Association Between Schizophrenia-Related Polygenic Liability and the Occurrence and Level of Mood-Incongruent Psychotic Symptoms in Bipolar Disorder

This case-control study examines the association between polygenic liability for schizophrenia and psychotic presentations of bipolar disorder.

5][16][17][18][19][20][21][22] Bipolar disorder is a phenomenologically heterogeneous construct, and within the diagnostic category, individuals with BD may have quite different symptom profiles.][25] Delusions and hallucinations are common in BD, 26,27 with approximately one-third of all psychotic features judged to be mood incongruent. 28,29Mood-incongruent psychotic features are associated with poor prognosis and poor lithium response and are qualitatively similar to the prototypic symptoms of SCZ, [30][31][32] suggesting that BD with psychosis and particularly mood-incongruent psychotic features may specify a subgroup or stratum with stronger etiological links to SCZ.Stratified linkage and candidate-gene studies of BD associations with chromosomal regions and genes implicated in SCZ show stronger effects in psychosis and mood-incongruent subsamples, [33][34][35][36] providing some support for this causal heterogeneity hypothesis; however, lack of consistency in earlier linkage and candidate-gene studies renders the overall support weak.
Genome-wide association studies (GWAS) have found a substantial polygenic component to both BD and SCZ risks, with a large proportion of the disorders' genetic variance explained by common alleles partially shared by the 2 disorders. 20This polygenic risk can be calculated for individuals with a single summary measure: the polygenic risk score (PRS; with higher scores indicating a higher burden of risk alleles), which allows us to examine the genetic basis of symptom domains within and across the 2 disorders [37][38][39] with greater power than do the historical linkage and candidate-gene approaches.The PRS-SCZ differentiates BD cases from controls, 16,20 and there are differential PRS associations across subtypes with schizoaffective bipolar disorder (SABD) (an intermediate subtype characterized by admixture of SCZ and BD symptoms) having a relatively larger burden of SCZ risk, compared with other BD subtypes. 15,40To date, lack of power in well-phenotyped samples has hindered fine-scale examination of the association of SCZ polygenic-risk with psychotic symptoms in BD.
This study aimed to examine the association between polygenic liability for SCZ and psychotic presentations of BD using the PRSs generated from the most powerful SCZ-GWAS discovery set currently available. 21Measures relevant to the occurrence and nature of psychotic symptoms were considered.We hypothesized that BD with psychosis would be associated with higher polygenic risk for SCZ and that this association would be stronger when moodincongruent psychotic features were present, given their phenotypic similarity to the psychotic symptoms of prototypic SCZ.

Sample Ascertainment
Bipolar Disorder Sample In total, data from 4436 cases of BD with deep phenotypic information, European ancestry, and domicile in the United Kingdom were collected between January 1, 2000, and December 31, 2013, via the UK Bipolar Disorder Research Network using recruitment methods reported previously. 15,41,42The sample contained 1399 cases not included in previous publications of the Bipolar Disorder Research Network. 15,40All participants were assessed using a consistent protocol, which included the Schedules for Clinical Assessment in Neuropsychiatry (SCAN) interview 43 administered by trained research psychologists and psychiatrists, with very good to excellent interrater reliability for all domains of psychopathology. 44Using information from the SCAN interview and case note review, we completed the Operational Criteria Checklist. 45Research Diagnostic Criteria (RDC) 3 diagnoses, which differentiate individuals on the basis of their pattern of mood and psychotic symptoms better 40 than either the DSM-5 2 or the International Statistical Classification of Diseases and Health-Related Disorders, Tenth Revision Classification of Mental and Behavioural Disorders, 1 were made with consensus lifetime best-estimate method informed by all available information. 46

Schizophrenia Sample
To allow the comparison of BD to SCZ, we included a subset (n = 4976) of the CLOZUK (treatment-resistant schizophrenia, treated with clozapine) study sample collected via the Zaponex Treatment Access System, which was detailed in a previous report. 47All patients in the sample were prescribed clozapine for treatment-resistant SCZ and are independent of and unrelated (pi-hat <0.2) with individuals in the discovery

Key Points
Question What is the association between schizophrenia-related polygenic liability and the occurrence and level of mood-incongruence of psychotic symptoms in bipolar disorder?Findings In this case-control study involving 4436 cases of bipolar disorder, 4976 cases of schizophrenia, and 9012 controls, there was an exposure-response gradient of polygenic risk.Schizophrenia had the strongest association, followed by bipolar disorder with prominent mood-incongruent psychotic features, bipolar disorder with mood-congruent psychotic features, and bipolar disorder with no psychosis; all differential associations were statistically significant.

Meaning
This study shows a gradient of genetic liability across schizophrenia and bipolar disorder, indexed by the occurrence of psychosis and level of mood incongruence.
GWAS. 21In principle, treatment-resistant SCZ may carry higher polygenic risk burden; however, the PRSs in the CLOZUK sample are similar to the PRSs in other SCZ samples used by the Psychiatric Genomics Consortium. 21The CLOZUK procedures and methods were approved by the National Research Ethics and were in line with the UK Human Tissue Act regulations in partnership with the Leyden Delta.

Control Samples
The controls came from 2 UK sources: (1) the Type 1 Diabetes Genetics Consortium study, which comprised unscreened controls (n = 2532) recruited through the 1958 Birth Cohort, 48 and (2) a subsample (n = 6480) of the Generation Scotland study screened for psychiatric disorders. 49Controls were not associated (pi-hat <0.2) with individuals in the Psychiatric Genomics Consortium-SCZ discovery set and were matched ancestrally to our case data sets. 47The Generation Scotland Access Committee approved this application to use Generation Scotland as controls.
Genotyping, Quality Control, Phasing, and Imputation Bipolar Cases Genotypic data for the BD cases were processed in 3 batches, each on a different platform.To mitigate against potential bias from batch effects, 50 stringent quality control (QC) was performed on each platform separately prior to merging.Singlenucleotide polymorphisms (SNPs) were excluded if the call rate was less than 98%, the minor allele frequency (MAF) was less than 0.01, or the SNPs deviated from the Hardy-Weinberg equilibrium (HWE) at P <1×10 −6 .Individuals were excluded if they had minimal or excessive autosomal homozygosity (F|>0.1),high pairwise relatedness (pi-hat >0.2), or mismatch between recorded and genotypic sex.Following QC, the data for each platform were phased using SHAPEIT, 51 version 3.4.0.1023 (Olivier Delaneau), and imputed with IMPUTE2, 52 version 2.3.0 (University of Oxford), using the 1000 Genomes Project reference panel (phase 3).Imputed data were converted into the most probable genotypes (probability >0.9) and merged on shared SNPs.After QC, 4399 BD cases remained.

CLOZUK Cases and Controls
The CLOZUK and control samples went through strict QC separately before being phased and imputed simultaneously as part of a larger SCZ study. 47

Merging Imputed Genotypic Data Sets
After SNPs with stand ambiguity were excluded, BD, CLOZUK, and control samples were merged and the imputed markers underwent a second QC filter. 50This second QC excluded SNPs with a missingness rate of more than 5% of individuals, an information content score lower than 0.8, an MAF of less than 0.01, or deviation from HWE at P <1×10 −6 .

Principal Component Analysis
To adjust for potential confounding from population structure, we performed principal components analysis.We used PLINK, version 1.9 (Christopher Chang), after pruning the linkage disequilibrium and frequency filtering the SNPs from the merged sample, keeping the eigenvectors for the first 10 principal components to use as covariates in the association analysis.

Polygenic Risk Scores
We generated the PRSs 20 using the 2014 Psychiatric Genomics Consortium-SCZ meta-analysis as our discovery set 21 calculated for each individual on the basis of a set of alleles with association P < .05.This decision was informed by the Psychiatric Genomics Consortium leave-one-cohort-out PRS analyses for all SNP selection P value thresholds, which found the median and the mode was P = .05,which represents the association that best optimizes the balance of false and true risk alleles at the current discovery sample size. 21The most informative and independent markers were selected to minimize statistical noise where possible, by using P value-informed clumping at r 2 < 0.2 with 1-MB windows and by excluding the extended major histocompatibility complex (chromosome 6: position 25-35 MB) because of its complex linkage disequilibrium structure.

Subtypes of BD
The RDC subtypes were used as categorical outcomes in casecontrol analyses.The RDC 3 and the DSM-5, 2 although not the International Statistical Classification of Diseases and Health-Related Disorders, Tenth Revision, Classification of Mental and Behavioural Disorders, 1 subdivides BD into bipolar I disorder (BD I) and bipolar II disorder (BD II) depending on the nature of the mood states, mania in BD I, and hypomania in BD II.All classification systems recognize SABD.Psychotic symptoms are most prominent in SABD and then BD I and are least prominent in BD II. 53,54polar Affective Disorder Dimension Scale Outcome measures were generated from the Bipolar Affective Disorder Dimension Scale (BADDS) subscales of psychosis and mood incongruence, which provide an ordered, but not necessarily linear, measure of lifetime symptom domain severity.55 An interrater reliability exercise for this sample demonstrates excellent interclass correlation: (psychosis) 0.91 and (mood incongruence) 0.89. 1.A binary categorical outcome measure for lifetime occurrence of psychosis, defined as an unambiguous episode of positive and/or disorganized psychotic symptoms, generated by dichotomizing the psychosis domain scale at a score higher than 9. 55 2. A binary categorical outcome measure for lifetime occurrence of predominant mood-incongruent psychotic features, defined as high or low prominence of mood incongruence, generated by dichotomizing the mood incongruence domain scale at a score higher than 19. 3.An ordinal measure of mood-incongruent psychotic features that assesses the overall balance between moodcongruent and mood-incongruent psychosis across the lifetime, rated using all available information according to the Bipolar Disorder Research Network protocol (see eNote 1 in the eAppendix in the Supplement).

Statistical Analysis
A multinomial logit model was used to estimate differential associations of standardized PRSs, adjusted for the first 10 principal components and genotyping platforms across the categories of cases and controls.We report the estimated coefficient transformed to relative risk ratio (RR), defined as the exponentiated regression coefficient.In addition, PRS associations across levels of mood-incongruent psychotic features using ordinal logistic regression were estimated.To examine whether SABD subtypes were driving observed PRS associations with moodincongruent psychotic features, we did a sensitivity analysis that excluded SABD cases.Postestimation predicted probabilities were plotted to aid the interpretation of PRS associations across RDC subtypes of BD. 56 To correct for multiple comparisons of PRS associations across different phenotypic strata within each model, we generated bootstrapped SEs and 95% CIs as an approximation to exact permutation methods 57 (see eNote 2 in the eAppendix in the Supplement).Possible familywise, type I error proliferation was controlled using the Bonferroni method, calculated by multiplying the bootstrapped P values by 4. 58 Post hoc analyses used a multinomial logit model casecontrol design to examine differential associations across composite phenotypic categories defined by BD I and BD II subtypes and stratified by psychosis status.Complementary logistic regression analyses were conducted to compare the PRS association with lifetime occurrence of psychosis across BD I and BD II subtypes.To examine the distribution of RDC-defined cases across PRS levels, we converted the PRSs to deciles and generated a stacked bar chart (SCZ [CLOZUK], SABD, BD I, BD II), by decile.Analyses were performed using PLINK, version 1.9 59

Sample Description, Genotyping, and Quality Control
Of the 4436 cases of BD, 2966 (67%) were female patients, and the mean (SD) age at the SCAN interview was 46 [12] years.After BD, CLOZUK, and control imputed-genotyped samples were merged and further QC was performed, 18 387 cases and controls (eTable1intheSupplement) with 3 451 354 SNPs, with an information content score higher than 0.8 and a MAF greater than 1%, were available for analysis.Within the BD sample, 2296 cases (52%) endorsed lifetime occurrence of definite psychosis, with less than a 1% missingness rate in this variable (n = 25).Of the BD cases with definite psychosis, 981 (43%) were classified as having high lifetime mood-incongruent psychotic features.There was a 9% missingness rate (n = 214) for the mood-incongruence variable within the BD cases with psychosis.

Case-Control PRS Associations
As expected, the PRSs discriminated CLOZUK from control samples (Table 1).The PRSs in those with a diagnosis of SABD or BD I, but not BD II, were significantly higher than the PRSs in controls.Across clinical phenotypes, there was an exposureresponse gradient, with the strongest PRS association for

PRS Associations Within Cases
The PRSs discriminated SCZ from all BD subtypes (Table 2).Within BD, the PRSs discriminated BD II from both BD I and SABD (Figure 1).The percentage of CLOZUK cases increased monotonically with increasing decile of PRS, while the percentage of bipolar subtypes decreased (Figure 2).

PRS Associations With Psychotic BD
Compared with controls, the PRSs were higher in BD, regardless of whether there was a history of psychosis (Table 1 and Figure 2).However, the PRSs were significantly higher in BD with psychosis, compared with BD without psychosis (Table 1 and Figure 3).Within BD cases, PRSs discriminated those with and without psychosis (RR = 1.25; 95% CI, 1.16-1.33;P < .001).
Post hoc analyses showed the association between PRS and psychosis was present in BD I (odds ratio [OR] = 1.21; 95% CI, 1.10-1.32)but was not statistically significant in BD II (OR = 0.98; 95% CI, 0.80-1.18).The composite subgroup, defined as BD I with psychosis, had higher PRSs compared with the PRSs in controls (RR = 1.38; 95% CI, 1.31-1.46).This association was significantly stronger than that of the composite BD I without psychosis (RR = 1.16; 95% CI, 1.08-1.25).Within BD II, there was no differential association across subgroups, defined by presence or absence of psychosis, as compared with the differential association in controls (eTable 2 in the Supplement).

PRS Associations With Mood-Incongruent Psychotic Features
Psychotic BD characterized by high mood incongruence had a higher SCZ polygenic risk burden than that in controls, with a 1-SD increase in PRS increasing the RR of being in the high mood-incongruence category by 46% (RR = 1.46; 95% CI, 1.36-1.57)(Figure 3 and Table 1).Although the association was significantly weaker than for the high moodincongruent group, SCZ risk alleles were enriched in those with low mood-incongruence, compared with controls (RR = 1.24; 95% CI, 1.17-1.33).Sensitivity analysis excluding the SABD group from analyses found comparable results (Table 1).Finally, a within-BD case analysis, measuring mood incongruence on an ordinal scale, found the odds of having higher levels of mood incongruence increased with increasing PRS (OR = 1.17; 95% CI, 1.08-1.27;P < .001).Analyses excluding the SABD sample found comparable results (OR = 1.20; 95% CI, 1.09-1.32;P < .001).

Discussion
Higher PRS-SCZ in BD 20,60 is well established.Here, we replicate and extend this observation, demonstrating a gradient of PRS associations across SCZ and BD subtypes (CLOZUK>SABD>BD I with psychosis>BD I without psychosis>BD II).In addition, we show that BD cases with psychosis carry a higher burden of SCZ risk alleles, compared with BD without a history of psychosis (RR = 1.09; 95% CI, 1.04-1.15).Furthermore, individuals with psychotic BD characterized by prominent mood-incongruent psychotic features carry the highest burden of schizophrenia risk alleles.There is a clear exposure-response gradient, with increasing PRS associated with psychotic BD and increasing mood incongruence (mood incongruent > mood congruent > no psychosis), supporting our hypothesis that mood-incongruence indexes phenotypic features linked to SCZ liability.
Previously published work examining the PRSs for SCZ across BD, stratified by psychosis, did not find significant discrimination, 22,40 although a trend was observed that is consistent with the findings presented here.The most likely explanations for the enhanced signal in the current analysis are as follows: the PRSs were constructed using alleles derived from a larger SCZ-GWAS discovery set, which reduces the measurement error and improves power from both this sample and the larger BD sample. 61This group has shown that PRS-SCZ significantly differentiates SABD from non-SABD subtypes, while finding no statistically significant differential between BD stratified by psychosis, 40 suggesting it is the nature of the psychotic symptoms rather than their presence that better indexes the liability shared with SCZ. 62The current analysis supports the proposition that it is the level of mood incongruence rather than the presence of psychosis that better specifies a shared biologically validated dimensional trait, which is captured, although with less precision, by the SABD diagnostic category.
1][32] It is possible the transdiagnostic exposureresponse gradient for the PRS, with the occurrence and nature of psychotic symptoms presented here, could be the result of a general psychopathological factor that cuts across psychiatric disorders and influences the severity of psychopathology generally as well as, or rather than, a psychosisspecific domain.The PRS derived from SCZ-GWAS may be indexing a general liability for psychopathological severity (at least in part) 63 rather than a (SCZ) disease-specific liability.

Implications
Our study supports the hypothesis that, within BD, positive and disorganized psychotic symptoms-particularly, moodincongruent psychotic features-represent a dimensionally defined stratum with underpinning biological validity.These features are not only phenotypically similar to those observed in prototypal SCZ but also index a greater shared-genetic liability, which suggests BD and SCZ share more pathophysiological features. 64Notably, in those diagnosed with BD I with no history of psychosis, the association with SCZ liability was weaker but still higher than in the control group, while there was no over-lap with SCZ liability in the BD II subsample.We are not suggesting that psychotic features are the best or the only index of shared pathophysiological features, but having established stronger genetic links between the risk for SCZ and BD characterized by the occurrence of psychosis and level of mood incongruence, we now have a basis to refine this signal.These findings represent a step toward the goal of reconceptualizing phenotypic definitions using richer clinical signatures, measured across quantitative or qualitative domains, including symptom loading and biomarker expression, outlined in the rationale for the RDC 65,66 and the Road Map for Mental Health Research in Europe 67 projects.[70] Limitations Phenotypic misclassification is a potential methodological concern.However, the phenotypic ratings used in the current analyses are based on both the SCAN interview and case-note review by raters with excellent interrater reliability, which is expected to minimize rates of missing data and differential misclassification due to recall bias of psychotic symptoms. 70It is possible that differential misclassification of mood incongruence may still be present.The psychosis phenotypes examined in this study are broadly defined and likely to represent imperfect measurements of a phenotype that may be continuously distributed 71 ; imposing categorical constraints as we have done may reduce power.Multiple testing can produce spurious results; thus, to reduce this likelihood we generated PRSs using a single discovery-set threshold of P < .05.Bootstrap resampling approaches were used within each of the 4 independent analyses to deal with multiple comparisons across different phenotypic strata.Bonferroni correction was used to adjust for possible familywise type I error proliferation.The PRSs were generated using the most probable genotypes that can potentially reduce power due to a (nondifferential) loss of information at some markers, making our results conservative.Cases and controls were collected independently, which can result in confounding due to population stratification and potential batch effects across the cases and controls.We mitigated against this by partialling out the first 10 principal components and genotyping platforms from the PRS, but some confounding is still possible.Finally, we have only examined the effect of common variants, as rare variants are not captured by current GWAS.

Conclusions
To our knowledge, this study is the first to show a gradient of polygenic liability across SCZ and BD, indexed by the occurrence and level of mood incongruence of positive and disorganized psychotic symptoms.These results highlight the usefulness of genetic data to dissect clinical heterogeneity within and across disorders and suggest further research could potentially aid in defining patient stratifiers with improved biological precision and validity, moving us tentatively toward precision medicine in psychiatry.

Figure 2 .
Figure 2. Percentage of Bipolar Subtype as a Function of Polygenic Risk Scores for Schizophrenia, Grouped by Decile

Figure 1 .
Figure 1.Probability of RDC Bipolar Subtype as a Function of Polygenic Risk Scores (PRSs) Associated With Schizophrenia 1.0

Table 1 .
Differential Association of Polygenic Risk Scores Across Variously Defined Bipolar Disease Strata (Controls as Comparator Category)

Table 2 .
Polygenic Risk Scores for Schizophrenia Associations Among Cases a Adjusted for polygenic risk score for the first 10 principal components and genotyping platforms.Downloaded From: by a Cardiff University User on 11/16/2018