Genome-wide analyses reveal novel opioid use disorder loci and genetic overlap with schizophrenia, bipolar disorder, and major depression.

Opioid use disorder (OUD) and mental disorders are often comorbid, with increased morbidity and mortality. The causes underlying this relationship are poorly understood. Although these conditions are highly heritable, their shared genetic vulnerabilities remain unaccounted for. We applied the conditional/conjunctional false discovery rate (cond/conjFDR) approach to analyse summary statistics from independent genome wide association studies of OUD, SCZ, BD and MD. Next, we characterized the identified shared loci using biological annotation resources. OUD data was obtained from the Million Veteran Program (15,756 cases 99,039 controls). SCZ (53,386 cases 77,258 controls), BD (41,917 cases 371,549 controls) and MD (170,756 cases 329,443 controls) data was provided by the Psychiatric Genomics Consortium. We discovered genetic enrichment for OUD conditional on associations with SCZ, BD, MD and vice versa, indicating polygenic overlap with identification of 14 novel OUD loci at condFDR<0.05 and 7 unique loci shared between OUD and SCZ (n=2), BD (n=2) and MD (n=7) at conjFDR<0.05 with concordant effect directions, in line with estimated positive genetic correlations. Two loci were novel for OUD, one for BD and one for MD. Three OUD risk loci were shared with more than one psychiatric disorder, at DRD2 on chromosome 11 (BD and MD), at FURIN on chromosome 15 (SCZ, BD and MD), and at the major histocompatibility complex region (SCZ and MD). Our findings provide new insights into the shared genetic architecture between OUD and SCZ, BD, and MD, indicating a complex genetic relationship, suggesting overlapping neurobiological pathways.


INTRODUCTION
Opioid use disorder (OUD) causes substantial morbidity and mortality worldwide 1 . Some countries are more affected, such as the USA with the opioid crisis 2 . Although OUD is less prevalent than other substance use disorders, OUD is an enormous public health burden with high rates of overdose deaths, which increased during the COVID-19 pandemic 3 .
OUD is relatively prevalent in patients with severe mental disorders, such as schizophrenia (SCZ), bipolar disorder (BD) and major depression (MD) [4][5][6] , all of which are associated with affective and psychotic symptoms to various degrees. Comorbidities lead to greater suffering compared to having either disorder alone 7 , and often impede treatment 4 .
In patients with severe mental disorders the estimated prevalence of co-occurring OUD was 2.6-5.3% 4 .
Among adults with OUD the prevalence of a co-occurring severe mental disorder was estimated to be 27 % 8 . At a sub-diagnostic threshold, an increased risk of nonmedical opioid use was reported for patients who had already been diagnosed with a severe mental disorder 9 . Similarly, an increased risk of developing severe mental disorders for patients with existing nonmedical opioid use was also found 9 . Further, there are several lines of evidence suggesting an overlapping neurobiological substrate related to the reward system and dopamine across mental illness in general 10 , OUD 11 , mood disorders 12 , and psychotic disorders 13 . E.g., can negative symptoms in SCZ, a category of internal heterogeneity, be conceptualized as a reward processing impairment 14 . Better understanding of the mechanisms underlying these comorbidities is crucial for improving treatment and quality of life of affected individuals.
OUD and severe mental disorders can increase the risk of developing one another, have common environmental and genetic risk factors 15 . Both OUD and severe mental disorders have moderate to high heritability estimates from twin and family studies, with OUD at 0.50 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 9, 2022. ; https://doi.org/10.1101/2022.09.09.22279755 doi: medRxiv preprint 16 SCZ at 0.80 17 , BD at 0.70 18 and MDD at 0.35 19 . Recent progress in genotyping technology and international collaborations assembling large genome-wide association studies (GWAS) have provided novel insights into the genetic architecture of these complex disorders. A key discovery is that these disorders are highly polygenic 20 , i.e., associated with many genetic risk variants, each with a small effect. Genetic studies have also suggested shared genetic aetiologies across severe mental disorders 15 , including recent research which has found significant concordant genetic correlations (r g ) between substance use disorders and severe mental disorders 16 with genetic correlation estimates between OUD and SCZ at 0.29, for BD at 0.16 and MD at 0.35 21 , suggesting common genetic risk factors. However, genetic correlation is a genome-wide measure, which includes the effects of all Single Nucleotide Polymorphisms (SNPs), and thus does not identify overlap at the individual locus level. Moreover, genetic correlation is unable to capture genetic overlap in the presence of shared genetic variants with a mixture of same and opposite effect directions as they "cancel each other out" 22 . In recent years, genome-wide analyses demonstrated genetic overlap with mixed effect directions among a wide range of human traits and disorders, regardless of their genetic correlations 23 . Therefore, investigating genetic overlap beyond genetic correlation is required to further elucidate the genetic architecture of complex disorders and their genetic relationships. Here, we aimed to reveal more of the shared genetic architecture between OUD and the mental disorders SCZ, BD and MD by applying analytical tools designed for polygenic architectures 24 . We applied the conditional/conjunctional false discovery rate (cond/conjFDR) tools to boost discovery of genetic variants and to identify overlapping genetic loci between OUD, SCZ, BD and MD, beyond genetic correlation 24 .

GWAS Samples
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 9, 2022.  21 were used. In the MVP cohort, case and control status were defined using electronic health record-data. There were 8,529 affected European American (EA) individuals and 71,200 opioid-exposed EA controls, 4,032 affected African American (AA) individuals and 26,029 opioid-exposed AA controls. The Yale-Penn and SAGE datasets included 2,015 EA cases and 963 controls and 1,180 AA cases and 847 controls from a previous GWAS 25 . To ensure compatibility of linkage disequilibrium (LD) pattern, our main analysis on OUD focused on individuals of EA ancestry and included a total number of 10,544 cases and 72,163 opioid-exposed controls.
GWAS data on SCZ (53,386 cases and 77,258 controls) 26 , BD (41,917 cases and 371,549 controls) 27 and MD (170,756 cases and 329,443 controls) 28 were obtained from the Psychiatric Genetics Consortium (PGC) and were of European ancestry. See Supplementary methods for details.

Statistical Analyses
We visualized cross-trait enrichment using conditional quantile-quantile (Q-Q) plots, where distribution of p-values of all SNPs of a primary phenotype and for strata defined by the pvalues for association with a secondary phenotype are represented. Cross-trait enrichment is evident if enrichment of statistical associations in the primary phenotype increases with increased significance of association with the secondary phenotype 29 . We then applied the conditional FDR (condFDR), which leverages cross-trait enrichment between two phenotypes to improve genetic discovery. CondFDR readjusts the test-statistics in the primary phenotype by conditioning on SNP associations with the secondary phenotype, returning a condFDR . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 9, 2022. ; https://doi.org/10.1101/2022.09.09.22279755 doi: medRxiv preprint value. OUD loci were identified at condFDR<0.05. We then applied the conjunctional FDR (conjFDR) tool to increase discovery of shared genetic loci jointly associated with OUD and in turn SCZ, BD and MD. The conjFDR approach is an extension of condFDR 29 . Reversing the order of the phenotypes gives the condFDR value for the second phenotype conditioned on the first phenotype. ConjFDR then identifies SNPs which are significantly associated with both phenotypes 24 . Shared loci were found at conjFDR<0.05, in line with prior literature 30 .
To control for spurious enrichment, random pruning was averaged over 500 iterations, and one SNP in each LD block (r 2 >0.1) was randomly selected for each iteration. We excluded SNPs within the major histocompatibility complex (MHC) (chr6:25000000-33000000), the chromosomal region 8p23.1 (location 7200000-12500000) and the gene MAPT (chr17:40000000-47000000) before fitting the FDR model given their complex LD structures that may bias FDR estimation 31 . We performed conjFDR analyses in independent samples for OUD (5,212 cases and 26,876 controls) 21 , SCZ (22,778 cases and 35,362 controls) 32 , BD (4,501 cases and 192,220 controls) [https://r5.finngen.fi/], and MD (75,607 cases and 231,747 controls) 33 (Table 1). See Supplementary methods for details. We then conducted a version of the binomial test, i.e., sign concordance test 34 . Finally, we tested for variants nominally significant at p<0.05 in each independent sample. All analyses were conducted in Oslo, Norway.

Genetic locus definition
We defined independent genetic loci according to the FUMA protocol 35 . Briefly, independent significant genetic variants were identified as variants with conjFDR<0.05 and LD r2<0.6 with each other. A subset of these independent significant variants with LD r2<0.1 were then selected as lead variants. For each lead variant all candidate variants were identified as variants with LD r2≥0.6 with the lead variant. For a given lead variant the . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 9, 2022. ; https://doi.org/10.1101/2022.09.09.22279755 doi: medRxiv preprint borders of the genetic locus were defined as min/max positional coordinates over all corresponding candidate variants. Loci were then merged if they were separated by less than 250kb. LD information was calculated from the 1000 Genomes Project European-ancestry reference panel 36 . Directional effects of loci were evaluated by comparing z-scores of the lead SNPs. We investigated all discovered loci for overlap with previously identified loci using our internal database which includes updated GWAS on all the traits investigated in this study.

Functional annotation of loci shared between OUD and severe mental disorders
The FUMA application SNP2GENE was run to identify genes mapped to the candidate SNPs with a conjFDR value <0.1 35 . All analyses were corrected for multiple comparisons. To minimize false positive and negative findings, we implemented a gene mapping strategy of qualification of at least two out of three of the following criteria: physical proximity of SNPs to genes, expression Quantitative Trait Loci (eQTL) and chromatin interaction mapping. SNP deleteriousness was measured by a Combined Annotation-Dependent Depletion (CADD)score above 12.37 37 . For complementarity we also used the open-source Open Targets Genetics application, Variant to Gene (V2G) 38 to map lead SNPs to genes. V2G utilizes physical proximity of SNPs to genes, molecular phenotype quantitative trait loci investigations (QTL) which maps SNPs to genes where expression level is influenced by allelic variation at the SNP level (eQTL), and protein Quantitative Trait Loci (pQTL) which maps SNPs to genes where protein level is influenced by allelic variation at the SNP level, and chromatin interaction where 3D DNA-DNA interactions are considered. V2G leverages this information in machine learning algorithms on the input of lead SNPs.

Cross-trait enrichment
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 9, 2022. ; https://doi.org/10.1101/2022.09.09.22279755 doi: medRxiv preprint To visualize cross-trait enrichment, we constructed stratified quantile-quantile (Q-Q) plots that show successive increments of SNP enrichment for OUD conditional on increasing levels of SNP associations with SCZ, BD and MD (Figure 1 A-C). We then reversed the stratified Q-Q plots displaying the SNP associations for SCZ, BD and MD conditional on SNP associations with OUD ( Figure 1 D-F). In all cases we observed successive upward and leftward deflection of the Q-Q plots after conditioning the primary phenotype on the conditional phenotype, indicative of genetic overlap between the phenotypes.

CondFDR OUD loci
We applied condFDR to boost discovery of OUD loci conditional on mental disorders, and identified a total of 20 loci for OUD, including 14 novel (condFDR<0.05) ( Table 2). This

Shared loci between OUD and mental disorders (conjFDR)
To identify shared loci between OUD and the psychiatric disorders we performed conjFDR analyses with a threshold at <0.05 ( Figure 2). We identified in total seven loci, two of which are novel for OUD and shared with MD at RN7SKP157 and B3GALTL (Table 3). One locus was novel for BD at DRD2 and one was novel for MD at PPPC6 (Table 3). For all lead SNPs in the shared loci, the effect directions in OUD and each psychiatric disorder were concordant, i.e., risk of OUD was linked to higher risk of mental illness (Supplementary   Tables 4-6). As expected, we found a locus shared between OUD, SCZ and MD located to the MHC region which has previously been implicated in these disorders at the genome wide significant level [39][40][41] . This overlapping genetic signal suggests involvement of the immune . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 9, 2022. ; https://doi.org/10.1101/2022.09.09.22279755 doi: medRxiv preprint system in the shared genetic risk of these disorders. However, the extended MHC region has highly complex LD structure spanning a large number of genes. We cannot reliably infer whether this overlapping signal reflects separate or shared loci, or causal genes.  Table 10). Thirty of these were left after filtering on two out of three mapping criteria (positional, eQTL, and chromatin interaction information). We identified 19 genes shared between OUD and BD in two loci (Supplementary Table 11). Five of these passed filtering with two out of three mapping strategies. Finally, we identified 145 genes shared between OUD and MD in seven loci (Supplementary Table 12). Thirty-six genes passed the filtering strategy.

Validation analyses
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

DISCUSSION
In the current study, we used the cond/conjFDR analytical method and leveraged cross-trait enrichment between OUD, and SCZ, BD and MD to improve the statistical power for discovery of shared genetic loci to shed light on their genetic relationships. We identified polygenic overlap beyond genetic correlation. Specifically, we discovered 14 novel OUD loci and 7 novel loci jointly associated with OUD and psychiatric disorders (Figure 2, Table 2 and 3). There was a concordant allelic direction of effects in all the shared loci, indicating that these genetic variants increase the risk of both OUD and the respective psychiatric disorders.
Several lines of evidence suggest a pleiotropic nature of mental traits and disorders 44 . Our findings are in line with the observation that multiple genetic variants with small effect sizes . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 1 1 influence many traits to different degrees 17 . The currently identified shared loci explain only a small fraction of the genetic architecture of OUD and severe mental disorders and, thus, the liability to these disorders. We identified 14 novel loci for OUD, but there are still numerous undetected common variants, which will be identified with larger GWAS samples 17  We identified a pleiotropic locus on chromosome 15 (lead SNP rs4702) which was shared between OUD, SCZ, BD and MD at conjFDR<0.05. It has previously been associated with SCZ 40 , BD 27 and MD 41 . This locus was not identified in the original OUD GWAS 21 , but reached genome-wide significance in subsequent GWAS on OUD, validating this finding 39,42,43 . The lead SNP rs4702 is shown to influence the expression of its nearest gene FURIN in neurons derived from human induced pluripotent stem cells, and influence synaptic function in synergy with other SCZ risk variants 50 . FURIN encodes a cleaving and activating enzyme involved in cleavage of the endogenous opioid proenkephalin 51 , which is found in most parts of the brain and nervous system 52 . We discovered a novel OUD locus at chromosome 5 shared with MD, which has previously been associated with both SCZ 32 and MD 53 . The nearest gene to the lead SNP is pseudogene RN7SKP157, while the most likely causal genes according to the V2G algorithm were KIF2A and DIMT1. We also discovered a novel OUD locus at chromosome 13, implicating three genes (HSPH1, RXFP2 and . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted September 9, 2022. ; https://doi.org/10.1101/2022.09.09.22279755 doi: medRxiv preprint B3GALTL). This locus is shared with MD and has previously been implicated in SCZ 32 and MD 28 . FUMA and V2G both identified the nearest gene and most likely implicated gene as B3GALTL, respectively. We identified another OUD risk locus on chromosome 11 which was shared with both BD and MD. The locus is novel for BD and has previously been implicated in OUD 43 , MD 41 , as well as for alcohol use disorder 54  NCAM1 has previously been found to be associated with SCZ, BD, MD, alcohol use disorder and cannabis use disorder 54 . Sequence variants in the DRD2 gene has been associated with SCZ 58 , Parkinson's disease 59 and opioid addiction 60 .
The identified genetic variants seem to support neurobiological hypotheses about the reward system in substance use disorders (SUD) and severe mental disorders (SMDs). The identified loci include variants that may result in changed function of the mesocorticolimbic circuit, influenced by altered metabolism of opioids (FURIN variants) and altered dopamine receptor function (DRD2 variants). Such functional modification of the reward system may lead to an increased risk of SUDs and SMDs, for example via adapted dopamine transmission leading to reduced reward processing 14 which may underlie negative psychotic symptoms and depressive mood, and increased susceptibility for exogenous opioids due to elevated dopamine release 61 resulting in higher risk of opioid use disorder. Our findings warrant experimental validation to elucidate the pathophysiological mechanisms of the reward circuitry underlying comorbid SUDs and SMDs.
The current study has some limitations. We cannot exclude sporadic, non-systematic sample overlap; however, we calculated linkage disequilibrium score regression genetic covariance intercepts and neither is significant (all p-values>0.05), suggesting no significant effect of . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 1 3 overlap between the samples. We included datasets from a diverse set of populations, but larger datasets for other cohorts than Europeans are needed to uncover more of the genetic underpinnings of OUD and SMDs.
To conclude, our study suggests overlapping genetic architecture between OUD and SCZ, BD, and MD, indicating similarities in the genetic architectures of these debilitating disorders. We highlight loci previously associated with SUDs and suggest underlying molecular pathways for OUD. Uncovering the underlying genetics of these disorders can lead to improvements in patient stratification and identification of novel targets for drug development.

ACKNOWLEDGMENTS
We thank the Psychiatric Genomics Consortium, the Million Veteran Program, FinnGen biobank and 23andMe for access to data, and the many people who provided DNA samples.

DECLARATIONS OF INTERESTS
OAA has received speaker's honorarium from Sunovion and Lundbeck and is a consultant for Healthlytix. AMD is a founder of and holds equity interest in CorTechs Labs and serves on its scientific advisory board. He is also a member of the Scientific Advisory Board of Healthlytix and receives research funding from General Electric Healthcare (GEHC). The . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 9, 2022. ; https://doi.org/10.1101/2022.09.09.22279755 doi: medRxiv preprint Table 2. The most strongly associated single nucleotide polymorphisms (SNPs) in genomic loci associated with opioid use disorder (OUD) at conditional false discovery rate (condFDR) < 0.05 given association with schizophrenia (SCZ), bipolar disorder (BD) or major depression (MD) after merging regions < 250 kb apart into a single locus are shown. The table presents chromosomal position (Chr.), conditional false discovery rate (FDR) value, conditioned trait and OUD novelty status. For more details including genes mapped to these loci and the full list of all loci associated with OUD at condFDR < 0.05, see Supplemental  Tables 1-3. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 1 Table 3. Summary data of shared loci identified between opioid use disorder (OUD) and severe mental disorders (SMDs); schizophrenia (SCZ), bipolar disorder (BD) and major depression (MD) at conjunctional false discovery rate (conjFDR) < 0.05. Bp = base pair; SNP = single nucleotide polymorphism; Chr = chromosome, Bold p-value refers to the listed SMD z-score.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 9, 2022. ; https://doi.org/10.1101/2022.09.09.22279755 doi: medRxiv preprint Figure 1. a-c: Conditional Q-Q plots of observed versus expected opioid use disorder (OUD) −log 10 p-values (corrected for inflation) below the standard GWAS threshold of p < 5×10 −8 as a function of the significance of the association with schizophrenia (SCZ), bipolar disorder (BD) and major depression (MD), at the level of p . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

FIGURES
The copyright holder for this preprint this version posted September 9, 2022. ; . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 9, 2022. ; https://doi.org/10.1101/2022.09.09.22279755 doi: medRxiv preprint