New Results

Cannabis use, depression and self-harm: phenotypic and genetic relationships

K Hodgson, JRI Coleman, SP Hagenaars, KL Purves, K Glanville, SW Choi, P O’Reilly, G Breen, Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium, CM Lewis
doi: https://doi.org/10.1101/549899

Abstract

Background and Aims The use of cannabis has previously been linked to both depression and self-harm, however the role of genetics in this relationship are unclear. We aimed to examine the phenotypic and genetic relationships between these traits.

Design Genetic and cross-sectional phenotypic data collected through UK Biobank, together with consortia genome-wide association study summary statistics. These data were used to assess the phenotypic and genetic relationship between cannabis use, depression and self harm.

Setting UK, with additional international consortia data

Participants N=126,291 British adults aged between 40 and 70 years, recruited into UK Biobank

Measurements Genome-wide genetic data, phenotypic data on lifetime history of cannabis use, depression and self-harm.

Findings In UK Biobank, cannabis use is associated with increased likelihood of depression (OR=1.64, 95% CI=1.59-1.70, p=1.19×10−213) and self-harm (OR=2.85, 95% CI=2.69-3.01, p=3.46×10−304). The strength of this phenotypic association is stronger when more severe trait definitions of cannabis use and depression are considered. Additionally, significant genetic correlations are seen between cannabis use and depression using consortia summary statistics (rg=0.289, SE=0.036, p=1.45×10−15). Polygenic risk scores for cannabis use and depression both explain a small but significant proportion of variance in cannabis use, depression and self harm within a UK Biobank target sample. However, two-sample Mendelian randomisation analyses were not significant.

Conclusions Cannabis use is both phenotypically and genetically associated with depression and self harm. Future work dissecting the causal mechanism linking these traits may have implications for cannabis users.

Introduction

Cannabis is one of the most commonly used psychoactive drugs1 and depression is the second leading cause of disability2 worldwide. Given the known comorbidity between these traits3–7, understanding the nature of their relationship is highly relevant from a public health perspective. This is particularly true in the context of changing legislation and public perceptions of the risks of using cannabis8.

Several theories regarding the mechanisms for associating cannabis use and depression have been proposed. One trait might causally lead to the other, or both traits could result from common factors (either genetic or environmental). There is some longitudinal evidence that cannabis use is associated with increased likelihood of later depression, however there are complexities in accounting for potential confounding factors5,9. Indeed, when all covariates are accounted for, cannabis use may follow rather than precede depression10. Thus, longitudinal attempts to disentangle causality remain inconclusive. However, this work highlights the potential impact of common risk factors that increase the likelihood of both using cannabis and suffering from depression.

Genetics is a useful tool by which to understand these common risk factors. Both cannabis use11 and depression12 are known to be heritable and substantial increases in sample sizes mean that there has been progress in understanding the genetic basis of these phenotypes individually. For cannabis use, the largest genome-wide association (GWA) study to date reported 8 significant loci13 and for depression, 44 genome-wide significant loci were identified by the Psychiatric Genomics Consortium Major Depressive Disorder (PGC-MDD) working group and 23andMe14. These developments mean efforts to understand the interplay between comorbidity traits from a genetic perspective are now much more powerful.

But there is mixed evidence for a shared genetic aetiology between cannabis use and depression. A significant genetic correlation between cannabis use and depression was observed in an extended pedigree sample15 but with a discordant twin design, genetic factors played only a small role in the link between early or frequent cannabis use and depression or suicidal ideation16. In unrelated samples, polygenic risk for depression was associated with non-problematic cannabis use17, however using genome-wide summary statistics to consider the genetic correlations between cannabis use and a range of genetic traits, the estimated correlation with depression was modest but not significant13. In the latter two analyses, the underpowered (and since superseded) PGC-MDD 2013 depression sample18 was used, which is likely to have impacted upon results. The recent availability of larger datasets enables us to examine the genetic relationship between cannabis use and depression with substantially more statistical power.

Cannabis use has also been associated with increased rates of self-harm; both for non-suicidal19–21 and suicidal16,22,23 behaviours. This phenotypic association may be stronger than that observed between cannabis and depression but whilst some studies report evidence of a shared genetic aetiology23,24 and others find that genetic factors are less important16,25. Therefore, additional work exploring how cannabis use related to self-harm is also needed.

Here we have chosen to focus on the use, rather than abuse of cannabis. Problematic cannabis use is more heritable11 and also potentially more strongly associated with depression5. But cannabis is not highly addictive26; estimates suggest only 10% of those who use cannabis become dependent27 meaning that cannabis use disorders are relatively rare. Further, they are an undesirable outcome in themselves. In the context of changing perceptions of the risks associated with cannabis use, understanding if and how mental health traits relate to non-disordered use of cannabis is arguably a more useful question from a public health perspective.

Here we use data from UK Biobank28,29 in addition to summary statistics available from genome-wide analyses by the PGC-MDD working group14 and the International Cannabis Consortium30 to explore the relationship between cannabis use, depression and self-harm. Many GWA studies of cannabis use consider all individuals who have ever tried cannabis use. Nevertheless, cross-sectional evidence suggests that more frequent (non-disordered) cannabis use is more strongly associated with depression7. Using the phenotypic detail available in UK Biobank, we investigate the risks associated with initial and continued cannabis use, as well as single episode and recurrent depression. We assess phenotypic associations and use genetic approaches31–33 to consider overlapping genetic influences on these traits. Finally, we use Mendelian randomisation (MR) techniques to examine the direction of causality between cannabis use and depression.

Methods

UK Biobank Sample

UK Biobank (http://www.ukbiobank.ac.uk) includes measures of many health-related phenotypes for approximately 500,000 British adults aged between 40 and 70 years28. UK Biobank received ethical approval from the Research Ethics Committee (11/NW/0382). The work presented here uses data from a subset of 157,366 participants who completed an online follow-up mental health questionnaire (category 136 on http://biobank.ctsu.ox.ac.uk), which included items on cannabis use, depression and self-harm34.

Phenotypic Measurements

Cannabis

Participants were asked about their lifetime cannabis use (data field 20453): “Have you taken cannabis (marijuana, grass, hash, ganja, blow, draw, skunk, weed, spliff, dope), even if it was a long time ago?”. Those who responded “No” were classified as controls and those endorsing “Yes” options were classified as cannabis users. We separated these users into two groups: those reporting initial cannabis use (“Yes, 1-2 times”, “Yes, 3-10 times”) and continued cannabis use (“Yes, 11-100 times”, “Yes, more than 100 times”).

Depression

Using the work from Davis et al.34 as a framework, depression was defined using lifetime criteria based on questions derived from the Composite International Diagnostic Interview (CIDI). Exclusions were made if depression cases reported previous diagnoses of schizophrenia, other psychoses or bipolar disorder. However we did not exclude controls based on current prescriptions for antidepressant medications, given the use of these drugs for diagnoses other that depression. The reported number of depressed periods was used to classify individuals with single episode or recurrent depression.

Self harm

Participants were asked “Have you deliberately harmed yourself, whether or not you meant to end your life?” (data field 20480).

Other phenotypes

To consider potential confounding due to the use of substances other than cannabis, we used the “Ever smoked” item (derived data field 20160) as well as the mental health questionnaire item on “Frequency of drinking alcohol” (data field 20414).

Genetic data

Genome-wide data is available for the UK Biobank cohort29. Genotyping was performed using two highly-overlapping custom genotyping arrays, covering ∼800,000 markers. This data underwent centralised quality control to remove genotyping errors before being imputed in a two-stage imputation to the Haplotype Reference Consortium35 and UK10K36 reference panels. Additional genotypic and sample quality controls thresholds were then applied and the sample was limited to individuals of White European ancestry (Supplementary Materials Section 1.2).

This gave a sample of N=126,291 individuals with both genetic and mental health questionnaire data available.

Statistical methods

Univariate analyses within UK Biobank

Phenotypic analyses

For each of the seven phenotypes of interest (cannabis use, initial cannabis use, continued cannabis use, depression, single episode depression, recurrent depression and self-harm), chi-squared models were used to compare differences between cases and controls in terms of sex, smoking and frequency of alcohol drinking.

Genome-wide association analyses

Using BGENIE v1.229, genome-wide association (GWA) analyses were completed for each of the seven phenotypes, after residualising for assessment centre, batch and the first 6 principal components, as calculated in the European-only subset of the data using flashpca237. Manhattan and QQ plots were made using the qqman38 R package. All positions are given in GRCh37 coordinates and results were clumped in plink 1.9039 (www.coggenomics.org/plink/1.9/) using clumping thresholds as in the PGC-MDD analysis14.

Heritability of phenotypes

Using the generated GWA results and LD-score regression31,32, the SNP-heritability of each of the seven traits was calculated. For conversion to the liability scale, sample prevalence was defined as [Ncases/(Ncases + Ncontrols)], whilst population prevalence was defined as [Ncases]/[Ntotal completing MHQ]. To consider if the UK Biobank may have over-or under-sampled cases, we also calculated the heritability using population prevalence estimates plus or minus 20% of the cases observed.

Bivariate analyses

Phenotypic relationships within UK Biobank

To explore the association between cannabis, depression and self-harm phenotypes, odds ratios were calculated using logistic regression models. To assess the impact of differences in sex, smoking and frequency of alcohol drinking, stratified analyses were perfomed.

Genetic Correlations

The genetic relationships between phenotypes were examined using bivariate LD-score regression31,32.

Maximum available sample size

To consider the relationship between cannabis use and depression in the largest available sample, we used consortium-based summary statistics. For cannabis use, the International Cannabis Consortium (ICC, N=32,33030) is the largest publically available dataset. A meta-analysis combining the ICC with UK Biobank (UKB) was performed, total sample size N=158,621. Summary statistics from the ICC and UKB samples were cleaned using the EasyQC software40 and then meta-analysed using a p-value-based approach, with genomic control in METAL41. For this ICC+UKB combined cohort, the sample prevalence of cannabis use was 28%. For depression, the summary statistics from the PGC-MDD GWA were used, including both the UK Biobank and 23andMe cohorts (N=480,359, sample prevalence 28%14).

Detailed phenotypes within UK Biobank

To consider whether genetic correlations vary according the severity of the phenotype considered, bivariate LD-score analyses within UK Biobank were also performed, considering each combination of the seven traits of interest.

Polygenic risk scoring

Consortium-based GWA summary statistics were also used to build polygenic risk scores for prediction into UK Biobank. Polygenic risk scoring can be biased by overlap between discovery and target samples. Therefore, when building a depression polygenic risk score (PRSDEP), UK Biobank was excluded from the PGC-MDD and 23andMe summary statistics in the discovery dataset (N=450,619). To maximise the discovery sample size available when building a polygenic risk score for cannabis use (PRSCANN), but retain an independent target sample for prediction, UK Biobank was split equally into two using the caret R package42 to create a discovery and a target subsample. The ICC data was then meta-analysed (using the same methods as described above) with the discovery 50% of UK Biobank to give a total discovery dataset for cannabis use of N=95,384 (ICC+50%UKB).

The target 50% of UK Biobank (N=63,054) was then used as the target sample for both polygenic risk scoring analyses, to facilitate comparison between phenotypes.

PRS scores were built using PRSice2 (https://choishingwan.github.io/PRSice/)33,43. For computational reasons, in the UK Biobank target dataset only directly genotyped variants were used. Full details are in Supplementary Materials Section 4, but briefly, 11 p-value thresholds were used when building the polygenic risk scores and a glm logistic model was used to test these. All R2 values shown have been converted to the liability scale44.

Mendelian randomisation

To explore the direction of causality between cannabis use and depression, we used two-sample MR, using published genetic variants for cannabis use and depression as instrumental variables. MR can be biased by sample overlap. To obtain the largest sample size with available GWA summary statistics, whilst minimising sample overlap, we used the PGC-MDD with 23andMe (N=480,359) and ICC (N=32,330) datasets. Both of these datasets contain samples from the Netherlands Twin Register/NESDA45,46 and Queensland Institute of Medical Research Berghofer cohorts18,47. This gives a maximum possible overlap of 1.4% (N=6,719). Whilst results should be considered in the context of this overlap, Burgess et al.48 conclude that the bias from an overlap of this magnitude is minimal.

We used the R package TwoSampleMR49 to test casuality in both directions, using both inverse-variance weighted (IVW) and MR-Egger models (to consider evidence for pleiotropy). This package includes harmonisation steps to ensure allelic alignment between exposure and outcome datasets.

Testing the causal effect of depression on cannabis use

For the PGC-MDD and 23andMe dataset (SNP-h2=8.7%), 44 genome-wide significant independent loci have been previously identified14. The logOR and SE for the 39 of these SNPs were available in the ICC dataset.

Testing the causal effect of cannabis use on depresion

For the ICC dataset (SNP-h2=13%), no SNPs reached genome-wide signficance30. Instead, we focus on those SNPs reaching suggestive significance (p<1×105, N=152 SNPs). After clumping (using clump_kb=3000 and clump_r2=0.1 to match PGC thresholds), 19 independent SNPs were retained. The logOR and SE for these 19 SNPs were extracted from the PGC-MDD data for use as genetic instruments when testing the causal effect of cannabis use (the exposure) on depression (the outcome).

As both cannabis use and depression are binary outcomes, the effect estimates represent the odds for the outcome per unit increase in the log OR for the exposure. To assist with interpretation, these effect estimates were converted (by multiplying log ORs by 0.693 and then exponentiating) to give the OR per doubling in odds of the binary exposure50.

Results

Univariate analyses within UK Biobank

Of the 126291 individuals with both mental health questionnaire and genetic data available, 43.8% are male, 60.2% have ever smoked and 60.9% are high frequency alcohol drinkers. Table 1 shows the prevalence of each of the seven phenotypes of interest. For cannabis use phenotypes, cases are more likely than controls to be male and high frequency alcohol users. For depression and self-harm phenotypes, cases are less likely than controls to be male and high frequency alcohol users. For all phenotypes, cases more frequently report smoking (Supplementary Materials Section 1)

View this table:
Table 1: Prevalence of traits in UK Biobank

We performed GWA analyses for each of the seven traits within UK Biobank (Supplementary Materials Section 1) and then using these statistics estimated the SNP-heritability of each trait, using LD-score regression. These results are reported on the liability scale. For both cannabis use and depression, the more severe phenotype (continued cannabis use and recurrent depression, respectively) have higher SNP-heritability estimates. (Supplementary Materials Section 1 includes estimates for over-or under-sampling scenarios).

Bivariate analyses

Phenotypic relationships within UK Biobank

The prevalence of all depression and self-harm phenotypes is significantly higher amongst cannabis users than non-users (Table 2). Furthermore, the associations are stronger for more severe phenotypes (continued vs initial cannabis use and recurrent vs single episode depression), as illustrated in Figure 2. Cannabis use phenotypes are most strongly associated with self-harm. There is a strong relationship between depression and self-harm (OR=10.13, 95% CI=9.40-10.93). These patterns of association were also generally seen in each of the stratified analyses (Supplementary Materials Section 2 has details and exceptions).

View this table:
Table 2: Phenotypic relationship between traits within UK Biobank
Figure 1:
Figure 1:

Bivariate Genetic Methods Overview. PGC-MDD = Psychiatric Genomics Consortium Major Depressive Disorder, including 23andMe; ICC = International Cannabis Consortium; UKB = UK Biobank

Figure 2:
Figure 2:

Phenotypic relationship between traits within UK Biobank

The phenotypic findings do not consider the direction of effects; it cannot be determined whether one factor is on the causal pathway to another or if there are upstream factors acting on all of the phenotypes under consideration.

Genetic correlations

Maximum available sample size

We find a genetic correlation between cannabis use and depression of rg=0.289, SE=0.036, p=1.45×10−15 using our meta-analysis of ICC+UKB (N=158,621, SNP-h2=0.116, SE=0.009) and summary statistics made available by the PGC-MDD and 23andMe (N=480,359, SNP-h2=0.087, SE=0.004).

Detailed phenotypes within UK Biobank

We observe high within-trait genetic correlations between initial and continued cannabis use (rg=0.885, SE=0.083, p=1.03×10−26) and single episode and recurrent depression (rg=0.941, SE=0.098, p=1.02×10−21). In cross-trait analyses, as with phenotypic observations, all cannabis use traits are modestly and significantly correlated with depression traits (rg=0.255-0.345, p=1.16×10−2-2.36×10−7) and self-harm (rg=0.298-0.308, p=1.17×10−3-5.44×10−5). Depression phenotypes show a high genetic correlation with self-harm (rg=0.681-0.745, p=8.21×10−7-1.68×10−20). However, in contrast to our phenotypic findings, cross-trait genetic correlations do not show the same pattern of varying magnitude across different definitions of cannabis use and depression within UK Biobank (Figure 3, Supplementary Materials Section 3).

Figure 3:
Figure 3:

Genetic Correlations within UK Biobank, calculated using LD-score regression

Polygenic risk scores

The two calculated polygenic risk scores for depression and cannabis use were tested against the seven phenotypes of interest in the 50% UK Biobank target sample (Figure 4). The results report Pthreshold=0.3 as a representative value, full findings in Supplementary Materials Section 4. A Bonferroni correction for the two PRS and 7 traits (which does not consider the correlation between these traits) gives a significance threshold of p=0.05/(2×7), or p=3.57×10−3.

Figure 4:
Figure 4:

Polygenic risk scoring in UK Biobank

Polygenic risk for cannabis

Looking within-trait, PRS(CANN) accounts for a significant proportion of the variance in cannabis use in the UK Biobank target sample (R2=0.82%, p=6.071×10−62). PRS(CANN) explains a greater proportion of variance in continued cannabis use (R2=1.2%, p=4.697×10−45) than initial cannabis use (R2=0.53%, p=2.609×10−32). In cross-trait analyses, PRS(CANN) accounts for a small proportion of variance in depression (R2=0.043%, p=5.058×10−4) and does not reach significance when considering single episode (R2=0.037%, p=1.211×10−2, NS) or recurrent depression (R2=0.037%, p=5.180×10−3, NS). PRS(CANN) predicts a greater proportion of variance for self-harm than depression (R2=0.21%, p=2.337×10−7).

Polygenic risk for depression

PRS(DEP) accounts for a significant proportion of variance in depression in the UK Biobank target sample (R2=1.7%, p=1.176×10−104) and accounts for more variance in recurrent depression (R2=2.1%, p=9.173×10−98) than single episode depression (R2=0.78%, p=2.792×10−30). PRS(DEP) also accounts for R2=1.1% of the variance in self-harm (p=2.427×10x−31). Looking at cannabis phenotypes, PRS(DEP) accounts for a significant but smaller proportion of the variance in cannabis use (R2=0.16%, p=2.144×10−13), initial cannabis use (R2=0.14%, p=7.589×10−10) or continued cannabis use (R2=0.13%, p=4.393×10−6).

Results are consistent with PRS(CANN) being less well powered than PRS(DEP) due to smaller discovery sample size, but overall, findings broadly reflect those patterns identified using phenotypic and LD-score approaches, showing that there is evidence of association between cannabis and depression phenotypes and each of these traits are also related to self-harm.

Mendelian randomisation

Two-sample MR analyses results are shown in Table 3, and Supplementary Materials Section 5.

View this table:
Table 3: Two-sample Mendelian randomisation results

Causal effect of depression on cannabis use

An MR analysis using 39 SNPs previously association with depression (reaching genome-wide significance in the published PGC-MDD and 23andMe GWA) indicated no evidence of association with cannabis use, using the ICC GWA cannabis use summary statistics. Both IVW and Egger models did not reach significance (p>0.01), with no evidence of directional horizontal pleiotropy, as indicated by the MR-Egger intercept.

Causal effect of cannabis use on depression

Nineteen independent SNPs associated with cannabis use (reaching suggestive significance p<1×10−5 in the published ICC cannabis use GWA) were used to test for evidence of cannabis use having a causal effect on depression using the PGC-MDD and 23andMe GWA summary statistics. Again, both IVW and Egger models did not reach significance (p>0.01), with no evidence of directional horizontal pleiotropy.

Discussion

Summary of findings

In UK Biobank, cannabis use is associated with an increased likelihood of depression and particularly self-harm. There is also a genetic overlap between these traits, as demonstrated using cross-trait LD-score regression. Polygenic risk for depression is associated with depression, self-harm and cannabis use, however cross-trait results using polygenic scores for cannabis use are more equivocal. We explored how these phenotypic and genetic relationships vary for initial or continued cannabis use and for single episode or recurrent depression. Phenotypically, cross-trait associations are stronger for more severe trait subtypes. However, this pattern is not evident in the genetic analyses. Finally, as these analyses cannot determine the direction of effects, Mendelian randomisation was implemented using published GWA summary statistics. We did not find evidence to support causality between cannabis use and depression in either direction, although we note that these MR analyses are underpowered.

Phenotypic and genetic relationships between cannabis use, depression and self-harm

Phenotypic associations between cannabis use and depression are significant but the association between cannabis use phenotypes and self-harm is stronger than those observed with depression, mirroring reports in other populations16,19,23. Stratified analyses suggest that these associations are not driven by differences in sex, smoking or alcohol consumption.

Significant genetic correlations between cannabis use, depression and self-harm are also in line with some15,23,24, but not all16 previous findings in family studies. The genetic correlation between cannabis use and depression estimated here is higher than those previously published using the same methodology13, likely reflecting the increased power of more recent depression GWA analyses. Indeed, the genetic correlation observed here between cannabis use and depression also exceeds that reported between cannabis use and schizophrenia (rg=0.245, SE=0.031, p=5.81×10−15)13.

Polygenic risk scoring methods show that PRS(DEP) accounts for a small but significant proportion of the variance in each of the cannabis phenotypes. The PRS(CANN) is less powerful (shown by within-trait prediction) and whilst significantly predicting depression, fails to reach significance for either single episode or recurrent depression. Both PRS(DEP) and PRS(CANN) account for a significant proportion of the variation in self-harm.

Significant PRS show the importance of cross-trait polygenic effects, however they explain a very small proportion of trait variance. This is expected; PRS is an out-of-sample prediction method and current GWA samples have limited power to capture the small genetic effects for complex traits like cannabis use and depression. As GWA sample sizes continue to grow, we predict polygenic risk scores should explain an increasing proportion of trait variance, and so will become more useful as a tool for individual prediction. In the meantime, the convergent genetic results from both LD-score regression and polygenic risk scoring methods both indicate shared genetic influences on cannabis use, depression and self-harm that warrant further attention.

The impact of trait definitions

Whilst all phenotypic relationships tested are significant, our observation that associations are stronger for more severe trait definitions is consistent with previous reports16,19. Our SNP-h2 estimates of depression and cannabis use are also inline with previous findings13,14, but we show that continued cannabis use is a more heritable trait than initial cannabis use and that there is a strong genetic correlation between these two. This suggests that focussing on continued cannabis use may be beneficial for genetic analyses.

Unlike phenotypic patterns, the strength of genetic relationship does not vary with the trait definition, using LD-score or PRS cross-trait approaches. This may reflect a difference in the phenotypic versus the genetic relationships for these traits, however it may also be due to limitations in the statistical power of genetic analyses to differentiate between subtypes.

Irrespective, there is still substantial evidence of both phenotypic and genetic associations across all traits tested; even initial cannabis use is associated with an increased risk of depression and self-harm, underscoring the importance of work examining the causal mechanisms linking these phenotypes.

Directions of causality between cannabis use and depression

The direction of causality is of particular interest, given current debates surrounding the legalisation and safety of using cannabis. Longitudinal approaches have been inconclusive5,9 and MR offers an alternative genetic-based method to test causality. Here we found no evidence for causality in either direction, but our analyses were underpowered, relying on suggestive but not significant SNPs as instrumental variables for cannabis use. This is in part because we were unable to include the genome-wide significant variants implicated in most recent and largest cannabis GWAS13, given issues of bias and sample overlap.

Determining the direction of causality linking cannabis use and depression is key to understanding the potential risks of cannabis use, with associated health policy implications. It is most likely that there is a network of both genetic and environmental factors linking cannabis use, depression and self-harm which will vary between individuals.

Conclusions

Here we show that cannabis use is phenotypically and genetically associated with depression and self-harm. Whilst the phenotypic relationship is stronger for more severe phenotypes, there is a significant increase in risk for those who use cannabis less than ten times. However, we are unable to draw conclusions regarding the direction of causality linking cannabis use and depression. As has previously been noted, there is more commonly a focus on the relationship between cannabis and psychosis-related phenotypes9. This work highlights that when investigating the associations of cannabis use with mental health, both depression and self-harm are important and prevalent traits to consider.

Acknowledgements

The study represents independent research funded by the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care. High performance computing facilities were funded with capital equipment grants from the GSTT Charity (TR130505) and Maudsley Charity (980). Analysis using UK Biobank data was conducted under UK Biobank application number 18177. We would like to thank the research participants and employees of 23andMe for making this work possible. We also thank the members of the International Cannabis Consortium for making summary data available, enabling this work.

Footnotes

  • Declaration of competing interests: No competing interests to be declared.

References

  1. 1.↵
    United Nations Office on Drugs and Labor. World Drug Report 2018. S.l.: UNITED NATIONS, 2018.
  2. 2.↵
    Ferrari AJ, Charlson FJ, Norman RE, et al. Burden of Depressive Disorders by Country, Sex, Age, and Year: Findings from the Global Burden of Disease Study 2010. PLOS Medicine 2013; 10: e1001547.
    OpenUrl
  3. 3.↵
    Martins SS, Gorelick DA. Conditional substance abuse and dependence by diagnosis of mood or anxiety disorder or schizophrenia in the U.S. population. Drug Alcohol Depend 2011; 119: 28–36.
    OpenUrlCrossRefPubMedWeb of Science
  4. 4.
    Blanco C, Hasin DS, Wall MM, et al. Cannabis Use and Risk of Psychiatric Disorders: Prospective Evidence From a US National Longitudinal Study. JAMA Psychiatry 2016; 73: 388–395.
    OpenUrl
  5. 5.↵
    Lev-Ran S, Roerecke M, Le Foll B, et al. The association between cannabis use and depression: A systematic review and meta-analysis of longitudinal studies. Psychological medicine 2014; 44: 797–810.
    OpenUrlCrossRefPubMed
  6. 6.
    Chen C-Y, Wagner FA, Anthony JC. Marijuana use and the risk of Major Depressive Episode. Epidemiological evidence from the United States National Comorbidity Survey. Social psychiatry and psychiatric epidemiology 2002; 37: 199–206.
    OpenUrlCrossRefPubMedWeb of Science
  7. 7.↵
    Horwood LJ, Fergusson DM, Coffey C, et al. Cannabis and depression: An integrative data analysis of four Australasian cohorts. Drug Alcohol Depend 2012; 126: 369–378.
    OpenUrlCrossRefPubMedWeb of Science
  8. 8.↵
    Okaneku J, Vearrier D, McKeever RG, et al. Change in perceived risk associated with marijuana use in the United States from 2002 to 2012. Clin Toxicol (Phila) 2015; 53: 151–155.
    OpenUrlCrossRefPubMed
  9. 9.↵
    Gage SH, Hickman M, Heron J, et al. Associations of Cannabis and Cigarette Use with Depression and Anxiety at Age 18: Findings from the Avon Longitudinal Study of Parents and Children. PLOS ONE 2015; 10: e0122896.
    OpenUrlCrossRefPubMed
  10. 10.↵
    Feingold D, Weiser M, Rehm J, et al. The association between cannabis use and mood disorders: A longitudinal study. J Affect Disord 2015; 172: 211–218.
    OpenUrlCrossRefPubMed
  11. 11.↵
    Verweij KJH, Zietsch BP, Lynskey MT, et al. Genetic and environmental influences on cannabis use initiation and problematic use: A meta-analysis of twin studies. Addiction 2010; 105: 417–430.
    OpenUrlCrossRefPubMed
  12. 12.↵
    Sullivan PF, Neale MC, Kendler KS. Genetic epidemiology of major depression: Review and meta-analysis. The American journal of psychiatry 2000; 157: 1552–62.
    OpenUrlCrossRefPubMedWeb of Science
  13. 13.↵
    Pasman JA, Verweij KJH, Gerring Z, et al. GWAS of lifetime cannabis use reveals new risk loci, genetic overlap with psychiatric traits, and a causal influence of schizophrenia. Nature Neuroscience 2018; 21: 1161–1170.
    OpenUrl
  14. 14.↵
    Wray NR, Ripke S, Mattheisen M, et al. Genome-wide association analyses identify 44 risk variants and refine the genetic architecture of major depression. Nat Genet 2018; 50: 668–681.
    OpenUrlCrossRefPubMed
  15. 15.↵
    Hodgson K, Almasy L, Knowles EEM, et al. The genetic basis of the comorbidity between cannabis use and major depression. Addiction 2017; 112: 113–123.
    OpenUrl
  16. 16.↵
    Agrawal A, Nelson EC, Bucholz KK, et al. Major depressive disorder, suicidal thoughts and behaviours, and cannabis involvement in discordant twins: A retrospective cohort study. The Lancet Psychiatry 2017; 4: 706–714.
    OpenUrl
  17. 17.↵
    Carey CE, Agrawal A, Bucholz KK, et al. Associations between Polygenic Risk for Psychiatric Disorders and Substance Involvement. Front Genet; 7. Epub ahead of print 2016. DOI: 10.3389/fgene.2016.00149.
  18. 18.↵
    Psychiatric GWAS Consortium MDDWG of the. A mega-analysis of genome-wide association studies for major depressive disorder. Mol Psychiatry; 18. Epub ahead of print April 2013. DOI: 10.1038/mp.2012.21.
  19. 19.↵
    Rossow I, Hawton K, Ystgaard M. Cannabis use and deliberate self-harm in adolescence: A comparative analysis of associations in England and Norway. Arch Suicide Res 2009; 13: 340–348.
    OpenUrlPubMed
  20. 20.
    Giletta M, Scholte RH, Engels RC, et al. Adolescent non-suicidal self-injury: A cross-national study of community samples from Italy, the Netherlands and the United States. Psychiatry Res 2012; 197: 66–72.
    OpenUrlCrossRefPubMedWeb of Science
  21. 21.↵
    Mars B, Heron J, Crane C, et al. Clinical and social outcomes of adolescent self harm: Population based birth cohort study. BMJ 2014; 349: g5954.
    OpenUrlAbstract/FREE Full Text
  22. 22.↵
    van Ours JC, Williams J, Fergusson D, et al. Cannabis use and suicidal ideation. J Health Econ 2013; 32: 524–537.
    OpenUrl
  23. 23.↵
    Delforterie M, Lynskey M, Huizink A, et al. The relationship between cannabis involvement and suicidal thoughts and behaviors. Drug and Alcohol Dependence 2015; 150: 98–104.
    OpenUrl
  24. 24.↵
    Few LR, Grant JD, Nelson EC, et al. Cannabis Involvement and Nonsuicidal Self-Injury: A Discordant Twin Approach. J Stud Alcohol Drugs 2016; 77: 873–880.
    OpenUrl
  25. 25.↵
    Lynskey MT, Glowinski AL, Todorov AA, et al. Major depressive disorder, suicidal ideation, and suicide attempt in twins discordant for cannabis dependence and early-onset cannabis use. Arch Gen Psychiatry 2004; 61: 1026–1032.
    OpenUrlCrossRefPubMedWeb of Science
  26. 26.↵
    Nutt D, King LA, Saulsbury W, et al. Development of a rational scale to assess the harm of drugs of potential misuse. The Lancet 2007; 369: 1047–1053.
    OpenUrl
  27. 27.↵
    Hall W, Pacula RL. Cannabis Use and Dependence. Cambridge, UK: Cambridge University Press, https://www.rand.org/pubs/commercial_books/CB405.html (2003, accessed 18 September 2018).
  28. 28.↵
    Allen NE, Sudlow C, Peakman T, et al. UK biobank data: Come and get it. Sci Transl Med 2014; 6: 224ed4.
    OpenUrlFREE Full Text
  29. 29.↵
    Bycroft C, Freeman C, Petkova D, et al. Genome-wide genetic data on ∼500,000 UK Biobank participants. bioRxiv 2017; 166298.
  30. 30.↵
    Stringer S, Minică CC, Verweij KJH, et al. Genome-wide association study of lifetime cannabis use based on a large meta-analytic sample of 32 330 subjects from the International Cannabis Consortium. Transl Psychiatry 2016; 6: e769.
    OpenUrl
  31. 31.↵
    Bulik-Sullivan B, Finucane HK, Anttila V, et al. An atlas of genetic correlations across human diseases and traits. Nat Genet 2015; 47: 1236–1241.
    OpenUrlCrossRefPubMed
  32. 32.↵
    Bulik-Sullivan BK, Loh P-R, Finucane HK, et al. LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nature Genetics 2015; 47: 291–295.
    OpenUrlCrossRefPubMed
  33. 33.↵
    Euesden J, Lewis CM, O’Reilly PF. PRSice: Polygenic Risk Score software. Bioinformatics 2015; 31: 1466–1468.
    OpenUrlCrossRefPubMed
  34. 34.↵
    Davis KAS, Coleman JRI, Adams M, et al. Mental health in UK Biobank: Development, implementation and results from an online questionnaire completed by 157 366 participants. BJPsych Open 2018; 4: 83–90.
    OpenUrlCrossRef
  35. 35.↵
    McCarthy S, Das S, Kretzschmar W, et al. A reference panel of 64,976 haplotypes for genotype imputation. Nat Genet 2016; 48: 1279–1283.
    OpenUrlCrossRefPubMed
  36. 36.↵
    Consortium TU. The UK10K project identifies rare variants in health and disease. Nature 2015; 526: 82–90.
    OpenUrlCrossRefPubMed
  37. 37.↵
    Abraham G, Qiu Y, Inouye M. FlashPCA2: Principal component analysis of Biobank-scale genotype datasets. Bioinformatics 2017; 33: 2776–2778.
    OpenUrlCrossRefPubMed
  38. 38.↵
    Turner SD. Qqman: An R package for visualizing GWAS results using Q-Q and manhattan plots. The Journal of Open Source Software. Epub ahead of print 19 May 2018. DOI: 10.21105/joss.00731.
  39. 39.↵
    Chang CC, Chow CC, Tellier LC, et al. Second-generation PLINK: Rising to the challenge of larger and richer datasets. Gigascience 2015; 4: 7.
    OpenUrlCrossRefPubMed
  40. 40.↵
    Winkler TW, Day FR, Croteau-Chonka DC, et al. Quality control and conduct of genome-wide association meta-analyses. Nat Protoc 2014; 9: 1192–1212.
    OpenUrlCrossRefPubMed
  41. 41.↵
    Willer CJ, Li Y, Abecasis GR. METAL: Fast and efficient meta-analysis of genomewide association scans. Bioinformatics 2010; 26: 2190–2191.
    OpenUrlCrossRefPubMedWeb of Science
  42. 42.↵
    Kuhn M. Building Predictive Models in R Using the Caret Package. Journal of Statistical Software; 28. Epub ahead of print 2008. DOI: 10.18637/jss.v028.i05.
  43. 43.↵
    Choi SW, Mak TSH, O’Reilly P. A guide to performing Polygenic Risk Score analyses. bioRxiv 2018; 416545.
  44. 44.↵
    Lee SH, Goddard ME, Wray NR, et al. A Better Coefficient of Determination for Genetic Profile Analysis. Genetic Epidemiology 2012; 36: 214–224.
    OpenUrlCrossRefPubMed
  45. 45.↵
    Nivard MG, Dolan CV, Kendler KS, et al. Stability in symptoms of anxiety and depression as a function of genotype and environment: A longitudinal twin study from ages 3 to 63 years. Psychol Med 2015; 45: 1039–1049.
    OpenUrlCrossRefPubMed
  46. 46.↵
    Penninx BWJH, Beekman ATF, Smit JH, et al. The Netherlands Study of Depression and Anxiety (NESDA): Rationale, objectives and methods. Int J Methods Psychiatr Res 2008; 17: 121–140.
    OpenUrlCrossRefPubMedWeb of Science
  47. 47.↵
    Cuellar-Partida G, Springelkamp H, Lucas SEM, et al. WNT10A exonic variant increases the risk of keratoconus by decreasing corneal thickness. Hum Mol Genet 2015; 24: 5060–5068.
    OpenUrlCrossRefPubMed
  48. 48.↵
    Burgess S, Davies NM, Thompson SG. Bias due to participant overlap in two-sample Mendelian randomization. Genet Epidemiol 2016; 40: 597–608.
    OpenUrlCrossRefPubMed
  49. 49.↵
    Hemani G, Zheng J, Elsworth B, et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife; 7. Epub ahead of print 30 May 2018. DOI: 10.7554/eLife.34408.
  50. 50.↵
    Gage SH, Jones HJ, Burgess S, et al. Assessing causality in associations between cannabis use and schizophrenia risk: A two-sample Mendelian randomization study. Psychol Med 2017; 47: 971–980
    OpenUrlCrossRefPubMed
Back to top
Posted February 14, 2019.
Download PDF
Email
Cannabis use, depression and self-harm: phenotypic and genetic relationships
K Hodgson, JRI Coleman, SP Hagenaars, KL Purves, K Glanville, SW Choi, P O’Reilly, G Breen, Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium, CM Lewis
bioRxiv 549899; doi: https://doi.org/10.1101/549899
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area