Characterizing the Genetic Overlap Between Psychiatric Disorders and Sleep-Related Phenotypes

BACKGROUND: A range of sleep disturbances are commonly experienced by patients with psychiatric disorders, and genome-wide genetic analyses have shown some signi ﬁ cant genetic correlations between these traits. Here, we applied novel statistical genetic methodologies to better characterize the potential shared genetic architecture between sleep-related phenotypes and psychiatric disorders. METHODS: Using the MiXeR method, which can estimate polygenic overlap beyond genetic correlation, the shared genetic architecture between major psychiatric disorders (bipolar disorder [ N = 51,710], depression [ N = 480,359], and schizophrenia [ N = 77,096]) and sleep-related phenotypes (chronotype [ N = 449,734], insomnia [ N = 386,533] and sleep duration [ N = 446,118]) were quanti ﬁ ed on the basis of genetic summary statistics. Furthermore, the conditional/conjunctional false discovery rate framework was used to identify speci ﬁ c shared loci between these phenotypes, for which positional and functional annotation were conducted with FUMA. RESULTS: Extensive genetic overlap between the sleep-related phenotypes and bipolar disorder (63% – 77%), depression (76% – 79%), and schizophrenia (64% – 79%) was identi ﬁ ed, with moderate levels of congruence between most investigated traits (47% – 58%). Speci ﬁ c shared loci were identi ﬁ ed for all bivariate analyses, and a subset of 70 credible genes were mapped to these shared loci. CONCLUSIONS: The current results provide evidence for substantial polygenic overlap between psychiatric disorders and sleep-related phenotypes, beyond genetic correlation (| r g | = 0.02 to 0.42). Moderate congruency within the shared genetic components suggests a complex genetic relationship and potential subgroups with higher or lower genetic concordance. This work provides new insights and understanding of the shared genetic etiology of sleep-related phenotypes and psychiatric disorders and highlights new opportunities and avenues for future investigation. Unique mapped to lead SNPs, of shared loci between psychiatric disorders and sleep-related phenotypes, by the three employed mapping strategies: quantitative trait locus

underlying causal relationship has been difficult to establish from clinical and epidemiological data. The direction of effect is still unknown-whether sleep abnormalities are a causal factor or a result of the psychiatric disorder, or whether there are shared underlying mechanisms leading to both (13,15). An important reason for the difficulty in disentangling cause-and-effect is that sleep disturbances are both heterogeneous and vary over time and are present in different combinations of multiple phenotypes (16).
In support of the phenotypic relationship between sleep disturbances and psychiatric disorders, molecular studies suggest common underlying mechanisms. Initial studies of variants within candidate circadian/clock genes identified associations with psychiatric disorders, such as BIP (17), MDD (18), and SCZ (19). More recently, however, genome-wide association studies (GWASs) of sleep-related phenotypes, including chronotype (20), insomnia (21), and sleep duration (22), have shown that sleep-related phenotypes are complex traits, modulated by more than just the core circadian/clock genes. In support of the shared clinical picture and overlapping candidate genes, genome-wide genetic overlap has also been identified using standard tools. In major psychiatric disorders, there are reports of significant genetic correlations for chronotype with MDD and SCZ (20), for sleep duration with BIP and SCZ (22), and between insomnia and MDD (21). Moreover, Mendelian randomization results suggest that reduced sleep duration, evening type chronotype, and insomnia are potentially causal for SCZ, while bidirectional effects were identified between insomnia and depression (DEP) (20)(21)(22). Increased understanding of the genetic overlap between psychiatric disorder and sleep-related phenotypes may aid in the development of novel treatment strategies for psychiatric patients with comorbid sleep disturbances.
The aim of the present study was therefore to further characterize the overlapping genetic architecture [the number of genomic loci contributing to risk, the distribution of their allelic frequencies and effect sizes, and the interactions of alleles in and between genes (23,24)] of sleep-related phenotypes and psychiatric disorders, beyond genetic correlation, by utilizing results from large GWASs and relevant statistical genetics tools. To do so, we employed a stepwise approach. First, MiXeR (25) was used to quantify the trait-specific and shared polygenic architecture of sleep-related phenotypes and psychiatric disorders. Second, we utilized the conditional false discovery rate/conjunctional false discovery rate (condFDR/ conjFDR) approach (26)(27)(28) to detect novel phenotype-specific variants and shared loci between sleep-related phenotypes and psychiatric disorders. Finally, we performed functional annotation and gene mapping of these identified specific loci. While genetic correlation provides an aggregate measure for the balance of variants with the same and opposite effects on a given pair of traits, both MiXeR and condFDR/conjFDR are able to identify genetic overlap regardless of effect direction (25,28,29). These methods, therefore, complement genetic correlation to provide a more comprehensive overview of the genetic relationships between traits.

GWAS Data
GWAS summary statistics for BIP and SCZ were obtained from the Psychiatric Genomics Consortium. The BIP sample comprised 20,352 cases and 31,358 controls (30). Data for the SCZ sample included 33,640 cases and 43,456 controls (31). Data on DEP were obtained from the Psychiatric Genomics Consortium and 23andMe, Inc., and included 135,458 cases and 344,901 controls (32). Detailed description of each sample is provided in the corresponding publications (30)(31)(32).
GWAS data for insomnia, chronotype, and sleep duration were obtained from publicly available summary statistics, derived from the UK Biobank (detailed definition of each phenotype is provided in Supplement 1). The insomnia sample included 109,402 cases and 277,131 controls (21), and the chronotype sample included 449,734 individuals (20). Owing to data sharing restrictions, the insomnia (944,477) and chronotype (248,100) cohorts from 23andMe could not be included (20,21). The chronotype GWAS was analyzed such that reported allelic effects are specific to morningness. Finally, sleep duration was treated as a continuous variable and included a sample of 446,118 individuals (22).
For all GWAS samples used, summary statistics were generated only from individuals of European ancestry. All GWASs investigated in the present study were approved by the relevant ethics committees, and informed consent was obtained from all participants. The Norwegian Institutional Review Board for the South-East Norway Region has evaluated the current protocol and found that no additional institutional review board approval was needed because no individual data were used.

Data Analysis
We generated conditional quantile-quantile plots to visually assess the cross-phenotype polygenic enrichment, conditioning each of the psychiatric disorders on the sleep phenotypes and vice versa. After observing cross-phenotype polygenic enrichment, we sought to quantify the genetic overlap between investigated phenotypes. To do so, we applied causal mixture models (25) to the GWAS summary statistics, using MiXeR (version 1.3). MiXeR provides univariate estimates of the number of trait-influencing loci for each trait of interest as well as bivariate estimates of genome-wide genetic overlap between pairs of traits. In these cross-trait analyses, MiXeR models additive genetic effects as a mixture of four components, representing single nucleotide polymorphisms (SNPs) not affecting either trait, SNPs affecting only one of the traits, and SNPs affecting both traits. These components are then plotted in Venn diagrams. After fitting parameters of the model, the Dice coefficient (DC) (the proportion of SNPs shared by two traits out of the total number of SNPs estimated to influence both traits) and genetic correlation were also calculated. In addition to the MiXeR analyses of psychiatric disorders and sleep-related phenotypes, we included two additional comparisons as positive and negative controls (Table S1 in Supplement 2). To illustrate substantial genetic overlap estimates between two traits with high genetic correlation (positive control), we included MiXeR estimates for overlap between SCZ and BIP. Similarly, to illustrate little genetic overlap estimates between two traits with low genetic correlation (negative control), we included MiXeR estimates for the overlap between SCZ and height (33). Further details are provided in the Supplemental Methods in Supplement 1 and in the original publication (25), and all code is available online (https://github.com/precimed/mixer).
Next, we employed the condFDR/conjFDR method (26)(27)(28), which allows for the identification of specific shared loci between pairs of traits, thus complementing the genome-wide genetic overlap observed with MiXeR. The condFDR method utilizes genetic association summary statistics from a trait of interest (psychiatric) together with those of a conditional trait (sleep-related) to estimate the posterior probability that an SNP has no association with the primary trait, given that the p values for that SNP in both the primary and conditional traits are lower than the observed p values. This method increases the power to identify loci associated with the primary trait by leveraging associations with conditional traits, thereby reranking SNPs compared with the original GWAS p value Genetic Overlap Between Psychiatric and Sleep Traits ranking. The conjFDR statistic is defined as the maximum of the two mutual condFDR values and is a conservative estimate of the posterior probability that an SNP has no association with either trait, given that the p values for that SNP in both the primary and conditional traits are lower than the observed p values. The conjFDR method thus allows the identification of loci associated with both traits. An FDR level of .01 per pairwise comparison was set for condFDR and conjFDR, corresponding to 1 false positive per 100 reported associations. We excluded SNPs around the extended major histocompatibility complex region and chromosome 8p23.1 (genome build 19 locations chr6:25119106-33854733 and chr8:7200000-12500000, respectively) before fitting the FDR model, because their intricate regional linkage disequilibrium (LD) may bias condFDR/conjFDR estimation (34). All code used to perform the described analyses is available online (https:// github.com/precimed/pleiofdr). More details about the condFDR/conjFDR methods can be found in the original publications (26,27) and subsequent review (28).

Genomic Loci Definition
Independent genomic loci were defined according to the FUMA protocol (Supplemental Methods in Supplement 1) (35). We evaluated the directional effects of shared loci by comparing z scores from the respective GWAS summary statistics.

Functional Annotation
Positional and functional annotation of all candidate SNPs, in the genomic loci with a conjFDR value , .10 having an LD r 2 $ 0.6 with one of the independent significant SNPs, was performed using multiple tools, implemented in FUMA (Supplemental Methods in Supplement 1) (35). In addition, we linked lead SNPs to genes using three gene-mapping strategies (35): 1) positional mapping to align SNPs to genes based on their physical proximity, 2) expression quantitative trait locus (eQTL) mapping to match cis-eQTL SNPs to genes whose expression is associated with allelic variation at the SNP level, and 3) chromatin interaction mapping to link SNPs to genes based on three-dimensional DNA-DNA interactions between each SNP's genomic region and nearby or distant genes. All gene-mapping strategies were limited to brain tissues, otherwise all other default settings in FUMA were used (35). Finally, we queried SNPs for known QTLs in brain tissues using the GTEx portal (GTEx, version 8) (36) and PsychEncode database (37).

Cross-trait Polygenic Enrichment
We observed cross-trait polygenic enrichment within stratified conditional quantile-quantile plots when conditioning the investigated psychiatric disorders on each of the sleep-related phenotypes and vice versa ( Figure S1 in Supplement 1), indicative of polygenic overlap.

Quantification of Genetic Overlap With MiXeR
MiXeR estimated a wide range (|r g | = 0.04 to 0.42) in the measure of genetic correlation between each of the psychiatric disorders and sleep-related phenotypes (Figure 1), consistent with previous studies (20)(21)(22). Despite this range in correlation strengths, considerable genetic overlap was observed between all psychiatric and sleep-related traits ( Figure 1 and Table 1). The largest amount of genetic overlap was observed between SCZ and chronotype (DC = 80%), while the set of variants influencing DEP was shown to fully encompass the variants estimated to influence sleep duration. Both SCZ (DCs: SCZ and chronotype = 80%, SCZ and insomnia = 76%, SCZ and sleep duration = 78%) and BIP (DCs: BIP and chronotype = 64%, BIP and insomnia = 78%, BIP and sleep duration = 76%) showed considerable overlap with all sleeprelated phenotypes despite a wide range of genetic correlation results (r g = 20.12 to 0.19). The highest genetic correlation was observed between DEP and insomnia (r g = 0.42); however, MiXeR was unable to accurately quantify the genetic overlap between DEP and insomnia, as evidenced by the negative Akaike information criterion scores indicative of poor model fit. This is likely due to the high polygenicity and low heritability estimated for these traits (Table S1 in Supplement 2). Despite identifying large genetic overlap between the investigated sleep-related phenotypes and psychiatric disorders, moderate congruency of variant effects within the shared component was observed for most bivariate analyses ( Table 1). As expected, MiXeR estimated considerable overlap between the SCZ and BIP (DC = 87%) and little overlap between SCZ and height (DC = 18%) (Table S1 in Supplement 2). Log-likelihood and bivariate density plots, illustrating the relationship between the GWAS test statistics, for each pairwise comparison are presented in the Supplemental data ( Figure S2 in Supplement 1).
When considering the number of shared SNPs as a proportion of the total polygenicity of each trait (Table 1), nearly all SNPs affecting both chronotype and sleep duration also influence DEP. Similar patterns were observed between SCZ and chronotype, SCZ and sleep duration, and BIP and sleep duration. The lowest proportion for any sleep-related phenotype shared with a psychiatric disorder was observed for BIP and chronotype. In contrast, the highest proportion of SNPs affecting psychiatric disorders that also influence sleep-related phenotypes was identified for BIP and insomnia.

Shared and Novel Loci for Psychiatric Disorders and Sleep-Related Phenotypes
For the conjFDR analyses of BIP and the sleep-related phenotypes, 9, 1, and 6 LD-independent loci were significantly (conjFDR , .01) associated with, and shared between, BIP and chronotype, insomnia, and sleep duration, respectively (Tables S2 and S3 in Supplement 2). Evaluation of these loci revealed that none were novel for BIP (Table S2 in Supplement 2). A Manhattan plot from these conjFDR analyses is presented in Figure 2A. Twenty-two credible genes were mapped to these loci using all three mapping strategies (Table S4 in Supplement 2).
A total of 22, 12, and 21 LD-independent loci were significantly (conjFDR , .01) associated with, and shared between, DEP and chronotype, insomnia, and sleep duration, respectively (Tables S6 and S7   SCZ) and sleep-related phenotypes (right circle; chronotype, insomnia, and sleep duration) and unique (colors) to either of them. The number of trait-influencing variants in thousands is shown, with the standard error in thousands provided in parentheses. The size of the circles reflects the polygenicity of each phenotype, with larger circles corresponding to greater polygenicity and vice versa. The estimated genetic correlation for each pair is also shown below the corresponding Venn diagram, with an accompanying directional scale (negative, blue shades; positive, red shades). *MiXeR was unable to accurately quantify the genetic overlap between DEP and insomnia because of their low heritability and high polygenicity (see Table S1 in Supplement 2 and Figure S2 in Supplement 1 for further details). BIP, bipolar disorder; DEP, depression; SCZ, schizophrenia. The number of overlapping single nucleotide polymorphisms in each bivariate analysis are presented as a proportion of the total polygenicity of each trait. MiXeR could not accurately quantify the polygenic overlap between depression and insomnia, and as a result, these values were nd. BIP, bipolar disorder; DEP, depression; nd, not determined; Psych, psychiatric; SCZ, schizophrenia.
Genetic Overlap Between Psychiatric and Sleep Traits DEP and sleep duration are novel risk loci for DEP (Table 2; Table S6 in Supplement 2). A Manhattan plot from these conjFDR analyses is presented in Figure 2B. When considering all DEP-associated loci, 29 credible genes were identified using all three mapping strategies (Table S8 in Supplement 2).  Tables 1 and 2, and Tables S3, S7, and S11 in Supplement 2. BIP, bipolar disorder; conjFDR, conjunctional false discovery rate; DEP, depression; SCZ, schizophrenia; SLEEP, sleep duration; SNP, single nucleotide polymorphism.

Genetic Overlap Between Psychiatric and Sleep Traits
Finally, conjFDR analyses revealed 43, 7, and 34 LDindependent loci significantly (conjFDR , .01) associated with, and shared between, SCZ and chronotype, insomnia, and sleep duration, respectively (Tables S10 and S11 in Supplement 2). Three loci shared between SCZ and chronotype and 3 loci shared between SCZ and sleep duration are novel risk loci for SCZ ( Table 2; Table S10 in Supplement 2). A Manhattan plot from these conjFDR analyses is presented in Figure 2C. Gene mapping analysis using lead SNPs within the SCZ-associated loci implicated 32 credible genes.
Of the genes mapped to each of the shared loci described above, 70 unique credible genes were mapped by all three gene-mapping strategies (Table 4; Tables S4, S8, and S13 in Supplement 2). The LOC100127955 gene was mapped to lead SNPs of shared loci for all three psychiatric disorders, as well as chronotype and sleep duration (Table 4). A further 19 credible genes were mapped to loci for at least three of the investigated phenotypes ( Table 4). The remaining 50 credible genes were unique to specific bivariate analyses (Table 4). Results from network analysis identified multiple relationships between these credible genes, including physical interactions, gene coexpression, genetic interactions, and predicted and shared protein domains ( Figure S3 in Supplement 1 and Table S15 in Supplement 2).
In total, all the shared loci identified using the conjFDR methodology are represented by 146 unique lead SNPs, of which 49 were identified as eQTLs for 115 unique genes within 12 different brain tissues in the GTEx database (Table S12 in Supplement 2). Implicated brain regions include the basal ganglia and related structures, as well as the cortex, hippocampus, and cerebellum. These brain tissues were identified by lead SNPs within shared loci between chronotype and sleep duration and all investigated psychiatric disorders. Moreover, 76 lead SNPs were identified as QTLs in the PsychEncode database (Table S16 in Supplement 2). No significant eQTLs were identified for lead SNPs shared between psychiatric disorders and insomnia. Further information related to the functional annotation and gene-mapping results of the specific identified loci is available in Supplement 1.

DISCUSSION
The current study employed GWAS summary statistics from sleep-related phenotypes and psychiatric disorders to better understand the shared underlying genetic architecture. Using the novel MiXeR tool (25), our results provide further evidence of extensive genetic overlap between sleep-related phenotypes and psychiatric disorders, which is consistent with the strong phenotypic correlations and comorbidity observed in the clinic, as well as previous genetic findings. We identified substantial genetic overlap between major psychiatric disorders (BIP, DEP, and SCZ) and sleep-related phenotypes (chronotype, insomnia, and sleep duration), beyond genetic correlation. Further, we identified 146 specific shared loci between pairs of sleep-related phenotypes and psychiatric disorders, of which 42 were novel for at least one of the investigated phenotypes. Finally, we annotated these specific The most strongly associated SNPs in independent genomic loci shared between indicated psychiatric disorders and sleep-related phenotypes at conjFDR , .01 after merging regions ,250 kb apart into a single locus. The table presents Chr, credible mapped gene, and nearest gene, as well as p values and z scores from the original genome-wide association study summary statistics. A1, allele 1; A2, allele 2; Chr, chromosomal position; conjFDR, conjunctional false discovery rate; DEP, depression; SCZ, schizophrenia; SNP, single nucleotide polymorphism.
a Novel for the indicated sleep-related phenotype as well. For more details and a list of all candidate variants in these loci, see Tables S6, S7, S10, and S11 in Supplement 2. shared loci and identified a subset of 70 credible genes and numerous eQTLs for multiple brain tissues, including the basal ganglia and related structures, as well as the cortex, hippocampus, and cerebellum. The MiXeR results showed similar polygenicity for the majority of traits investigated (~8000-10,000 trait-influencing variants), except for depression, which had a much greater polygenicity (~14,000 trait-influencing variants), highlighting the genetic complexity of all of these traits. Moreover, the results identified extensive genetic overlap between sleeprelated phenotypes and psychiatric disorders, consistent with previous literature. In a scenario with a substantial overlap of genetic loci and moderate genetic correlation, there is likely a mix of agonistic and antagonistic effect directions among the shared variants, consistent with the concordance rates of approximately 50% observed between traits. This implicates shared molecular biological processes between sleep-related phenotypes and psychiatric disorders, which have mixed effect directions within and between phenotypes. These results are in line with the growing evidence for shared genetic architecture between related phenotypes (38)(39)(40) indicating mixed effect directions. These results suggest that the genetic architecture and biological processes shared by sleep phenotypes and psychiatric disorders are present in one type of trait (e.g., sleep phenotype) even in the absence of the other type (e.g., psychiatric disorder). The current findings may indicate that the clinical observations of sleep problems increasing risk for psychiatric disorders and vice versa (15) may still have a genetic influence even in the absence of genetic correlation. Heterogeneity among individuals affected by the same psychiatric disorder is observed through wide distributions of polygenic risk scores (that overlap largely with unaffected control individuals) (41) as well as in brain morphology (42) and is masked at the group level. Thus, patterns of concordance and discordance, within specific biological processes and pathways, may interplay with the environment to drive the phenotypic comorbidities within particular disorder subgroups. For example, we may speculate that altered brain morphology and neurotransmitter signaling are both biological processes that might influence phenotypic comorbidity between psychiatric disorders and sleep-related phenotypes. Moreover, the interplay of these and other biological processes are also likely to modulate effects at the phenotypic level. Specific subgroups of individuals may have high concordance/discordance within genes implicated in one or both of these biological processes, resulting in variance in related traits and comorbidities. These complex relationships are not detected at the group level, as is the case for the participants included in the GWAS summary statistics used in this study. As GWAS sample sizes increase, deep phenotyping of included participants for known comorbidities will be essential to investigate potential subgroups.

Genetic Overlap Between Psychiatric and Sleep Traits
Leveraging the extensive polygenic overlap between sleeprelated phenotypes and psychiatric disorders, we used the condFDR/conjFDR method to identify the specific genetic loci involved. Numerous shared loci were identified between each pair of investigated phenotypes. Among these, novel loci were identified for DEP (n = 8), SCZ (n = 6), chronotype (n = 12), insomnia (n = 3), and sleep duration (n = 29), demonstrating the utility of this statistical framework to increase discovery and The most strongly associated SNPs in independent genomic loci shared between indicated psychiatric disorders and sleep-related phenotypes at conjFDR a Novel for the indicated psychiatric disorder as well. For more details and a list of all candidate variants in these loci, see Tables S2, S3, S6, S7, S10, and S11 in Supplement 2.
Genetic Overlap Between Psychiatric and Sleep Traits uncover polygenic overlap between complex phenotypes (28).
In-silico analyses of these novel loci indicate potential regulatory functionality and deleterious effects on protein structure and function, as well as QTL effects in a number of brain tissues for identified lead SNPs. Moreover, we determined a subset of 70 unique credible genes that map to specific identified shared loci based on three independent gene-mapping strategies. Network analysis highlights that this subset of genes are predominantly linked by physical interactions and gene coexpression. One gene, the noncoding RNA LOC100127955, was mapped to lead SNPs associated with BIP, DEP, SCZ, chronotype, and sleep duration. Variants within this gene have also been implicated in other behavioral phenotypes, including autism spectrum disorder (43), risk-taking behavior (44), and age of smoking initiation (45). The function of this gene is currently unknown, and gene coexpression results indicate that it is most likely coexpressed with other noncoding RNA genes (46). However, LOC100127955 shows higher expression in brain tissue when compared with other tissues, specifically within regions of the cortex, the amygdala, and the hippocampus ( Figure S4 in Supplement 1). It is tempting to speculate that the expression of LOC100127955 in these brain regions provides some clue as to the possible mechanisms through which it may modulate sleep-related phenotypes and development of psychiatric disorders; however, functional studies are required to better understand the role of this gene and the possible biological processes that it may contribute to.
Another 19 genes were mapped to lead SNPs for at least three phenotypes. Interestingly, however, the majority of credible mapped genes were specific to particular bivariate analyses of sleep-related phenotypes and psychiatric disorders. These results further emphasize the complex underlying genetic architecture and biology of these phenotypes, suggesting the involvement of common shared processes, mechanisms, and pathways, as well as phenotype-specific aspects. Although further research is required to validate these results, future studies should interrogate the genes identified here and their associated mechanisms and pathways to improve our understanding of the shared genetic etiology of sleep-related phenotypes and psychiatric disorders. Moreover, it should be noted that the gene mapping performed in this study, based on identified specific loci, represents only a small fraction of the genetic architecture of the traits investigated. At any rate, while these findings may not be representative of the complete genetic etiology of these traits, they are informative and provide some basis to further explore the underlying shared genetic architecture of sleep-related phenotypes and psychiatric disorders.
This study applied MiXeR to quantify the number of unique and shared genetic variants between psychiatric disorders and sleep-related traits and the congruency of variants within the shared component. Following this, a subset of specific variants within this shared component were identified and annotated. Despite this, the exact role of variants within this shared component, their interplay with environmental and lifestyle factors, and how this may modulate phenotypic comorbidities require further investigation. Furthermore, it is also possible that common variants cannot fully explain the overall comorbidity, but that it is the psychiatric disorder itself that is causative or that other environmental and lifestyle factors or rare Unique genes mapped to lead SNPs, of shared loci between psychiatric disorders and sleep-related phenotypes, by the three employed mapping strategies: 1) positional mapping, 2) expression quantitative trait locus mapping, and 3) chromatin interaction mapping. Additional details are provided in Tables S2, S6, and S10 in Supplement 2. BIP, bipolar disorder; DEP, depression; SCZ, schizophrenia; SNP, single nucleotide polymorphism. The main limitation to the condFDR/conjFDR method, which it inherits from the GWAS it draws upon, is that it is agnostic to specific causal variants underlying the shared loci. These shared loci could arise because of both shared or separate causal variants, or mediated pleiotropy, where one phenotype is causative for the other (28,47). Whenever separate causal variants underlie a specific shared locus, this has implications for the functionality of that locus in each of the associated traits because separate variants may influence different genes or other biological processes. Mediated pleiotropy may result in identification of a false positive shared locus where the causal variant is only truly associated with one of the traits (28). Moreover, both condFDR/conjFDR and MiXeR are limited to bivariate analyses, so it is not possible to determine whether the identified genetic overlap is influenced by additional factors, such as social stratification and socioeconomic status, as has recently been shown (48)(49)(50). A final limitation is that patients with self-reported psychiatric disorders were not excluded from the sleep phenotype GWAS samples (20)(21)(22). Although the number of individuals with BIP and SCZ is low in the UK Biobank (n~1000), there are a large number of individuals with self-reported or diagnosed depression, which may result in inflated MiXeR shared components and/or false positive shared loci between sleep phenotypes and DEP.
In conclusion, we have demonstrated novel and substantial polygenic overlap between psychiatric disorders (BIP, DEP, and SCZ) and sleep-related phenotypes (chronotype, insomnia, and sleep duration). These findings highlight a greater quantity of genetic overlap than that indicated by genetic correlation. Interestingly, the majority of bivariate analyses showed only moderate congruency of effect direction for genetic variants within the shared component, suggestive of a complex genetic relationship and potential subgroups with higher or lower genetic concordance. Future studies and deep phenotyping may allow for analysis of these subgroups to improve risk stratification of individuals with comorbid sleep disturbances and psychiatric disorders. Moreover, we have identified numerous shared loci and a subset of credible genes likely to play a role in the underlying genetic etiology of these phenotypes, which may represent novel drug targets and thus opportunities for personalized approaches to treatment.