Introduction
Schizophrenia is a highly heritable, severe psychiatric disease with typical symptoms including positive symptoms such as hallucinations, delusions and thought disorder, negative symptoms such as apathy and avolition, and cognitive dysfunction. Genome-wide association studies (GWAS) provide strong evidence of multiple independent loci contributing to the etiology of schizophrenia (Pardiñas et al., Reference Pardiñas, Holmans, Pocklington, Escott-Price, Ripke, Carrera and Walters2018; Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014). However, whilst individual loci have small effects on risk, multi-locus approaches show that cumulatively, even moderately associated alleles explain at least a third of schizophrenia genetic risk (Purcell et al., Reference Purcell, Wray, Stone, Visscher, O'Donovan and Sullivan2009; Ripke et al., Reference Ripke, O'Dushlaine, Chambert, Moran, Kähler, Akterin and Sullivan2013). Based on these alleles, an individual's genetic liability can be quantified using a polygenic score, a valuable tool when investigating shared genetics between disorders and how genetic risk manifests throughout the life course (Hubbard et al., Reference Hubbard, Tansey, Rai, Jones, Ripke, Chambert and Zammit2016; Jones et al., Reference Jones, Stergiakouli, Tansey, Hubbard, Heron, Cannon and Zammit2016).
Cannabis use is more common in individuals with schizophrenia than in the general population (Green, Young, & Kavanagh, Reference Green, Young and Kavanagh2005), and a large body of evidence from observational (Moore et al., Reference Moore, Zammit, Lingford-Hughes, Barnes, Jones, Burke and Lewis2007) and experimental (D'Souza et al., Reference D'Souza, Perry, MacDougall, Ammerman, Cooper, Wu and Krystal2004) studies support a causal effect of cannabis use on psychosis. However, some recent studies (Carey et al., Reference Carey, Agrawal, Bucholz, Hartz, Lynskey, Nelson and Bogdan2016; Power et al., Reference Power, Verweij, Zuhair, Montgomery, Henders, Heath and Martin2014; Reginsson et al., Reference Reginsson, Ingason, Euesden, Bjornsdottir, Olafsson, Sigurdsson and Stefansson2017; Verweij et al., Reference Verweij, Abdellaoui, Nivard, Cort, Ligthart, Draisma and Vink2017), though not all (Guloksuz et al., Reference Guloksuz, Pries, Delespaul, Kenis, Luykx, Lin and van Os2019), have found that genetic liability to schizophrenia (as captured by polygenic scores) is associated with cannabis use, suggesting that the association between cannabis and schizophrenia might be partly genetically confounded, or represent a pathway from schizophrenia risk to cannabis use. The latter may result from early manifestations of schizophrenia liability that may increase an individual's likelihood to start using cannabis, for example, experiencing difficulties with peers (Cannon et al., Reference Cannon, Walsh, Hollis, Kargin, Taylor, Murray and Jones2001; Malmberg, Lewis, David, & Allebeck, Reference Malmberg, Lewis, David and Allebeck1998). If a bi-directional relationship does exist, then identifying the mechanisms by which schizophrenia genetic risk increases the risk of cannabis use could provide important insights about targets to prevent cannabis use in the population, and particularly in those at genetically high risk for schizophrenia where cannabis reduction is likely to lead to the greatest benefit in reducing population levels of schizophrenia.
Schizophrenia is also associated with a higher prevalence of tobacco smoking behaviors compared with the general population (de Leon & Diaz, Reference de Leon and Diaz2005; Dickerson et al., Reference Dickerson, Stallings, Origoni, Vaughan, Khushalani, Schroeder and Yolken2013). As such, the possibility that cigarette smoking might increase the risk for schizophrenia has gained attention (Gurillo, Jauhar, Murray, & MacCabe, Reference Gurillo, Jauhar, Murray and MacCabe2015), although recent work shows that evidence consistent with causal effects on psychotic experiences are much stronger for cannabis use than they are for tobacco use (Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018). Schizophrenia polygenic risk and a schizophrenia GWAS hit in the CHRNA5-A3-B4 gene cluster, are associated with cigarette smoking phenotypes, including initiation, dependence and heaviness (Reginsson et al., Reference Reginsson, Ingason, Euesden, Bjornsdottir, Olafsson, Sigurdsson and Stefansson2017; Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014) which, similar to the findings for cannabis use, raises questions regarding a shared genetic etiology and direction of effect between cigarette use and schizophrenia, and the potential to gain insights into mechanisms leading to cigarette use in the population.
We identified, a priori to any analyses, a number of potentially modifiable pathways through which genetic risk for schizophrenia could theoretically lead to adolescent substance use. Based on the evidence of association with both genetic/familial risk for schizophrenia, and with substance use, these included peer-victimization, poorer social relationships, deficits in cognitive ability and impulsivity, and emotional or behavioral problems during childhood (Courtney, Mejia, & Jacobus, Reference Courtney, Mejia and Jacobus2017; Varese et al., Reference Varese, Smeets, Drukker, Lieverse, Lataster, Viechtbauer and Bentall2012; Welham, Isohanni, Jones, & McGrath, Reference Welham, Isohanni, Jones and McGrath2009).
Whilst understanding whether genetic risk for schizophrenia is associated with specific patterns of substance use, and the pathways involved in these relationships, could provide important insights into the etiology of both schizophrenia and substance use disorders, disentangling such associations may be hindered by measurement error in the outcomes, the high correlation between cigarette and cannabis use that makes it difficult to study independent effects of these substances, and by experimental and fluctuating use over time which are difficult to capture with single time-point assessments. To overcome some of these difficulties, we previously used longitudinal latent class analysis (LLCA) of repeated measurements of adolescent cigarette and cannabis use to identify subgroups of individuals based on their use or co-use of cigarettes and cannabis and capture information on persistent use as opposed to brief experimentation with these substances (Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018). The current study therefore aims to use these latent classes to: (i) examine whether schizophrenia genetic risk is associated with patterns of cigarette and cannabis use in adolescence, and (ii) examine whether genetic effects on substance use are mediated via cognitive, social, emotional or behavioral pathways during childhood.
Methods
Participants
The sample consisted of participants from the Avon Longitudinal Study of Parents and Children (ALSPAC) (see online Supplementary Methods) (Boyd et al., Reference Boyd, Golding, Macleod, Lawlor, Fraser, Henderson and Davey Smith2013; Fraser et al., Reference Fraser, Macdonald-Wallis, Tilling, Boyd, Golding, Davey Smith and Lawlor2013). Details of available data are accessible through a searchable data dictionary and variable search tool (http://www.bristol.ac.uk/alspac/researchers/data-access/data-dictionary). Ethical approval for the study was obtained from the ALSPAC Ethics and Law Committee and the Local Research Ethics Committees (http://www.bristol.ac.uk/alspac/researchers/research-ethics/). Consent for biological samples has been collected in accordance with the Human Tissue Act (2004). Informed consent for the use of data collected via questionnaires and clinics was obtained from participants following the recommendations of the ALSPAC Ethics and Law Committee at the time.
Genetic data
Genetic data were acquired using the Illumina HumanHap550 quad genome-wide single nucleotide polymorphism (SNP) genotyping platform from 9912 participants. Following quality control assessment and imputation (see online Supplementary Methods), genetic data were available for 7977 ALSPAC individuals.
Measures
Polygenic scores
Polygenic scores for schizophrenia were constructed for each ALSPAC individual using data from the most recent schizophrenia GWAS 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. Following quality control (see online Supplementary Methods), polygenic scores were calculated using the PLINK (v1.9) (Chang et al., Reference Chang, Chow, Tellier, Vattikuti, Purcell and Lee2015; Purcell et al., Reference Purcell, Neale, Todd-Brown, Thomas, Ferreira, Bender and Sham2007) ‘score’ command following the methodology described previously (Purcell et al., Reference Purcell, Wray, Stone, Visscher, O'Donovan and Sullivan2009).
For the primary analysis, scores were constructed using a list of SNPs with a GWAS training set p value threshold ⩽0.05, which optimally captures phenotypic variance in schizophrenia (Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014). Scores were weighted by the logarithm of the odds ratio (OR) for schizophrenia reported by the training set.
For sensitivity analyses, additional polygenic scores were created based on different GWAS training set p value thresholds [p ⩽ 0.5, 1e−5 and 5e−8 (genome-wide significant)] and after excluding the CHRNA5-A3-B4 nicotinic receptor gene cluster (chromosome 15: 78–79.5 Mb), a locus which is strongly associated with smoking cigarette quantity and nicotine dependence (Saccone et al., Reference Saccone, Wang, Breslau, Johnson, Hatsukami, Saccone and Bierut2009; Tobacco Genetics Consortium, 2010) and also genome-wide significantly associated with schizophrenia (Pardiñas et al., Reference Pardiñas, Holmans, Pocklington, Escott-Price, Ripke, Carrera and Walters2018; Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014). Weighted polygenic scores were also constructed for cigarette smoking initiation and cannabis use initiation using results from the GWAS and Sequencing Consortium of Alcohol and Nicotine use (GSCAN) tobacco and alcohol use GWAS (n = 1 232 091) (Liu et al., Reference Liu, Jiang, Wedow, Li, Brazel, Chen and Vrieze2019) and from a cannabis use GWAS meta-analysis using data from the International Cannabis Consortium (ICC) and UK Biobank (n = 184 765) (Pasman et al., Reference Pasman, Verweij, Gerring, Stringer, Sanchez-Roige, Treur and Vink2018), respectively, using SNPs meeting a p value threshold ⩽0.5 in the training set GWAS. As ALSPAC was a part of the GSCAN and ICC GWAS samples (Liu et al., Reference Liu, Jiang, Wedow, Li, Brazel, Chen and Vrieze2019; Stringer et al., Reference Stringer, Minica, Verweij, Mbarek, Bernard, Derringer and Vink2016), the SNPs and log ORs used to generate and weight the polygenic scores were from results after removal of ALSPAC and, due to data permissions, 23andMe from the GWAS meta-analyses.
Repeated measures of cigarette and/or cannabis use
Repeated measures of cigarette and/or cannabis use in ALSPAC were collected from clinic visits and questionnaires between approximate ages 14 and 19 years. At each time point, individuals were asked questions relating to their current use and frequency of use (see online Supplementary Methods for more detail). At each time point, data on cigarette and cannabis use were combined into a 3-category nominal variable: ‘Non-users,’ ‘Cigarette-only users’ and ‘Cannabis users (either with or without cigarettes)’ as previously described (Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018).
Potential mediators
Potential mediators were selected on the basis that they are all premorbid antecedents of schizophrenia, associated with a familial risk of schizophrenia, and/or associated with substance use. These included: IQ [assessed via the Wechsler Intelligence Scale for Children (Wechsler, Golombok, & Rust, Reference Wechsler, Golombok and Rust1992) at age 8 years], victimization [from the Bullying and Friendship Interview Schedule (Wolke, Woods, Bloomfield, & Karstadt, Reference Wolke, Woods, Bloomfield and Karstadt2000) at age 8 years], emotional problems [Strengths and Difficulties Questionnaire (SDQ) (Goodman, Reference Goodman1999) sub-scale score at age 9 years], antisocial behavior (assessed via a short structured interview at age 10 years), impulsivity [number of incorrect items on the stop-signal task (150 ms delay) (Handley, Capon, Beveridge, Dennis, & Evans, Reference Handley, Capon, Beveridge, Dennis and Evans2004) at age 10 years], friendship quality [based on 5 items from the Cambridge Friendship Questionnaire (Baron-Cohen & Wheelwright, Reference Baron-Cohen and Wheelwright2003) at age 12 years], and psychotic experiences [Psychosis-Like Symptom Interview (PLIKSi) (Horwood et al., Reference Horwood, Salvi, Thomas, Duffy, Gunnell, Hollis and Harrison2008) at age 12 years]. For more information, see online Supplementary Methods.
Statistical analysis
Longitudinal latent class analysis
Using the above described 3-category nominal variables of cigarette and cannabis use, LLCA was used to derive distinct behavior patterns of cigarette and/or cannabis use (Howe et al., Reference Howe, Trela-Larsen, Taylor, Heron, Munafò and Taylor2017; Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018; Taylor et al., Reference Taylor, Collin, Munafò, MacLeod, Hickman and Heron2017) (see online Supplementary Methods). Briefly, Individuals were included in the analysis if they had cigarette and cannabis use data present for 3 or more time points. Starting with one class, additional classes were added, and each time the model fit assessed using the proportion of individuals in each class, sample size adjusted Bayesian Information Criterion (SSABIC) and Lo-Mendell-Rubin likelihood ratio test (LMR-LRT), until the optimal number of classes that explained the variation within the data was achieved. LLCA was performed using MPlus version 8 (Muthén & Muthén, Reference Muthén and Muthén1998–2017).
Using the same data, our previous study found that a 5-class solution adequately describes the combined cigarette and cannabis use data between ages 14 and 19 years (Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018). The classes were defined as: non-users, early-onset cigarette-only users, early-onset cannabis users (with or without cigarette use), late-onset cigarette-only users and late-onset cannabis users (with or without cigarette use)(Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018). Based on the patterns of class membership across time (see Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018), early-onset and late-onset substance use are approximately defined as a higher probability of use between approximate ages 14 and 16 years and a higher probability of use between approximate ages 16 and 19 years, respectively (see online Supplementary Fig. 1).
Association analyses
Multinomial logistic regression was used to assess whether polygenic scores predicted latent class membership. Associations were assessed using a manual implementation of the bias-adjusted three-step method in MPlus [see online Supplementary Methods and Heron, Croudace, Barker, and Tilling (Reference Heron, Croudace, Barker and Tilling2015) for more detail]. Association analyses were conducted using individuals who had cigarette and cannabis use data present for 3 or more time points and genetic data.
To investigate whether associations between schizophrenia polygenic scores and latent class membership were influenced by the genetic overlap between variants associated with both schizophrenia and cannabis or cigarette use, analyses were also adjusted for the cigarette smoking initiation and cannabis use initiation polygenic scores.
As it was not possible to incorporate information on the frequency of substance use in our 5-class latent class approach as this resulted in an unstable model (Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018), we examined whether schizophrenia polygenic scores were associated with frequency of cannabis or cigarette use using data from single time-points. To aid future meta-analyses, the association between schizophrenia polygenic scores and cannabis and cigarette ever v. never use were also investigated. The association between schizophrenia polygenic scores and ever/never use and frequency of use was assessed using logistic regression and ordered logistic regression, respectively, in Stata statistical software (version 15; StataCorp LLC).
Mediation analysis
Mediation models were used to assess the direct effects of polygenic risk for schizophrenia on latent class membership and indirectly through each potential mediator. Mediation models were run in MPlus using a maximum likelihood estimator, and standard errors for indirect effects were calculated using a non-parametric bootstrapping approach with 100 replications. As two of the mediators were dichotomous measures, a counterfactual approach was implemented to allow for incorporation of the dichotomous mediators with easily interpretable effect estimates (Valeri & VanderWeele, Reference Valeri and VanderWeele2013). However, it is noted that for the models incorporating continuous mediators, this approach simplifies to the product of coefficient strategy as we did not allow for interaction between exposure and mediator.
Class reparameterization
As the main analyses were performed using multinomial logistic regression, the effect estimates are interpreted as the strength of association between the exposure and each outcome class in relation to a reference class, rather than the effect of the exposure on class membership in the whole population. To address whether this influenced our results, we repeated all analyses after reparametrizing the longitudinal latent classes (maintaining uncertainty in class membership) to examine, primarily, the effects for late-onset cannabis use as compared to all other classes combined in logistic regression. Effects from these analyses, therefore, represent odds for membership in late-onset cannabis use class compared to membership in any other latent class.
Results
There was strong evidence that genetic risk for schizophrenia differed across the combined cigarette use and cannabis use latent classes (omnibus p = 0.004; Table 1). The schizophrenia polygenic score based on SNPs meeting a training sample p threshold of 0.05 was associated with late-onset cannabis use as compared to non-use (OR = 1.23; 95% CI = 1.08, 1.41). There was also weak evidence of association with decreased odds of late-onset cigarette-only use (OR = 0.87; 95% CI = 0.76, 1.00) as compared to non-use, but little evidence of association with increased odds of early-onset cigarette-only use (OR = 1.13; 95% CI = 0.94, 1.36) or early-onset cannabis use (OR = 1.08; 95% CI = 0.87, 1.33). These associations persisted after adjusting for cigarette smoking initiation and cannabis use initiation polygenic scores (Table 1) which both showed evidence of association with the cigarette use and cannabis use latent classes (omnibus p < 0.001; online Supplementary Table 1).
Table 1. Associations between polygenic score for schizophrenia and subsequent cigarette and/or cannabis use as compared to non-use (N = 3925)
Note: SNPs, single nucleotide polymorphisms; OR, odds ratio; 95% CI, 95% confidence interval; p, omnibus p-value for the association between polygenic score and cigarette/cannabis use classes.
a Class proportions for latent class membership based on the estimated model.
b Compared to non-use class (class proportion for latent class membership based on the estimated model: 65.3%).
c Adjusted for polygenic scores for cigarette smoking initiation and cannabis use initiation (P T = 0.5).
Results were similar when excluding the CHRNA5-A3-B4 gene cluster, and when using a more relaxed p value threshold for inclusion of SNPs into the schizophrenia polygenic score (p value threshold ⩽0.5). However, evidence was weaker when using polygenic scores based on more stringent p value thresholds (p ⩽ 1e−5 or p ⩽ 5e−8) for SNP inclusion, that captures very little variance in liability to schizophrenia (online Supplementary Tables 2 and 3).
Results were also similar following reparameterization of classes with evidence of increased genetic liability for schizophrenia (p value threshold ⩽0.05) being associated with a 1.2-fold increase in odds (95% CI = 1.05, 1.37) of late-onset cannabis use as compared to all other classes combined (online Supplementary Table 4).
Evidence of the association between the schizophrenia polygenic score and ever/never substance use as well as the frequency of substance use was generally stronger for cannabis use than for cigarette use, and also stronger for measures of frequency of use in late adolescence and early adulthood than for measures of use in early adolescence (online Supplementary Tables 5 and 6).
There was weak evidence that genetic risk for schizophrenia was associated with lower quality of friendships (higher score indicates a lower friendship quality) (Beta = 0.06; 95% CI = −0.01, 0.13), and lower IQ score in childhood (Beta = −0.05; 95% CI = −0.07, −0.02), but less so with emotional symptoms, victimization, antisocial behavior or impulsivity (online Supplementary Figs. 2 and 3). There was evidence that higher IQ and engagement in antisocial behavior were associated with an increased odds of late-onset cannabis use (IQ: OR = 1.39; 95% CI = 1.18, 1.64; antisocial behavior: OR = 1.62; 95% CI = 1.02, 2.56). There was weaker evidence that a higher emotional symptoms score was associated with a reduction in late-onset cannabis use (OR = 0.91; 95% CI = 0.82, 1.01) (online Supplementary Fig. 2).
In the mediation analysis, there was weak evidence that the effect of schizophrenia polygenic score on IQ score at age 8 years acts to reduce the effect of schizophrenia genetic risk on late-onset cannabis use (indirect effect through IQ at age 8 years: OR = 0.99; 95% CI = 0.97, 1.00), but little evidence that any other mediators affected this pathway (Table 2). Results were also similar following reparameterization of classes (online Supplementary Table 7).
Table 2. Total effect, direct effect and indirect effect of schizophrenia polygenic score (P T = 0.05) on late-onset cannabis with/without cigarette use as compared to non-use through a range of potential mediators
Note: OR, odds ratio; 95% CI, 95% confidence interval; P T, p value threshold for inclusion of SNPs into the polygenic score. Within the mediation models, higher emotional, impulsivity and friendship quality scores indicate more emotional problems, a higher level of impulsivity and worse friendship quality, respectively.
Discussion
We examined whether genetic risk for schizophrenia was associated with cigarette and cannabis use during adolescence within a general population cohort and, where appropriate, tested for mediating effects of a range of factors measured prior to our outcome measures. Our primary outcome measures were latent classes summarizing the use of cigarettes and cannabis between ages 14 and 19 years. As previously reported (Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018), our data was best summarized by 5 classes comprising individuals with early-onset cigarette-only use, late-onset cigarette-only use, early-onset cannabis use, late-onset cannabis use, and no use of either substance. In our primary analysis, using a training sample p threshold of 0.05 that optimally captures variance in schizophrenia liability, we found that schizophrenia polygenic risk was most strongly associated with late-onset cannabis use. Early-onset cigarette and cannabis use class estimates were compatible with the late-onset cannabis use estimate. However, these estimates were less precise as the classes were substantially smaller and therefore analyses had lower power. Interestingly, we found that schizophrenia polygenic risk was also associated with a decreased odd of late-onset cigarette only use, however, this weak association did not survive after class reparameterization.
Our findings are consistent with other studies showing that schizophrenia polygenic risk is associated with cannabis use (Carey et al., Reference Carey, Agrawal, Bucholz, Hartz, Lynskey, Nelson and Bogdan2016; Power et al., Reference Power, Verweij, Zuhair, Montgomery, Henders, Heath and Martin2014; Reginsson et al., Reference Reginsson, Ingason, Euesden, Bjornsdottir, Olafsson, Sigurdsson and Stefansson2017; Verweij et al., Reference Verweij, Abdellaoui, Nivard, Cort, Ligthart, Draisma and Vink2017). Furthermore, results from both our primary results and sensitivity analyses provide evidence that genetic risk of schizophrenia is more strongly associated with cannabis use than with cigarette use.
One interpretation of our findings is that genetic risk for schizophrenia confers a risk of substance use that is more specific for some drug classes than others, perhaps due to pleiotropic effects on more substance-specific biological pathways than ones that are common across addictive behaviors. However, as almost all individuals within the cannabis use class also use tobacco, this class could just index a more severe phenotype. Therefore, genetic risk for schizophrenia could confer a risk of multiple substance use, for example through dopaminergic or opioid function biological pathways that are strongly implicated across all addictive behaviors (Koob & Volkow, Reference Koob and Volkow2016).
It is also possible that the association with late-onset cannabis use is not due to pleiotropic effects of addiction-related biological pathways, but due to behavioral manifestations of schizophrenia genetic risk leading to adolescent use of cannabis. To explore this possibility we examined if the strongest association we observed in our primary analysis, between schizophrenia genetic risk and late-onset cannabis use, was mediated by lower childhood IQ, emotional problems, victimization, engagement in antisocial behavior, impulsivity or poorer social relationships, all of which are characteristics associated with increased risk of schizophrenia incidence or cannabis use (Courtney et al., Reference Courtney, Mejia and Jacobus2017; Varese et al., Reference Varese, Smeets, Drukker, Lieverse, Lataster, Viechtbauer and Bentall2012; Welham et al., Reference Welham, Isohanni, Jones and McGrath2009). Our results suggested that little to none of this association was mediated through these pathways, and indeed that ‘direct’ effects of schizophrenia genetic risk on late-onset cannabis use may be stronger than first observed. However, this does not exclude the possibility that other variables that we did not test mediate this relationship.
Whilst the cognitive, emotional, and behavioral characteristics we examined did not mediate the relationship between schizophrenia genetic risk and cannabis use, identifying mediating phenotypes expressed in childhood or adolescence is important not just for understanding the mechanisms underlying addictive behavior, but also to inform potential targets for early intervention to prevent substance use and harmful consequences of this. The mediators we examined were measured in childhood, to ensure they occurred prior to substance use, hence minimizing bias in our models. However, a potential limitation of this is that our results might not adequately reflect the relationship of schizophrenia genetic risk with the same characteristics in adolescence when they might have a more immediate effect on substance use behavior.
The association we observe here between schizophrenia genetic risk and cannabis use suggests either that the association between cannabis use and psychosis observed consistently in epidemiological studies (Gage, Hickman, & Zammit, Reference Gage, Hickman and Zammit2016; Moore et al., Reference Moore, Zammit, Lingford-Hughes, Barnes, Jones, Burke and Lewis2007) is, at least in part, due to pleiotropy, or that cannabis has a causal effect on schizophrenia [and therefore risk variants for cannabis use will also be identified as risk variants for schizophrenia (Gage, Davey Smith, Ware, Flint, & Munafò, Reference Gage, Davey Smith, Ware, Flint and Munafò2016) in adequately powered GWASs where there would be many more cannabis users among cases than controls]. In fact, despite the finding from this and other studies that schizophrenia genetic risk is associated with cannabis use, there is little evidence that shared genetic effects confound associations between cannabis use and risk of psychotic outcomes in epidemiological studies. For example, in a recent study, we found strong evidence that classes of cannabis use were associated with subsequent risk of psychotic experiences, and that this was not attenuated after adjusting for family history of schizophrenia (Jones et al., Reference Jones, Gage, Heron, Hickman, Lewis, Munafo and Zammit2018) or the schizophrenia genetic risk score (online Supplementary Table 8).
One approach that has been used to examine the causal effects of cannabis use on schizophrenia and assess the presence of genetic confounding (horizontal pleiotropy) is Mendelian randomization (MR). Evidence consistent with a causal effect of schizophrenia risk on the likelihood of cannabis initiation, as well as weak evidence of a causal effect from cannabis initiation to schizophrenia has been reported (Gage, et al., Reference Gage, Jones, Burgess, Bowden, Davey Smith, Zammit and Munafò2017; Pasman et al., Reference Pasman, Verweij, Gerring, Stringer, Sanchez-Roige, Treur and Vink2018). Similarly, analyses have reported a bidirectional relationship between a measure of lifetime cigarette smoking (capturing smoking duration, heaviness and cessation) and schizophrenia risk (Wootton et al., Reference Wootton, Richmond, Stuijfzand, Lawn, Sallis, Taylor and Munafò2019). However, when there is little understanding of the biological effects of the genetic instruments used in MR analyses, bidirectional relationships such as these can be difficult to interpret (Davey Smith & Hemani, Reference Davey Smith and Hemani2014), and therefore neither MR studies to date, nor our results here, lead to substantially stronger conclusions about the causal effects of cannabis and cigarettes on psychosis than those from more traditional epidemiology designs.
Whilst our findings cannot address whether cannabis use has a causal effect on schizophrenia, our results show that schizophrenia genetic liability does not lead to increased cannabis use through the mechanisms examined here, and that the investigation of other pathways is required to identify potentially modifiable targets to reduce the public health burden of cannabis use in the population.
Strengths and limitations
One of the strengths of our study is that we use a large, population-based cohort, with multiple measures of cigarette and cannabis use data over the whole adolescent period, and thus our results are much less prone to measurement error than if we had used single time-point measures of substance use, although it likely still exists to some extent. Furthermore, using a latent class approach with longitudinal data allows us to maximize the use of data for individuals even where participation and question response has been sporadic, and hence minimize potential selection bias, despite the considerable levels of attrition over time. We also used the largest, most recent published GWASs of schizophrenia, cigarette use and cannabis use as training sets for the derivation of our polygenic scores. Nevertheless, there are a number of limitations in our study.
Whilst our use of latent classes derived from information on the combined use of cigarettes and cannabis use is useful for teasing out independent effects of schizophrenia genetic risk on these outcomes, it was not possible to define a class of individuals who use cannabis without tobacco, as most cannabis users smoke cannabis in combination with tobacco (Amos, Wiltshire, Bostock, Haw, & McNeill, Reference Amos, Wiltshire, Bostock, Haw and McNeill2004), even when they self-report as being cigarette non-smokers (Gage et al., Reference Gage, Hickman, Heron, Munafo, Lewis, Macleod and Zammit2014). Furthermore, we have previously found that a substantial proportion of the people who smoke cigarettes most heavily also use cannabis (Gage et al., Reference Gage, Hickman, Heron, Munafo, Lewis, Macleod and Zammit2014), and thus the cigarette-only class might not include those who have been most heavily exposed to tobacco. Therefore, we cannot rule out whether the associations observed between schizophrenia genetic risk and the late-onset cannabis use class is driven by heavier cigarette use in these individuals than in those within the early-onset cigarette only or late-onset cigarette only classes (although this would not be consistent with our sensitivity analyses).
Another limitation is that it was not possible to incorporate information on the frequency of substance use per time point within the combined cannabis and cigarette use model due to model instability. We, therefore, also examined the frequency of cigarette use and cannabis use using single time-point measures and found no consistent evidence of association with the genetic liability of schizophrenia, with the exception of increase odds of cannabis use frequency at ages 17–19 years.
Furthermore, although we attempted to minimize genetic confounding by adjusting for cigarette and cannabis initiation polygenic scores, heterogeneity between training set GWAS samples (i.e. differing ages of participants) and substance use measures (i.e. measures combined experimental and regular users into a single group) may have reduced their power to detect genetic associations. Furthermore, polygenic scores for cigarette and cannabis use initiation explain only a small proportion of the variance for these phenotypes in independent samples. Hence, adjusting for cigarette and cannabis initiation polygenic scores may have not adequately removed confounding effects resulting from pleiotropy. It is also possible that our mediation effects are underestimated due to residual confounding.
Finally, as our cohort only included data up to 19 years of age, it was not possible to examine the effects of schizophrenia genetic risk on longer-term patterns, or long-term cumulative use of cannabis or cigarettes. Addressing these model limitations may become more tractable in the future.
Conclusion
In conclusion, our study provides evidence that genetic risk for schizophrenia is associated with patterns of cannabis use during adolescence, and that this is not mediated through other measured phenotypic manifestations of genetic risk for schizophrenia during childhood, including lower IQ, victimization, increased emotional difficulties, antisocial behavior, impulsivity, or poorer social relationships. Evidence of an association between genetic risk for schizophrenia and cigarette use was weaker. Further studies need to examine longer-term patterns of use of these substances over time to minimize measurement error in allocation of substance use classes, and to establish the mechanisms by which these associations arise to inform substance use reduction strategies.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291720001798.
Acknowledgements
We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. The UK Medical Research Council (MRC) and Wellcome (grant number 102215/2/13/2) and the University of Bristol provided core support for ALSPAC. A comprehensive list of grant funding is available on the ALSPAC website. Measures used were specifically funded by the MRC (grant number G0800612/86812) and the Wellcome Trust (grant numbers 076467/Z/05/Z, 086684, 092731). GWAS data were generated by Sample Logistics and Genotyping Facilities at the Wellcome Trust Sanger Institute and LabCorp (Laboratory Corporation of America) using support from 23andMe. This study was supported by MRC (grant number MR/M006727/1) and the National Institute for Health Research (NIHR) Biomedical Research Centre at the University Hospitals Bristol National Health Service Foundation Trust and the University of Bristol. The views expressed in this publication are those of the author(s) and not necessarily those of the National Health Service, the National Institute for Health Research or the Department of Health. HJJ, GH and MRM are members of the MRC Integrative Epidemiology Unit at the University of Bristol funded by the MRC (grant numbers MC_UU_00011/1, MC_UU_00011/7). GH is funded by a Sir Henry Wellcome Postdoctoral Fellowship (grant number 209138/Z/17/Z). CW is funded by a Cancer Research UK Population Research Postdoctoral Fellowship (grant number C60153/A23895).
Conflict of interest
Professor O'Donovan received a consultancy fee from Roche in July 2015. All other authors have declared no conflicts of interest.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
