Polygenic Risk and Episode Polarity Among Individuals With Bipolar Disorder

Data for this study were obtained from the Lundbeck Foundation Initiative for Integrative Psychiatric Research iPSYCH2015 case-cohort sample, which is a newly updated version of the iPSYCH2012 sample (19). The iPSYCH2015 sample was drawn from a base population of all singletons born in Denmark between 1981 and 2008 who survived to their first birthday and had known mothers. The sample combines genetic and register-based data for all individuals within the base population who were diagnosed with certain mental disorders (affective disorder, schizophrenia, schizophrenia spectrum disorder, attention deficit hyperactivity disorder, autism, anorexia nervosa, or postpartum mental disorder) at Danish psychiatric hospitals during the period from January 1, 1995, to December 31, 2015 (N=93,608) (20). The sample also includes a randomly selected subset of 50,615 individuals drawn from the Danish base population. Members of the iPSYCH2015 case cohort were identified using data from the Danish Civil Registration System (21), which contains demographic and vital status information on all individuals born or residing in Denmark as well as the unique 10-digit personal registration number, which allows for individual linkage of information from all population-based registers in Denmark. Data on psychiatric diagnoses were obtained from the Danish Central Psychiatric Research Register (DCPRR), which includes information on all psychiatric diagnoses assigned in relation to inpatient treatment at psychiatric hospitals in Denmark since 1969 as well as outpatient and emergency treatment since 1995 (22). Diagnoses in the DCPRR are based on ICD-10 and are assigned by the treating psychiatrist as part of routine clinical care.

DNA for members of the iPSYCH2015 case cohort was obtained from dried blood spots collected at birth as part of routine clinical testing for congenital disorders and stored in the Danish Newborn Screening Biobank (23). iPSYCH2012 samples were genotyped using the Illumina Infinium Psychchip v1.0 array (Illumina, San Diego), and iPSYCH2015 samples were genotyped using the Illumina Global Screening array. In both samples, genotyping was carried out according to the manufacturer’s protocols. Quality control was performed as follows: First, samples with a call rate ≤95% were excluded. Second, the genetically obtained sex was compared to the sex recorded in the Danish Civil Registration System. All samples that failed to match with the recorded sex without valid reason (e.g., aneuploidy or change of gender) were excluded from the sample. Imputation was conducted using the RICOPILI pipeline (24). After removing single-nucleotide polymorphisms (SNPs) with minor allele frequency <0.01 and Hardy-Weinberg p value <10⁻⁶ (χ² [df=1] test statistic p_HWE), we restricted SNPs to the HapMap3 variants. The final data set consisted of 1,184,443 SNPs and was then used for the genetic analyses. We then calculated the first five genetic principal components using the autoSVD algorithm available in the R package bigsnpr (25), following the guidelines outlined by Privé et al. (26).

A flowchart of the sample selection process is shown in Figure S1 in the online supplement. We identified all individuals from the iPSYCH2015 study who 1) received a main, secondary, or “underlying cause” diagnosis of bipolar disorder (ICD-10 codes F30–F31, F38.00) at age 10 or later in relation to inpatient or outpatient treatment at a psychiatric hospital in Denmark between January 1, 1995, and March 31, 2017; 2) had no prior diagnosis of schizophrenia or related disorders (ICD-10 codes F20–F29); and 3) were successfully genotyped. The majority of individuals in the iPSYCH2015 sample are of European ancestry. For the sake of statistical power, we did not exclude individuals of non-European ancestry, as previous research in this cohort has demonstrated that such exclusions have limited impact on the results (13, 27). Instead, we controlled for ancestry using principal components in the analyses and also performed sensitivity analyses excluding individuals of non-European ancestry. The final study sample comprised 2,705 individuals.

Information on episode polarity was obtained from the DCPRR (22). Episodes were defined on the basis of either an inpatient contact (admission) or a series of outpatient contacts. Episode polarity (depressive, manic, mixed, unspecified) was defined according to the ICD-10 diagnostic codes registered in the DCPRR. For post hoc analyses, we also created subcategories for psychotic symptoms in any polarity, psychotic mania, and psychotic depression. The ICD-10 codes used to define polarities are listed in Table S1 in the online supplement. Unipolar depressive episodes were included in the (bipolar) depression polarity category, as patients with bipolar disorder frequently receive a unipolar depression diagnosis prior to their first bipolar diagnosis, and these episodes can reasonably be considered bipolar depressive episodes in retrospect.

The DCPRR describes admissions and series of outpatient contacts that occasionally overlap in time or immediately follow each other. In the event of two separate but overlapping or back-to-back admissions or outpatient contacts with a diagnosis of bipolar disorder, we chose to merge those records with the same polarity that were separated by a period of less than 8 weeks. However, if the records had different polarities, we considered them separate episodes regardless of the interval between them. For this process of merging hospital contacts, we considered psychotic mania and psychotic depression as separate categories.

PRSs for bipolar disorder, major depression, and schizophrenia were calculated using a meta-PRS method (28). The meta-PRS method combines two independently constructed PRSs, namely, a PRS trained on external (not including iPSYCH) genome-wide association study (GWAS) summary statistics and an internal PRS derived from individual-level genetic data (28). The external PRSs were generated using the most recent GWAS summary results from the Psychiatric Genomics Consortium (PGC) meta-analyses for bipolar disorder (14), major depression (29), and schizophrenia (30) using LDpred2-auto (31). The internal PRS was trained on the iPSYCH sample using BOLT-LMM (32), with a fivefold cross-validation scheme to avoid overfitting the model. The external and the internal PRSs were then combined into a meta-PRS with the formula w₁*PRS_internal+w₂*PRS_external. The weights w₁ and w₂ were calculated from the individual-level data with a linear regression model using nested cross-validation, as described in the meta-PRS method from Albiñana et al. (28). This method was chosen because it increases polygenic prediction accuracy compared to methods using external data alone (28). The final PRSs were standardized (mean=0, SD=1) according the mean and standard deviation from the iPSYCH2015 random subcohort.

Associations between episode polarity and PRS variables were examined in three ways. First, we ran binary logistic regression models to estimate the associations between PRS variables and the odds of having a given polarity versus not having that polarity at the first recorded hospital contact for bipolar disorder. Second, we ran binary logistic regression models examining associations between PRS variables and the odds of having a given polarity at any hospital contact for bipolar disorder versus never having it. Third, we ran negative binomial regression models to examine the associations between PRS variables and the number of episodes with a given polarity (or with psychotic symptoms) recorded for each individual. All models were adjusted for gender, age at first bipolar diagnosis, the first five principal components, and genotyping platform. Logistic regression models were additionally adjusted for the total number of episodes. All three PRS variables were included in the models to control for shared genetic liabilities between the disorders. Individual associations between PRS variables and polarity (adjusted for the first five principal components and genotyping platform) are shown in Figures S2 and S3 and Table S2 in the online supplement. Analyses were performed using SAS, version 9.4.0.

To assess how our sampling choices affected the association between PRSs and episode polarity, we ran two separate sensitivity analyses. First, we examined the association only among the fraction of the sample who had parents born in Denmark. Second, we examined the associations between PRSs and polarity of episodes in the 3-year period following initial diagnosis among the subset of the sample with at least 3 years of follow-up, to ensure equal follow-up time for all patients.

The Danish Scientific Ethics Committee, the Danish Health Data Authority, the Danish Data Protection Agency, the Danish Neonatal Screening Biobank steering committee, and Statistics Denmark approved this study (20).

The demographic and clinical characteristics of the study sample are summarized in Table 1. The patients in the sample were young (median age, 22 years) and predominantly female (65%). Follow-up time ranged from 1 day to 22 years, with a median of 5 years (interquartile range, 3–8 years). Most patients in the sample (83%) were followed for 3 or more years. The majority of the patients (72%) were treated in an outpatient setting, and almost half (45%) either had a depressive episode as the first registered episode of bipolar disorder or were registered with a diagnosis of unipolar depression prior to the first diagnosis of bipolar disorder.

Approximately one-third (N=971, 36%) of the sample had only one episode of bipolar disorder treated in a hospital setting during the study period. Among the patients with two or more episodes, 16% (N=278) had episodes of the same polarity type, 63% (N=1,090) had episodes of two different polarities, 18% (N=309) had episodes of three polarities, and 3% (N=57) had episodes in all four polarity categories. Figure 1 shows the distribution of polarity patterns among the 1,734 patients with bipolar disorder who were registered with two or more episodes during the study period. Depressive plus unspecified bipolar episodes constituted by far the most common polarity pattern observed in the sample.

The mean observed PRSs for major depression (0.36 [SD=0.95]), bipolar disorder (0.29 [SD=0.96]), and schizophrenia (0.14 [SD=0.95]) were each elevated in patients with bipolar disorder relative to the background population in which mean=0 and SD=1 (standardized values). As in past studies (14), we found statistically significant correlations between all three PRS variables (see Table S3 in the online supplement), with the strongest correlation between bipolar disorder PRS and schizophrenia PRS (Pearson’s r=0.52, p<0.0001). Similar to a recent study (33), we found that higher bipolar disorder PRS (β=0.05, p=0.009) and schizophrenia PRS (β=0.03, p=0.0484) but not major depression PRS (β=−0.01, p=0.46) were associated with more episodes of bipolar disorder, regardless of polarity.

Figure 2 shows the associations between PRSs and episode polarity. Higher bipolar disorder PRS (odds ratio per standard deviation increase in bipolar disorder PRS, 1.23, 95% CI=1.09–1.38, p=0.0007) and schizophrenia PRS (odds ratio=1.13, 95% CI=1.01–1.27, p=0.04) were associated with statistically significantly increased odds of experiencing any episodes with manic polarity during the study time frame. Higher major depression PRS was associated with statistically significantly increased odds of experiencing any episode of depressive (odds ratio=1.11, 95% CI=1.01–1.23, p=0.03) and mixed polarity (odds ratio=1.15, 95% CI=1.03–1.28, p=0.01) but decreased odds of manic polarity (odds ratio=0.76, 95% CI=0.69–0.84, p<0.0001). There were no statistically significant associations between any of the three PRSs and unspecified polarity. The pattern of results was similar for episode polarity at the first diagnosis of bipolar disorder (Figure 2A) and episode polarity at any episode (Figure 2B), although the confidence intervals were wider for first diagnosis.

The pattern of results was similar when we looked at the number of episodes of a given polarity (Table 2; see also Figure S4 in the online supplement). Higher bipolar disorder PRS (β=0.21, p<0.0001) and schizophrenia PRS (β=0.14, p=0.01) were associated with more manic episodes. Higher major depression PRS was associated with more mixed episodes (β=0.09, p=0.048) and fewer manic episodes (β=−0.25, p<0.0001). Although the association with number of depressive episodes trended in the same direction, it was not statistically significant (β=0.04, p=0.09). Again, there were no statistically significant associations between any of the three PRSs and episodes of unspecified polarity.

Figure 3 shows the associations between PRSs and manic or depressive episodes with and without psychotic symptoms. Higher bipolar disorder PRS was associated with increased odds of manic polarity, both with (odds ratio=1.27, 95% CI=1.06–1.54, p=0.01) and without (odds ratio=1.22, 95% CI=1.07–1.38, p=0.002) psychotic symptoms. Schizophrenia PRS, however, was associated with mania only when psychotic symptoms were present (odds ratio=1.27, 95% CI= 1.05–1.54, p=0.01). In contrast, major depression PRS was positively associated with depressive polarity when psychotic symptoms were absent (odds ratio=1.13, 95% CI=1.02–1.24, p=0.01) and negatively associated with depressive polarity in the presence of psychotic symptoms (odds ratio=0.80, 95% CI=0.66–0.96, p=0.02). Similar patterns were observed for the number of episodes (see Table S4 in the online supplement).

In the subsample of patients with Danish-born parents (N=2,366), the findings were similar to those of the main analyses reported above (see Tables S5–S8 in the online supplement). In the subsample of patients with at least 3 years of follow-up (N=1,926), all findings were similar to those of the main analyses except that schizophrenia PRS was also significantly associated with increased odds of nonpsychotic mania (see Tables S9–S12 in the online supplement).