Supplementary Table S2 shows information on reference data used to generate PRSs.
Statistical analyses
To explore associations of phenotype dimension scores with PRSs of six mental illnesses, linear regression analyses were conducted with each dimension score as a dependent variable and each PRS as an independent variable. Age, sex, and 10 principal component scores were entered as additional covariates. A p value of <0.05 was considered statistically significant. All statistical analyses were performed using R version 4.02.
Results
Table 2 displays the results of linear regression analyses using PRS from East Asian ancestry.
Linear regression analyses on the association between polygenic risk score for mental illnesses from East Asian ancestry and six phenotype dimensions
PRS, polygenic risk score; SCZ-EA, East Asian Schizophrenia; BD-EA, East Asian bipolar disorder; MDD-EA, East Asian major depressive disorder.
Dependent variable: each dimension score; independent variable: PRS for each mental illnesses stated on each row; age and sex were entered as covariates.
β value is presented with 95% confidence interval.
Bold font indicates statistical significance.
The cyclicity dimension showed a significant negative association with PRS-SCZ-EA (β = −0.25, p = 0.044), indicating BD patients with higher polygenic risk to SCZ had lower episode frequencies of manic and depressive episodes and less rapid cycling courses than those with lower polygenic risk to SCZ. In contrast, the psychotic/irritable mania dimension showed a significant positive association with the PRS-SCZ-EA (β = 0.28, p = 0.001), demonstrating patients with a higher genetic predisposition to SCZ presented with more psychotic and mixed features in their manic episodes.
The elation dimension revealed a significant negative association with PRS-MDD-EA (β = −0.7, p = 0.003). This reflects that patients having a higher score for this dimension with pure manic features of increased energy and elevated mood had lower polygenic risks for MDD. On the contrary, PRS-MDD-EA was significantly associated with the comorbidity dimension (β = 0.28, p = 0.28), indicating that BD patients with higher polygenic risk for MDD demonstrated diverse features of other psychiatric conditions such as anxiety disorders, eating disorders, and OCD.
Table 3 and Fig. 1 summarize the results of linear regression analyses using PRS from the largest GWAS generated with mixed ethnic groups. Supplementary Table S2 summarizes ancestry of the GWAS in detail. The psychotic/irritable mania dimension showed a similar positive association with PRS-SCZ (β = 0.37, p < 0.001) as with PRS-SCZ-EA. Additionally, a significant negative association was found between the depression dimension and PRS-SCZ (β = −0.44, p = 0.031). Unlike the analysis with East Asian PRS data, PRS-BD and PRS-BD-I (β = 0.65, p = 0.03; β = 0.24, p = 0.007, respectively) showed a significant association with the psychotic/irritable mania dimension score.
Linear regression analyses on the association between polygenic risk score for mental illnesses from the largest available GWAS and six phenotype dimensions
PRS, polygenic risk score; SCZ, schizophrenia; BD, bipolar disorder; BD-I, bipolar disorder type I; BD-II, bipolar disorder type II; MDD, major depressive disorder; ANX, anxiety disorder; OCD, obsessive-compulsive disorder; ADHD, attention deficit hyperactivity disorder.
Dependent variable: each dimension score; independent variable: PRS for each mental illnesses stated on each row; age and sex were entered as covariates.
β value is presented with 95% confidence interval.
Bold font indicates statistical significance.
Comparisons of beta and 95% confidence interval from the linear regression analyses with polygenic risk score (PRS) for mental illnesses and six lifetime phenotype dimensions. The X axis denotes reference diagnosis of polygenic risk score; The Y axis denotes beta value. *p value <0.05. BD, bipolar disorder; BD-EA, East Asian bipolar disorder; BD-I, bipolar disorder type I; BD-II, bipolar disorder type II; SCZ, schizophrenia; SCZ-EA, East Asian schizophrenia; MDD, major depressive disorder; MDD-EA, East Asian major depressive disorder; OCD, obsessive compulsive disorder; ANX, anxiety disorder; ADHD, attention deficit hyperactivity disorder.
We did not find a significant association with any PRS score of other psychiatric disorders in the atypical vegetative symptoms dimension. Also, PRS-BD-II, PRS-ANX, PRS-OCD, and PRS-ADHD were not associated with any phenotype dimension.
Figure 1 summarizes the beta values of PRSs in each phenotype dimension. PRSs generated using EA samples generally showed the same directions (plus v. minus) of beta values as PRSs generated using other ethnic samples. Also, analyses using EA samples identified the association quite efficiently despite much smaller sample sizes compared to the largest available GWAS data.
Discussion
The present study demonstrates that phenotype dimensions of BD created using the lifetime clinical characteristics have distinct genomic risk sharing with other mental disorders. This result supports clinical and genetic heterogeneity in BD. At the same time, it suggests the possibility that polygenic overlap between psychiatric illnesses, i.e., inter-diagnoses commonality or pleiotropy, could be identified at the level of specific phenotype dimensions across diagnostic boundaries.
In exploring the genetic basis of psychiatric illnesses, within-diagnosis heterogeneity and inter-diagnoses commonality have been regarded as an important issue to be clarified. Genomic data for various endophenotypes and subtypes within single diseases have been accumulated (Guglielmo, Miskowiak, & Hasler, Reference Guglielmo, Miskowiak and Hasler2021). Also, recent genomic studies started to reveal significant inter-disease correlations between psychiatric illnesses (Docherty et al., Reference Docherty, Mullins, Ashley-Koch, Qin, Coleman, Shabalin and Ruderfer2023). However, research questions or designs incorporating the two issues into a single study have yet to be attempted much. For this approach, BD is an optimal complex genetic disorder, given its diversity in clinical manifestation and highly overlapping clinical features with other psychiatric illnesses, such as psychotic disorders, anxiety disorders, ADHD, eating disorders, and substance abuse (Nierenberg et al., Reference Nierenberg, Agustini, Kohler-Forsberg, Cusin, Katz, Sylvia and Berk2023).
The application of deep phenotyping in genomic studies for BD has been tried only recently. So, a direct comparison of the current results with previous studies might be difficult. However, our study findings are in line with several prior study results with similar research questions.
The PRS-SCZ and PRS-SCZ-EA both showed significant positive associations with the psychotic/irritable mania dimension. This result is consistent with European study findings demonstrating associations between psychosis dimensions in BD and PRS-SCZ (Coombes et al., Reference Coombes, Markota, Mann, Colby, Stahl, Talati and Biernacka2020; Markota et al., Reference Markota, Coombes, Larrabee, McElroy, Bond, Veldic and Biernacka2018). In a study that generated phenotype dimensions across SCZ, BD, and MDD, the psychotic feature was also associated with PRS-SCZ (David et al., Reference David, Stein, Andlauer, Streit, Witt, Herms and Forstner2023). Together with these findings, the current result suggests that psychotic, irritable, and mixed features of mania might be a phenotypic constellation sharing the genetic risk with SCZ.
In contrast, the cyclicity dimension showed a significant negative association with PRS-SCZ-EA. This suggests that BD patients with more genetic predisposition for SCZ are less likely to show high cyclicity in their illness courses. In our previous study, the cyclicity dimensions were significantly more presented in BD-II compared to BD-I (Baek et al., Reference Baek, Ha, Kim, Cho, Yang, Choi and Hong2019). It might be compatible with recent genomic study findings showing a higher genetic correlation with SCZ in BD-I than in BD-II (Li, Li, & Chen, Reference Li, Li and Chen2022).
In the present study, PRS-MDD-EA demonstrated a significant negative association with the elation dimension (p = 0.003). Subjects with high scores in this dimension would present higher and more dominant manic features. Therefore, the negative correlation of this dimension with PRS-MDD-EA suggests that genetic loci contributing to the development of MDD might have a symptom- or course-modifying role in BD.
Regarding the comorbidity dimension, the term ‘comorbidity’ might be a misnomer. Though our previous and current studies used this term, this dimension could, in fact, reflect miscellaneous BD features overlapping with eating disorders, anxiety disorders, and OCD. A significant association of this dimension with PRS-MDD-EA might be compatible with the high prevalence of anxiety, obsessive-compulsive, and eating disorder symptoms in MDD (Kessler et al., Reference Kessler, Gruber, Hettema, Hwang, Sampson and Yonkers2008; Overbeek, Schruers, Vermetten, & Griez, Reference Overbeek, Schruers, Vermetten and Griez2002; Welch et al., Reference Welch, Jangmo, Thornton, Norring, von Hausswolff-Juhlin, Herman and Bulik2016). Depression is also known as having intricate genetic architecture (Flint & Kendler, Reference Flint and Kendler2014) with high genetic heterogeneity. It would be worthwhile to evaluate the association between this dimension and PRS for eating disorders, anxiety disorders, and OCD. However, we could not find reference GWAS data on EA populations for those diagnoses. We tried to use European data, which showed no significant correlations. Of note, a prior European study indicated a noteworthy association between comorbid ANX in BD and PRS-ANX (Lopes et al., Reference Lopes, Zhu, Purves, Song, Ahn, Hou and McMahon2020).
The atypical vegetative symptoms dimension did not show any significant associations with PRS examined in our study. Notably, this dimension was quite independent, not showing any significant association with other dimensions in our previous study (Baek et al., Reference Baek, Ha, Kim, Cho, Yang, Choi and Hong2019). In a recent study with UK biobank data, atypical depression showed a significant positive association with PRS for immune-metabolic traits compared to non-atypical depression (Badini et al., Reference Badini, Coleman, Hagenaars, Hotopf, Breen, Lewis and Fabbri2022).
This study highlights the need for establishing genomic data sets of diverse ancestral origins. Analysis using European reference data did not show the same results as those using EA reference data (Table 2, Table 3). For BD-I and BD-II subtypes, anxiety disorders, OCD, and ADHD, East Asian populations GWAS data with sufficient sample sizes were not available. Also, there were no Korean GWAS data to be used as reference data for the current analysis.
Our study findings have several limitations. First, the sample size of the subjects might not be enough to examine heterogenous phenotype dimensions. Second, as mentioned previously, limited large-scale GWAS data from East Asian or Korean populations could also affect the study findings. In particular, large-scale GWAS for Korean patients with mental illnesses were not available. Third, despite significant associations observed in our study, the amount of variance explained using PRS was still small. Lastly, we did not have validation samples.
Notwithstanding these limitations, this study provided insight into how intra-diagnosis heterogeneity and inter-diagnoses commonality issues can be integrated in understanding genetic basis of psychiatric disorders. The strength of this study is a deep phenotyping and dimensional approach in defining phenotypes. In future genomic studies, deep phenotyping across psychiatric illnesses beyond diagnostic boundaries is required for diverse ancestral populations. Through these efforts, we will be able to approach closer to the goal of biology-based reclassification of mental illnesses.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S003329172400120X.
Acknowledgements
We would like to thank Jinseob Kim (Zarathu Co., Ltd., Seoul, Republic of Korea) for statistical advice.
Authors' contributions
JHB and KSH designed the overall study and generated the hypothesis. JHB, THH, KH, and HKS participated in participant recruitment. HWJ took part in data collection. EYC collected genomic samples and took part in the genomic analysis part. DL conducted statistical analyses. JHB wrote the manuscript. All authors took part in the final manuscript editing process.
Funding statement
This study was supported by the Original Technology Research Program for Brain Science through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (grant number: NRF-2019M3C7A1030624).
Competing interests
The authors have no relevant financial or non-financial interests to disclose.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Open access
