Wainberg et al.

Translational Psychiatry (2021)11:211

https://doi.org/10.1038/s41398-021-01330-w Translational Psychiatry

ARTICLE Open Access

Cannabis, schizophrenia genetic risk, and psychotic

experiences: a cross-sectional study of 109,308
participants from the UK Biobank
1
Michael Wainberg , Grace R. Jacobs1,2, Marta di Forti3,4 and Shreejoy J. Tripathy1,2,5,6

Abstract
Cannabis is known to produce acute, transient psychotic-like experiences. However, it is unclear whether cannabis
disproportionately increases the risk of specific types of psychotic experiences and whether genetic predisposition
influences the relationship between cannabis use and psychotic experiences. In this cross-sectional study of 109,308
UK Biobank participants, we examined how schizophrenia polygenic risk modulates the association between self-
reported cannabis use and four types of self-reported psychotic experiences (auditory hallucinations, visual
hallucinations, persecutory delusions, and delusions of reference). Cohort-wide, we found a strong, dose-dependent
relationship between cannabis use and all four types of psychotic experiences, especially persecutory delusions.
Cannabis users’ psychotic experiences tended to be earlier-onset and cause greater distress than non-users’, but were
not more likely to lead to help-seeking. Participants with high schizophrenia polygenic risk scores showed stronger
associations between cannabis use and auditory hallucinations, visual hallucinations, and delusions of reference, as
well as psychotic experiences overall. For instance, cannabis ever-use was associated with 67% greater adjusted odds
1234567890():,;
1234567890():,;
1234567890():,;
1234567890():,;

of delusions of reference among individuals in the top fifth of polygenic risk, but only 7% greater adjusted odds
among the bottom fifth. Our results suggest that cannabis use is a predictive risk factor for psychotic experiences,
including early-onset and distressing experiences. Individuals genetically predisposed to schizophrenia may be
especially vulnerable to psychotic experiences as a result of using cannabis, supporting a long-postulated hypothesis.
This study exemplifies the utility of population-scale biobanks for elucidating gene-by-environment interactions
relating substance use to neuropsychiatric outcomes and points to the translational potential of using polygenic risk
scores to inform personalized harm reduction interventions.

Introduction experiences appears largely genetically mediated, albeit with

Substantial epidemiological evidence associates cannabis some environmental contribution6–8.
use with psychosis1 and accelerated age of onset of psy- Multinational population-based surveys suggest self-
chosis2,3, although the causality of these relationships have reported psychotic experiences—whether caused by can-
long been debated1,4. What is incontrovertible is that can- nabis or not—are several times more common than
nabis can induce acute psychotic-like experiences in healthy diagnosed psychotic disorders9. Despite most not being
individuals, although these are generally mild and tran- sufficiently severe to merit a diagnosis, psychotic experi-
sient5. The association of cannabis use with psychotic-like ences nonetheless predict poor outcomes including psy-
chotic disorders10–12, supporting their clinical relevance. A
recent meta-analysis found a dose-dependent relationship
between cannabis use and a variety of psychosis-related
Correspondence: Shreejoy J. Tripathy (shreejoy.tripathy@camh.ca)
1 outcomes, including self-reported psychotic experiences13.
Centre for Addiction and Mental Health, Toronto, ON, Canada
2
Institute of Medical Sciences, University of Toronto, Toronto, ON, Canada
Full list of author information is available at the end of the article

© The Author(s) 2021

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction
in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if
changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If
material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain
permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
Wainberg et al. Translational Psychiatry (2021)11:211 Page 2 of 9

Cannabis use may have different relationships with phenotyping on ~500,000 British individuals, aged 40–69
different types of psychotic experiences. Although can- years at recruitment. In total, 157,348 participants com-
nabis was historically classified as a hallucinogen based on pleted an online Mental Health Questionnaire23, of which
the acute perceptual changes it tends to induce14, this 109,308 participants (61,047 female and 48,261 male) of
categorization is considered controversial15 and case unrelated White British ancestry (defined using the same
reports of bona fide cannabis-induced hallucinations are criteria as a previous study24) met the inclusion criteria.
rare15,16. On the other hand, delusions resulting from Specifically, these participants answered questions on
cannabis use appear much more common: cannabis both cannabis use and psychotic experiences, and lacked a
readily induced delusional thinking in multiple instances diagnosis of any psychotic disorder (ICD-10 codes
within a single randomized control trial5 and cannabis use F20–F29) according to linked inpatient, primary care, or
has been linked to persecutory ideation17 and paranoia18. death records (e.g., according to “Source of report of F20
Although these observations relate primarily to the acute (schizophrenia)”, Data-Field 130875). No individuals had
effects of cannabis intoxication, the same dichotomy an ICD code for F19.15 or F19.95 (drug-induced psy-
between hallucinations and delusions may also apply in chosis); thus, this was not used as a criterion for exclusion.
the longer term. For instance, each year of regular mar- This cohort formed the basis of our analyses.
ijuana use among adolescent boys was associated with an
odds ratio of 1.92 for hallucinations, but an even greater Definitions of self-reported cannabis use and psychotic
odds ratio of 2.33 for paranoia19. experiences
Genetic studies have implicated the endocannabinoid Self-reported cannabis use was defined by the “Ever
system in psychotic experiences: the strongest association taken cannabis” question (Data-Field #20453; Table 1).
with distressing psychotic experiences in a recent genome- Use frequency was defined by the “Maximum frequency
wide association study (GWAS) was the CNR2 (cannabi- of taking cannabis” question (#20454: ever-use = 1, N =
noid receptor 2) locus10. Genetics may also modulate 14,642; monthly use = 2, N = 2671; weekly use = 3, N =
susceptibility to cannabis-related harm: the association 3582; daily use = 4, N = 1403); individuals answering no
between cannabis use and decreased cortical thickness, a to “Ever taken cannabis” were not asked this question and
risk factor for psychotic experiences, among early adoles- assigned a value of 0 (N = 87,010).
cent males is strongest among those with the highest Self-reported psychotic experiences were defined as
polygenic risk for schizophrenia20. However, to our auditory (#20463) or visual (#20471) hallucinations, or
knowledge, no study has directly tested the hypothesis21,22 delusions of persecution (#20468) or reference (#20474),
that genetic predisposition to schizophrenia makes can- as in previous studies of the UK Biobank10,25. The number
nabis users especially liable to psychotic experiences. of individuals of the 109,308 with non-missing data for
Here we sought to investigate the association between each individual psychotic experience ranged from 108,174
cannabis use and self-reported psychotic experiences in a for visual hallucinations to 109,104 for persecutory delu-
population-scale cohort, the UK Biobank, with several sions. Self-reported age of onset was ascertained from the
questions in mind. First, whether cannabis use is differ- “Age when first had unusual or psychotic experience”
entially associated with different categories of self-reported question (#20461), distress from “Distress caused by
psychotic experiences. The UK Biobank asks about four unusual or psychotic experiences” (#20462), and help-
types—auditory hallucinations, visual hallucinations, per- seeking from “Ever talked to a health professional about
secutory delusions, and delusions of reference—and we set unusual or psychotic experiences” (#20477).
out to characterize cannabis’s associations with each one,
with appropriate correction for multiple testing. Second, Schizophrenia polygenic risk score
whether cannabis users’ self-reported psychotic experi- A polygenic risk score (PRS) for schizophrenia was
ences are different from non-users’ in terms of age of computed for each participant, based on a recent
onset2,3, distress, and likelihood of help-seeking. Third, GWAS26 (walters.psycm.cf.ac.uk/clozuk_pgc2.meta.sum-
whether the association of cannabis use with any of the stats.txt.gz) from an independent cohort. First, samples
four types of self-reported psychotic experiences is more were subset to genetically defined White British (“Genetic
pronounced among individuals genetically predisposed to ethnic grouping” [Data-Field #22006] is 1) without sex
schizophrenia, once again rigorously correcting for multi- chromosome aneuploidy (missing value for “Sex chro-
ple testing. mosome aneuploidy” [#22019]), who were used to com-
pute genotype principal components (non-missing value
Methods for “Used in genetic principal components” [#22020]).
Participants Second, variants from the UK Biobank’s imputed geno-
Participants were included from the UK Biobank, a types were subset to non-duplicate, autosomal single-
prospective cohort study with genetics and deep nucleotide polymorphisms with call rate > 95%,
Wainberg et al. Translational Psychiatry (2021)11:211 Page 3 of 9

Table 1 Cannabis use and psychotic experience questions from the UK Biobank Mental Health Questionnaire.

Categories of self-reported psychotic experiences

Auditory hallucinations Did you ever hear things that other people said did not exist, like strange voices coming from inside your head talking to you
or about you, or voices coming out of the air when there was no one around?
Visual hallucinations Did you ever see something that wasn’t really there that other people could not see?
Persecutory delusions Did you ever believe that there was an unjust plot going on to harm you or to have people follow you, and which your
family and friends did not believe existed?
Delusions of reference Did you ever believe that a strange force was trying to communicate directly with you by sending special signs or signals
that you could understand but that no one else could understand (e.g., through the radio or television)?
Qualities of self-reported psychotic experiences
Age of onset How old were you (approximately) when you first had one of these experiences?
Distress How distressing did you find having any of these experiences?
Help-seeking Did you ever talk to a doctor, counselor, psychiatrist, or other health professionals about any of these experiences?
Self-reported cannabis use
Cannabis ever-use Have you taken cannabis (marijuana, grass, hash, ganja, blow, draw, skunk, weed, spliff, dope), even if it was a long time ago?
Cannabis use frequency Considering when you were taking cannabis most regularly, how often did you take it?

Hardy–Weinberg equilibrium p-value > 1 × 10−10, allele analysis, cannabis use frequency (coded as never-use = 0,
frequency > 0.1%, and imputation info score > 0.8. Third, ever-use = 1, monthly use = 2, weekly use = 3, daily
GWAS summary statistics were harmonized with the UK use = 4) was used in the logistic regression instead of
Biobank with respect to reference/alternate allele and cannabis ever-use. Covariates consisted of birth year
strand; ambiguous variants (A/T, C/G, G/C, and T/A) and (Data-Field #34), sex (#31), educational qualifications
variants missing from UK Biobank were excluded. Fourth, (#6138), pre-tax household income (#738), employment
variants were filtered to p < 0.05. This p < 0.05 threshold status (#6142), Townsend deprivation index (#189), Index
led to a better prediction of schizophrenia across the of Multiple Deprivation (#26410, #26426, and #26427),
unrelated White British individuals (area under the curve smoking status (#20116), alcohol intake frequency
[AUC] = 0.677) than stricter p-value thresholds of (#1558), UK Biobank assessment center (#54), and the top
0.005 (AUC = 0.648), 0.0005 (AUC = 0.612), or 0.00005 ten genotype principal components (#22009). Categorical
(AUC = 0.586). (The AUC, also known as the area under covariates were coded as indicator variables.
the receiver operating characteristic curve or concordance P-values for additive interactions with schizophrenia
statistic, is the fraction of the time that the PRS would genetic risk were calculated using the glm function by
rank a randomly chosen case higher than a randomly performing linear regression of psychotic experiences on
chosen control.) Fifth, linkage disequilibrium pruning to three variables (plus covariates)—the exposure (either
r2 < 0.5 was performed using a 500 kb sliding window. The cannabis ever-use or cannabis use frequency), the schi-
effect sizes (log odds ratios) of the remaining variants zophrenia PRS, and the product of the two (i.e., the
constituted the weights of the PRS. The PRS was scored interaction term)—then performing a χ2-test using the
on each individual in the cohort by summing, across the anova function in R, to compare this model to a simpler
variants in the PRS, the variant’s weight times the indi- two-variable model (plus covariates) lacking the inter-
vidual’s number of effect alleles of that variant; missing action term. To properly control for confounding27,
genotypes were mean-imputed. covariate-by-exposure and covariate-by-PRS interaction
terms were also included as covariates in both models.
Statistical analysis Multiple testing correction was performed using
Raw prevalences of each self-reported psychotic Benjamini–Hochberg correction at a standard false dis-
experience were tabulated among ever- and never-users. covery rate (FDR) of 10%.
Covariate-corrected adjusted odds ratios (AORs) and P-values for the difference D between pairs of logistic
associated 95% confidence intervals were also calculated, regression coefficients (log odds ratios) were calculated by
via logistic regression of each psychotic experience on computing a SE for the difference as the root sum of
cannabis ever-use (coded as binary variables) and cov- squares of the coefficients’ SEs, dividing D by this SE to
ariates, using the glm function in R. As a sensitivity yield a Z-score, then inverse-normal transforming.
Wainberg et al. Translational Psychiatry (2021)11:211 Page 4 of 9

Table 2 Self-reported psychotic experiences are strongly associated with cannabis use frequency.

Prevalence of self-reported psychotic experiences in the UK Biobank, stratified by self-reported AOR per risk unit
cannabis use frequency

Never Ever (1 risk unit) Monthly Weekly Daily

(0 risk units) (2 risk units) (3 risk units) (4 risk units)

Any psychotic 4.1% 7.0% 8.4% 8.8% 9.6% AOR = 1.20

experience AOR = 1.54 AOR = 1.69 AOR = 1.69 AOR = 1.79 [1.16, 1.24]
[1.43, 1.65] [1.54, 1.87] [1.51, 1.89] [1.52, 2.20]
Auditory hallucinations 1.3% 2.7% 3.4% 3.6% 3.6% AOR = 1.21
AOR = 1.57 AOR = 1.84 AOR = 1.85 AOR = 1.79 [1.15, 1.27]
[1.40, 1.77] [1.57, 2.16] [1.55, 2.21] [1.38, 2.33]
Visual hallucinations 2.8% 4.6% 5.4% 5.5% 6.1% AOR = 1.21
AOR = 1.58 AOR = 1.69 AOR = 1.66 AOR = 1.76 [1.16, 1.26]
[1.45, 1.73] [1.50, 1.91] [1.45, 1.91] [1.44, 2.15]
Persecutory delusions 0.6% 1.3% 1.8% 2.2% 2.6% AOR = 1.24
AOR = 1.59 AOR = 1.95 AOR = 2.14 AOR = 2.44 [1.15, 1.33]
[1.34, 1.89] [1.56, 2.44] [1.68, 2.74] [1.75, 3.40]
Delusions of reference 0.6% 0.9% 1.2% 1.3% 1.4% AOR = 1.18
AOR = 1.39 AOR = 1.57 AOR = 1.65 AOR = 1.67 [1.08, 1.28]
[1.15, 1.68] [1.21, 2.03] [1.24, 2.19] [1.10, 2.53]

Square brackets denote 95% confidence intervals. It is noteworthy that AORs are adjusted for covariates, while percentages are not.
The bold values are purely for visual emphasis.

Results Cannabis use is particularly associated with persecutory

Self-reported psychotic experiences strongly correlate delusions
with cannabis use frequency Considering each of the four types of psychotic
When considering the cohort as a whole, we found a experiences individually (Table 2), we again found strong
strong and consistent relationship between self-reported associations with cannabis ever-use (AOR = 1.39–1.59),
cannabis use frequency and all types of self-reported psy- and even stronger ones with monthly (AOR =
chotic experiences (Table 2). Although 4.1% of cannabis 1.57–1.95), weekly (AOR = 1.65–2.14), and daily
never-users reported one of the four types of experiences (AOR = 1.67–2.44) use. Persecutory delusions were
surveyed (auditory hallucinations, visual hallucinations, especially strongly correlated (AOR = 2.44 [1.96, 3.64]
persecutory delusions, or delusions of reference), this rose for daily users; AOR = 1.24 [1.15, 1.33] per risk unit).
to 7.0% among ever-users (AOR = 1.54 [1.43, 1.65]) and Thus, when considering the cohort as a whole, we found
rose further to 8.4% among those reporting ever using a strong, dose-dependent relationship between cannabis
cannabis at least monthly (AOR = 1.69 [1.54, 1.87]), 8.8% use and all four types of psychotic experiences, parti-
among ever-weekly users (AOR = 1.69 [1.51, 1.89]), and cularly persecutory delusions.
9.6% among ever-daily users (AOR = 1.79 [1.52, 2.20]).
Defining never-use as 0 “risk units”, ever-use as 1 “risk unit”, Cannabis ever-users report earlier-onset and more
and so forth up to 4 “risk units” for daily use, we found that distressing experiences than never-users
the odds of any of the four psychotic experiences increased Cannabis users were also especially likely to report early-
by 20% per risk unit (AOR = 1.20 [1.16, 1.24]). onset (<18 years old) psychotic experiences (Table 4).
A sensitivity analysis stratifying by sex (Table 3) indi- Cannabis ever-users reported adult-onset psychotic
cated that the association of cannabis ever-use with psy- experiences at greater rates than never-users (AOR = 1.52
chotic experiences was significantly stronger among [1.38, 1.66]), but even greater rates of early-onset experi-
females than among males (AOR = 1.59 vs. 1.44). Two ences (AOR = 1.90 [1.64, 2.20]). (When calculating the
particular types of psychotic experiences, auditory hallu- association with adult-onset psychotic experiences, indi-
cinations (AOR = 1.69 vs. 1.40) and delusions of reference viduals with early-onset experiences were excluded, and
(AOR = 1.67 vs. 1.20), also had significantly stronger vice versa.) Thus, although cannabis use was associated
associations among females. with both adult- and early-onset psychotic experiences, its
Wainberg et al. Translational Psychiatry (2021)11:211 Page 5 of 9

Table 3 Sex differences in associations between psychotic experiences and cannabis ever-use.

Prevalence of self-reported psychotic experiences (ever- vs. never-users)

Any psychotic experience Auditory hallucinations Visual hallucinations Persecutory delusions Delusions of reference

Female 8.0% vs. 4.5% 3.0% vs. 1.5% 5.5% vs. 3.1% 1.0% vs. 0.5% 0.9% vs. 0.5%
AOR = 1.59 AOR = 1.69 AOR = 1.64 AOR = 1.58 AOR = 1.67
[1.45, 1.75] [1.45, 1.97] [1.47, 1.84] [1.21, 2.08] [1.28, 2.19]
Male 6.1% vs. 3.7% 2.3% vs. 1.2% 3.6% vs. 2.2% 1.3% vs. 0.6% 0.8% vs. 0.5%
AOR = 1.44 AOR = 1.40 AOR = 1.51 AOR = 1.59 AOR = 1.20
[1.30, 1.61] [1.17, 1.67] [1.32, 1.74] [1.25, 2.02] [0.90, 1.60]
Difference FDR = 6% FDR = 5% FDR = 20% FDR = 97% FDR = 5%
(p = 0.04) (p = 0.01) (p = 0.16) (p = 0.97) (p = 0.02)

FDRs are derived from Benjamini–Hochberg correction for five tests.

FDR false discovery rate.
The bold values are purely for visual emphasis.

Table 4 Cannabis ever-users report earlier-onset and Schizophrenia polygenic risk modulates the association of
more distressing psychotic experiences than never-users. cannabis use with psychotic experiences
Finally, we considered whether a PRS for schizophrenia
Prevalence of self-reported psychotic
modulated the strengths of association between cannabis
experiences with particular qualities (ever- vs.
use and self-reported psychotic experiences (Table 5). We
never-users)
found that schizophrenia polygenic risk significantly
interacted with cannabis ever-use to predict rates of
Early-onset Distressing Associated with
auditory hallucinations (p = 0.02, FDR = 9%), delusions of
(<18 help-seeking
reference (p = 0.04, FDR = 9%), and psychotic experi-
years old)
ences overall (p = 0.05, FDR = 9%). Cannabis use fre-
Had a psychotic 1.9% vs. 0.8% 3.0% vs. 1.5% 1.6% vs. 0.8% quency was better powered to detect interactions with
experience that AOR = 1.90 AOR = 1.62 AOR = 1.45 schizophrenia polygenic risk, with significant results not
was ________ [1.64, 2.20] [1.45, 1.81] [1.25, 1.70] only for auditory hallucinations (p = 0.01, FDR = 2%),
Had a psychotic 4.1% vs. 2.5% 4.1% vs. 2.6% 5.6% vs. 3.3%
delusions of reference (p = 0.0007, FDR = 0.4%), and
experience that was AOR = 1.52 AOR = 1.50 AOR = 1.55
psychotic experiences overall (p = 0.01, FDR = 2%), but
not ________ [1.38, 1.66] [1.37, 1.64] [1.43, 1.68]
also for visual hallucinations (p = 0.06, FDR = 7%).
To better interpret these interactions, we stratified
Difference FDR = 0.5% FDR = 9% FDR = 81%
individuals into quintiles (20-percentile bins) based on
(p = 1 × 10−5) (p = 0.1) (p = 0.3)
their schizophrenia polygenic risk and computed asso-
FDRs are derived from Benjamini–Hochberg correction for three tests; square ciations within each quintile between cannabis ever-use
brackets denote 95% confidence intervals. and each psychotic experience. We found that ever-use
FDR false discovery rate.
The bold values are purely for visual emphasis. was associated with 1.58-fold [1.36, 1.58] greater adjusted
odds of psychotic experiences among the one-fifth of
individuals with the highest PRSs, compared to only 1.39-
association with early-onset experiences was significantly fold [1.16, 1.65] greater adjusted odds among the one-fifth
more pronounced (p = 1 × 10−5, FDR = 0.5%). Ever-users with the lowest PRSs. This pattern also held true for
also reported disproportionately greater rates of distres- auditory hallucinations (AOR = 1.73 [1.36, 2.22] among
sing psychotic experiences (AOR = 1.62 [1.45, 1.81]) the top quintile vs. 1.36 [1.02, 1.82] among the bottom
compared to non-distressing ones (AOR = 1.50 [1.37, quintile) and visual hallucinations (AOR = 1.66 [1.38,
1.64]), but no greater rates of psychotic experiences lead- 1.99] vs. 1.46 [1.18, 1.81]), and was particularly pro-
ing to help-seeking (AOR = 1.45 [1.25, 1.70]) compared to nounced for delusions of reference, which had no sig-
ones not leading to help-seeking (AOR = 1.55 [1.43, 1.68]). nificant association with cannabis use except among those
Thus, ever-users’ psychotic experiences tended to be in the top two-fifths of genetic risk. Thus, cannabis use
earlier-onset and more distressing than never-users’, but was disproportionately highly correlated with psychotic
no more likely to lead to help-seeking. experiences among individuals at high genetic risk of
Wainberg et al. Translational Psychiatry (2021)11:211 Page 6 of 9

Table 5 Schizophrenia PRS modulates the association of cannabis use with self-reported psychotic experiences.

Prevalence of self-reported psychotic experiences (ever- vs. never-users), stratified by PRS-by-ever- PRS-by-
schizophrenia PRS use interaction frequency
interaction
0–20th 20–40th 40–60th 60–80th 80–100th
Percentile Percentile Percentile Percentile Percentile
SCZ PRS SCZ PRS SCZ PRS SCZ PRS SCZ PRS

Any psychotic 5.8 vs. 3.6% 6.7 vs. 4.0% 7.0 vs. 4.2% 7.5 vs. 4.3% 8.2 vs. 4.7% FDR = 9% FDR = 2%
experience AOR = 1.39 AOR = 1.48 AOR = 1.52 AOR = 1.64 AOR = 1.58 (p = 0.02) (p = 0.01)
[1.16, 1.65] [1.26, 1.74] [1.30, 1.78] [1.40, 1.91] [1.36, 1.83]
Auditory 2.2 vs. 1.2% 2.4 vs. 1.4% 2.7 vs. 1.5% 2.8 vs. 1.3% 3.2 vs. 1.5% FDR = 9% FDR = 2%
hallucinations AOR = 1.36 AOR = 1.32 AOR = 1.54 AOR = 1.91 AOR = 1.73 (p = 0.04) (p = 0.01)
[1.02, 1.82] [1.01, 1.73] [1.20, 1.98] [1.47, 2.48] [1.36, 2.22]
Visual 3.7 vs. 2.4% 4.6 vs. 2.7% 4.5 vs. 2.7% 4.9 vs. 3.0% 5.2 vs. 3.0% FDR = 17% FDR = 7%
hallucinations AOR = 1.46 AOR = 1.63 AOR = 1.52 AOR = 1.59 AOR = 1.66 (p = 0.1) (p = 0.06)
[1.18, 1.81] [1.34, 1.98] [1.25, 1.85] [1.31, 1.91] [1.38, 1.99]
Persecutory 0.8 vs. 0.4% 1.3 vs. 0.4% 1.1 vs. 0.6% 1.4 vs. 0.6% 1.7 vs. 0.9% FDR = 99% FDR = 29%
delusions AOR = 1.60 AOR = 2.30 AOR = 1.65 AOR = 1.39 AOR = 1.36 (p = 1) (p = 0.3)
[0.98, 2.60] [1.53, 3.45] [1.11, 2.44] [0.96, 2.00] [0.97, 1.89]
Delusions of 0.5 vs. 0.4% 0.7 vs. 0.5% 0.8 vs. 0.6% 1.1 vs. 0.6% 1.5 vs. 0.7% FDR = 9% FDR = 0.4%
reference AOR = 1.07 AOR = 1.29 AOR = 1.07 AOR = 1.54 AOR = 1.68 (p = 0.05) (p = 0.0007)
[0.63, 1.82] [0.82, 2.05] [0.69, 1.67] [1.04, 2.28] [1.18, 2.38]

FDRs for each type of interaction are derived from Benjamini–Hochberg correction for five tests. Square brackets denote 95% confidence intervals. It is noteworthy
that our interaction tests (two right-most columns) treat polygenic risk as a continuous variable and do not rely on discretization into quintiles.
FDR false discovery rate.
The bold values are purely for visual emphasis.

schizophrenia and less correlated among individuals at population-scale biobanks for elucidating gene-by-
lower risk. environment interactions.
Notably, self-reported delusions of persecution and
Discussion reference each have a special association with cannabis, but
Our results suggest four main findings (Fig. 1). First, we in different ways. Persecutory delusions have the strongest
confirm a strong, dose-dependent association between association with cannabis use frequency of all psychotic
cannabis use and self-reported psychotic experiences, experiences surveyed, whereas delusions of reference are
consistent across all four types of psychotic experiences the sole type of delusion to interact with schizophrenia
surveyed—an important replication supported by the UK polygenic risk vis-a-vis cannabis use. It has been suggested
Biobank’s large sample size. Second, we find a particularly that although both types of delusions involve frontostriatal
pronounced association of cannabis use with persecutory prediction errors29, some brain regions involved may be
delusions. Third, we provide the first evidence of an specific to one type or the other. Paranoid delusions may
association between cannabis use and earlier psychotic specifically involve regions involved in theory of mind, in
experience onset, extending prior studies showing such an particular the right temporoparietal junction and right
association with diagnosed psychotic disorders3,28. posterior superior temporal sulcus, as well as the amygdala,
Although cannabis ever-users report more distressing which is involved in paranoia-associated fear and hypervi-
psychotic experiences than never-users, rates of help- gilance, whereas delusions of reference may instead involve
seeking were similar in both groups. Fourth, we discover a the associative striatum and nucleus basalis30. Cannabis is
strong modulatory effect of schizophrenia polygenic risk known to interact with the amygdala31, associative stria-
on cannabis’s association with multiple types of psychotic tum32, and nucleus basalis33, and it is conceivable that
experiences. The difference in AORs between those in the cannabis’s interactions with delusions-of-reference-specific
bottom and top fifth of polygenic risk is strikingly large for regions may be more modulated by schizophrenia-
delusions of reference (7% vs. 67%) and auditory halluci- associated genetic factors than its interactions with
nations (36% vs. 74%), exemplifying the added value of persecutory-delusions-specific regions.
Wainberg et al. Translational Psychiatry (2021)11:211 Page 7 of 9

3.5
Cannabis use
Never Psychotic experience
2.2
Ever that was ___
3.0 Psychotic experience
Monthly
Adjusted odds ratio [AOR]

2.0 that was not ___

Weekly

Adjusted odds ratio [AOR]

Daily
2.5
1.8

2.0 1.6

1.4
1.5
1.2

1.0
1.0
Any psychotic Auditory Visual Persecutory Delusions Early-onset Distressing Associated with
experience hallucinations hallucinations delusions of reference help-seeking

3.5 Schizophrenia PRS

0-20th percentile
20-40th percentile
3.0 40-60th percentile
Adjusted odds ratio [AOR]

60-80th percentile
80-100th percentile
2.5

2.0

1.5

1.0
Any psychotic Auditory Visual Persecutory Delusions
experience hallucinations hallucinations delusions of reference

Fig. 1 Summary of results. Cannabis use frequency correlates with all types of psychotic experiences (top left), cannabis ever-users report earlier-
onset and more distressing psychotic experiences (top right), and schizophrenia PRS modulates the association of cannabis use with most types of
psychotic experiences (bottom).

Our findings should be evaluated in the context of their The use of self-report data for both cannabis use and
limitations. In particular, it remains unclear to what extent psychotic experiences brings with it an additional layer of
cannabis plays a causal role in the results presented here. limitations. Self-reporting delusions requires a degree of
The causality of the relationship between cannabis and full- self-awareness that may be absent in some individuals42,43
blown psychotic disorders has long been controversial: Some people may be more prone than others to conceal
pleiotropy, reverse causality, bias, and confounding have or under-report both cannabis use and psychotic experi-
been proposed as alternative explanations1,4,34. Causal ences, for instance due to stigma or the illegal status of
inference studies of genetic variants associated with can- recreational cannabis in the UK; encouragingly, self-
nabis use and schizophrenia have found reduced35,36 or reported cannabis use has been shown to correlate rea-
nonsignificant37 effects of cannabis use on schizophrenia, sonably well with measurements of cannabinoids in the
while suggesting a reverse causal effect35,37. In other words, hair and urine in the UK44. Although the UK Biobank’s
an alternative explanation is that genetic predisposition to Mental Health Questionnaire asks at what age individuals
schizophrenia may lead individuals to use cannabis, perhaps last used cannabis, it does not ask at what age they first
due to dysfunction in reward circuitry induced by these used cannabis and this lack of data leaves open the pos-
genetic factors38 or as a means of self-medicating to reduce sibility that some individuals only started using cannabis
negative symptoms, anxiety, or insomnia39–41. after their first psychotic experience. Further, it is not
Wainberg et al. Translational Psychiatry (2021)11:211 Page 8 of 9

clear whether the reported psychotic symptoms occurred Publisher’s note

during cannabis use or months or years afterwards. The Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
lack of data on recreational use of other drugs implicated
in psychosis or psychotic experiences, such as psyche- Received: 18 January 2021 Revised: 25 February 2021 Accepted: 19 March
delics (e.g., lysergic acid diethylamide, psilocybin), dis- 2021
sociatives (e.g., ketamine), entactogens (e.g., ecstasy), and
stimulants (e.g., methamphetamine), may further con-
found the results. Participants did not self-report the
potency of cannabis consumed45 nor its content of can- References
1. Gage, S. H., Hickman, M. & Zammit, S. Association between cannabis and
nabidiol, which has antipsychotic properties46–48 and
psychosis: epidemiologic evidence. Biol. Psychiatry 79, 549–556 (2016).
(although evidence is mixed49) may modulate the putative 2. Bagot, K. S., Milin, R. & Kaminer, Y. Adolescent initiation of cannabis use and
causal effects of cannabis on psychosis50. Also, because early-onset psychosis. Subst. Abus. 36, 524–533 (2015).
3. Large, M., Sharma, S., Compton, M. T., Slade, T. & Nielssen, O. Cannabis use and
cannabis-use frequency was ascertained as the maximum
earlier onset of psychosis: a systematic meta-analysis. Arch. Gen. Psychiatry 68,
frequency of ever taking cannabis, participants did not 555–561 (2011).
report variations in cannabis use frequency over time, 4. Gillespie, N. A. et al. High-potency cannabis and incident psychosis: correcting
the causal assumption. Lancet Psychiatry 6, 464 (2019).
which may further complicate interpretation of the
5. D’Souza, D. C. et al. The psychotomimetic effects of intravenous delta-9-
results. Finally, the four psychotic experiences ascertained tetrahydrocannabinol in healthy individuals: implications for psychosis. Neu-
in the UK Biobank Mental Health Questionnaire do not ropsychopharmacology 29, 1558–1572 (2004).
6. Nesvåg, R. et al. Genetic and environmental contributions to the association
represent the full range of possible positive symptoms—
between cannabis use and psychotic-like experiences in young adult twins.
let alone negative symptoms, which participants at clinical Schizophr. Bull. sbw101, https://doi.org/10.1093/schbul/sbw101 (2016).
high risk for psychosis may in any case be less aware of51 7. Karcher, N. R. et al. Genetic predisposition vs individual-specific processes in
the association between psychotic-like experiences and cannabis use. JAMA
and therefore less likely to self-report.
Psychiatry 76, 87 (2019).
Overall, we find that self-reported cannabis use is dose- 8. Vaissiere, J., Thorp, J. G., Ong, J.-S., Ortega-Alonso, A. & Derks, E. M. Exploring
dependently associated with self-reported psychotic phenotypic and genetic overlap between cannabis use and schizotypy -
Corrigendum. Twin Res. Hum. Genet. 23, 306 (2020).
experience frequency. Cannabis users’ psychotic experi-
9. McGrath, J. J. et al. Psychotic experiences in the general population: a cross-
ences tend to be earlier-onset and more distressing than national analysis based on 31,261 respondents from 18 countries. JAMA Psy-
non-users, but not more likely to lead to help-seeking. For chiatry 72, 697–705 (2015).
10. Legge, S. E. et al. Association of genetic liability to psychotic experiences with
the first time, we show that genetic predisposition to
neuropsychotic disorders and traits. JAMA Psychiatry, https://doi.org/10.1001/
schizophrenia strongly modulates the association of can- jamapsychiatry.2019.2508 (2019).
nabis with most types of psychotic experiences. At a time 11. Zammit, S. et al. Psychotic experiences and psychotic disorders at age 18 in
relation to psychotic experiences at age 12 in a longitudinal population-based
when the spread of laws legalizing cannabis for medicinal
cohort study. Am. J. Psychiatry 170, 742–750 (2013).
or recreational use has been accompanied by more 12. Fisher, H. L. et al. Specificity of childhood psychotic symptoms for predicting
relaxed attitudes towards cannabis, our results support schizophrenia by 38 years of age: a birth cohort study. Psychol. Med. 43,
2077–2086 (2013).
the notion that genetics may make some cannabis users
13. Marconi, A., Di Forti, M., Lewis, C. M., Murray, R. M. & Vassos, E. Meta-analysis of
more prone to certain psychotic experiences than others, the association between the level of cannabis use and risk of psychosis.
which could enable targeted harm reduction interventions Schizophr. Bull. 42, 1262–1269 (2016).
14. Keeler, M. H., Ewing, J. A. & Rouse, B. A. Hallucinogenic effects of marijuana as
focused on protecting those at the highest risk.
currently used. Am. J. Psychiatry 128, 213–216 (1971).
15. Barrett, F. S., Schlienz, N. J., Lembeck, N., Waqas, M. & Vandrey, R. ‘Hallucinations’
Acknowledgements following acute cannabis dosing: a case report and comparison to other
M.W. and S.J.T. acknowledge support from the Kavli, Krembil, and McLaughlin hallucinogenic drugs. Cannabis Cannabinoid Res. 3, 85–93 (2018).
Foundations, CAMH Discovery Fund, NSERC (RGPIN-2020-05834 and DGECR- 16. Sherif, M., Radhakrishnan, R., D’Souza, D. C. & Ranganathan, M. Human
2020-00048), and CIHR (NGN-171423). G.R.J. acknowledges support from an laboratory studies on cannabinoids and psychosis. Biol. Psychiatry 79, 526–538
Ontario Graduate Scholarship and the Ontario Student Opportunity Trust Fund. (2016).
M.d.F. acknowledges support from MRC (MR/T007818/1). We acknowledge 17. Freeman, D. et al. Persecutory ideation and a history of cannabis use. Schi-
Romina Mizrahi, Andreea Diaconescu, and Jibran Khokhar for helpful feedback. zophr. Res. 148, 122–125 (2013).
This research was conducted under the auspices of UK Biobank application 18. Freeman, D. et al. How cannabis causes paranoia: using the intravenous
61530, “Multimodal subtyping of mental illness across the adult lifespan administration of Δ9-tetrahydrocannabinol (THC) to identify key cognitive
through integration of multi-scale whole-person phenotypes”. mechanisms leading to paranoia. Schizophr. Bull. 41, 391–399 (2015).
19. Bechtold, J., Hipwell, A., Lewis, D. A., Loeber, R. & Pardini, D. Concurrent and
Author details sustained cumulative effects of adolescent marijuana use on subclinical psy-
1
Centre for Addiction and Mental Health, Toronto, ON, Canada. 2Institute of chotic symptoms. Am. J. Psychiatry 173, 781–789 (2016).
Medical Sciences, University of Toronto, Toronto, ON, Canada. 3Institute of 20. French, L. et al. Early cannabis use, polygenic risk score for schizophrenia and
Psychiatry, Psychology and Neuroscience, King’s College London, London, UK. brain maturation in adolescence. JAMA Psychiatry 72, 1002–1011 (2015).
4
South London and Maudsley NHS Mental Health Foundation Trust, London, 21. Henquet, C., Di Forti, M., Morrison, P., Kuepper, R. & Murray, R. M. Gene-
UK. 5Department of Psychiatry, University of Toronto, Toronto, ON, Canada. environment interplay between cannabis and psychosis. Schizophr. Bull. 34,
6
Department of Physiology, University of Toronto, Toronto, ON, Canada 1111–1121 (2008).
22. van Winkel, R. & Kuepper, R. Epidemiological, neurobiological, and genetic
Conflict of interest clues to the mechanisms linking cannabis use to risk for nonaffective psy-
The authors declare no competing interests. chosis. Annu. Rev. Clin. Psychol. 10, 767–791 (2014).
Wainberg et al. Translational Psychiatry (2021)11:211 Page 9 of 9

23. Davis, K. A. S. et al. Mental health in UK Biobank – development, imple- 37. Pasman, J. A. et al. GWAS of lifetime cannabis use reveals new risk loci, genetic
mentation and results from an online questionnaire completed by 157 366 overlap with psychiatric traits, and a causal effect of schizophrenia liability. Nat.
participants: a reanalysis. BJPsych Open 6 (2020). Neurosci. 21, 1161–1170 (2018).
24. Tanigawa, Y. et al. Components of genetic associations across 2,138 pheno- 38. Khokhar, J. Y., Dwiel, L. L., Henricks, A. M., Doucette, W. T. & Green, A. I. The link
types in the UK Biobank highlight adipocyte biology. Nat. Commun. 10, 4064 between schizophrenia and substance use disorder: a unifying hypothesis.
(2019). Schizophr. Res. 194, 78–85 (2018).
25. Alloza, C. et al. Psychotic-like experiences, polygenic risk scores for schizo- 39. Compton, M. T., Furman, A. C. & Kaslow, N. J. Lower negative symptom scores
phrenia, and structural properties of the salience, default mode, and central- among cannabis-dependent patients with schizophrenia-spectrum disorders:
executive networks in healthy participants from UK Biobank. Transl. Psychiatry preliminary evidence from an African American first-episode sample. Schi-
10, 122 (2020). zophr. Res. 71, 61–64 (2004).
26. Pardiñas, A. F. et al. Common schizophrenia alleles are enriched in mutation- 40. Schofield, D. Reasons for cannabis use in psychosis. Aust. N. Z. J. Psychiatry 40,
intolerant genes and in regions under strong background selection. Nat. 570–574 (2006).
Genet. 50, 381–389 (2018). 41. Gill, K. E. et al. Reasons for cannabis use among youths at ultra high risk for
27. Keller, M. C. Gene × environment interaction studies have not properly con- psychosis. Early Interv. Psychiatry 9, 207–210 (2015).
trolled for potential confounders: the problem and the (simple) solution. Biol. 42. Gilleen, J., Greenwood, K. & David, A. S. Domains of awareness in schizo-
Psychiatry 75, 18–24 (2014). phrenia. Schizophr. Bull. 37, 61–72 (2011).
28. Di Forti, M. et al. Daily use, especially of high-potency cannabis, drives the 43. Bruno, N., Sachs, N., Demily, C., Franck, N. & Pacherie, E. Delusions and
earlier onset of psychosis in cannabis users. Schizophr. Bull. 40, 1509–1517 metacognition in patients with schizophrenia. Cogn. Neuropsychiatry 17, 1–18
(2014). (2012).
29. Fletcher, P. C. & Frith, C. D. Perceiving is believing: a Bayesian approach to 44. Curran, H. V. et al. Which biological and self-report measures of cannabis use
explaining the positive symptoms of schizophrenia. Nat. Rev. Neurosci. 10, predict cannabis dependency and acute psychotic-like effects? Psychol. Med.
48–58 (2009). 49, 1574–1580 (2019).
30. Corlett, P. R., Taylor, J. R., Wang, X.-J., Fletcher, P. C. & Krystal, J. H. Toward a 45. Forti, M. D. et al. High-potency cannabis and the risk of psychosis. Br. J.
neurobiology of delusions. Prog. Neurobiol. 92, 345–369 (2010). Psychiatry 195, 488–491 (2009).
31. Phan, K. L. et al. Cannabinoid modulation of amygdala reactivity to social 46. Davies, C. & Bhattacharyya, S. Cannabidiol as a potential treatment for psy-
signals of threat in humans. J. Neurosci. 28, 2313–2319 (2008). chosis. Ther. Adv. Psychopharmacol. 9, 2045125319881916 (2019).
32. van de Giessen, E. et al. Deficits in striatal dopamine release in cannabis 47. Bhattacharyya, S. et al. Effect of cannabidiol on medial temporal, midbrain, and
dependence. Mol. Psychiatry 22, 68–75 (2017). striatal dysfunction in people at clinical high risk of psychosis: a randomized
33. Cenni, G. et al. Differential effect of cannabinoid agonists and endocannabi- clinical trial. JAMA Psychiatry 75, 1107–1117 (2018).
noids on histamine release from distinct regions of the rat brain. Eur. J. Neu- 48. McGuire, P. et al. Cannabidiol (CBD) as an adjunctive therapy in schizophrenia:
rosci. 24, 1633–1644 (2006). a multicenter randomized controlled trial. Am. J. Psychiatry 175, 225–231
34. Gillespie, N. A. & Kendler, K. S. Use of genetically informed methods to clarify (2018).
the nature of the association between cannabis use and risk for schizo- 49. Hindley, G. et al. Psychiatric symptoms caused by cannabis constituents: a
phrenia. JAMA Psychiatry https://doi.org/10.1001/jamapsychiatry.2020.3564 systematic review and meta-analysis. Lancet Psychiatry 7, 344–353 (2020).
(2020). 50. Schubart, C. D. et al. Cannabis with high cannabidiol content is associated
35. Gage, S. H. et al. Assessing causality in associations between cannabis use and with fewer psychotic experiences. Schizophr. Res. 130, 216–221 (2011).
schizophrenia risk: a two-sample Mendelian randomization study. Psychol. 51. Selten, J. P., Gernaat, H. B., Nolen, W. A., Wiersma, D. & van den Bosch, R. J.
Med. 47, 971–980 (2017). Experience of negative symptoms: comparison of schizophrenic patients to
36. Vaucher, J. et al. Cannabis use and risk of schizophrenia: a Mendelian ran- patients with a depressive disorder and to normal subjects. Am. J. Psychiatry
domization study. Mol. Psychiatry 23, 1287–1292 (2018). 155, 350–354 (1998).