Hostname: page-component-89b8bd64d-5bvrz Total loading time: 0 Render date: 2026-05-10T18:27:12.600Z Has data issue: false hasContentIssue false
  • English
  • Français

Investigating the polygenic relationship between heavy cannabis use and schizophrenia in the All of Us Research Program

Published online by Cambridge University Press:  17 December 2025

Isabelle Austin-Zimmerman ,
Hayley H.A. Thorpe ,
John J. Meredith ,
Jibran Y. Khokhar Open the ORCID record for Jibran Y. Khokhar [Opens in a new window] ,
Tian Ge ,
Marta Di Forti ,
Arpana Agrawal ,
Emma C. Johnson Open the ORCID record for Emma C. Johnson [Opens in a new window]  and
Sandra Sanchez-Roige Open the ORCID record for Sandra Sanchez-Roige [Opens in a new window]
Isabelle Austin-Zimmerman
Affiliation:
Social, Genetic, and Developmental Psychiatry Centre, Institute of Psychiatry, Psychology, and Neuroscience, King’s College London, London, UK
Hayley H.A. Thorpe
Affiliation:
Department of Anatomy and Cell Biology, Schulich School of Medicine and Dentistry, Western University, London, ON, Canada
John J. Meredith
Affiliation:
Department of Psychiatry, University of California San Diego, San Diego, CA, USA
Jibran Y. Khokhar
Affiliation:
Department of Anatomy and Cell Biology, Schulich School of Medicine and Dentistry, Western University, London, ON, Canada
Tian Ge
Affiliation:
Department of Psychiatry, Harvard Medical School, Boston, MA, USA Psychiatric and Neurodevelopmental Genetics Unit, Center for Genomic Medicine, Massachusetts General Hospital, Boston, MA, USA Center for Precision Psychiatry, Department of Psychiatry, Massachusetts General Hospital, Boston, MA, USA Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard, Cambridge, MA, USA
Marta Di Forti
Affiliation:
Social, Genetic, and Developmental Psychiatry Centre, Institute of Psychiatry, Psychology, and Neuroscience, King’s College London, London, UK South London and Maudsley NHS Foundation Trust, London, UK
Arpana Agrawal
Affiliation:
Department of Psychiatry, Washington University School of Medicine, St. Louis, MO, USA
Emma C. Johnson*
Affiliation:
Department of Psychiatry, Washington University School of Medicine, St. Louis, MO, USA
Sandra Sanchez-Roige*
Affiliation:
Department of Psychiatry, University of California San Diego, La Jolla, CA, USA Institute for Genomic Medicine, University of California San Diego, La Jolla, CA, USA Department of Medicine, Division of Genetic Medicine, Vanderbilt University Medical Center, Nashville, TN, USA
*
Corresponding authors: Sandra Sanchez-Roige and Emma C Johnson; Emails: sanchezroige@ucsd.edu; emma.c.johnson@wustl.edu
Corresponding authors: Sandra Sanchez-Roige and Emma C Johnson; Emails: sanchezroige@ucsd.edu; emma.c.johnson@wustl.edu
Rights & Permissions [Opens in a new window]

Abstract

Background

Decades of research have identified a strong association between heavy cannabis use and schizophrenia (SCZ), with evidence of correlated genetic factors. However, many studies on the genetic relationship between cannabis use and psychosis have lacked data on both phenotypes within the same individuals, creating challenges due to unmeasured confounding. We aimed to address this by using multimodal data from the All of Us Research Program, which contains genetic data as well as information on SCZ diagnosis and cannabis use.

Methods

We tested the association between cannabis use disorder (CUD) and SCZ polygenic scores (PGSs) with SCZ and heavy cannabis use. We tested models where both CUD and SCZ PGSs were included as joint predictors of heavy cannabis use and SCZ case status. We defined three sets of cases based on comorbidities: relaxed (assessing for only the primary condition), strict (excluding comorbidity), and dual-comorbidity.

Results

CUD and SCZ polygenic liability were independently associated with heavy cannabis use; the SCZ PGS effect was very modest. In contrast, both SCZ and CUD PGSs were independently associated with SCZ, with independent significant effects of CUD PGS. Polygenic liability to CUD was associated with SCZ in individuals without a documented history of cannabis use, suggesting widespread pleiotropy.

Conclusions

These findings underscore the need for comprehensive models that integrate genetic risk factors for heavy cannabis use to advance our understanding of SCZ etiology.

Keywords

cannabisGWASpolygenic risk scorepsychosisschizophrenia

Information

Type
Original Article
Information
Psychological Medicine , Volume 55 , 2025 , e381
Check for updates
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
© The Author(s), 2025. Published by Cambridge University Press

Introduction

Chronic cannabis use is associated with greater psychosis risk (Di Forti, Marconi, & Carra, Reference Di Forti, Marconi and Carra2015; Di Forti, Quattrone, & Freeman, Reference Di Forti, Quattrone and Freeman2019; Hasan, von Keller, Friemel, et al., Reference Hasan, von Keller and Friemel2020; Marconi et al., Reference Marconi, Di Forti, Lewis, Murray and Vassos2016), psychotic-like experiences (Bagot, Milin, & Kaminer, Reference Bagot, Milin and Kaminer2015; Cheng, Parker, Karadag, et al., Reference Cheng, Parker and Karadag2023; Large et al., Reference Large, Sharma, Compton, Slade and Nielssen2011), and an earlier age of onset for psychotic symptoms (Bagot et al., Reference Bagot, Milin and Kaminer2015; Large et al., Reference Large, Sharma, Compton, Slade and Nielssen2011). Many prior studies have argued that heavy cannabis use increases the risk of psychosis (Di Forti et al., Reference Di Forti, Quattrone and Freeman2019; Hasan et al., Reference Hasan, von Keller and Friemel2020; Marconi et al., Reference Marconi, Di Forti, Lewis, Murray and Vassos2016).

A complementary hypothesis posits that common genetic mechanisms underlie cannabis use and psychosis (Cheng et al., Reference Cheng, Parker and Karadag2023; Johnson, Austin-Zimmerman, Thorpe, et al., Reference Johnson, Austin-Zimmerman and Thorpe2024; Johnson, Hatoum, Deak, et al., Reference Johnson, Hatoum and Deak2021; Khokhar et al., Reference Khokhar, Dwiel, Henricks, Doucette and Green2018). Genome-wide association studies (GWASs) and twin studies support that schizophrenia (SCZ), the most common psychotic disorder, is heritable (Bigdeli, Genovese, Georgakopoulos, et al., Reference Bigdeli, Genovese and Georgakopoulos2020; Trubetskoy, Pardiñas, Qi, et al., Reference Trubetskoy, Pardiñas and Qi2022; twin-estimated heritability ~ 80%; Owen, Sawa, & Mortensen, Reference Owen, Sawa and Mortensen2016) as are lifetime cannabis use (Thorpe, Fontanillas, Meredith, et al., Reference Thorpe, Fontanillas and Meredith2025) and cannabis use disorder (CUD; Johnson, Demontis, Thorgeirsson, et al., Reference Johnson, Demontis and Thorgeirsson2020; Levey, Galimberti, Deak, et al., Reference Levey, Galimberti and Deak2023; twin-estimated heritability ~40–48% and ~51–59%, respectively; Agrawal & Lynskey, Reference Agrawal and Lynskey2006; Verweij, Zietsch, & Lynskey, Reference Verweij, Zietsch and Lynskey2010). Cannabis use traits and schizophrenia are positively genetically correlated (rg range = 0.25–35; Cheng et al., Reference Cheng, Parker and Karadag2023). The most recent Mendelian randomization studies support a bidirectional causal relationship between SCZ and CUD (Elkrief, Lin, Marchi, et al., Reference Elkrief, Lin and Marchi2023; Levey et al., Reference Levey, Galimberti and Deak2023), with a greater magnitude of causal effect from CUD to SCZ. Further, with the advent of polygenic scores (PGSs), which aggregate the effect sizes from GWAS across genome-wide variants to quantify genetic liability for a trait, it has been observed that SCZ PGS are associated with psychotic-like experiences in cannabis users (Wainberg, Jacobs, di Forti, & Tripathy, Reference Wainberg, Jacobs, di Forti and Tripathy2021). This indicates that genetic risk for SCZ might make one particularly vulnerable to the psychotogenic effects of cannabis use. Thus, delineating the genetic relationship between cannabis use and SCZ may inform psychosis prevention, particularly in the context of expanding cannabis legalization and accessibility.

Another complicating factor is that while the prevalence of cannabis use and CUD in individuals with SCZ is high (Abush, Ghose, Van Enkevort, et al., Reference Abush, Ghose and Van Enkevort2018; Di Forti et al., Reference Di Forti, Marconi and Carra2015, Reference Di Forti, Quattrone and Freeman2019; Hjorthøj, Compton, Starzer, et al., Reference Hjorthøj, Compton and Starzer2023; Hjorthøj, Posselt, & Nordentoft, Reference Hjorthøj, Posselt and Nordentoft2021; Marconi et al., Reference Marconi, Di Forti, Lewis, Murray and Vassos2016), no GWASs of SCZ account for this co-occurrence. Likewise, GWASs of CUD rarely account for comorbid psychosis unless such conditions are exclusionary criteria for study recruitment. Thus, it is plausible that the genetic correlation between SCZ and cannabis use/CUD is partially confounded by their co-occurrence in the population.

In the present study, we availed of a unique opportunity to disarticulate the SCZ–cannabis comorbidity by leveraging the multimodal data of the National Institute of Health’s All of Us (AoU) Research Program, which provides phenotypic data relevant to cannabis and SCZ, as well as genetic data, for up to ~250,000 participants in the version 7 release. Using data from prior well-powered GWASs, we examined the association between CUD (Levey et al., Reference Levey, Galimberti and Deak2023) and SCZ (Bigdeli et al., Reference Bigdeli, Genovese and Georgakopoulos2020; Trubetskoy et al., Reference Trubetskoy, Pardiñas and Qi2022) PGS, and cannabis-related traits and SCZ in the AoU cohort. We hypothesized that polygenic liability to both CUD and SCZ would exert independent effects on cannabis use and SCZ. However, we also expected that the associations would be attenuated in the absence of the comorbid condition (e.g., SCZ PGS would not strongly relate to cannabis-related traits without comorbid psychosis).

Methods

Participants

Participants enrolled in AoU (v7) were included in our analysis. AoU is a diverse database of electronic health records (EHR), survey responses, physical measurements, and whole-genome sequencing (WGS) data from blood or saliva samples; the v7 release included data for over ~250,000 adults in the United States. For details on recruitment and genomic data, please see previous publications from the AoU Research Program (All of Us Research Program Investigators et al., Reference Denny and Rutter2019; All of Us Research Program Investigators, 2024; Ramirez, Sulieman, Schlueter, et al., Reference Ramirez, Sulieman and Schlueter2022).

Unrelated participants who were assigned male or female at birth, had applicable EHR and/or survey-level data, had WGS data, and whose genomes were statistically correlated with the genomes of European (EUR) or African (AFR) reference populations were included in the study. Genetic similarity was based on principal components (PCs) provided by AoU. Related participants (kinship score > 0.1) were removed using a relatedness flagged sample list provided by AoU kinship estimation that minimized the number of samples needing to be removed (All of Us Research Program Investigators, 2024).

Heavy cannabis use and SCZ were derived using multimodal data, including diagnostic codes (SNOMED Clinical Terms), survey data, and prescription medication records (Table 1). We defined heavy cannabis use as anyone with a diagnosis of CUD, as well as anyone who reported daily cannabis use within the last 3 months. We included daily cannabis use based on its strong correlations with CUD (Thorpe et al., Reference Thorpe, Fontanillas and Meredith2025) and with SCZ (Di Forti et al., Reference Di Forti, Marconi and Carra2015; Marconi et al., Reference Marconi, Di Forti, Lewis, Murray and Vassos2016). We defined SCZ as anyone with SCZ based on diagnostic codes and/or survey data. Participants were excluded from the control groups if they had a record of antipsychotic prescriptions. See Supplementary Figure 1 for a participant inclusion flowchart.

Table 1.

All of Us concept codes used to assign binary identifiers for schizophrenia, heavy cannabis use, heavy tobacco smoking, and antipsychotic medication statuses. Participants across different Concept IDs may overlap. See Supplementary Figure 1 for sample details.

We classified participants into different case/control groups (Supplementary Figure 1). First, we used ‘relaxed’ definitions that assigned individuals to case status regardless of comorbidity (e.g., SCZ with or without a history of heavy cannabis use). Second, to account for the comorbid phenotypic manifestation of SCZ and heavy cannabis use, we created ‘strict’ definitions that excluded comorbidities (i.e., cases with heavy cannabis use histories that excluded those with a SCZ diagnosis; SCZ cases excluding those with a history of heavy cannabis use). Finally, we also defined a comorbid case group, which included those with both a history of heavy cannabis use and a SCZ diagnosis.

Polygenic analysis

All analyses were conducted using the AoU Researcher Workbench cloud computing environment. Single-nucleotide polymorphisms (SNPs) were curated using the AoU short-read WGS Allele Count/Allele Frequency call set for each ancestral subpopulation, and further filtered to only biallelic SNPs present in HapMap3 from the EUR and AFR 1,000 Genomes Linkage Disequilibrium Scores. Cohort-specific PGSs were calculated using polygenic risk score continuous shrinkage (PRS-CS) ‘auto’ v1.1.0 (Ge et al., Reference Ge, Chen, Ni, Feng and Smoller2019) based on SNPs identified by publicly available GWAS summary statistics of SCZ (AFR N cases = 7,509, N controls = 8,337; Bigdeli et al., Reference Bigdeli, Genovese and Georgakopoulos2020; EUR N cases = 67,390, N controls = 94,015; Trubetskoy et al., Reference Trubetskoy, Pardiñas and Qi2022) and CUD (AFR N cases = 19,065, N controls = 104,143; EUR N cases = 42,281, N controls = 843,744; Levey et al., Reference Levey, Galimberti and Deak2023) and that intersected with SNPs in the AoU database. The ‘auto’ mode of PRS-CS allows the shrinkage parameter (phi) to be learned from the data automatically. We used the default values for Markov Chain Monte Carlo (1,000) and burn-in (500) iterations, which are recommended for stable posterior estimation. SCZ and CUD PGSs were created from up to 973,634 SNPs using the allelic-scoring function, score, in PLINK (v1.9; Ge et al., Reference Ge, Chen, Ni, Feng and Smoller2019).

Statistical analyses

We examined the independent and joint effect of CUD and SCZ PGSs on both heavy cannabis use and SCZ case status from unrelated individuals (All of Us Research Program Investigators, 2024) using various case/control groups based on relaxed, strict, and comorbid definitions (Supplementary Figure 1). All models were adjusted for age, sex, and the first 10 ancestry-specific genomic PCs. Tobacco smoking is also prevalent among those with cannabis use and SCZ (Agrawal, Budney, & Lynskey, Reference Agrawal, Budney and Lynskey2012; Ziedonis, Hitsman, Beckham, et al., Reference Ziedonis, Hitsman and Beckham2008); we performed sensitivity analyses using heavy tobacco smoking as a covariate.

Liability scale R 2 was calculated according to Lee et al (Lee, Goddard, Wray, & Visscher, Reference Lee, Goddard, Wray and Visscher2012) using the R package fmsb (v0.7.6), the NagelkerkeR2 function, and the estimated population-level prevalence of 0.90% for SCZ (Perälä, Suvisaari, Saarni, et al., Reference Perälä, Suvisaari and Saarni2007) and 6.27% for CUD (Hasin, Kerridge, Saha, et al., Reference Hasin, Kerridge and Saha2016) in US adults.

Results

Participant demographics are described in Supplementary Table 1. For the case/control composition, see Supplementary Figure 1, and for the distribution of PGSs across groups, see Supplementary Figures 2 and 3.

CUD-PGS and SCZ-PGS associations with heavy cannabis use:

Relaxed case definitions. In the EUR sample, both CUD-PGS and SCZ-PGS were significantly associated with heavy cannabis use when modeled independently, with a larger magnitude of effect for CUD-PGS, as expected (CUDind-PGS odds ratio [OR] = 2.479, 95% confidence interval [CI]: 2.267–2.711; SCZind-PGS OR = 1.464, 95% CI: 1.33–1.611; Table 2 and Figure 1). In the model including both PGSs, CUD-PGS remained strongly associated with heavy cannabis use (CUDjoint-PGS OR = 2.391, 95% CI: 2.182–2.620), with only a 3.51% reduction in the effect size when modeled alongside SCZ-PGS, while SCZ-PGS retained a significant but attenuated effect (SCZjoint-PGS OR = 1.198, 95% CI: 1.086–1.321), with an 18.24% reduction in the effect size compared to modeling SCZ-PGS alone.

Table 2.

Summary of polygenic associations with heavy cannabis use and schizophrenia across various case group definitions. a: analyses include all heavy cannabis users regardless of schizophrenia status or all schizophrenia cases regardless of heavy cannabis use patterns (“relaxed”) or either exclude all comorbid cases or include only comorbid cases (“strict”). b: ind=Independently modelled; joint=both PGSs modelled jointly.

Figure 1.

CUD and schizophrenia (SCZ) polygenic score associations with heavy cannabis use and schizophrenia. Associations shown for European and African samples under Relaxed (cases only screened for the primary outcome), Strict (cases with comorbid schizophrenia or heavy cannabis use removed), and comorbid (including both schizophrenia and heavy cannabis use) case definitions. Analyses conducted with CUD and SCZ PGSs considered independently (indep) or jointly (joint).

In the AFR sample, CUD-PGS was also associated with heavy cannabis use, both when modeled independently (CUDind-PGS OR = 1.164, 95% CI: 1.118–1.210) and when modeled jointly with SCZ-PGS (CUDjoint-PGS OR = 1.164, 95% CI: 1.118–1.210; Table 2 and Figure 1). SCZ-PGS was not significantly associated with heavy cannabis use when modeled independently (SCZind-PGS OR = 1.006, 95% CI: 0.971–1.043) nor when modeled jointly (SCZjoint-PGS OR = 1.000, 95% CI: 0.964–1.036).

Strict case definitions. We next investigated associations between CUD-PGS and SCZ-PGS and heavy cannabis use when all cases with comorbid SCZ were excluded. In the EUR sample, CUD-PGS remained significantly associated with heavy cannabis use (CUDind-PGS OR = 2.436, 95% CI: 2.222–2.672) while SCZ-PGS showed a significant but somewhat attenuated association (SCZind-PGS OR = 1.385, 95% CI: 1.255–1.529). When both PGSs were included, CUD-PGS was associated with heavy cannabis use (CUDjoint-PGS OR = 2.375, 95% CI: 2.162–2.610), while the SCZ-PGS association was further attenuated (OR = 1.136, 95% CI: 1.027–1.257, p = 0.013).

In the AFR sample, CUD-PGS was also associated with heavy cannabis use (CUDind-PGS OR = 1.163, 95% CI: 1.116–1.213; Table 2 and Figure 1). SCZ-PGS was not significantly associated with heavy cannabis use when modeled independently (SCZind-PGS OR = 1.004, 95% CI: 0.967–1.042) or when modeled jointly (SCZjoint-PGS OR = 0.997, 95% CI: 0.960–1.035).

CUD-PGS and SCZ-PGS associations with SCZ:

Relaxed case definitions. In the EUR sample, the SCZ-PGS and CUD-PGS were significantly associated with SCZ when modeled independently, with a larger magnitude of effect for the within-trait association (SCZind-PGS OR = 3.014, 95% CI: 2.555–3.555; CUDind-PGS OR = 2.131, 95% CI: 1.828–2.485). The inclusion of both PGSs in a joint model resulted in a slight reduction in effect size for both PGSs (SCZjoint-PGS OR = 2.643, 95% CI: 2.232–3.131; CUDjoint-PGS OR = 1.748, 95% CI: 1.494–2.045; Table 2 and Figure 1).

In the AFR sample, the SCZ-PGS was associated with SCZ, both when modeled independently (SCZind-PGS OR = 1.127, 95% CI: 1.064–1.193) and when modeled jointly (SCZjoint-PGS OR = 1.127, 95% CI: 1.064–1.193). CUD-PGS were not associated with SCZ when modeled independently (CUDind-PGS OR = 1.009, 95% CI: 0.948–1.074) nor when modeled jointly (CUDjoint-PGS OR = 1.005, 95% CI: 0.944–1.069; Table 2 and Figure 1).

Strict case definitions. We considered whether the explanatory power of SCZ-PGS and CUD-PGS would differ when all SCZ cases with a history of heavy cannabis use were excluded (Table 2 and Figure 1). Among EUR participants, SCZ-PGS was highly associated with SCZ in both the independent and joint models (SCZind-PGS OR = 2.975, 95% CI: 2.455–3.603; SCZjoint-PGS OR = 2.679, 95% CI: 2.201–3.260), while CUD-PGS had a smaller but significant effect (CUDind-PGS OR = 1.913, 95% CI: 1.601–2.285; CUDjoint-PGS OR = 1.568, 95% CI: 1.307–1.881). In the AFR sample, SCZ-PGS was significantly associated with SCZ (SCZind-PGS OR = 1.168, 95% CI: 1.087–1.256; SCZjoint-PGS OR = 1.170, 95% CI: 1.088–1.258). CUD-PGS had a weaker and nonsignificant effect (CUDind-PGS OR = 0.959, 95% CI: 0.886–1.038; CUDjoint-PGS OR = 0.953, 95% CI: 0.881–1.031).

CUD-PGS and SCZ-PGS associations with comorbid heavy cannabis use and SCZ:

Finally, we considered the explanatory value of both SCZ-PGS and CUD-PGS on cases with comorbid SCZ and heavy cannabis use (Table 2 and Figure 1). In the EUR sample, both PGSs were associated with these comorbid cases when modeled independently (CUDind-PGS OR = 3.275, 95% CI: 2.328–4.608; SCZind-PGS OR = 3.387, 95% CI: 2.346–4.889), and the effect sizes were slightly reduced for both PGSs when modeled jointly (SCZjoint-PGS OR = 2.667, 95% CI: 1.829–3.888; CUDjoint-PGS OR = 2.655, 95% CI: 1.871–3.768). In the AFR sample, we observed a similar pattern, although the SCZ-PGS was not significantly associated with these comorbid cases in either the independent or joint models (SCZind-PGS OR = 1.039, 95% CI: 0.936–1.152; SCZjoint-PGS OR = 1.033, 95% CI: 0.931–1.146; CUDind-PGS OR = 1.144, 95% CI: 1.021–1.282; CUDjoint-PGS OR = 1.143, 95% CI: 1.019–1.281).

Discussion

Decades of research have identified a strong association between heavy cannabis use and SCZ, with evidence of correlated genetic factors. However, the genetic characterization of this comorbidity remains incomplete because prior studies have not fully accounted for the genetic correlation between CUD and SCZ or the phenotypic co-occurrence of these traits in target samples. In this study, we modeled CUD-PGS and SCZ-PGS jointly and found that while SCZ-PGS explained little additional variance in heavy cannabis use beyond CUD-PGS (0.1% increase), CUD-PGS accounted for greater (although still modest) variance explained for SCZ (0.7% increase beyond the SCZ-PGS). We hypothesized that more stringent case definitions would improve specificity and help delineate the extent to which underlying genetic risk contributed to CUD/SCZ case status. However, we showed that there was little difference in the effect sizes between the relaxed (i.e., case status regardless of comorbidity) versus strict (i.e., case status excluding comorbidity) and comorbid definitions (Figure 1). For instance, the association between CUD-PGS and both heavy cannabis use and SCZ, regardless of their comorbidity, was statistically equivalent (range: 1.91–3.28). Even though CIs were wide, outcome associations between CUD and SCZ PGSs were notably higher for the comorbid definition (R 2 up to 7.5%) compared to the relaxed and strict definitions (~3%; Figure 2). These findings have important implications for SCZ etiology, and demonstrate that CUD genetic liability is associated with SCZ risk even in individuals who do not report heavy cannabis use. Taken together, these findings suggest that prior genetic correlations between CUD and SCZ are partially due to true horizontal pleiotropy rather than confounding due to co-occurring diagnoses. In other words, regardless of comorbidity, the etiologies of CUD and SCZ are partially due to the same genetic pathways.

Figure 2.

Liability-scale R2 for PGS associations with heavy cannabis use (CU) and schizophrenia (SCZ) in EUR estimated from a population-level prevalence of 0.90% for schizophrenia (Perälä, Suvisaari, Saarni, et al., Reference Perälä, Suvisaari and Saarni2007) and 6.27% for CUD (Hasin, Kerridge, Saha, et al., Reference Hasin, Kerridge and Saha2016).

When investigating the relationships between CUD-PGS and SCZ-PGS with heavy cannabis use status using relaxed case definitions, we identified that, as expected, within-trait PGS associations (i.e., CUD-PGS and heavy cannabis use) provided the strongest explanatory power (3% in the EUR sample and 0.3% in the AFR sample). Excluding individuals with SCZ case status did not increase explanatory power (2.9% in the EUR sample and 0.3% in the AFR sample). We hypothesize that this may be the case because many contributing cohorts in the CUD GWAS have already been excluded for psychotic disorders. In contrast, SCZ-PGS showed a weak association with heavy cannabis use in the relaxed case analysis, and this was further reduced when CUD-PGS was included in joint models. This finding suggests that previous reports of SCZ-PGS predicting cannabis use may partially result from unmeasured comorbidity between SCZ risk and cannabis use (Elkrief et al., Reference Elkrief, Lin and Marchi2023; Hiemstra, Nelemans, Branje, et al., Reference Hiemstra, Nelemans and Branje2018; Verweij, Abdellaoui, Nivard, et al., Reference Verweij, Abdellaoui and Nivard2017).

When investigating the relationship between CUD-PGS and SCZ-PGS and SCZ status, SCZ-PGS was associated with SCZ regardless of case definition (2.2–2.33% in the EUR sample and 0.5–0.6% in the AFR sample). Similarly, CUD-PGS was associated with SCZ regardless of case definition (0.9–1.3% in the EUR sample and 0.05% in the AFR sample). When modeled jointly, CUD-PGS explained substantial phenotypic variance of SCZ above and beyond that of SCZ-PGS (28.14% increase). While it is possible that individuals with higher CUD-PGS are more likely to use cannabis, and that cannabis use could increase SCZ risk, our findings suggest that the association between CUD-PGS and SCZ is not solely explained by cannabis use. Instead, this reflects potential genetic liability shared between the two conditions, consistent with horizontal pleiotropy (Johnson et al., Reference Johnson, Hatoum and Deak2021). Mendelian randomization studies suggest bidirectional causal effects between CUD and SCZ (Johnson et al., Reference Johnson, Demontis and Thorgeirsson2020; Levey et al., Reference Levey, Galimberti and Deak2023; Lin, Pries, Sarac, et al., Reference Lin, Pries and Sarac2022). Although this study was not designed to interrogate the causal nature of the observed associations, future analyses could build upon our work to calculate PGSs that are derived from multivariate methods (i.e., genomic structural equation modeling) that parse CUD and SCZ genetics, or select variants based on Mendelian randomization results (Garfield & Anderson, Reference Garfield and Anderson2024). These findings also caution future Mendelian randomization studies to carefully estimate and account for horizontal pleiotropy (e.g., MR-PRESSO; Verbanck, Chen, Neale, & Do, Reference Verbanck, Chen, Neale and Do2018) when investigating the causal relationships between cannabis use and SCZ.

There are limitations worth noting. PGSs are inherently limited by the GWAS data used to build them. While we used PRS-CSx (Ge et al., Reference Ge, Chen, Ni, Feng and Smoller2019) to improve power by leveraging GWAS findings from both EUR and AFR populations (Supplementary Table 3), differences in predictive power of PGS in the AFR sample are likely driven by poor PGS generalizability rather than true biological or environmental effects (Atkinson, Bianchi, Ye, et al., Reference Atkinson, Bianchi and Ye2022). Even in the EUR sample, it is important to note that genetic liability for these traits explains a relatively small proportion of the variance in both outcomes. Therefore, these PGSs are not yet clinically meaningful. Moreover, CUD is genetically correlated with a variety of other behaviors (e.g., depressed mood, executive functioning, and impulsivity; Johnson et al., Reference Johnson, Demontis and Thorgeirsson2020; Levey et al., Reference Levey, Galimberti and Deak2023), which may co-occur with SCZ, and thus CUD-PGS may have indexed pleiotropic risk that extends beyond associations with heavy cannabis use. Of note, while tobacco use is both correlated with CUD and SCZ, and may be partly responsible for this residual association, adjusting for tobacco use did not alter our findings (Supplementary Table 3; Johnson et al., Reference Johnson, Austin-Zimmerman and Thorpe2024). Our reliance on EHR and/or self-report data may underestimate the true prevalence of cannabis and tobacco use disorders in our cohort (e.g., ~6.7% or past-year CUD and ~8.5% for nicotine dependence in the general US population vs. ~2.0 and ~6.2% in our sample, respectively). Similar underreporting has been noted in prior studies using AoU data (Barr, Bigdeli, & Meyers, Reference Barr, Bigdeli and Meyers2022). In the UK Biobank, participants with SCZ diagnoses have been observed to have less severe symptomatology and better health outcomes than clinically ascertained samples, attributable to volunteer bias (Legge, Pardiñas, Woolway, et al., Reference Legge, Pardiñas and Woolway2024). It may be that participants with cannabis or tobacco use disorders in AoU are similarly less severe than clinical cohorts. We also noted higher rates of SCZ in the AFR ancestry sample than in the EUR ancestry sample. Similar patterns have been reported previously (Schwartz & Blankenship, Reference Schwartz and Blankenship2014) and may reflect clinician bias or differential access to healthcare. However, because AoU is not a representative sample of the US population, further epidemiological studies are needed to accurately characterize SCZ prevalence across ancestries. Finally, the SCZ-PGS association with heavy cannabis use may be due to undetected CUD in SCZ GWAS. The same is true for CUD GWAS, but less apparent due to the low SCZ prevalence. Without accounting for phenotypic overlap in the original GWASs, it will be difficult to confirm the extent to which this explains cross-trait associations.

In conclusion, we find compelling evidence of common underlying genetic mechanisms between CUD and SCZ. Future studies with more granular phenotyping will be essential to better understand the true extent of this putative horizontal pleiotropy.

Supplementary material

The supplementary material for this article can be found at http://doi.org/10.1017/S0033291725102717.

Data availability statement

The AoU workspace used for this project will be made available upon request to registered and eligible AoU researchers through the AoU Research Workbench.

Acknowledgments

We gratefully acknowledge AoU participants for their contributions, without whom this research would not have been possible. We also thank the National Institutes of Health’s AoU Research Program for making available the participant data examined in this study. This work was supported by the National Institute on Drug Abuse (NIDA DP1DA054394 JJM, S.S.-R NIDA K01DA051759 E.C.J) and the California Tobacco-Related Disease Research Program (T32IR5226). J.Y.K. is supported by a Canada Research Chair in Translational Neuropsychopharmacology. A.A. receives funding from R01DA054869 and R21DA061592. M.D.-F. and I.A.-Z.were supported by MRC SRF (MRC MR/T007818/1). H.H.A.T. is supported by a Canadian Institutes of Health Research Postdoctoral Fellowship (MFE-187919) and a Canadian Consortium for the Investigation of Cannabinoids (CCIC) Neuroscience Fellowship in Cannabis and Cannabinoid Research made possible by the Canada Brain Research Fund (CBRF), an innovative arrangement between the Government of Canada (through Health Canada) and Brain Canada Foundation, and by the M. Wayne and J. Coleman Family Fund.

Competing interests

The authors declare none.

Footnotes

E.C.J. and S.S.-R. senior co-corresponding authors.

I.A.-Z. and H.H.A.T. first co-authors.

References

Abush, H., Ghose, S., Van Enkevort, E. A., et al. (2018). Associations between adolescent cannabis use and brain structure in psychosis. Psychiatry Research: Neuroimaging, 276, 53–64. https://doi.org/10.1016/j.pscychresns.2018.03.008.CrossRefGoogle ScholarPubMed
Agrawal, A., Budney, A. J., & Lynskey, M. T. (2012). The co-occurring use and misuse of cannabis and tobacco: A review. Addiction, 107(7), 1221–1233. https://doi.org/10.1111/j.1360-0443.2012.03837.x.CrossRefGoogle ScholarPubMed
Agrawal, A., & Lynskey, M. T. (2006). The genetic epidemiology of cannabis use, abuse and dependence. Addiction, 101(6), 801–812.10.1111/j.1360-0443.2006.01399.xCrossRefGoogle ScholarPubMed
All of Us Research Program Genomics Investigators. (2024 ). Genomic data in the all of us research program. Nature, 627(8003), 340–346. https://doi.org/10.1038/s41586-023-06957-xCrossRefGoogle Scholar
All of Us Research Program Investigators, Denny, J. C., Rutter, J. L., et al. (2019). The “All of Us” research program. The New England Journal of Medicine, 381(7), 668–676. https://doi.org/10.1056/NEJMsr1809937.Google Scholar
Atkinson, E. G., Bianchi, S. B., Ye, G. Y., et al. (2022). Cross-ancestry genomic research: Time to close the gap. Neuropsychopharmacology, 47(10), 1737–1738. https://doi.org/10.1038/s41386-022-01365-7.CrossRefGoogle ScholarPubMed
Bagot, K. S., Milin, R., & Kaminer, Y. (2015). Adolescent initiation of cannabis use and early-onset psychosis. Substance Abuse, 36(4), 524–533. https://doi.org/10.1080/08897077.2014.995332.CrossRefGoogle ScholarPubMed
Barr, P. B., Bigdeli, T. B., & Meyers, J. L. (2022). Prevalence, comorbidity, and Sociodemographic correlates of psychiatric disorders reported in the all of us research program. JAMA Psychiatry, 79(6), 622–628. https://doi.org/10.1001/jamapsychiatry.2022.0685.CrossRefGoogle ScholarPubMed
Bigdeli, T. B., Genovese, G., Georgakopoulos, P., et al. (2020). Contributions of common genetic variants to risk of schizophrenia among individuals of African and Latino ancestry. Molecular Psychiatry, 25(10), 2455–2467. https://doi.org/10.1038/s41380-019-0517-y.CrossRefGoogle ScholarPubMed
Cheng, W., Parker, N., Karadag, N., et al. (2023). The relationship between cannabis use, schizophrenia, and bipolar disorder: A genetically informed study. Lancet Psychiatry, 10(6), 441–451. https://doi.org/10.1016/S2215-0366(23)00143-8.CrossRefGoogle ScholarPubMed
Di Forti, M., Marconi, A., & Carra, E. (2015). Proportion of patients in South London with first-episode psychosis attributable to use of high potency cannabis: A case-control study. Lancet Psychiatry, 2(3), 233–238.10.1016/S2215-0366(14)00117-5CrossRefGoogle ScholarPubMed
Di Forti, M., Quattrone, D., & Freeman, T. (2019). The contribution of cannabis use to variation in the incidence of psychotic disorder across Europe (EU-GEI): A multicentre case-control study. Lancet Psychiatry, 6(5), 427–436.10.1016/S2215-0366(19)30048-3CrossRefGoogle ScholarPubMed
Elkrief, L., Lin, B., Marchi, M., et al. (2023). Independent contribution of polygenic risk for schizophrenia and cannabis use in predicting psychotic-like experiences in young adulthood: Testing gene × environment moderation and mediation. Psychological Medicine, 53(5), 1759–1769. https://doi.org/10.1017/S0033291721003378.CrossRefGoogle Scholar
Garfield, V., & Anderson, E. L. (2024). A brief comparison of polygenic risk scores and Mendelian randomisation. BMC Medical Genomics, 17(1), 10. https://doi.org/10.1186/s12920-023-01769-4.CrossRefGoogle ScholarPubMed
Ge, T., Chen, C. Y., Ni, Y., Feng, Y. C. A., & Smoller, J. W. (2019). Polygenic prediction via Bayesian regression and continuous shrinkage priors. Nature Communications, 10(1), 1776. https://doi.org/10.1038/s41467-019-09718-5.CrossRefGoogle ScholarPubMed
Hasan, A., von Keller, R., Friemel, C. M., et al. (2020). Cannabis use and psychosis: A review of reviews. European Archives of Psychiatry and Clinical Neuroscience, 270(4), 403–412. https://doi.org/10.1007/s00406-019-01068-z.CrossRefGoogle ScholarPubMed
Hasin, D. S., Kerridge, B. T., Saha, T. D., et al. (2016). Prevalence and correlates of DSM-5 cannabis use disorder, 2012-2013: Findings from the National Epidemiologic Survey on alcohol and related conditions-III. The American Journal of Psychiatry, 173(6), 588–599. https://doi.org/10.1176/appi.ajp.2015.15070907.CrossRefGoogle ScholarPubMed
Hiemstra, M., Nelemans, S. A., Branje, S., et al. (2018). Genetic vulnerability to schizophrenia is associated with cannabis use patterns during adolescence. Drug and Alcohol Dependence, 190, 143–150. https://doi.org/10.1016/j.drugalcdep.2018.05.024.CrossRefGoogle ScholarPubMed
Hjorthøj, C., Compton, W., Starzer, M., et al. (2023). Association between cannabis use disorder and schizophrenia stronger in young males than in females. Psychological Medicine, 1–7. https://doi.org/10.1017/S0033291723000880.Google Scholar
Hjorthøj, C., Posselt, C. M., & Nordentoft, M. (2021). Development over time of the population-attributable risk fraction for cannabis use disorder in schizophrenia in Denmark. JAMA Psychiatry, 78(9), 1013. https://doi.org/10.1001/jamapsychiatry.2021.1471.CrossRefGoogle ScholarPubMed
Johnson, E. C., Austin-Zimmerman, I., Thorpe, H. H. A., et al. (2024). Cross-ancestry genetic investigation of schizophrenia, cannabis use disorder, and tobacco smoking. Neuropsychopharmacology, 49(11), 1655–1665. https://doi.org/10.1038/s41386-024-01886-3.CrossRefGoogle ScholarPubMed
Johnson, E. C., Demontis, D., Thorgeirsson, T. E., et al. (2020). A large-scale genome-wide association study meta-analysis of cannabis use disorder. Lancet Psychiatry, 7(12), 1032–1045. https://doi.org/10.1016/S2215-0366(20)30339-4.CrossRefGoogle ScholarPubMed
Johnson, E. C., Hatoum, A. S., Deak, J. D., et al. (2021). The relationship between cannabis and schizophrenia: A genetically informed perspective. Addiction, 116(11), 3227–3234. https://doi.org/10.1111/add.15534.CrossRefGoogle ScholarPubMed
Khokhar, J. Y., Dwiel, L. L., Henricks, A. M., Doucette, W. T., & Green, A. I. (2018). The link between schizophrenia and substance use disorder: A unifying hypothesis. Schizophrenia Research, 194, 78–85. https://doi.org/10.1016/j.schres.2017.04.016.CrossRefGoogle ScholarPubMed
Large, M., Sharma, S., Compton, M. T., Slade, T., & Nielssen, O. (2011). Cannabis use and earlier onset of psychosis: A systematic meta-analysis. Archives of General Psychiatry, 68(6), 555–561. https://doi.org/10.1001/archgenpsychiatry.2011.5.CrossRefGoogle ScholarPubMed
Lee, S. H., Goddard, M. E., Wray, N. R., & Visscher, P. M. (2012). A better coefficient of determination for genetic profile analysis: A better coefficient of determination. Genetic Epidemiology, 36(3), 214–224. https://doi.org/10.1002/gepi.21614.CrossRefGoogle ScholarPubMed
Legge, S. E., Pardiñas, A. F., Woolway, G., et al. (2024). Genetic and phenotypic features of schizophrenia in the UK biobank. JAMA Psychiatry, 81(7), 681–690. https://doi.org/10.1001/jamapsychiatry.2024.0200.CrossRefGoogle ScholarPubMed
Levey, D. F., Galimberti, M., Deak, J. D., et al. (2023). Multi-ancestry genome-wide association study of cannabis use disorder yields insight into disease biology and public health implications. Nature Genetics. https://doi.org/10.1038/s41588-023-01563-z.CrossRefGoogle ScholarPubMed
Lin, B. D., Pries, L. K., Sarac, H. S., et al. (2022). Nongenetic factors associated with psychotic experiences among UK biobank participants: Exposome-wide analysis and Mendelian randomization analysis. JAMA Psychiatry, 79(9), 857–868. https://doi.org/10.1001/jamapsychiatry.2022.1655.CrossRefGoogle ScholarPubMed
Marconi, A., Di Forti, M., Lewis, C. M., Murray, R. M., & Vassos, E. (2016). Meta-analysis of the association between the level of cannabis use and risk of psychosis. Schizophrenia Bulletin, 42(5), 1262–1269. https://doi.org/10.1093/schbul/sbw003.CrossRefGoogle ScholarPubMed
Owen, M. J., Sawa, A., & Mortensen, P. B. (2016). Schizophrenia. The Lancet, 388, 86–97.10.1016/S0140-6736(15)01121-6CrossRefGoogle ScholarPubMed
Perälä, J., Suvisaari, J., Saarni, S. I., et al. (2007). Lifetime prevalence of psychotic and bipolar I disorders in a general population. Archives of General Psychiatry, 64(1), 19–28. https://doi.org/10.1001/archpsyc.64.1.19.CrossRefGoogle ScholarPubMed
Ramirez, A. H., Sulieman, L., Schlueter, D. J., et al. (2022). The all of us research program: Data quality, utility, and diversity. Patterns, 3(8), 100570. https://doi.org/10.1016/j.patter.2022.100570.CrossRefGoogle ScholarPubMed
Schwartz, R. C., & Blankenship, D. M. (2014). Racial disparities in psychotic disorder diagnosis: A review of empirical literature. World Journal of Psychiatry, 4(4), 133–140. https://doi.org/10.5498/wjp.v4.i4.133.CrossRefGoogle ScholarPubMed
Thorpe, H. H. A., Fontanillas, P., Meredith, J. J., et al. (2025). Genome-wide association studies of lifetime and frequency cannabis use in 131,895 individuals. Mol Psychiatry. https://doi.org/10.1038/s41380-025-03219-2. https://www.nature.com/articles/s41380-025-0CrossRefGoogle ScholarPubMed
Trubetskoy, V., Pardiñas, A. F., Qi, T., et al. (2022). Mapping genomic loci implicates genes and synaptic biology in schizophrenia. Nature, 604(7906), 502–508. https://doi.org/10.1038/s41586-022-04434-5.CrossRefGoogle ScholarPubMed
Verbanck, M., Chen, C. Y., Neale, B., & Do, R. (2018). Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nature Genetics, 50(5), 693–698. https://doi.org/10.1038/s41588-018-0099-7.CrossRefGoogle ScholarPubMed
Verweij, K. J. H., Abdellaoui, A., Nivard, M. G., et al. (2017). Short communication: Genetic association between schizophrenia and cannabis use. Drug and Alcohol Dependence, 171, 117–121. https://doi.org/10.1016/j.drugalcdep.2016.09.022.CrossRefGoogle ScholarPubMed
Verweij, K. J. H., Zietsch, B. P., & Lynskey, M. T. (2010). Genetic and environmental influences on cannabis use initiation and problematic use: A meta-analysis of twin studies. Addiction, 105(3), 417–430.10.1111/j.1360-0443.2009.02831.xCrossRefGoogle ScholarPubMed
Wainberg, M., Jacobs, G. R., di Forti, M., & Tripathy, S. J. (2021). Cannabis, schizophrenia genetic risk, and psychotic experiences: A cross-sectional study of 109,308 participants from the UK biobank. Translational Psychiatry, 11(1), 211. https://doi.org/10.1038/s41398-021-01330-w.CrossRefGoogle Scholar
Ziedonis, D., Hitsman, B., Beckham, J. C., et al. (2008). Tobacco use and cessation in psychiatric disorders: National Institute of Mental Health report. Nicotine & Tobacco Research, 10(12), 1691–1715. https://doi.org/10.1080/14622200802443569.CrossRefGoogle ScholarPubMed
Figure 0

Table 1. All of Us concept codes used to assign binary identifiers for schizophrenia, heavy cannabis use, heavy tobacco smoking, and antipsychotic medication statuses. Participants across different Concept IDs may overlap. See Supplementary Figure 1 for sample details.

Figure 1

Table 2. Summary of polygenic associations with heavy cannabis use and schizophrenia across various case group definitions. a: analyses include all heavy cannabis users regardless of schizophrenia status or all schizophrenia cases regardless of heavy cannabis use patterns (“relaxed”) or either exclude all comorbid cases or include only comorbid cases (“strict”). b: ind=Independently modelled; joint=both PGSs modelled jointly.

Figure 2

Figure 1. CUD and schizophrenia (SCZ) polygenic score associations with heavy cannabis use and schizophrenia. Associations shown for European and African samples under Relaxed (cases only screened for the primary outcome), Strict (cases with comorbid schizophrenia or heavy cannabis use removed), and comorbid (including both schizophrenia and heavy cannabis use) case definitions. Analyses conducted with CUD and SCZ PGSs considered independently (indep) or jointly (joint).

Figure 3

Figure 2. Liability-scale R2 for PGS associations with heavy cannabis use (CU) and schizophrenia (SCZ) in EUR estimated from a population-level prevalence of 0.90% for schizophrenia (Perälä, Suvisaari, Saarni, et al., 2007) and 6.27% for CUD (Hasin, Kerridge, Saha, et al., 2016).

Supplementary material: File

Austin-Zimmerman et al. supplementary material 1

Austin-Zimmerman et al. supplementary material
Download Austin-Zimmerman et al. supplementary material 1(File)
File 849.6 KB
Supplementary material: File

Austin-Zimmerman et al. supplementary material 2

Austin-Zimmerman et al. supplementary material
Download Austin-Zimmerman et al. supplementary material 2(File)
File 264.3 KB