Genetic association study of psychotic experiences in UK Biobank

Psychotic experiences, such as hallucinations and delusions, are reported by approximately 5%-10% of the general population, though only a small proportion of individuals develop psychotic disorders such as schizophrenia or bipolar disorder. Studying the genetic aetiology of psychotic experiences in the general population, and its relationship with the genetic aetiology of other disorders, may increase our understanding of their pathological significance. Using the population-based UK Biobank sample, we performed the largest genetic association study of psychotic experiences in individuals without a psychotic disorder. We conducted three genome-wide association studies (GWAS) for (i) any psychotic experience (6123 cases vs. 121,843 controls), (ii) distressing psychotic experiences (2143 cases vs. 121,843 controls), and (iii) multiple occurrence psychotic experiences (3337 cases vs. 121,843 controls). Analyses of polygenic risk scores (PRS), genetic correlation, and copy number variation (CNV) were conducted to assess whether genetic liability to psychotic experiences is shared with schizophrenia and/or other neuropsychiatric disorders and traits. GWAS analyses identified four loci associated with psychotic experiences including a locus in Ankyrin-3 (ANK3, OR=1.16, p=3.06 × 10−8) with any psychotic experience and a locus in cannabinoid receptor 2 gene (CNR2, OR=0.66, p=3.78×10−8) with distressing psychotic experiences. PRS analyses identified associations between psychotic experiences and genetic liability for schizophrenia, major depressive disorder, and bipolar disorder, and these associations were stronger for distressing psychotic experiences. Genetic correlation analysis identified significant genetic correlations between psychotic experiences and major depressive disorder, schizophrenia, autism spectrum disorder and a cross-disorder GWAS. Individuals reporting psychotic experiences had an increased burden of CNVs previously associated with schizophrenia (OR=2.04, p=2.49×10−4) and of those associated with neurodevelopmental disorders more widely (OR=1.75, p=1.41×10−3). In conclusion, we identified four genome-wide significant loci in the largest GWAS of psychotic experiences from the population-based UK Biobank sample and found support for a shared genetic aetiology between psychotic experiences and schizophrenia, but also major depressive disorder, bipolar disorder and neurodevelopmental disorders.


Introduction
Psychotic experiences, such as hallucinations and delusions, are features of psychiatric disorders, for example schizophrenia and bipolar disorder, but they are also reported by approximately 5%-10% of the general population 1,2 . Psychotic experiences are only considered to be symptoms of psychiatric illness if they co-occur with other features of that disorder, including some aspect of psychosocial impairment. It is currently unclear whether psychotic experiences in the general population are (i) on a spectrum that at the extreme relates specifically to schizophrenia, (ii) largely unrelated to the psychotic symptoms experienced in schizophrenia and other major mental disorders, or (iii) related to liability to major mental disorders more generally.
Twin studies and genome-wide association studies (GWAS) have provided evidence that psychotic experiences are heritable (30-50% from twin studies 3-5 , 3-17% for SNP-heritability estimates 6,7 ), indicating that common genetic variants play a role in their aetiology. There have been three GWAS of psychotic experiences to date, all conducted in adolescent samples [7][8][9] with relatively small sample sizes (largest total sample n=10,098) and no reported genome-wide significant findings. These studies have focused on adolescence under the hypothesis that psychotic experiences at this age could be associated with an increased risk of mental health disorders in later life 10 . Whilst there was an initial assumption that psychotic experiences in adolescence would specifically increase the risk for schizophrenia, evidence suggests a nonspecific increased risk for a broader psychopathology 10 , suggesting that research on psychotic experiences may have an important role in understanding the pathway to a wide array of clinical diagnoses 5 . However, to date no study has found strong evidence for association between genetic liabilities for schizophrenia or any other mental disorder with psychotic experiences 7,8,[11][12][13] .
Although many individuals with a lifetime history of psychotic experiences have their first experience in adolescence, nearly a quarter of first onset psychotic experiences occur after 40 years of age 14 . It is possible that psychotic experiences first occurring in adulthood have a different aetiology to those occurring in adolescence. Our aims were to use the UK Biobank to (i) identify genetic loci associated with psychotic experiences reported by adults in a populationbased study, and (ii) to determine whether genetic liability to psychotic experiences is shared with schizophrenia and/or other neuropsychiatric disorders and traits. We selected three primary phenotypes for GWAS: (i) any psychotic experience defined as a positive response to any of the four symptom questions (UKB field IDs: 20471, 20463, 20474, 20468), (ii) a distressing psychotic experience, defined as any psychotic experience that was rated as 'a bit', 'quite' or 'very' distressing (UKB field ID: 20462), (iii) multiple occurrence psychotic experiences defined as any psychotic experience that occurred on more than one occasion (UKB field IDs: 20473, 20465, 20476, 20470). As a comparator group, we included individuals who provided a negative response to all four psychotic experience symptom questions. Supplementary Figure 1 details the overlap in samples between these phenotypes.

Genetic data
Genetic data for the study participants was provided by UK Biobank and the imputation and quality control procedures are fully described elsewhere 16 . The data release contained 488,377 participants assayed on either the UK Biobank Axiom or the UK BiLEVE Axiom purpose-built arrays at the Affymetrix Research Services Laboratory. Standard quality control procedures were applied prior to imputation using Haplotype Reference Consortium (HRC) 17

GWAS analysis
To identify genetic risk variants for psychotic experiences, association analysis was performed in SNPTEST (v2.5.4) 22 using bgen v1.2 imputed dosage data 23 . Over 7.5 million SNPs were included in each GWAS. An additive logistic regression model was used including as covariates genotyping array, the top five principal components (as recommended for most GWAS approaches 24 ), and any additional principal components from the first 20 that were nominally associated (p < 0.05) with the GWAS phenotype in a logistic regression. To obtain relatively independent index SNPs, linkage disequilibrium (LD) clumping was performed in PLINK 25 (r 2 < 0.1, p < 1x10 -4 , window size < 3MB) for each GWAS using a reference panel of 1000 randomly selected UK Biobank individuals with confirmed European ancestry. Functional annotation was conducted using FUMA 26 . LDSC 27 was used to calculate the LD score intercept and heritability on the observed scale using the summary statistics from each psychotic experience GWAS.

Validation analyses in ALSPAC cohort
To assess the reproducibility of the psychotic experience GWAS, we used the summary statistics from the any psychotic experience GWAS to target psychotic experiences in the Avon Longitudinal Study of Parents and Children (ALSPAC) cohort 28,29 , which have been previously described 11,30 . A psychotic experience PRS was generated for each ALSPAC participant using the accepted method 31

Polygenic risk scores
To examine the relationship between psychotic experiences and genetic risk for psychiatric and personality traits from publicly available GWAS datasets (that did not include UK Biobank where possible) included; schizophrenia 32 , bipolar disorder 33 , major depressive disorder 34 , neuroticism 35 , and intelligence 36 . PRSs were generated using the method described by the PGC 31 and detailed in Supplementary Methods. The intelligence GWAS excluded UK Biobank participants (n=74,214 individuals remaining, summary statistics specifically derived for this study). As it was not possible to obtain the major depressive disorder GWAS summary statistics excluding UK Biobank participants, we conducted sensitivity analyses removing individuals with a diagnosis of depression to control for overlap with the discovery dataset. The primary analysis used standardised scores generated from SNPs with a discovery sample p-value threshold of p ≤ 0.05, but associations at 10 other p-value thresholds were also tested. A logistic regression model was used to test the association of each PRS with various psychotic experience phenotypes, covarying for the first five principal components and genotyping array.
Genetic correlations LDSC 27,37 was used to calculate the genetic correlation between each psychotic experience GWAS and psychiatric and personality traits. External GWAS datasets used to generate the correlations were the same as those used for PRS analysis (detailed above) but also included attention deficit/hyperactivity disorder (ADHD) 38 , autism spectrum disorder 39 , and the PGC cross-disorder analysis 40 . A Bonferroni correction was applied to control for multiple testing.

Copy number variation
CNV calling, which has been described in detail elsewhere 41 , was carried out using biallelic markers common to both genotyping platforms using PennCNV-Affy 42

GWAS
The GWAS of any psychotic experience in 6123 cases and 121,843 controls (following QC, exclusions detailed in Supplementary Methods) identified two variants that were associated at the genome-wide significance (GWS) level of p < 5x10 -8 (Figure 1, for heritability or genetic correlation analyses with LDSC.

Validation analyses in ALSPAC
There was evidence of association between the PRS calculated using the any psychotic experience GWAS from UK Biobank at the p-value threshold of ≤0. 5  . These findings were consistent across most p-value thresholds (Figure 2 and Supplementary Figure 7). Supplementary Figure 8 compares the average PRS scores (SNP inclusion threshold p<0.05) for each psychotic experience phenotype. Individuals with distressing psychotic experiences had significantly higher PRS scores for schizophrenia, bipolar disorder and major depressive disorder than those with non-distressing psychotic experiences (Supplementary Table 6).
We also considered individual psychotic symptoms and found that PRS for schizophrenia was more strongly associated with delusions of persecution than the other psychotic symptoms measured (Supplementary Table 6). This pattern was similar for PRS related to bipolar disorder and major depressive disorder and was consistent across p-value thresholds (Supplementary  Table 7). The association with major depressive disorder PRS remained significant when individuals with a diagnosis of depression were removed to control for any potential overlap with the discovery dataset (Supplementary Table 8).

Discussion
We conducted the largest GWAS of psychotic experiences using the population-based UK Biobank sample, and identified four genome-wide significant loci; an intronic variant to ANK3, an intronic variant to CNR2, and two intergenic loci. PRS, genetic correlation and CNV analyses indicated a shared genetic aetiology between psychotic experiences and schizophrenia, but also neurodevelopmental disorders and other psychiatric disorders such as major depressive disorder and bipolar disorder. The shared genetic aetiology between psychotic experiences and major mental health conditions was stronger for psychotic experiences rated as distressing.
The primary GWAS findings are related to intronic variants in ANK3 and CNR2. The GWAS of any psychotic experience identified two significant loci, the most significant of which was indexed by rs10994278, an intronic variant to ANK3 (OR=1.16, p=3.06x10 -8 ). The ANK3 gene encodes ankyrin-G, a protein that has been shown to regulate the assembly of voltage-gated sodium channels and is essential for normal synaptic function 46 . ANK3 is one of strongest and most replicated genes for bipolar disorder 33 , and variants within ANK3 have also been associated in the PGC cross-disorder GWAS 40 , and in a rare variant analysis of autism spectrum disorder 47 .
The GWAS of distressing psychotic experiences also identified two significant loci, the most significant of which was indexed by rs75459873, an intronic variant to CNR2 (OR=0.66, p=3.78x10 -8 ). CNR2 encodes for CB2, one of two well-characterised cannabinoid receptors (CB1 being the other). CB2 receptors they have abundant expression in the immune system, microglia and some neurons 48 . Several lines of evidence have implicated the endocannabinoid system in psychiatric disorders including schizophrenia 49,50 and depression 51 . The main psychoactive agent of cannabis, Δ 9 -tetrahydrocannabinol, can cause acute psychotic symptoms and cognitive impairment 52 . Given that cannabis use is strongly associated with psychotic experiences, we tested, but found no evidence for, a mediating or moderating effect of cannabis use on the association of rs75459873 and distressing psychotic experiences. However, whilst no evidence was found in this study, a mediating effect of cannabis use cannot be ruled out given there may be questionable reliability in self-reported illicit cannabis use.
We found that psychotic experiences shared genetic risk with several psychiatric disorders, which included, but was not specific to, schizophrenia. PRS analyses identified associations between psychotic experiences and genetic liability for schizophrenia, major depressive disorder and bipolar disorder but not for intelligence or neuroticism. The magnitude of associations in the PRS analysis was comparable between psychiatric disorders and was very small (maximum r 2 value<1% and AUC<0.55). We found particular enrichment of schizophrenia, major depressive disorder and bipolar disorder PRS in cases that found the psychotic experiences distressing and for delusions of persecution. Genetic correlation analysis identified significant genetic correlations between psychotic experiences and major depressive disorder (r g =0.46), schizophrenia (r g =0.21), autism spectrum disorder (r g =0.39), and for the PGC cross-disorder GWAS (r g =0. 30). Furthermore, we found an increased burden of CNVs previously associated with schizophrenia (OR=2.04) and neurodevelopmental disorders more widely (OR=1.75) in individuals with distressing psychotic symptoms. There was also an increased burden of CNVs in those reporting any psychotic experience but the association was stronger for distressing psychotic experiences. All schizophrenia-associated CNVs are also associated with neurodevelopmental disorders such as intellectual disability and autism spectrum disorder, and in fact penetrance is higher in these disorders 43 . Furthermore, CNVs in UK Biobank have been associated with a range of outcomes including cognitive performance 41,53 and depression 54 , adding strength to our findings of a lack of specificity for psychotic experiences genetic risk.
A number of studies have demonstrated that psychopathology in the population is best described by a bifactor model with a common latent trait as well as specific traits, and that psychotic experiences index the more severe end of the common or shared trait 56,57 . Our findings of non-specificity of genetic risk for psychotic experiences with risk for other disorders are consistent with previous studies 7,8,11,12 . Nonetheless, despite lacking specificity, our results suggest that incorporating questions about frequency and, in particular, distress of psychotic experiences to self-reported assessments may allow a more valid identification of experiences that index schizophrenia and major mental health disorder liability.
Given the heritability estimate was very low (1.7%) for the any psychotic experience GWAS and the low amount of variance explained in our PRS analysis, and those of others 7,11 , our findings indicate that understanding the genetics of psychotic experiences is unlikely to have an important impact on understanding the genetics of schizophrenia specifically. However, given that the small heritability of psychotic experiences is substantially shared with other psychiatric disorders, and that non-genetic aetiology of psychotic experiences has been shown to be shared 55 ; then studies of psychotic experience aetiology may be informative for psychiatric disorders in general.

Strengths and limitations
Strengths of this study include the large sample size (approximately 10x that of previous studies), which is required for genetic association studies, the use of an adult cohort, and the use of multiple psychotic experience phenotypes, which increase confidence in our findings.
Furthermore, individuals with psychotic disorders were removed from our analyses, and thus the results will not be driven by this smaller subset of individuals.
One limitation of this study relates to the retrospective measurement of lifetime psychotic experiences by self-report from an online questionnaire, as this increases the likelihood of measurement error. A further limitation is the evidence of a 'healthy volunteer' bias for the participants recruited to UK Biobank and the sample cannot be therefore considered representative of the general population 58 . We also found that the participants that completed the MHQ had significantly higher intelligence and lower schizophrenia, depression and neuroticism PRS compared to UK Biobank participants who did not complete the MHQ (Supplementary Table 9). It is possible therefore that our results are affected by selection bias.
Lastly, we were not able to entirely de-duplicate the UK Biobank individuals from all of the external datasets used for the polygenic risk score analysis. We excluded datasets that explicitly included the UK Biobank where possible, and failing that removed cases in sensitivity analyses.

Conclusions
In the largest GWAS of psychotic experiences from the population-based UK Biobank sample, we identified four genome-wide significant loci associated with psychotic experiences including an intronic variant to ANK3, an intronic variant to CNR2, and two intergenic loci. We found support for a shared genetic aetiology between psychotic experiences and several psychiatric disorders including schizophrenia, major depressive disorder, bipolar disorder, and neurodevelopmental disorders indicating that psychotic experiences are not specifically related to schizophrenia, but rather to a general risk for mental health disorders.

Figure 1
Manhattan plot for GWAS analyses of any psychotic experience, distressing psychotic experiences and multiple occurrence psychotic experiences. Dashed line represent the genomewide significance level of p < 5 x 10 -8 and the dotted line represents p < 1 x 10 -5 .

Figure 2
Polygenic risk score analysis. Points represent odds ratio (OR) and error bars are 95% confidence intervals. A plot with r 2 and p-values is presented in Supplementary Figure 7.

Figure 3
Genetic correlation analysis. Colour corresponds to the strength of the correlation (rg) and stars correspond to the statistical significance of the correlation (* P < 0.003, ** P < 0.0001, *** P < 0.00001). Positive correlations are shown in blue and negative correlations in red.  Intergenic ST18 Genome-wide significant associations (p < 5  Association of psychotic experience phenotypes with copy number variants (CNVs) previously associated with (i) schizophrenia and (ii) neurodevelopmental disorders (note all schizophreniaassociated CNVs are also included in neurodevelopmental disorders). Columns represent; phenotype (PE = psychotic experience), rate of CNVs in cases (those reporting a psychotic experience), rate of CNVs in controls (the comparator group, those who did not report a psychotic experience), odds ratio (OR) and 95% confidence intervals (CI), and p-value (P).