Introduction
Cannabis use is a well-studied risk factor for psychosis, schizophrenia spectrum disorders and psychopathology in general. Meta-analysis and systematic reviews have consistently shown that there is a higher incidence of psychotic outcomes among cannabis users (Moore et al., Reference Moore, Zammit, Lingford-Hughes, Barnes, Jones, Burke and Lewis2007; Semple, McIntosh, & Lawrie, Reference Semple, McIntosh and Lawrie2005) and that this relationship is dose dependent (Marconi, Di Forti, Lewis, Murray, & Vassos, Reference Marconi, Di Forti, Lewis, Murray and Vassos2016). Using cannabis during adolescence further increases risk for psychosis (Kelley et al., Reference Kelley, Wan, Broussard, Crisafio, Cristofaro, Johnson and Compton2016; Mustonen et al., Reference Mustonen, Niemelä, Nordström, Murray, Mäki, Jääskeläinen and Miettunen2018), earlier onset of psychotic symptoms (Galvez-Buccollini et al., Reference Galvez-Buccollini, Proal, Tomaselli, Trachtenberg, Coconcea, Chun and Delisi2012) and worsened prognosis (Manrique-Garcia et al., Reference Manrique-Garcia, Zammit, Dalman, Hemmingsson, Andreasson and Allebeck2014). Although the epidemiological evidence, along with some experimental evidence (D'Souza et al., Reference D'Souza, Perry, MacDougall, Ammerman, Cooper, Wu and Krystal2004), suggests a causal link between cannabis use and psychosis, the nature of this relationship remains the focus of fierce debate (Forti, Morgan, Selten, Lynskey, & Murray, Reference Forti, Morgan, Selten, Lynskey and Murray2019; Sommer & van den Brink, Reference Sommer and van den Brink2019). Generally, three different hypotheses are used to explain the mechanisms of the cannabis–schizophrenia association: (1) the relationship is fully causal, i.e. cannabis use causes schizophrenia, (2) the relationship may be partially confounded by shared genetic and environmental confounders and/or reverse causation and (3) this link is entirely non-causal (Gillespie & Kendler, Reference Gillespie and Kendler2021; Hiemstra et al., Reference Hiemstra, Nelemans, Branje, van Eijk, Hottenga, Vinkers and Boks2018).
Considering that part of the aetiology of cannabis use and psychosis can be explained through heritable processes (Cardno et al., Reference Cardno, Marshall, Coid, Macdonald, Ribchester, Davies and Murray1999; Verweij et al., Reference Verweij, Zietsch, Lynskey, Medland, Neale, Martin and Vink2010), recent large scale genome-wide association studies (GWASs) have demonstrated that multiple single-nucleotide polymorphisms (SNPs) are associated with risk for schizophrenia (Pardiñas et al., Reference Pardiñas, Holmans, Pocklington, Escott-Price, Ripke, Carrera and Walters2018), and predict cannabis use behaviours (Johnson et al., Reference Johnson, Demontis, Thorgeirsson, Walters, Polimanti, Hatoum and Agrawal2020; Pasman et al., Reference Pasman, Verweij, Gerring, Stringer, Sanchez-Roige, Treur and Vink2018). Researchers can summarise the genetic risk for a disease through polygenic risk score (PRS) calculations, derived from the summary statistics generated in these large-scale GWASs. Although most PRSs for psychiatric diseases can currently only account for a small portion of the variance of disease [approximately <10% (Murray et al., Reference Murray, Lin, Austin, McGrath, Hickie and Wray2021)], PRS can inform about shared genetic aetiology among complex traits, and can also be used to estimate the genetic risk to a trait at the individual level (Choi, Mak, & O'Reilly, Reference Choi, Mak and O'Reilly2020). In view of the purported cannabis–psychosis link, researchers have examined the link between polygenic risk score for schizophrenia (PRS-Sz) and cannabis use. PRS-Sz has been consistently associated with varying levels of cannabis use across numerous cohorts (Carey et al., Reference Carey, Agrawal, Bucholz, Hartz, Lynskey, Nelson and Bogdan2016; Guloksuz et al., Reference Guloksuz, Pries, Delespaul, Kenis, Luykx, Lin and van Os2019; Hartz et al., Reference Hartz, Horton, Oehlert, Carey, Agrawal, Bogdan and Bierut2017; Hiemstra et al., Reference Hiemstra, Nelemans, Branje, van Eijk, Hottenga, Vinkers and Boks2018; Jones et al., Reference Jones, Hammerton, McCloud, Hines, Wright, Gage and Zammit2020; Verweij et al., Reference Verweij, Abdellaoui, Nivard, Sainz Cort, Ligthart, Draisma and Vink2017). Consequently, some have concluded that the relationship between PRS-Sz and cannabis use represents a pathway from genetic risk for schizophrenia to cannabis use (Jones et al., Reference Jones, Hammerton, McCloud, Hines, Wright, Gage and Zammit2020), or that sensitivity to exposure to cannabis use is moderated by PRS-Sz (Guloksuz et al., Reference Guloksuz, Pries, Delespaul, Kenis, Luykx, Lin and van Os2019). In contrast, one highly powered study reported that PRS-Sz was not associated with cannabis use disorder in healthy controls, or patients with psychiatric disorders other than schizophrenia (Hjorthøj et al., Reference Hjorthøj, Uddin, Wimberley, Dalsgaard, Hougaard, Børglum and Nordentoft2021). Furthermore, they report that the association between prior cannabis use disorder and later development for schizophrenia was not altered after adjustment for PRS-Sz and PRS of other psychiatric disorders (Hjorthøj et al., Reference Hjorthøj, Uddin, Wimberley, Dalsgaard, Hougaard, Børglum and Nordentoft2021), suggesting that the association between cannabis use and development of schizophrenia is not explained by common genetic vulnerability (Hjorthøj et al., Reference Hjorthøj, Uddin, Wimberley, Dalsgaard, Hougaard, Børglum and Nordentoft2021). Nevertheless, the results of most studies utilising PRS-Sz, along with experiments employing discordant relative designs (Giordano, Ohlsson, Sundquist, Sundquist, & Kendler, Reference Giordano, Ohlsson, Sundquist, Sundquist and Kendler2015) and studies using Mendelian randomisation (MR) techniques (Gage et al., Reference Gage, Jones, Burgess, Bowden, Smith, Zammit and Munafò2017; Pasman et al., Reference Pasman, Verweij, Gerring, Stringer, Sanchez-Roige, Treur and Vink2018) support the second hypothesis – mainly that the relationship between schizophrenia and cannabis use is confounded by shared genetic vulnerability and reverse causation.
The relationship between cannabis use and psychosis development is particularly interesting in the adolescent ‘clinical high-risk for psychosis’ (Fusar-Poli, Reference Fusar-Poli2017) population. These individuals are at a high risk for psychosis in the presence of sub-clinical psychotic symptoms, functional decline and/or genetic risk (Fusar-Poli, Reference Fusar-Poli2017). As such, research on the developmental origins of psychosis risk has focused on the emergence of psychotic-like experiences (PLEs) during the adolescent period and how cannabis might influence such trajectories.
PLEs are highly prevalent sub-clinical psychotic symptoms (Ronald et al., Reference Ronald, Sieradzka, Cardno, Haworth, McGuire and Freeman2014), reported in up to 7% of individuals (Linscott and van Os, Reference Linscott and van Os2013). Similar to the current models of symptomatology in patients along the psychotic spectrum, these sub-clinical symptoms have been further subdivided into various dimensions, such as positive, negative and affective symptoms (van Os & Reininghaus, Reference van Os and Reininghaus2016). Although these sub-clinical experiences are transitory in about 80% of individuals, PLEs are persistent in 20% of individuals (van Os et al., Reference van Os, van der Steen, Islam, Gülöksüz, Rutten and Simons2017). Moreover, the presence of PLEs in community samples is associated with increased odds for any mental disorder [odds ratio (OR) 3.08, 95% confidence interval (95% CI) 2.26–4.21], and psychotic disorders (OR 3.96, 95% CI 2.03–7.73) (Healy et al., Reference Healy, Brannigan, Dooley, Coughlan, Clarke, Kelleher and Cannon2019).
Considering the close relationship of PLEs to psychotic disorders, many have tested the hypothesis that cannabis use also increases one's risk for PLEs (see Ragazzi, Shuhama, Menezes, & Del-Ben, Reference Ragazzi, Shuhama, Menezes and Del-Ben2018 for a systematic review). One study found that cannabis use is significantly associated with the positive PLEs (β = 0.061, p < 1 × 10−4), even after controlling for numerous confounding factors (van Gastel et al., Reference van Gastel, Wigman, Monshouwer, Kahn, van Os, Boks and Vollebergh2012). Another study found that the relationship between PLEs and cannabis use is increased in the heaviest of cannabis consumers (Schubart et al., Reference Schubart, van Gastel, Breetvelt, Beetz, Ophoff, Sommer and Boks2011); in those who spend >€25/week on cannabis (i.e. heaviest users), there was an increased odds for various domains of PLE such as negative symptoms (OR 3.4, 95% CI 2.9–4.1), positive symptoms (OR 3.0, 95% CI 2.4–3.6) and depressive symptoms (OR 2.8, 95% CI 2.3–3.3) (Schubart et al., Reference Schubart, van Gastel, Breetvelt, Beetz, Ophoff, Sommer and Boks2011). Furthermore, cannabis use has also been shown to temporally precede PLE in adolescent cohorts (Bourque, Afzali, & Conrod, Reference Bourque, Afzali and Conrod2018), but PLEs in childhood do not predict cannabis use (Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafò and Zammit2018a). Overall, the study of PLE in cohorts of cannabis users may be an interesting avenue to understand the nature and potential directionality of the cannabis–psychosis relationship.
PRS-Sz are also related to PLE. Although initial studies reported no relationship between PRS-Sz and PLEs (Derks et al., Reference Derks, Vorstman, Ripke, Kahn, Consortium and Ophoff2012; Zammit et al., Reference Zammit, Hamshere, Dwyer, Georgiva, Timpson, Moskvina and O'Donovan2014), more recent studies – with greater power – have found that PRS-Sz is associated with PLEs (Jones et al., Reference Jones, Stergiakouli, Tansey, Hubbard, Heron, Cannon and Zammit2016, Reference Jones, Heron, Hammerton, Stochl, Jones, Cannon and Zammit2018b; Pain et al., Reference Pain, Dudbridge, Cardno, Freeman, Lu, Lundstrom and Ronald2018; Taylor et al., Reference Taylor, Martin, Lu, Brikell, Lundström, Larsson and Lichtenstein2019). But, there remains contradictory evidence in this field. For example, some have reported that PRS-Sz is related to the negative and affective symptom domains (Jones et al., Reference Jones, Stergiakouli, Tansey, Hubbard, Heron, Cannon and Zammit2016; Jones et al., Reference Jones, Heron, Hammerton, Stochl, Jones, Cannon and Zammit2018b), but not positive symptoms (hallucinations, paranoia and thought disturbance), whereas others have reported an association between PRS-Sz and positive symptoms (Pain et al., Reference Pain, Dudbridge, Cardno, Freeman, Lu, Lundstrom and Ronald2018; Taylor et al., Reference Taylor, Martin, Lu, Brikell, Lundström, Larsson and Lichtenstein2019).
Thus, although the relationship between polygenic risk for schizophrenia and cannabis use has been consistently described in the literature, and the link between cannabis use and psychotic-like symptoms is shown to be significant, the relationship between all three factors (polygenic risk, cannabis use and PLE) is not yet fully understood. Although other studies have attempted to find environmental factors that mediate the relationship between PRS-Sz and cannabis use (Jones et al., Reference Jones, Hammerton, McCloud, Hines, Wright, Gage and Zammit2020), to our knowledge no study has examined if cannabis use mediates the relationship between PRS-Sz and PLEs. Thus, considering that PRS-Sz may be directly or indirectly linked to cannabis use, the current study aims to investigate whether or not the pathway from genetic vulnerability to psychosis symptoms, is at least partially mediated by an indirect pathway through cannabis use.
In addition to the mediation hypothesis, we also test a moderation hypothesis, in which cannabis use might exacerbate genetic vulnerability to schizophrenia, and in turn increase the frequency of PLE. Clarifying the moderating role of genetic vulnerability on the relationship between cannabis and psychosis would also help to inform decision making with regards to guidelines for recreational cannabis in which individuals with a certain risk profile could be advised accordingly, in addition to the existing literature. These two hypotheses will be contrasted against a null hypothesis, which postulates that despite any potential common genetic vulnerability to cannabis use and psychosis risk, the relationship between cannabis use and psychosis risk holds, and is independent of (or in addition to) a common genetic vulnerability (i.e. cannot be explained by common genetic vulnerability). To test all hypotheses, we use a developmentally informed approach that focuses on temporal precedence to confirm mediation between variables. The current study uses data from two independent European cohorts: we use data from the IMAGEN (Schumann et al., Reference Schumann, Loth, Banaschewski, Barbot, Barker and Büchel2010) study, a longitudinal study of over 2000 European adolescents, as a discovery sample, and aim to replicate those results in an independent European sample, the Utrecht cannabis cohort (Schubart et al., Reference Schubart, van Gastel, Breetvelt, Beetz, Ophoff, Sommer and Boks2011). The use of the IMAGEN cohort is ideal considering that it allows for a longitudinal view of cannabis use and PLE development, during the critical years of adolescence. Furthermore, this cohort is relatively well powered to detect mediation effects, as similarly sized cohorts have attempted to discern such effects using similar phenotypes (Jones et al., Reference Jones, Hammerton, McCloud, Hines, Wright, Gage and Zammit2020). This is compared with the cross-sectional Utrecht cannabis cohort, which is a cohort that has been enriched for PLE and heavy cannabis use; heavy cannabis use being a particularly strong risk factor for development of psychosis and PLE.
Methods
Participants
IMAGEN sample
The IMAGEN study is a longitudinal imaging genetics study of over 2000 healthy adolescents, mostly of European descent. Detailed descriptions of this study, genotyping procedures and data collection have previously been published (Schumann et al., Reference Schumann, Loth, Banaschewski, Barbot, Barker and Büchel2010). The current study uses data for the 2087 who contributed their genetic data. The multicentric IMAGEN project had obtained ethical approval by the local ethics committees (at their respective sites) and written informed consent from all participants and their legal guardians. The parents and adolescents provided written informed consent and assent, respectively at 14 and 16, and then participants gave full consent at 18 and 21 years of age.
Utrecht cannabis cohort
Data from the Utrecht cannabis cohort come from a subset (N = 1223) of a large (N = 17 698) cohort of young Dutch participants, for which genetic, cannabis use and PLE data were available. Detailed descriptions of recruitment methods, genotyping procedures and data collection were previously published (Boks et al., Reference Boks, He, Schubart, van Gastel, Elkrief, Huguet and de Witte2020; Schubart et al., Reference Schubart, van Gastel, Breetvelt, Beetz, Ophoff, Sommer and Boks2011). Participants gave online informed consent, and the study received approval by the University Medical Centre Utrecht medical ethical commission. Of note is the enrichment for the extremes in PLE and cannabis use data in the Utrecht cannabis cohort. To increase power for gene × environment interactions in previous studies (Boks et al., Reference Boks, Schipper, Schubart, Sommer, Kahn and Ophoff2007), data from individuals from the general population were combined with data of participants selected from the top or bottom quintile of total PLE scores, who are either non-users (<2 lifetime exposures to cannabis) or heavy users (i.e. current expenditure for personal cannabis use exceeded €10 weekly).
Phenotype measures
Cannabis use measures
IMAGEN participants were repeatedly assessed for cannabis use at 14, 16, 18 and 21 years of age using questions taken from the European School Survey of Alcohol and other Drugs (ESPAD) questionnaire. The ESPAD is a self-report questionnaire that measures the use of various drugs of abuse, including cannabis (Hibell et al., Reference Hibell, Andersson, Bjarnason, Kokkevi, Morgan, Narusk and Ahlström1997, Reference Hibell, Andersson, Bjarnason, Ahlström, Balakireva, Kokkevi and Morgan2004). With very few participants reporting cannabis use at 14 years of age, we focus our analyses on data that were collected at the 16-year-old assessment, using responses to the question ‘On how many occasions in your whole lifetime have you used marijuana (grass, pot) or hashish (hash, hash oil)?’. Answers are scored on a scale ranging from 0 to 6: ‘0’ = 0, ‘1–2 times’ = 1, ‘3–5 times’ = 2, ‘6–9 times’ = 3, ‘10–19 times’ = 4, ‘20–39 times’ = 5, ‘40 or more times’ = 6. In the Utrecht cannabis cohort, lifetime cannabis use data were reported according to the following categories: never = 0, ‘1 time’ = 1, ‘2 times’ = 2, ‘5–9 times’ = 3, ‘>10 times’ = 4.
PLE measures
PLE data for both cohorts were drawn from the Community Assessment of Psychic Experiences-42 (CAPE-42) questionnaire (Stefanis et al., Reference Stefanis, Hanssen, Smirnis, Avramopoulos, Evdokimidis, Stefanis and Van Os2002). CAPE-42 is a widely used self-report questionnaire that reliably measures lifetime PLEs (Mark & Toulopoulou, Reference Mark and Toulopoulou2016). The CAPE-42 has three subscales that measure positive, negative and depressive symptom dimensions. The CAPE-42 measures frequency of symptoms, along with distress caused by symptoms. We only analyse frequency scores as distress and frequency scores are highly correlated in these cohorts (r > 0.80). In the primary analyses, we use the sum total of frequency scores, whereas we look at the various sub-dimensions in the secondary analysis. Due to the skewed distribution of scores, the log-transformed sum score of each individual dimension and total score of the frequency of symptoms was used. We used CAPE-42 data from the 18-year-old follow-up for the IMAGEN cohort.
Genetic data
IMAGEN
The genotyping was conducted using the Illumina Quad 610 chip and 660Wq at the ‘Centre National de Genotypage’ (Paris, France). Non-imputed autosomal SNPs are used for this study (498 892 SNPs). Following all quality control steps, genetic data (468 170 SNPs) remained for 1740 individuals. Baseline quality control steps and principal component analysis to control for ancestry are described in online Supplementary materials.
Utrecht
The genotyping in this cohort was conducted using either the Illumina® HumanOmniExpress (733 202 SNPs; 576 individuals) or the Illumina® Human610-Quad Beadchip (620 901 SNPs; 768 individuals). The CannabisQuest cohort genetic dataset was also imputed, as described in Boks et al. (Reference Boks, He, Schubart, van Gastel, Elkrief, Huguet and de Witte2020), using the HapMap III release 24 via Beagle 5.2 imputation server (Browning and Browning, Reference Browning and Browning2009). As with the IMAGEN sample, quality control steps and principal component analysis for ancestry are described in online Supplementary materials. After all quality control steps, a total of 5 173 601 SNPs and 1126 individuals remained for analysis.
Analysis
Polygenic risk scores
PRS-Sz were constructed for each of the IMAGEN and Utrecht cannabis individuals, who passed genetic quality control. PRS-Sz were built using data from the most recent schizophrenia GWASs based on 40 675 cases and 64 643 controls (Pardiñas et al., Reference Pardiñas, Holmans, Pocklington, Escott-Price, Ripke, Carrera and Walters2018) as a training set (for description of base set, see online Supplementary materials). PRSs were built using PRScs (Ge, Chen, Ni, Feng, & Smoller, 2019) and PLINK1.9 (Purcell). PRScs is used to infer posterior SNP effect sizes, by placing a continuous shrinkage prior on SNP effect sizes reported in the most recent schizophrenia GWASs (Ge et al., Reference Ge, Chen, Ni, Feng and Smoller2019), as well as an external LD reference panel. Here, we use the publically available 1000 Genomes Project phase 3 panel (https://github.com/getian107/PRScs). To calculate posterior effect sizes, in both cohorts, we use the default settings of PRScs, described in more detail in online Supplementary materials. After calculation of posterior effect sizes, PRSs were calculated using the ‘--score’ function and SUM modifier in PLINK1.9. After quality control, 321 567 variants are used to calculate PRS in the IMAGEN cohort, and 763 754 SNPs in the Utrecht cannabis cohort.
We aligned our analyses closely to the replication study, using the same protocol to create PRS, and, in both IMAGEN and Utrecht cannabis cohorts. To ease interpretability of results, we scale the PRS, using the scale function in R (R Core Team, 2020). Using the same methods, we created a PRS for cannabis use (PRS-Can), using publicly available data from the GWAS studying lifetime cannabis use (Pasman et al., Reference Pasman, Verweij, Gerring, Stringer, Sanchez-Roige, Treur and Vink2018) (a detailed description of the base set can be found in online Supplementary materials), to be used as a potential confounder.
Statistical analysis
Multiple multinomial linear regressions were used to assess the relationships between PRS-Sz, cannabis use and PLEs in both cohorts. For our primary analyses, we examine four distinct models:
Model 1: The relationship between PRS-Sz, independent variable (IV), and total CAPE score (log-transformed), dependent variable (DV).
Model 2: The relationship between PRS-Sz (IV) and lifetime cannabis use (DV).
Model 3: The relationship between cannabis use (IV) and CAPE scores (DV), when accounting for PRS-SZ.
Model 4: The interaction between PRS-Sz and cannabis use as predictors of CAPE scores (moderation analysis).
We performed mediation analysis using maximum likelihood estimation (MLE) path analysis to assess the effect of PRS-Sz on PLE scores through the possible mediation effect of cannabis use. As alluded to above, our hypothesis is that cannabis use (M) mediates the relationship between PRS-Sz (independent variable; IV) and PLE (dependent variable; DV). We report the bias-corrected bootstrap 95% CI for the indirect effect, using an adjusted bootstrap percentile method (BCa), based on 5000 bootstrap samples. We use the MLE to handle missing data. All statistical analyses are performed in R (R Core Team, 2020). Mediation analysis was performed via the ‘lavaan’ (Rosseel et al., Reference Rosseel, Jorgensen, Rockwood, Oberski, Byrnes, Vanbrabant and Du2021) and ‘tidySEM’ packages (van Lissa, Reference van Lissa2021). An α = 0.05 was set for significance. Mediation analysis is only executed using the IMAGEN data as this dataset is the only sample that assessed cannabis use some years before the PLE assessment and therefore the only dataset that can provide a true estimate of a longitudinal relationship.
The association between PRS-Sz, cannabis use and the various sub-domains of the CAPE-42 questionnaire was analysed in secondary analyses, through linear regression, in both cohorts. For all statistical analyses of the IMAGEN cohort, we consider the following potential confounders: sex, the first-six genetic principal components (PCs). In the Utrecht cannabis cohort, we add age as a potential confounder. Finally, as a sensitivity analysis we include PRS-Can as a potential confounder in regression analyses for the IMAGEN cohort.
Results
Sample characteristics
Characteristics of participants who passed genetic QC and responded to cannabis use and PLE questionnaire data are detailed in online Supplementary Table S2. Data from a total of 1740 individuals were used to calculate PRS-Sz in IMAGEN, and 1223 individuals in Utrecht cannabis cohort. In both IMAGEN and Utrecht cannabis cohort samples, males report higher cannabis use compared to females (p < 0.001). The total frequency of CAPE-42 symptoms reported is significantly greater in males (p < 0.001) in the IMAGEN cohort. There was no difference in the reported total CAPE-42 symptoms between males and females in the Utrecht cannabis cohort (online Supplementary Table S2). Female participants in both cohorts report significantly higher scores in the depression symptom sub-scale of the CAPE-42 (p < 0.001). Finally, the mean age of the Utrecht cannabis cohort is 20.5 years.
Regression models
Model 1: association of PRS-Sz with PLE
PRS-Sz predicted PLE in both cohorts (Table 1), when accounting for covariates. PRS-Sz was significantly associated with CAPE-42 scores in both cohorts (β IMAGEN = 0.015 p = 0.004, R 2 = 0.019; β Utrecht cannabis = 0.021, p = 0.0003, R 2 = 0.016). The results for the full regression model, including covariates, are shown in online Supplementary Table S3.
β, main effect size; std. error, standard error. Bold values indicate p < 0.05.
This table shows results of the effects of the independent variables in linear regression models. Dependent variable is log-transformed CAPE-42 scores for models 1-3-4, whereas dependent variable for model 2 is lifetime cannabis use. We considered the first six PC and sex as covariates for all analyses and age is included for all analyses of the Utrecht cannabis cohort.
Model 2: association of PRS-Sz with cannabis use
After accounting for covariates, the PRS-Sz predicted cannabis use in both cohorts (β IMAGEN = 0.097, p = 0.027, R 2 = 0.041; β Utrecht cannabis = 0.24, p < 0.00001, R 2 = 0.17, Table 1). The results for the regression model, including covariates, are shown in online Supplementary Table S4.
Model 3: association of cannabis use with PLE
Cannabis use significantly predicted PLE in both cohorts (β IMAGEN = 0.0091, p = 0.007, R 2 = 0.028; β Utrecht cannabis = 0.017, p < 0.00001, R 2 = 0.037; Table 1), when considering PRS-Sz and all confounders. Moreover, PRS-Sz remained as a significant predictor (p < 0.05) within this model in both cohorts (online Supplementary Table S5).
Model 4: moderation analysis
In our moderation model, the interaction between cannabis use and PRS-Sz was also not significant (p > 0.05; online Supplementary Table S6), suggesting that both cannabis use and PRS-Sz independently predict PLE.
Mediation analysis
Although PRS-Sz predicted PLE at 18 years of age (Fig. 1, path c), and lifetime cannabis use at 16 years of age (Fig. 1, path a), and that cannabis use (16 years) was significantly associated with PLE (18 years) (Fig. 1, path b), there was no evidence that PRS-Sz influences PLE through previous cannabis use (β = 8.79 × 10−4, 95% CI −1.23 × 10−4 to 1.88 × 10−3, p = 0.08; Table 2). For full results of path analysis, including covariates, see online Supplementary Table S7. Taken together, these results suggest that both cannabis use and PRS-Sz independently predict PLE.
β, main effect size; s.e., standard error. Bold values indicate p < 0.05.
We report the results of the path analysis examining the link between PRS-Sz and PLE, and whether there is a significant indirect effect (mediation) through cannabis use. Here, we report, estimates (β), s.e., z statistics and p value. We considered the first six PC and sex as covariates for all analyses. CIs are calculated adjusted bootstrap percentile method (BCa) based on 5000 bootstrap samples.
Secondary analysis
The relationship PRS-Sz, cannabis use and the different sub-domains in the CAPE-42 questionnaire was studied. In the IMAGEN cohort (Table 3; online Supplementary Table S8), PRS-Sz was significantly associated with the depression subscale (β IMAGEN = 0.018, p = 0.01) and positive subscale (β IMAGEN = 0.011, p = 0.02). Moreover, cannabis use was predictive of the depressive subscale (β IMAGEN = 0.01, p = 0.02) and negative symptoms subscale (β IMAGEN = 0.013, p = 0.003). Path analysis was also performed, for the IMAGEN cohort, to detect mediation effects of cannabis onto the different subscales, but it was not significant (p < 0.05; online Supplementary Table S10). On the contrary, in the Utrecht cannabis cohort, cannabis use was significantly associated with all three sub-domains (p < 0.005, Table 3, online Supplementary Table S9), whereas PRS-Sz was significantly associated with the depressive and negative subscales (p < 0.005; online Supplementary Table S9)
β, main effect size; std. error, standard error. Bold values indicate p < 0.05.
This table shows results of the effects of the independent variables in linear regression models on the various sub-scales of the CAPE-42. Dependent variable is log-transformed CAPE-42 scores for each subscale. We considered the first six PC and sex as covariates for all analyses and age is included for all analyses of the Utrecht cannabis cohort.
Sensitivity analysis
In the IMAGEN cohort, the three significant linear models (models 1–3) were reassessed considering PRS-Can as a potential confounder. The lifetime cannabis use PRS (PRS-Can) did not predict cannabis use measures at 16 years of age, or CAPE scores at 18 years of age (p > 0.5). After including PRS-Cannabis a covariable in model 3, both PRS-Sz and cannabis use remained significant (β PRS-Sz = 0.014, p = 0.01, β Cannabis Use = 0.009, p = 0.008). Moreover, PRS-Sz significantly predicted cannabis use (model 2), after inclusion of PRS-Can into the regression (β = 0.095, p = 0.03, online Supplementary Table S11).
Discussion
In this study, we examine whether polygenic risk for schizophrenia predicts cannabis use, and higher levels of PLEs, in two independent European ancestry cohorts. Furthermore, we explore potential hypotheses through mediation and moderation analyses. Our results demonstrate that cannabis use can be reliably predicted by PRS-Sz, strengthening the existing literature (Carey et al., Reference Carey, Agrawal, Bucholz, Hartz, Lynskey, Nelson and Bogdan2016; Hartz et al., Reference Hartz, Horton, Oehlert, Carey, Agrawal, Bogdan and Bierut2017; Hiemstra et al., Reference Hiemstra, Nelemans, Branje, van Eijk, Hottenga, Vinkers and Boks2018; Jones et al., Reference Jones, Hammerton, McCloud, Hines, Wright, Gage and Zammit2020). The evidence of association between PRS-Sz and cannabis use in the IMAGEN cohort has previously demonstrated by French et al., that cannabis use at 14 years of age interacted with PRS-Sz in decreasing cortical thickness from 14.5 to 18.5 years old (French et al., Reference French, Gray, Leonard, Perron, Pike, Richer and Paus2015). Here, we extend these findings by showing that the PRS-Sz predicts PLE. This too is in line with other study, using a variety in PLE assessments in various sub-domains (Jones et al., Reference Jones, Stergiakouli, Tansey, Hubbard, Heron, Cannon and Zammit2016; Jones et al., Reference Jones, Heron, Hammerton, Stochl, Jones, Cannon and Zammit2018b; Pain et al., Reference Pain, Dudbridge, Cardno, Freeman, Lu, Lundstrom and Ronald2018; Taylor et al., Reference Taylor, Martin, Lu, Brikell, Lundström, Larsson and Lichtenstein2019). The current study confirms that PRS-Sz and cannabis use are linked to risk for PLEs overall and in the depressive and domains in both samples.
Considering the abundance of observational evidence showing temporal precedence of cannabis use in risk for psychosis, a reasonable alternative to a causal hypothesis is the proposal that cannabis use and PLE are explained through common genetic risk. However, our findings do not confirm this explanation, despite showing that PRS-Sz is correlated with both PLE and cannabis outcomes in both cohorts. This is in line with recent study that reported that various classes of cannabis use were associated with increased risk for psychotic experiences, even after adjusting for family history of schizophrenia (Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafò and Zammit2018a) and other study adjusting for PRS-Sz (Jones et al., Reference Jones, Hammerton, McCloud, Hines, Wright, Gage and Zammit2020). To our knowledge, this is the first study to examine if cannabis use mediates the relationship between PRS-Sz and PLE. Through our longitudinal design, our analyses did not find any evidence to support mediation nor moderation hypotheses that explain the relationship between lifetime cannabis use and PLE. Consequently, these null findings suggest that despite the common genetic vulnerability of psychotic experiences, cannabis use and schizophrenia (Barkhuizen, Pain, Dudbridge, & Ronald, Reference Barkhuizen, Pain, Dudbridge and Ronald2020; Pasman et al., Reference Pasman, Verweij, Gerring, Stringer, Sanchez-Roige, Treur and Vink2018), both PRS-Sz and cannabis use independently increase one's risk for PLE, leaving room for alternative explanations of the cannabis–psychosis relationship.
Two recent studies employed MR technique to investigate causal links between cannabis use and schizophrenia (Gage et al., Reference Gage, Jones, Burgess, Bowden, Smith, Zammit and Munafò2017; Vaucher et al., Reference Vaucher, Keating, Lasserre, Gan, Lyall, Ward and Holmes2018). In both of these studies, there was weak evidence to support the causal hypothesis in the direction schizophrenia to cannabis use, while the reverse relationship was strong (Gage et al., Reference Gage, Jones, Burgess, Bowden, Smith, Zammit and Munafò2017; Vaucher et al., Reference Vaucher, Keating, Lasserre, Gan, Lyall, Ward and Holmes2018). Although these studies are limited by the power of the respective GWASs used, recent study has called into question causal inferences made in MR studies of complex traits (O'Connor & Price, Reference O'Connor and Price2018), and suggest the use of a latent causal variable (LCV) instead. In LCV models, genetic correlation between ‘two traits is mediated by a latent variable which has a causal effect on each trait’ (O'Connor & Price, Reference O'Connor and Price2018). Accordingly, a recent study examined the causal link between schizophrenia and lifetime cannabis use employing LCV and found no evidence for a causal genetic link between the two (Jang et al., Reference Jang, Saunders, Liu, Jiang, Liu and Vrieze2020). Taken together, these reports do not preclude the possibility of a causal mechanism linking cannabis use to psychosis. Instead, they – along with the results presented above – suggest that psychosis or psychosis risk, and cannabis use may be linked through another environmental mediator rather than being linked through a common genetic predisposition.
The findings of the current study suggest that the variance in cannabis use that is most linked to PLEs is that which is not accounted for by PRS-Sz. This is interesting and suggests that future studies could focus on environmental factors influencing cannabis behaviours, such as the type of cannabis used, or available in a given population, the effects of advertisements endorsed by the cannabis industry, differing legalisation frameworks, and cannabis potency, when attempting to understand the link between cannabis and psychosis.
In secondary analysis of the current study, PRS-Sz was associated with depressive and positive sub-domains of the CAPE-42 in the IMAGEN cohort, whereas in the older Utrecht cannabis cohort, PRS-Sz was associated with depressive and negative sub-domains, with results trending towards significance for the positive sub-domain (p = 0.058). Some previous reports found no association between PLE-Sz and positive symptoms in adolescent populations (Jones et al., Reference Jones, Stergiakouli, Tansey, Hubbard, Heron, Cannon and Zammit2016, Reference Jones, Heron, Hammerton, Stochl, Jones, Cannon and Zammit2018b; Zammit et al., Reference Zammit, Hamshere, Dwyer, Georgiva, Timpson, Moskvina and O'Donovan2014). Nevertheless, one study has reported an association between PRS-Sz and positive PLE symptoms in their adolescent cohort (Pain et al., Reference Pain, Dudbridge, Cardno, Freeman, Lu, Lundstrom and Ronald2018), however, only when considering non-zero responders, i.e. those who have already manifested positive symptoms. Some have argued that the previously reported associations between PLE-Sz and the positive symptom domain in young adult populations can be explained by the fact that the genetic overlap between positive and negative psychotic experiences and schizophrenia might be stronger in adulthood than in adolescence (Barkhuizen et al., Reference Barkhuizen, Pain, Dudbridge and Ronald2020). As previously suggested by Jones et al. (Reference Jones, Stergiakouli, Tansey, Hubbard, Heron, Cannon and Zammit2016), this would imply that genetic risk for schizophrenia is in fact associated with positive PLE, but that this risk may be expressed in young adulthood rather than adolescence. Our results seem to be in line with this interpretation, considering that PLE in the IMAGEN cohort are reported at 18, i.e. the beginning of young adulthood. On the contrary, other environmental risk factors – such as cannabis use – may be what cause these same positive symptoms in adolescents.
Limitations
Although the findings of this study are consistent with previous independent studies, this study is not without limitations and results should be interpreted accordingly. First, PRSs can only explain small portions of the variances of the phenotypes they study (Murray et al., Reference Murray, Lin, Austin, McGrath, Hickie and Wray2021). In this study, PRS-Sz explained up to 17% of the variance of cannabis use and 1.9% of variance of PLE, when accounting for confounders. PRS also only incorporates data from common genetic variants; as such a significant portion of the genetic effects may not be captured through the PRS, such as the effects of rare variants and copy number variants, which also may play a role in the pathogenesis of schizophrenia (Malhotra & Sebat, Reference Malhotra and Sebat2012). Although a thorough imputation procedure was implemented to deal with missing data (Karahalios, Baglietto, Carlin, English, & Simpson, Reference Karahalios, Baglietto, Carlin, English and Simpson2012), the IMAGEN dataset had several missing data points, which could bias our results. Next, considering the self-report nature of our phenotypic measures, our results may be at risk for measurement error, due to underreporting of symptoms, leading to weakened power. Moreover, we use the PRS-Sz – which was built to predict outcomes of clinical schizophrenia in adults – to predict PLEs in adolescent and young adult populations. Although the PRS-Sz has been used to reliably predict PLEs (Jones et al., Reference Jones, Stergiakouli, Tansey, Hubbard, Heron, Cannon and Zammit2016, Reference Jones, Heron, Hammerton, Stochl, Jones, Cannon and Zammit2018b; Pain et al., Reference Pain, Dudbridge, Cardno, Freeman, Lu, Lundstrom and Ronald2018; Taylor et al., Reference Taylor, Martin, Lu, Brikell, Lundström, Larsson and Lichtenstein2019), the most discriminant SNPs for PLE may have not been captured by our PRS. However, considering the genetic overlap between schizophrenia and PLE (Barkhuizen et al., Reference Barkhuizen, Pain, Dudbridge and Ronald2020), our significant result remains informative. Although our results consistently show that PRS-Sz is associated with lifetime cannabis use, we cannot make any conclusions about the effects of recent or current cannabis use. This is an important limitation considering the recent literature which demonstrates that current cannabis use is significantly associated with psychotic experiences in the general population, but that lifetime cannabis use is not (Quattrone et al., Reference Quattrone, Ferraro, Tripoli, La Cascia, Quigley, Quattrone and Di Forti2020). Moreover, our lifetime cannabis measure cannot account for potency or dose effects. This is an important consideration, as previous meta-analysis has shown heavy cannabis use with high tetrahydrocannabinol (THC) content poses a particular risk for psychosis (Marconi et al., Reference Marconi, Di Forti, Lewis, Murray and Vassos2016). Thus, future studies could consider current use, and dose–response, when examining mediating and moderating effects of cannabis on the PRS-Sz–psychotic experience association.
Conclusion
In conclusion, although the current study could not confirm a mediated pathway between schizophrenia risk and PLE through cannabis use, the results contribute to the literature by showing the positive relationship between cannabis and future psychotic-like symptoms, while controlling for genetic vulnerability. Although we do not confirm any causal hypotheses, this result is important because, while cannabis producers would like to claim that cannabis use is only contra-indicated for individuals with a personal or family history of psychosis, the current findings suggest that cannabis use remains a risk factor for PLEs, over and above known genetic vulnerability for schizophrenia. Moreover, there was no evidence that genetically vulnerable individuals were more susceptible to the psychosis-related outcomes of adolescent onset cannabis use. As suggested by other authors (Jones et al., Reference Jones, Stergiakouli, Tansey, Hubbard, Heron, Cannon and Zammit2016), identifying a causal mechanism in the pathway from cannabis use to psychosis is extremely important for the development of targeted preventative interventions aimed at reducing cannabis use and/or schizophrenia risk.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291721003378
Acknowledgements
We thank all of the participants of both IMAGEN and Utrecht cannabis cohort studies. We thank the investigators of the Utrecht cannabis cohort for making data available.
Financial support
Dr Conrod is supported by a Tier 1 Canada Research Chair and this research was supported by a CIHR catalyst grant FRN155406 and the Canadian Cannabis and Psychosis Research Team Grant (CIHR FRN CA7170130) and National Institute on Drug Abuse, Grant/Award Numbeer: 1R01DA047119-01. This study received support from the following sources: the European Union-funded FP6 Integrated Project IMAGEN (Reinforcement-related behaviour in normal brain function and psychopathology) (LSHM-CT- 2007-037286), the Horizon 2020 funded ERC Advanced Grant ‘STRATIFY’ (Brain network-based stratification of reinforcement-related disorders) (695313), ERANID (Understanding the Interplay between Cultural, Biological and Subjective Factors in Drug Use Pathways) (PR-ST-0416-10004), BRIDGET (JPND: BRain Imaging, cognition Dementia and next generation GEnomics) (MR/N027558/1), Human Brain Project (HBP SGA 2, 785907), the FP7 project MATRICS (603016), the Medical Research Council Grant ‘c-VEDA’ (Consortium on Vulnerability to Externalising Disorders and Addictions) (MR/N000390/1), the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King's College London, the Bundesministeriumfür Bildung und Forschung (BMBF grants 01GS08152; 01EV0711; Forschungsnetz AERIAL 01EE1406A, 01EE1406B), the Deutsche Forschungsgemeinschaft (DFG grants SM 80/7-2, SFB 940/2), the Medical Research Foundation and Medical Research Council (grants MR/R00465X/1 and MR/S020306/1), the National Institutes of Health (NIH) funded ENIGMA (grants 5U54EB020403-05 and 1R56AG058854-01). Further support was provided by grants from: ANR (project AF12-NEUR0008-01 – WM2NA and ANR-12-SAMA-0004), the Eranet Neuron (ANR-18-NEUR00002-01), the Fondation de France (00081242), the Fondation pour la Recherche Médicale (DPA20140629802), the Mission Interministérielle de Lutte-contre-les-Drogues-et-les-Conduites-Addictives (MILDECA), the Assistance-Publique-Hôpitaux-de-Paris and INSERM (interface grant), Paris Sud University IDEX 2012, the fondation de l'Avenir (grant AP-RM-17-013); the National Institutes of Health, Science Foundation Ireland (16/ERCD/3797), USA (Axon, Testosterone and Mental Health during Adolescence; RO1 MH085772-01A1) and by NIH Consortium grant U54 EB020403, supported by a cross-NIH alliance that funds Big Data to Knowledge Centres of Excellence. Funding for the Utrecht cannabis cohort was provided by NWO, the Dutch council for scientific research (ZonMW TOP grant no. 91207039).
Conflict of interest
Dr Banaschewski served in an advisory or consultancy role for Lundbeck, Medice, Neurim Pharmaceuticals, Oberberg GmbH, Shire. He received conference support or speaker's fee by Lilly, Medice, Novartis and Shire. He has been involved in clinical trials conducted by Shire and Viforpharma. He received royalties from Hogrefe, Kohlhammer, CIP Medien and Oxford University Press. The current work is unrelated to the above grants and relationships.