Based on previous literature  we expect genetic overlap between cannabis use and drug use

Illicit drugs are substances that either stimulate or inhibit the central nervous system or cause hallucinogenic effects  to the effect that their nonmedical use has been prohibited globally . For some substances, like cannabis, the prohibition or legalization status varies widely over time and over different countries and states . In the present paper we focus on illicit drugs in a broad sense, including cannabis, ecstasy, stimulants, opioids. We do not consider substances that are legal in the Netherlands, such as nicotine and alcohol. Cannabis is one of the most widely consumed drugs worldwide, with 192.2 million past-year users in 2016, corresponding to 3.9 per cent of the global population aged 15–64 years . Despite the increasing use of cannabis for medicinal purpose and an ongoing debate about medicalization and decriminalization, associations with adverse health effects have been reported. These adverse health effects include development of dependence, cardiovascular disease, impaired respiratory function and mental health problems . Another increasingly popular drug is ecstasy, a psychoactive drug that consists of MDMA. The prevalence in the global population aged 15–64 years is estimated to be 0.4 % . In Europe, approximately 1.7 % of young adults  have used ecstasy, with estimates ranging from 0.3%–5.5% between countries . Other relatively popular illicit drugs include amphetamine and cocaine , with worldwide past year estimated prevalences of 0.77 %, and 0.35 % respectively . The past year prevalence of opioids  was 0.37 % worldwide in 2017 . For all illicit drug use together, the overall disease burden was estimated to be 27.8 million attributable disability-adjusted life-years  in 2017. DALYs reflect the number of years lost due to ill-health, disability or early death. The mortality rate due to illicit drugs was 6.9 deaths per 100,000 people in 2017 . Substance use, including cannabis use, is moderate to highly heritable ; Verweij et al., 2010, 2017. The largest genomewide association  study for cannabis use to date has successfully identified 35 genes  associated with lifetime cannabis use .

Two other genome-wide association studies identified genes for cannabis dependence and cannabis disorder . In the current study we have information on use , and will therefore use the GWA for cannabis use  as discovery sample. Epidemiological studies have consistently shown correlations between use of different substances, such that individuals that use one substance are more likely to also use another . The phenotypic correlations between substances are partly explained by common genetic influences . Many genetic variants, each with a small effect size, contribute to complex behaviors, such as substance use. With methodological advances in molecular genetics and increased sample sizes in GWA studies it has become viable to use many measured genetic variations in individuals to estimate their genetic vulnerability for a certain trait. To do this, polygenic scores  in individuals from a target dataset can be calculated based on their genome-wide genetic data and the genetic effect sizes estimated in large GWA studies . If the PGS in the target set, for example reflecting the genetic vulnerability for cannabis use, is associated with drug use, for example ecstasy, this would suggest that there is overlap in the genes underlying grow cannabis in containers and ecstasy use. In the present study, we used summary-level data from the largest GWA study for lifetime cannabis use to date  to generate PGSs in an independent sample of 8348 individuals registered at the Netherlands Twin Register . We tested the association of the PGS for lifetime cannabis use with ecstasy, stimulants  and a broad category of drug use, including stimulants, opioids and hallucinogens.A significant association  may indicate that there are common underlying genetic predispositions to the use of these substances, or can be the result of a causal association  between the use of the different substances. In that last case, use of cannabis may lead to use of ecstasy or other drugs, and therefore genes associated with cannabis use will also –indirectly- be associated with use of other drugs. The different explanations are not mutually exclusive and are difficult to distinguish.

If a significant association is found between the cannabis PGS and use of other drugs, we will explore the nature of this relationship by repeating the same analyses separately in cannabis users and non-users. If the association between the polygenic risk for cannabis and drug use is only significant in cannabis users and not in never users, this might indicate that causal effects play a role , although other explanations  are still possible. To further explore the causal role of cannabis in other drug use, we also explored drug use in monozygotic twins discordant for cannabis use.Prediction analyses were carried out using generalized estimation equations with a logit link function. To account for familial relatedness, this method uses an exchangeable covariance matrix, allowing for correlated residuals between family members. Analyses were run using robust standard errors for the parameter estimates. Sex, age, and 10 genetic principal components were included as covariates in all analyses. Principal components were included to correct for effects of population stratification. Age was negatively correlated with the outcome measures  and males had a higher prevalence of drug use than females. To explore possible sex differences we tested the interaction between the cannabis PGS and sex for ecstasy, stimulants and any illicit drug use . Estimates of the explained variance  were obtained from logistic regressions by subtracting the pseudo-R2 estimates of the model with only covariates from the model including both the PRSs and covariates. Odds ratios were also obtained through the regression analyses.To inspect how drug use varied with increasing cannabis PRS we used quintile plots. The cannabis PRS was divided in quintiles, and we calculated the odds ratio for respectively ecstasy use, stimulants and any illicit drug use within each quintile. For the twin analyses, we compared the prevalence of drug use in the cannabis using twins to that of their non-using co-twins with a McNemar test . In this design, genetic and common environmental influences are controlled for because MZ twins share all their genetic material and their  home environment. If the association between cannabis use and other drug use is solely explained by genes and/or shared environmental factors, then the twins who have used cannabis and their co-twins who have not should be equal in their use of other drugs. In contrast, if the association is to some extent causal or explained by environmental factors for which twin pairs are discordant, we would expect to find significantly higher prevalences in the cannabis users compared to their unaffected MZ co-twins.We showed that the genetic liability underlying cannabis use significantly explained variability in ecstasy, stimulant, and any illicit drug use.

When the sample was stratified for lifetime cannabis use, this association seemed to be stronger in cannabis users compared to nonusers for ecstasy and stimulants, but not for any drug use. However, this trend was not significant after correction for multiple testing. The observation that the PGSs for cannabis use were significantly associated with the examined drug use variables , suggests genetic overlap between the traits. The explained variance ranged between 0.5 and 1.2 %, which is quite low but consistent with other PGS studies of addictive phenotypes . As far as we know this is the first study exploring the genetic overlap of the genetic vulnerability for cannabis with other illicit drug use. Only a few studies explored genetic overlap across substances using a PGS method. A previous study showed genetic overlap between PGS for cigarettes per day with glasses of alcohol per week and cannabis initiation as well as between PGS for age at onset of smoking and age at regular drinking. However the PGSs for smoking initiation and smoking cessation did not significantly predict alcohol or cannabis use, possibly due to limited power . Demontis et al. showed that a PGS-for lifetime smoking was associated with cannabis use disorder . Recently, Chang et al. tested the association between 5 PGSs for licit substances  with 22 target phenotypes for illicit substance use and substance use disorders. Only 9 of the 110 tested associations were significant. Interestingly, the stimulants  showed some significant results, while associations with sedatives or pain killers were not significant. In particular, the PGS for smoking initiation significantly explained variation in the risk of cocaine, amphetamine, hallucinogens, ecstasy and pot for cannabis initiation, as well as DSM-5 alcohol use disorder . The PGS for drinks per week significantly explained variation in cocaine, amphetamine and ecstasy initiation . Taken together, these results indicate genetic overlap between the use of different substances, although in previous studies not all tested associations were significant. As explained in the introduction, genetic overlap may indicate that there are common underlying genetic predispositions to the use of these substances . In case of drug use, this could be genes involved in the vulnerability for reward , but could also reflect genetic vulnerability for more general personality traits, such as impulsivity, risk-taking behavior or sensation seeking which are also often associated with drug use  or educational attainment . On the other hand, genetic overlap can also be the result of a causal association . To explore whether cannabis use itself caused the use of ecstasy, stimulants or any drugs we tested the association between the PGS for cannabis and the outcome variables in cannabis users and never users separately. The association of the cannabis PGS with ecstasy and stimulant use seemed stronger in cannabis users compared to never users which could point to a causal relationship . This effect was only observed in people born after 1968, but given the fact that the prevalence is higher in this younger group there is probably more power to detect an association than in the older group. Since the association was not significant after correction for multiple testing we must be cautious with drawing conclusions. In addition, we explored the differences in drug use prevalence in MZ twin pairs discordant for cannabis use.

The twins who used cannabis had more often used drugs, compared to their MZ co-twins who never used cannabis. This is in accordance with previous research using the co-twin control methodology . This finding suggest that the differences in illicitit drug use between twins who used cannabis and their unaffected co-twins cannot solely be explained by genetic influences  but that individualspecific environmental factors such as cannabis use play a role. Together, this suggested that cannabis use could be a causal factor for other drug use. Future studies should explore causality with more advanced methods such as Mendelian Randomization , but larger samples sizes are needed than available in the current study to obtain enough power. In previous studies using two-sample bi-directional Mendelian Randomization analyses, no evidence was found for causal relationships between smoking, alcohol, caffeine, and cannabis  but these studies did not includeother illicit drugs. There might not be a sequential order of use for initiation of smoking, alcohol use or caffeine consumption since these substances are widely available and some people start with smoking while others start with drinking first. A gateway from licit substance use to illicit drug use or from one drug  to other drugs  might be more plausible. Ideally, causality should be tested in two directions, because some studies have also found evidence supporting a reverse-gateway hypothesis . For example, cannabis could influence ethanol  levels, although existing findings are inconclusive , and a recent MR study did not find evidence for a causal relationship ). A limitation of the PGS approach is that currently only large GWA studies are available for lifetime cannabis   and opiod use disorder , but not for other illicit drugs such as ecstasy. Large genome wide association studies for illicit drugs are needed as input to calculate reliable PGSs. A strength of the current study is the large discovery sample for cannabis . It is known that a larger discovery sample leads to a more reliable  PGS in the target sample. In the present study we showed as a proof of concept that the PGS for lifetime cannabis use was significantly associated with cannabis use in the target sample.