The current findings suggest that heavy and early-onset cannabis users are more likely to report unusual cannabis-related experiences and that higher genetic vulnerability to schizophrenia may place regular cannabis users at even greater risk for these experiences, especially cognitive difficulties. Even accounting for aspects of cannabis involvement, including polygenic risk for CUD, the schizophrenia PRS was significantly associated with all cannabis-related experiences except for hallucinations. This suggests a role for genetic vulnerability to schizophrenia in cannabis-related experiences, even in those without psychotic disorders. Interestingly, we find that accounting for a PRS for CUD does not dampen the associations between the PRS for schizophrenia and cannabis-related experiences, suggesting that these cannabis-related experiences are more strongly related to genetic liability for schizophrenia in this sample. Despite nonsignificant effects in the replication models, the magnitude of association for the schizophrenia PRS was comparable (supplementary table 6), implying that the much smaller sample size for CATS is the primary factor behind the null results in this replication sample. Notably, these findings are from samples that were partly ascertained for AUD or opioid use disorder and generally had high levels of other substance use; therefore, these results may not generalize to the overall population.

In COGA, there was especially robust support across ancestries for an association between the schizophrenia PRS and cognitive difficulties (meta-analyzed beta = 0.22, SE = 0.04, P = 5.2e−7). This is consistent with previous studies which have found that schizophrenia PRS are associated with increased disorganized symptom scores and lower cognitive ability,^17,18 while weaker associations are reported with other symptom domains. In the European ancestry sample, we created a categorical variable representing quartiles of polygenic risk for schizophrenia; when we regressed cognitive difficulties on this categorical variable of risk, we found that individuals in the top 25% of polygenic risk for schizophrenia had 2 times greater odds of reporting cognitive difficulties after using cannabis, relative to individuals in the lowest quartile of risk (supplementary figure 2). We note that while this may seem like a large increase in risk, there is great uncertainty in PRS models and individual-level prediction is unlikely to be accurate. Of the 5 cannabis-related experiences studied in this report, hallucinations were the least common and the least strongly correlated with the other reported experiences (supplementary figure 1). This is consistent with most reports of cannabis-related PLEs, which more often implicate delusions, paranoia, and cognitive dysfunction than hallucinations.^7,8,40,41

We found little evidence to support multiplicative interactions between the schizophrenia PRS and duration of daily use or age at first cannabis use (supplementary table 5). However, we note that the ascertained nature of the COGA sample and the fact that the overwhelming majority of individuals who endorse cannabis-related experiences in COGA have met the criteria for a lifetime CUD diagnosis (eg, 98% of individuals who report cognitive difficulties have met criteria for a CUD diagnosis) may have influenced our findings. The exclusion of individuals with schizophrenia, by study design, may also have impacted our results. It is likely that this range restriction (ie, few individuals who report cannabis-related experiences without heavy cannabis use and few individuals at the highest end of the spectrum of risk for schizophrenia), coupled with a small sample size, limited our statistical power to identify interactions. It is worth noting that the strongest interaction effect observed, which suggested that polygenic liability for schizophrenia has a greater effect on risk for any cannabis-related experiences in individuals with a CUD diagnosis, was actually stronger in the model that controlled for all covariate-by-PRS and covariate-by-CUD interactions³⁷ compared to the model without these terms (beta = 0.43, SE = 0.14, P = .003 vs beta = 0.29, SE = 0.12, P = .01). Our primary interaction models were on the multiplicative scale, a test of whether the combined effect of the PRS and moderator differs from the product of their individual effects. However, in light of these null findings, we subsequently tested interactions on the additive scale, which tests whether the combined effect of the PRS and moderator differs from the sum of their individual effects. Some have argued that departures from additivity may be more meaningful biologically than multiplicative models.^42,43 Supporting this theory, several of the interactions were stronger in the additive models (eg, PRS-by-age at first use predicting cognitive difficulties P = .004, PRS-by-CUD predicting cognitive difficulties P = .0011), though none of the interaction effects passed our statistical significance threshold (α = 9.8e−4). Our findings are most consistent with a multifactorial model,^44,45 wherein polygenic risk for schizophrenia combines additively with environmental risk factors, including early and heavy cannabis use, to increase an individual’s vulnerability to unusual cannabis-related experiences. Relatedly, while we found that CUD severity and polygenic risk for schizophrenia were both associated with the endorsement of a greater number of cannabis-related experiences, the schizophrenia PRS and CUD severity were not related, suggesting that these factors independently contribute to the risk of cannabis-related experiences in this sample.

There were few statistically significant differences in the prevalence of cannabis-related experiences across ancestries: self-reported cannabis-related paranoia was more common in the African ancestry sample than in the European ancestry sample, while cannabis-related social withdrawal was more commonly reported in the European ancestry sample. While there was no difference in associations with the measures of cannabis involvement (table 3), the schizophrenia PRS was more weakly associated with cannabis-related experiences in the African ancestry subsample of COGA relative to the European ancestry subsample, a divergence supported by significant heterogeneity tests in the meta-analysis of 2 of the 4 significant outcomes (table 4). One potential explanation for this divergence is the relatively small sample size of the African ancestry GWAS. While multi-ancestry methods like PRS-CSx have been shown to improve the predictive power of PRS in diverse samples,³² these methods still fall short of having a large, fully ancestry-matched discovery GWAS to construct SNP weights for PRS.⁴⁶ It may also be the case that the relative importance of different risk factors (eg, readiness to report hallucinations and other PLEs^47–49; the likelihood of living in an urban area^50,51; and the cumulative effects of racism and discrimination^52–54) for cannabis-related experiences varies across these 2 ancestry groups in COGA. However, our study was not designed to address this question; more studies of cannabis-related psychosis risk in multi-ancestry samples and large-scale discovery of GWAS in non-European ancestries are needed.

Our findings should be viewed with some limitations in mind. First, it is unclear how the timing of the self-reported cannabis-related experiences relate to the period of heaviest cannabis use, limiting causal inferences that can be drawn from our data. Second, although these experiences were self-reported to have occurred “because of” one’s marijuana use, assessments were retrospective, and it is unclear how tightly these experiences were temporally linked to cannabis use. We cannot determine whether individuals were acutely intoxicated when these experiences occurred. We did not have data available on the persistence of these experiences, nor whether they were considered distressing. However, the polygenic risk for schizophrenia was associated with a greater number of cannabis-related experiences endorsed (meta-analyzed beta = 0.11, SE = 0.02, P = 1.3e−6); this provides some evidence that genetic risk for schizophrenia is associated with greater severity of cannabis-related experiences as indexed by a greater number of experiences endorsed. Third, we analyzed a variable assessing duration of daily cannabis use—self-reported the longest period of time an individual used marijuana almost every day—but this may not be an ideal measure of the heaviness of use. Beyond limitations associated with self-report, this measure does not capture quantity nor potency, and any “noise” captured by this variable may have contributed to its weaker associations with cannabis-related experiences. Fourth, no data on the potency of cannabis used was available, and there is evidence that the ratio of THC to cannabidiol can vary widely in available cannabis across the US^55,56; there is a possibility that varying levels of cannabis potency and THC:cannabidiol ratio could have affected our results. Fifth, we only examined the genetic risk for schizophrenia, but some of the cannabis-related experiences analyzed—psychosis, anhedonia, and social withdrawal—are common features of other disorders, including bipolar disorder and depression, for which we did not examine polygenic liability. Sixth, while individuals with psychotic disorder diagnoses were excluded from the COGA cohort,²¹ new onsets of psychotic disorders may have been missed as the SSAGA does not evaluate these diagnoses. However, these are expected to be rare. Finally, the cannabis-related experiences were only queried in individuals who reported using cannabis at least 11 times (or 21 times in a year) in SSAGA assessments. Thus, our subsample excludes any individuals who may have had these experiences at lower levels of cannabis use.

In conclusion, we found that polygenic liability for schizophrenia was associated with an increased risk of cannabis-related paranoia, depression and anhedonia, decreased social contact, and cognitive difficulties, even when accounting for the effects of other pertinent risk factors including CUD, age at first use of cannabis, and lifetime use of other illicit substances. As noted above, our findings come from ascertained samples with a high prevalence of CUD and use of other substances; thus, it is unclear whether these findings will generalize to population-based samples. Our results support the hypothesis that individuals who have a preexisting genetic vulnerability for schizophrenia are more likely to report unusual experiences, especially cognitive difficulties, when using cannabis.