Anxiety and bipolar disorder (BP) are highly comorbid conditions1,2, but the basis of this comorbidity is uncertain. There is some overlap in diagnostic criteria, but this is not substantial.

It is possible that both conditions share environmental risk factors but these do not appear to be much more common in people with comorbid anxiety and BP than in those with BP alone1. Familial co-aggregation of anxiety and bipolar disorder3,4 suggests shared genetic risk factors, but this has not been tested using molecular genetic methods.

Comorbid anxiety has important clinical implications for people living with BP. Several studies have reported unfavorable outcomes in BP with comorbid anxiety, including more frequent mood episodes5,6, more severe depressive episodes7, higher rates of substance abuse6, less favorable treatment response6,8,9, and increased suicide attempts10,11,12. The United States National Comorbidity Survey found that the diagnosis of anxiety disorder is particularly frequent among individuals with suicide attempts13. A deeper understanding of the comorbidity between anxiety and BP could shed light on these clinical problems and could in future point toward new treatment approaches.

BP is a highly heritable disease (~80%), and many genome wide association studies (GWAS) have been published, together finding close to 40 replicated genetic loci associated with the disorder14,15. Molecular genetic studies of anxiety are still relatively uncommon16, with many studies having small samples and as a result inconsistently replicated results. Two recent large-scale GWAS of anxiety and anxiety disorders have made considerable progress in identifying robust genetic risk factors17,18.

Both studies found that anxiety is a highly polygenic trait, like most other psychiatric conditions, with thousands of associated alleles that each contribute a small share to the total risk. Polygenic risk scores (PRS) are a powerful tool that capitalize on this cumulative genetic risk. One way in which these scores can be used is to identify shared genetic liability—if people with one disorder have a high polygenetic risk score for a second disorder it may indicate a shared genetic liability. In the field of psychiatric disorders, the PRS was initially used by Purcell et al. in a study of schizophrenia19. The PRS approach has since been applied to BP20,21,22, Schizoaffective Disorder23, Schizophrenia21, and Major Depression Disorder24, among other traits. PRS have also demonstrated a shared genetic risk factors among many of these disorders25,26.

The primary aim of the present study is to examine the genetic relationship between genetic risk for anxiety disorders and anxiety comorbidity in bipolar disorder. We aim to answer the question of whether anxiety comorbidity in bipolar disorder reflects distinct genetic risk factors, such that this comorbidity occurs if an individual has high genetic liability for both bipolar disorder and anxiety disorder. Alternatively, this comorbidity may reflect an alternative clinical manifestation (pleiotropy) of the known genetic risk for bipolar disorder. Our secondary research question is whether genetic risk for anxiety contributes to the increase in suicide attempts among people with comorbid bipolar and anxiety disorders. The results shed light on the mechanisms that underlie variable clinical presentations of BP and might help inform clinical management of patients with comorbid BP and anxiety disorders.

Subjects and methods

Sample description

The present study used existing data from five independent samples: First, a GWAS of lifetime anxiety disorder undertaken in the UK Biobank18, referred to herein as the “discovery sample”. We also include three additional samples; the Genetic Association Information Network (GAIN) bipolar, the Translational Genomics Research Institute (TGEN), and the Swedish Bipolar Disorder Cohort (SWEBIC), which we combined and referred to herein as the “target samples.” Ascertainment, diagnosis, and genotyping of these samples have been previously described18,27,28. Further details of the target samples are presented in Table 1. Written informed consent was obtained from all participants. Each study was approved by a local Ethics Committee.

Table 1 Sample description.

The discovery sample comprised the largest available GWAS on anxiety disorders, including individuals of western European-ancestry who took part in the UK Biobank online mental health follow-up questionnaire. This sample represented people reporting a lifetime diagnosis by a professional of panic disorder (PD), agoraphobia, social phobia (SP), social anxiety disorder (SAD), or generalized anxiety disorder (GAD)18.

Target sample: GAIN/TGEN/SWEBIC

The GAIN, TGEN, and SWEBIC samples included 3369 individuals of European ancestry assessed by a comprehensive psychiatric interview. All individuals included in the present analysis had received a final DSM-III-R/IV diagnosis of Bipolar I (BPI) or Schizoaffective disorder—bipolar type (SA-BP).

The GAIN and TGEN participants (N = 1638) were assessed by a comprehensive psychiatric interview29 supplemented by family informant and medical record data. Phenotypic data was harmonized and compiled within the Bipolar Disorder Phenome Database30. Four individuals were excluded owing to missing genotype or phenotype data.

The SWEBIC participants (N = 1731) were assessed according to the DSM-IV criteria. Genotype data have been previously reported31.


Anxiety comorbidity

Anxiety comorbidity was defined as a lifetime diagnoses of PD, Agoraphobia, SP, SAD, and/or GAD occurring before or after a lifetime diagnosis of BP as assessed during comprehensive psychiatric interview described above.

Suicidal attempts

Suicide attempt was defined as a lifetime self-report of self-harm with lethal intent, reported during a structured interview according to previous studies32.

Polygenic risk score (PRS)

PRS for anxiety were calculated using summary statistics from the UK Biobank discovery sample18. We followed the guide for genomic profile risk scoring (Box 1 of reference25) to calculate the PRS. In the discovery sample, p value informed pruning of correlated SNPs was done from summary statistics by use of LD-based clumping (r2 threshold < 0.25 across a 500 kilobase window) as implemented in PLINK33, using the 1000 Genomes European-ancestry sample (excluding Finns) as a reference panel. The pruned SNPs were used to compute the anxiety PRS in the target samples based on 65,415 shared SNPs. Alleles were weighted by the effect sizes from the discovery sample (log[OR]), including all SNPs that were significant at p < 0.5.

We used the same approach to calculate a BP PRS in the target samples, based on summary statistics available from a recently-published BP GWAS34. P value threshold from the discovery sample was set to 0.2, since this captured the most phenotypic variance.

PRS were standardized using means and SDs from the respective distributions.

BP-PRS showed a small but significant correlation with anxiety-PRS (Pearson r = 0.07, p < 0.0001).

Statistical methods

Association between anxiety PRS and phenotypic information was tested by logistic regression, as implemented in SAS vs. 9.4. Population stratification was corrected with ancestry principal components analysis (PC) based on the variance-standardized relationship matrix in Plink33, using the first 5 PCs as covariates. Sex was used as a covariate for anxiety comorbidity, since we observed an association between Anxiety and Sex in the target (p < 0.001) sample, consistent with previous findings8,9. In addition, the source study (GAIN,TGEN, SWEBIC) was included as a covariate to control for any batch effects.

PRS association p values were Bonferroni-corrected for four different tests: Anxiety PRS and BP PRS versus comorbid anxiety or suicide attempts (p < 0.0125). Only one p value threshold was used for calculating PRS in each of the discovery samples (p < 0.5 for anxiety PRS, p < 0.2 for bipolar PRS, as noted above).

Power calculations

The power of PRS was carried out in AVENGEME35,36. The proportion of phenotypic variance explained by common SNPs was estimated from the target sample. The power of anxiety PRS and BP PRS to predict anxiety on BP sample was 90 and 74%, considering the h2SNP for anxiety (12%) obtained from the discovery sample18. The power of anxiety PRS and BP PRS to detect suicide attempt was 91 and 76%, respectively, given an h2SNP of 10% for suicide attempt32.


Sample characteristics

A total of 3369 patients with a diagnosis of BPI or SA-BP, from 3 different studies, comprised the target sample (Table 1). The studies were similar in terms of European ancestry and sex ratio. A lifetime diagnosis of any anxiety disorder (“Any Anxiety”) was present in 51.6% of the target sample (Table 1). 41.92% of the target sample reported (N = 1390) one or more suicide attempts (Table 1).

Anxiety comorbidity

Anxiety-PRS was significantly and positively associated with “Any Anxiety” in the target sample (Fig. 1a). Each unit increase in anxiety-PRS led to a 15.3% increase in the odds of a comorbid anxiety disorder (OR = 1.153, 95% CI: 1.048–1.269, p = 0.0034, Nagelkerke’s R2 = 0.08).

Fig. 1: Association of anxiety or bipolar disorder polygenic scores with anxiety comorbidity and suicide attempts.
figure 1

a The association between anxiety polygenic score (PRS) and anxiety comorbidity in patients with bipolar disorder. Anxiety comorbidity was evaluated as a categorical phenotype (“Any Anxiety”). b The association between anxiety PRS and suicide attempts in patients with bipolar disorder. c The association between bipolar PRS and anxiety comorbidity in patients with bipolar disorer. d The association between bipolar PRS and suicide attempts in patients with bipolar disorder. Logistic regression results are reported as odds ratios (OR). 95% confidence intervals (CI) are also shown. The PRS data are distributed in quintile.

In contrast, we detected no significant main effect of BP PRS on anxiety comorbidity in the target sample (Fig. 1c) (OR = 1.118, 95% CI: 1.000–1.248, p = ns, Nagelkerke’s R2 = 0.05).

Suicidal behavior

Anxiety PRS was significantly and positively associated with suicide attempts (SA) in the target sample (Fig. 1b) (OR = 1.106, 95% CI: 1.030–1.189, p = 0.0055, Nagelkerke’s R2 = 0.06). In contrast, we detected no significant association between BP PRS and SA in the target sample (OR = 1.078, 95% CI: 0.998–1.164, p = ns, Nagelkerke’s R2 = 0.02) (Fig. 1d).


To our knowledge, this is the first study to address the impact of genetic risk for anxiety on the clinical course of BP and to provide evidence for a molecular genetic distinction between bipolar disorder with and without comorbid anxiety. The results suggest that bipolar disorder with comorbid anxiety reflects a dual burden of bipolar and anxiety-related genes. Clinical approaches that address this dual genetic burden may help improve outcomes in people living with comorbid bipolar and anxiety disorders.

There were two main findings. First, anxiety disorder comorbidity in bipolar disorder was associated with anxiety PRS in the samples we studied. Second, anxiety PRS was also associated with suicide attempts in BP.

In previous population-based studies, about 60% of bipolar probands met criteria for comorbid anxiety disorders1,2,37. In the present study, we found that anxiety risk alleles play a significant role in this comorbidity. Our findings thus contribute to heavily-debated questions concerning the nosologic relationship between mood and anxiety disorders. Our data support the view that anxiety comorbidity in BP is due, at least in part, to the same common genetic risk variants as anxiety in general, while bipolar risk alleles do not. This finding is consistent with the known low genetic correlations between bipolar disorder and anxiety-related traits38, which suggests that the disorders share few genetic risk factors.

A second important result arising from the polygenic scoring analyses is the association between anxiety PRS and suicidal behavior. In our target sample we observed a positive association between suicide attempts and genetic risk for anxiety in individuals with a diagnosis of bipolar disorder. These results are consistent with the increased rates of suicidal behavior in people with anxiety disorders reported by population-based studies11 and call attention to the importance of monitoring suicide risk in people with BP and comorbid anxiety.

Our results show that anxiety PRS accounts for only a small proportion of overall anxiety comorbidity in BP, so other factors may be involved. One factor could be assortative mating where people with BP are more likely to partner with people with anxiety disorders, leading to increased comorbidity in the offspring. However, Nordsletten et al. have reported a maximal rate of assortative mating between anxiety and Bipolar I of 18%39. There is also a small but significant genetic correlation between anxiety and BP, which suggests that BP risk alleles may also contribute to comorbid anxiety18. Correlated nongenetic risks may also contribute to comorbidiy. Nongenetic risk factors that might contribute to both BD and anxiety disorders include social isolation, unstable relationships, socioeconomic disadvantage, and traumatic life events40. Some anxiety comorbidity may also arise as a complication of BD or its treatment. For example, according to the staging model41, anxiety may manifests as a residual symptom following an acute mood episode.

This study should be viewed in the light of several limitations. This is a cross-sectional study, that relied upon retrospective reports. Although the best-estimate diagnosis procedure considers convergent data from family informants and medical records, data on GAD was not available, and the results may underestimate any association between anxiety PGS and the full range of comorbid anxiety disorders. The target sample was underpowered to detect association between BP-PRS with both comorbid anxiety and suicide attempt, so the failure to detect a significant result in this study does not rule out a contribution of BP risk alleles to those trait. Moreover, the overall direction of the relationship between comorbid anxiety or suicide attempt is similar for both ANX-PRS and BP-PRS (Fig. 1c, d), suggesting that a larger target sample may uncover significant associations with BP-PRS. While we used the largest published BP GWAS available, most controls were not formally screened for anxiety. This would not create a false-positive result, but would further reduce the power of a BP-based PRS to detect anxiety in another sample. The results should be considered preliminary until replicated in an independent sample. The lack of a replication sample here reflects the scarcity of available samples worldwide that have been fully characterized for both BP and anxiety. Anxiety-PRS indexes only a small proportion of the variance in anxiety disorder risk. This is an inherent weakness with the PRS method but is expected to improve with increased size of the discovery samples. A further limitation of this study is the reliance of the anxiety sumstats on self-report of a past diagnosis made by an unknown professional or retrospective recall of lifetime symptoms, both of which may contain error. However, we note that Purves et al.18 report high genetic correlation between these summary statistics and several other anxiety phenotypes, reassuring us that this case selection approach has utility.

In conclusion, anxiety and suicidal behavior in bipolar disorder are influenced by genetic risk factors involved in anxiety disorders. Patients with comorbid bipolar and anxiety disorders thus carry a dual genetic burden, suggesting the need for clinical approaches that address both disorders. More research is needed to understand the interplay between genetic and non-genetic influences on the clinical presentation and course of BP. Better powered discovery samples for both BP and anxiety will be needed to further elucidate this relationship.