Genome‐wide analysis of adolescent psychotic‐like experiences shows genetic overlap with psychiatric disorders

This study aimed to test for overlap in genetic influences between psychotic‐like experience traits shown by adolescents in the community, and clinically‐recognized psychiatric disorders in adulthood, specifically schizophrenia, bipolar disorder, and major depression. The full spectra of psychotic‐like experience domains, both in terms of their severity and type (positive, cognitive, and negative), were assessed using self‐ and parent‐ratings in three European community samples aged 15–19 years (Final N incl. siblings = 6,297–10,098). A mega‐genome‐wide association study (mega‐GWAS) for each psychotic‐like experience domain was performed. Single nucleotide polymorphism (SNP)‐heritability of each psychotic‐like experience domain was estimated using genomic‐relatedness‐based restricted maximum‐likelihood (GREML) and linkage disequilibrium‐ (LD‐) score regression. Genetic overlap between specific psychotic‐like experience domains and schizophrenia, bipolar disorder, and major depression was assessed using polygenic risk score (PRS) and LD‐score regression. GREML returned SNP‐heritability estimates of 3–9% for psychotic‐like experience trait domains, with higher estimates for less skewed traits (Anhedonia, Cognitive Disorganization) than for more skewed traits (Paranoia and Hallucinations, Parent‐rated Negative Symptoms). Mega‐GWAS analysis identified one genome‐wide significant association for Anhedonia within IDO2 but which did not replicate in an independent sample. PRS analysis revealed that the schizophrenia PRS significantly predicted all adolescent psychotic‐like experience trait domains (Paranoia and Hallucinations only in non‐zero scorers). The major depression PRS significantly predicted Anhedonia and Parent‐rated Negative Symptoms in adolescence. Psychotic‐like experiences during adolescence in the community show additive genetic effects and partly share genetic influences with clinically‐recognized psychiatric disorders, specifically schizophrenia and major depression.


| I N TR ODU C TI ON
Psychotic-like experiences, also referred to as psychotic experiences, are traits in the community that at the extreme resemble symptoms of psychotic disorders, such as schizophrenia. Based on principal component analyses, psychotic-like experiences can be separated into replicable and specific domains, such as positive, cognitive, and negative domains Wigman et al., 2012). Psychotic-like experiences are common in the general population, particularly during adolescence (Freeman, 2006;Ronald et al., 2014). Evidence suggests that psychotic-like experiences are dimensional: they show varying degrees of severity and taxometric analyses support their continuous nature in adolescence (Ahmed, Buckley, & Mabe, 2012;Daneluzzo et al., 2009;Taylor, Freeman, & Ronald, 2016). Adolescence is just prior to a peak time of onset for several psychiatric disorders, particularly schizophrenia, bipolar disorder, and major depression (Laursen, Munk-Olsen, Nordentoft, & Mortensen, 2007). Psychotic-like experiences predict many types of psychiatric disorders and suicidal ideation with significant odds ratios of 1. 3-5.6 (Cederl€ of et al., 2016;Fisher et al., 2013;Kelleher et al., 2012Kelleher et al., , 2014McGrath et al., 2016;Werbeloff et al., 2012;Zammit et al., 2013).
One study to date has reported single nucleotide polymorphism (SNP)heritability estimates ranging from 0 to 32%. These results suggest that common genetic variation plays a role in at least some types of adolescent psychotic-like experiences (Sieradzka et al., 2015), but larger studies are needed to offer accurate estimates.
The one previous genome-wide association study (GWAS) of adolescent psychotic-like experiences assigned 3,483 individuals to high-or low-scoring groups based on a measure of positive psychoticlike experiences that combined paranoia, hallucinations, and delusions (Zammit et al., 2014). No locus achieved genome-wide significance.
The Psychiatric Genomics Consortium 2 schizophrenia GWAS has been previously used twice to test for an association between genetic risk of schizophrenia and adolescent psychotic-like experiences (these followed one study using Psychiatric Genomics Consortium 1 schizophrenia GWAS results; Zammit et al., 2014). The first of these studies used quantitative measures of individual psychotic-like experiences in adolescence in a sample of N 5 2,133-2,140 and reported no significant positive association between any of the psychotic-like experience domains and schizophrenia genetic risk . This study also reported no significant positive association with bipolar disorder genetic risk. A second study with N 5 3,676-5,444 tested for an association between schizophrenia genetic risk and dichotomous outcomes on several scales: a combined positive psychotic-like experiences scale (which included paranoia, hallucinations, delusions, and thought interference), a negative symptoms scale, a depressive disorder scale, and an anxiety disorder scale (Jones et al., 2016). This second study reported a significant positive association between schizophrenia genetic risk and both high negative symptoms and high anxiety disorder. There was no association with the positive psychotic-like experiences or depressive disorder scales.
Schizotypal traits are closely related to psychotic-like experiences.
Schizotypal traits focus on differences in personality that reflect liability to psychotic disorders rather than the presentation of subclinical psychotic-like experiences. Previous twin studies of schizotypal traits report modest heritability (Ericson et al., 2011;Lin et al., 2007), and studies are starting to explore the link between polygenic risk for schizophrenia and schizotypal traits (Hatzimanolis et al., 2017). In light of the differences in the constructs of schizotypy and psychotic-like experiences, we focused our brief review on psychotic-like experiences.
Our study aimed to test for overlap in genetic influences between specific psychotic-like experience domains in adolescence and psychiatric disorders-specifically, schizophrenia, bipolar disorder, and major depression. Four psychotic-like experience subscales were used, derived from principal component analysis: Paranoia and Hallucinations, Cognitive Disorganization, Anhedonia, and Parent-rated Negative Symptoms. We also present the largest GWAS to date of adolescent psychotic-like experiences using a community sample of 6,297-10,098 individuals (including siblings) from three European studies. As such our study also aimed to identify novel common genetic variants associated with specific psychotic-like experience domains and to estimate their SNP-heritability.

| Measures
We established comparable quantitative scales of psychotic-like experience domains across samples. The subscales in the Specific Psychotic Experiences Questionnaire, used in TEDS, were the starting point . The other two samples had used similar self-report items that were mapped onto domains of paranoia, hallucinations, cognitive disorganization, anhedonia, and parent-rated negative  Table 1.

| Phenotypic harmonization
Items were inspected to allow for harmonization of psychotic-like experience domains across samples via a two-stage process. First, two expert clinicians (AC and DF) matched items across samples that were capturing the same underlying construct (to ensure construct and content validity) based on their clinical knowledge and experience with psychotic-like experience measures. Second, the matched items within each sample were analyzed via principal components analysis, a psychometric approach used to determine individual components underlying the items. This process was used to identify the presence of distinct psychotic-like experience domains.
Individuals with >50% missingness for any psychotic-like experience domain were excluded from all analyses. The remaining missing phenotypic data was imputed using multiple imputation in R (Buuren & Groothuis-Oudshoorn, 2011). Imputation of item level data was carried out separately for each sample and psychotic-like experience domain. Individual scores of the resulting psychotic-like experience subscales were calculated using sum scores. To ensure the equal contribution of each item to the sum score, item response values were rescaled to values between 0 and 1. The phenotypic correlations between specific psychotic-like experiences within each sample and in all samples combined are in Supporting Information Table S6.
Sum scores for each psychotic-like experience subscale were normalized using inverse rank-based normalization (data ties ranked randomly) and then standardized. The median correlation between a given scale before and after normalization was 0.92 (Supporting Information Table S7), dependent on skew of the original scale. The normalized scores were then regressed against the following covariates: sex, age, age 2 , sex*age, sex*age 2 , study, and the top 8 principal components of ancestry. Principal components of ancestry were jointly calculated across all three samples using PLINK1.9 (https:// www.cog-genomics.org/plink2) (Chang et al., 2015) based only on observed genetic variation.

| Genotype imputation and quality control procedure
Genotypic data from each sample was imputed using the 1KG Phase 3 v5 to maximize genome-wide overlap between samples. After imputation, stringent variant and individual level quality control thresholds were applied in all samples before converting to hard-call genotypes (certainty threshold 5 0.9) and merging the three data sets for combined analysis. For further information see Supporting Information Note S3. Information on DNA collection and genotyping is given in Supporting Information Note S4.

| Estimation of SNP-heritability
SNP-heritability estimates were calculated within each sample and then meta-analyzed to provide meta-SNP-heritability estimates. The meta-analysis of SNP-heritability estimates approach was taken to avoid a downward bias due to possible genetic heterogeneity between samples. Inverse variance weighted meta-analysis was used.
Secondary analysis of mega-SNP-heritability, estimated across samples simultaneously, was also performed to estimate the proportion of phenotypic variance that could be explained by the mean genetic effects across samples.
Two methods were used to estimate SNP-heritability: genomicrelatedness-matrix restricted maximum-likelihood (GREML) in genomewide complex trait analysis (GCTA), and linkage disequilibrium (LD)score regression. Related individuals were included in both meta-and mega-GREML analyses (Zaitlen et al., 2013). LD-score regression (Bulik-Sullivan et al., 2015) was also performed within (meta-) and across (mega-) samples. The effective sample size was used in LD-score regression analyses, thus matching the sample in the GWAS. Effective sample size was calculated as follows: (2*sample size)/(1 1 correlation between siblings) (Minic a, Boomsma, Vink, Dolan, 2014). There was no evidence of confounding from genome-wide association results so the intercept was constrained to 1.

| Mega-GWAS
The three samples were combined to enable genome-wide association mega-analysis. Genome-wide association analysis of all four psychoticlike experience domains using related (i.e. monozygotic and dizygotic twin pairs) and unrelated individuals was performed in PLINK (http:// pngu.mgh.harvard.edu/purcell/plink/) (Purcell et al., 2007

| Replication analysis of rs149957215 association with Anhedonia
After the discovery sample had been prepared and analyzed for this project, an independent subsample of TEDS participants was genotyped. Of the newly genotyped individuals, 2,359 (incl. 635 MZ pairs) had reported on self-rated Anhedonia. This independent TEDS sample was used as a replication sample for the validation of the genome-wide significant association between rs149957215 and Anhedonia identified by the mega-GWAS in this study. rs149957215 genotypes were imputed (MACH r 2 5 0.93) using the haplotype reference consortium data via the Sanger imputation server (McCarthy et al., 2016). The genotypic data was converted to hard-call format (certainty threshold of 0.9) and analyzed in PLINK using the same GEE method to account for relatives. This replication analysis had a power of 0.86 to detect an association of the same magnitude (r 2 5 0.47%) at nominal significance.

| Gene-based association analysis
Two gene-based association analyses were performed. The first aggregates SNP associations within specific gene regions using the MAGMA program (de Leeuw, Mooij, Heskes, & Posthuma, 2015). The second analysis used PrediXcan (Gamazon et al., 2015) to predict frontal cortex gene expression differences using genotypic data. Further details can be found in Supporting Information Note S5.

| Genetic association between psychotic-like experience domains and psychiatric disorders
Using the software PRSice ( To estimate the genetic covariance between psychotic-like experience domains and schizophrenia, bipolar disorder, and major depression, both LD-score regression and additive variance explained and number of genetic effects method of estimation (AVENGEME) were used (Bulik-Sullivan et al., 2015;Palla & Dudbridge, 2015). AVENGEME uses the results of PRS analyses across multiple significance thresholds to estimate the model parameters including the genetic covariance. AVENGEME estimates 95% confidence intervals using profile likelihood method. There was no evidence of confounding or sample overlap in the mega-GWAS summary statistics, as such the heritabilityintercept was constrained to 1 and the genetic covariance intercept was set to 0 in LD-score regression. To improve the accuracy of the estimates of genetic covariance derived from the AVENGEME analysis, the SNP-heritability of liability for schizophrenia, bipolar disorder, and major depression were constrained to the LD-score regression estimates of SNP-heritability (see Supporting Information Table S8).  SNP-heritability estimates from meta-GREML and meta-LD-score regression were between 2.8-8.8% and 6.6-21.5%, respectively (Table 2). Results from secondary analysis of SNP-heritability using the mega-analysis approach are given in Supporting Information Table S11.  Table S12 and Figures S2-S5). There was no evidence of confounding with lambdas of 0.99-1.01 and LD-score regression intercept of 1.00 in all analyses (Supporting Information Figure S6).

| RE S U L TS
Regional gene-based tests identified no gene that was significantly associated with any psychotic-like experience domain after Bonferroni correction for multiple testing. The top ten most associated genes for   Table S15). Logistic regression comparing low and high groups of non-zero scoring individuals supported these findings (Supporting Information Table S18 and Figure S10).   Table 3, the genetic covariance between schizophrenia and anhedonia psychoticlike experience domain, and between major depression and parentrated negative symptoms domain, were not significant.

| D ISC USSION
A genetic relationship was identified between clinical schizophrenia and positive, cognitive, and negative psychotic-like experience trait domains in adolescence. A higher genetic risk for schizophrenia significantly predicted adolescents having more cognitive disorganization, anhedonia, and parent-rated negative symptoms, as well as more paranoia and hallucinations (this last finding was in the non-zero scorers). Thus our findings suggest that psychotic-like experience trait domains in adolescence comprise partly of the genetically-influenced phenotypic manifestation of schizophrenia. Furthermore, higher genetic risk for major depression significantly predicted having more self-rated anhedonia and parent-rated negative symptoms as a teenager. Our results discredit the hypothesis that psychotic-like experience trait domains in the general population are epiphenomena that do not share biological pathways with clinically-recognized psychiatric disorders.
Genetic risk for schizophrenia and major depression in adulthood limit of prediction is the degree to which PRSs for schizophrenia and major depression predict themselves (that is, their own phenotype) in an independent sample, which is 18.4% for schizophrenia (Ripke et al., 2014) and 0.6% for major depression (Ripke et al., 2013). It is known from epidemiological studies that the magnitude of phenotypic association between psychotic-like experience domains and schizophrenia is modest (McGrath et al., 2016;Zammit et al., 2013) and far more people report psychotic-like experiences in adolescence than develop disorders such as schizophrenia. Furthermore, there is considerable heterogeneity in schizophrenia and depression in terms of age of onset and symptom presentation. The associations were likely to be modest given their cross-phenotype and cross-age nature. The genetic association between psychiatric disorders and psychotic-like experience traits in the community may increase with age or with longitudinal assessments.
A notable result was that paranoia and hallucinations during adolescence are associated with schizophrenia common genetic risk if individuals report at least some degree of paranoia or hallucinations, that is, the association was only present in the non-zero scorers. Individuals reporting no paranoia and hallucinations can exist anywhere on the schizophrenia genetic liability spectrum. Previous studies did not find a genetic association between schizophrenia and positive psychotic-like experiences (Jones et al., 2016;Sieradzka et al., 2014;Zammit et al., 2014). The use of quantitative traits and a large sample allowed the identification of this non-linear effect here. An explanation for the non-linear effect may lie in the variable age of onset of paranoia and hallucinations. Our study was focused on psychotic-like experiences in mid to late adolescence. It is predicted that stronger positive associations between genetic risk for schizophrenia and positive psychotic-like experiences will be found in samples assessed over a longer time frame, when anyone who is going to have paranoia and hallucinations has manifested them.
The significant and positive genetic association between a selfrated anhedonia trait measure in adolescence and major depression in adulthood concurs with a previous report showing that subclinical depressive symptoms (including anhedonia) phenotypically predict major depressive episodes in adulthood (Pine, Cohen, Cohen, & Brook, 1999). Anhedonia is present as a symptom of both schizophrenia and depression in psychiatric diagnoses, and our research shows that as a trait dimension in adolescence it shares common genetic underpinnings with both schizophrenia and depression.
The significant negative association between paranoia and hallucinations and bipolar disorder genetic risk also deserves discussion.
Previous research reported the presence of paranoia, hallucinations, and delusions prior to the onset of bipolar disorder (McGrath et al., 2016). However, our results suggest that genetic influences on paranoia and hallucinations as traits during mid-adolescence are negatively associated with known common genetic risk associated with diagnosed bipolar disorder. The common genetic relationship between adolescent paranoia and hallucinations and bipolar disorder requires replication in an independent sample. Further insight into this relationship will also be gained as more powerful polygenic predictors for bipolar disorder become available.
Although in many cases the SNP-heritability estimates were not significantly non-zero, the point estimates of SNP-heritability indicate that common genetic variation influences psychotic-like experience domains during adolescence. This concurs with results from twin studies reporting significant twin heritability estimates (Zavos et al., 2014).
The SNP that achieved genome-wide significance in the mega-GWAS for anhedonia lies within the protein-coding gene IDO2. IDO2 is a key enzyme in the regulation of the kynurenine pathway, which upon stimulation by proinflammatory cytokines, converts tryptophan into kynurenine. It has been reported that increased metabolism of tryptophan to kynurenine is associated with increased depressive symptoms via the increased production of cytotoxic kynurenine metabolites (Dantzer, O'Connor, Lawson, & Kelley, 2011;Myint et al., 2007;Wichers et al., 2005). In fact, a previous study has reported a significant correlation between kynurenine production and anhedonia in an adolescent sample (Gabbay, Ely, Babb, & Liebes, 2012). These previous studies suggest the association between IDO2 and anhedonia is plausible. However, this association should be interpreted with caution as rs149957215 was imputed in all three samples, has a low minor allele frequency of 0.013, and appears to be uncorrelated with surrounding common genetic variation. Furthermore, this association failed to replicate in a sufficiently powered independent sample. has been estimated that a significance threshold of p 5 .001 is appropriate (Euesden et al., 2015). This threshold would be overly conservative in our study as we only tested a limited number of thresholds.
Genetic covariance estimates obtained by AVENGEME ( Collectively, these findings reveal novel evidence for some shared common genetic etiology between psychotic-like experience domains in mid to late adolescence and clinically-recognized psychiatric disorders in adulthood. Evidence is accruing that psychotic-like experiences manifested prior to adulthood form part of a wider phenotype or prodrome related to psychiatric disorders such as schizophrenia and depression. This study joins the existing evidence that individuals with a family history of schizophrenia score higher on psychotic-like experience scales (Jeppesen et al., 2014;Zavos et al., 2014), psychotic-like experiences are associated with the same environmental risk factors as schizophrenia (Linscott & Van Os, 2013), psychotic-like experiences are on a phenotypic and etiological continuum across the severity continuum (Taylor et al., 2016;Zavos et al., 2014), and psychotic-like experi- . The next step is to consider how psychotic-like experiences can be harnessed in a practical sense as a (small effect size) red flag for risk in early intervention and prevention strategies. Similar to family history, psychotic-like experiences might be a useful heuristic even if the majority of individuals with psychotic-like experiences will remain unaffected by psychiatric disorders.

ACKNOWLEDGMENTS
We thank the participants of TEDS, ALSPAC, and CATSS, and their research teams, which include interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. We also thank the Psychiatric