https://helda.helsinki.fi
GWAS of lifetime cannabis use reveals new risk loci, genetic
overlap with psychiatric traits, and a causal influence of schizophrenia
23andMe Res Team
2018-09
23andMe Res Team , Subst Use Disorders Working Grp Ps & Int Cannabis Consortium
2018 , ' GWAS of lifetime cannabis use reveals new risk loci, genetic overlap with psychiatric
traits, and a causal influence of schizophrenia ' , Nature Neuroscience , vol. 21 , no. 9 , pp.
1161-+ . https://doi.org/10.1038/s41593-018-0206-1
http://hdl.handle.net/10138/305510
https://doi.org/10.1038/s41593-018-0206-1
unspecified
publishedVersion
Downloaded from Helda, University of Helsinki institutional repository.
This is an electronic reprint of the original article.
This reprint may differ from the original in pagination and typographic detail.
Please cite the original version.
Articles
https://doi.org/10.1038/s41593-018-0206-1
C
annabis is a widely used psychoactive substance, and its use
is associated with various adverse mental health outcomes,
including psychosis and schizophrenia
1–3
. Successful preven-
tion and intervention efforts aimed at reducing cannabis use, mis-
use and related outcomes require a better understanding of why
some people use cannabis whereas others do not. Lifetime cannabis
use, defined as any use of cannabis during lifetime, is a heritable
trait: a meta-analysis of twin studies
4
estimated the heritability to
be approximately 45%. Twin studies have shown there is substantial
overlap in the genetic factors influencing cannabis use and those
underlying problematic cannabis use (abuse or dependence)
5,6
.
Several GWASs have tried to identify genetic variants underlying
cannabis use phenotypes
7–11
. Recently, Demontis et al.
11
performed
the largest GWAS for cannabis use disorder to date, with a discov
-
ery sample of 2,387 cases and almost 50,000 controls, plus a rep-
lication sample of 5,501 cases and ~300,000 controls. They found
one genome-wide significant risk locus for cannabis use disorder, a
single nucleotide polymorphism (SNP) that is a strong marker for
CHRNA2 expression. Their follow-up analyses showed that canna
-
bis-dependent individuals had a decreased expression of this gene
in the cerebellum, as well as in other brain regions.
The largest GWAS of lifetime cannabis use to date is from the
International Cannabis Consortium (ICC) and is based on a sample
size of 32,330 individuals in the discovery sample along with 5,627
individuals in the replication sample
10
. Although no individual
SNPs reached genome-wide significance, gene-based tests identi
-
fied four genes significantly associated with lifetime cannabis use:
NCAM1, CADM2, SCOC and KCNT2. Notably, NCAM1 has previ
-
ously been linked to other substance use phenotypes (for example,
refs
12,13
), and following publication of the study, CADM2 was found
to be associated with alcohol consumption
14
, personality
15
, repro-
ductive success and risk-taking behavior
16
in other GWASs. These
results indicate that CADM2 may play a role in a broader personal
-
ity profile of sensation-seeking and risk-taking behavior in general.
GWAS of lifetime cannabis use reveals new risk
loci, genetic overlap with psychiatric traits, and
a causal influence of schizophrenia
Joëlle A. Pasman
1,37
, Karin J. H. Verweij
1,2,37
, Zachary Gerring
3
, Sven Stringer
4
,
Sandra Sanchez-Roige
5
, Jorien L. Treur
6
, Abdel Abdellaoui
2
, Michel G. Nivard
7
,
Bart M. L. Baselmans
7
, Jue-Sheng Ong
3
, Hill F. Ip
7
, Matthijs D. van der Zee
7
, Meike Bartels
7
,
Felix R. Day
8
, Pierre Fontanillas
9
, Sarah L. Elson
9
, the 23andMe Research Team
10
, Harriet de Wit
11
,
Lea K. Davis
12
, James MacKillop
13
, The Substance Use Disorders Working Group of the
Psychiatric Genomics Consortium
14
, International Cannabis Consortium
15
, Jaime L. Derringer
16
,
Susan J. T. Branje
17
, Catharina A. Hartman
18
, Andrew C. Heath
19
, Pol A. C. van Lier
20
,
Pamela A. F. Madden
19
, Reedik Mägi
21
, Wim Meeus
17
, Grant W. Montgomery
22
, A. J. Oldehinkel
18
,
Zdenka Pausova
23
, Josep A. Ramos-Quiroga
24,25,26,27
, Tomas Paus
28,29
, Marta Ribases
24,25,26
,
Jaakko Kaprio
30
, Marco P. M. Boks
31
, Jordana T. Bell
32
, Tim D. Spector
32
, Joel Gelernter
33
,
Dorret I. Boomsma
7
, Nicholas G. Martin
3
, Stuart MacGregor
3
, John R. B. Perry
8
,
Abraham A. Palmer
5,34
, Danielle Posthuma
4
, Marcus R. Munafò
6,35
, Nathan A. Gillespie
3,36,38
,
Eske M. Derks
3,38
and Jacqueline M. Vink
1,38
*
Cannabis use is a heritable trait that has been associated with adverse mental health outcomes. In the largest genome-wide
association study (GWAS) for lifetime cannabis use to date (N = 184,765), we identified eight genome-wide significant inde-
pendent single nucleotide polymorphisms in six regions. All measured genetic variants combined explained 11% of the variance.
Gene-based tests revealed 35 significant genes in 16 regions, and S-PrediXcan analyses showed that 21 genes had different
expression levels for cannabis users versus nonusers. The strongest finding across the different analyses was CADM2, which
has been associated with substance use and risk-taking. Significant genetic correlations were found with 14 of 25 tested sub-
stance use and mental health–related traits, including smoking, alcohol use, schizophrenia and risk-taking. Mendelian ran-
domization analysis showed evidence for a causal positive influence of schizophrenia risk on cannabis use. Overall, our study
provides new insights into the etiology of cannabis use and its relation with mental health.
A full list of affiliations appears at the end of the paper.
NATURE NEUROSCIENCE | VOL 21 | SEPTEMBER 2018 | 1161–1170 | www.nature.com/natureneuroscience
1161
Articles
Nature NeuroscieNce
Here we present a GWAS of lifetime cannabis use on a substantially
larger sample, providing more power to identify genetic variants.
As mentioned above, cannabis use has been linked to a variety of
mental health outcomes, including substance abuse and dependence
and psychiatric disorders
3
. In particular, the relationship between
cannabis use and schizophrenia has been the subject of intensive
research and debate. It has long been established that the prevalence
of cannabis use is higher in patients with schizophrenia
17,18
. A sub-
stantial body of evidence supports the hypothesis that cannabis use
increases the risk for developing psychoses and schizophrenia
19
, but
other hypotheses (namely, schizophrenia increases the use of canna
-
bis, or the association is due to (genetic) pleiotropy) have also been
posed. Previous studies have shown that genetic risk factors for can
-
nabis use and schizophrenia are positively correlated
20,21
. However, a
genetic correlation does not provide insight in the direction of cau
-
sation. With Mendelian randomization it is possible to examine the
causality of the association between cannabis use and schizophrenia,
and recently it has become possible to apply this method using sum
-
mary statistics from GWASs
22
. Previous Mendelian randomization
studies have investigated the link between lifetime cannabis use and
schizophrenia, but findings were inconsistent. Vaucher et al.
23
tested
for causal effects from cannabis use to schizophrenia and found evi
-
dence for a causal influence of cannabis use on schizophrenia risk.
Gage et al.
24
tested bidirectional effects and found weak evidence for
a causal effect of cannabis use on schizophrenia and much stronger
evidence for a causal effect in the other direction. The results from
our GWAS provide more power to examine the causal association
between cannabis use and schizophrenia.
Here we report the largest GWAS yet for lifetime cannabis use.
We increased the sample size substantially by meta-analyzing
GWAS results from the ICC study (N = 35,297), along with new data
from the UK Biobank (N = 126,785) and 23andMe (N = 22,683).
The combined sample size of this study was N = 184,765, five times
as large as the previous largest GWAS on lifetime cannabis use. We
tested the association of millions of SNPs with lifetime cannabis use
and estimated the heritability of lifetime cannabis use based on all
SNPs. Tests of association for individual genetic variants were com
-
plemented with gene-based tests of association and S-PrediXcan
analysis. The latter was used to identify genes with differential
expression levels in cannabis users versus nonusers. We further
estimated the genetic correlation of lifetime cannabis use with
other traits, including use of other substances and mental health
traits, such as schizophrenia. Lastly, we performed bidirectional
two-sample Mendelian randomization analysis to examine whether
there was evidence for a causal relationship from cannabis use to
schizophrenia and vice versa.
Results
Genome-wide association meta-analysis. The meta-analysis
resulted in eight independent genome-wide significant SNP
associations (linkage disequilibrium (LD) R
2
< 0.1, window size
250 kb) on chromosomes 3, 7, 8, 11, 16 and 17 (Fig. 1, Table 1 and
Supplementary Table 1). The top SNP and two other independent
associations were located in CADM2 on chromosome 3 (rs2875907,
P = 9.38 × 10
–17
; rs1448602, P = 6.55 × 10
–11
; rs7651996, P = 2.37 × 10
–9
).
Other hits were located in ZNF704, SDK1, NCAM1, RABEP2 or
ATP2A1 and SMG6 (Fig. 2). All SNPs combined explained 11%
(
h
SNP
2
= 0.11, s.e. = 0.01) of the individual differences in lifetime
cannabis use. Supplementary Figs. 1–3 and Supplementary Table 2
provide information on results of the individual GWASs (ICC, UK
Biobank and 23andme).
Gene-based test of association and expression. Gene-based tests
of associations in MAGMA
25
identified 35 genes genome-wide
significantly associated with lifetime cannabis use (Fig. 3, Table 2,
Supplementary Fig. 4 and Supplementary Table 3). These genes
were located in 5 regions that were already identified in the SNP-
based analysis (including those containing CADM2 and NCAM1)
and in 11 other regions (Supplementary Fig. 5).
S-PrediXcan analysis
26
revealed 133 Bonferroni-corrected sig-
nificant associations across tissues targeting 21 unique genes
(Supplementary Tables 4 and 5). Eight genes were also signifi
-
cant in the gene-based test, whereas 13 were newly identified.
For genes identified in multiple tissues, directions of effects were
largely consistent across tissues (Supplementary Fig. 6). Again, the
most significant finding was CADM2; genetic variants associated
with increased liability to use cannabis are predicted to upregulate
expression levels of CADM2 in eight nonbrain tissues, including
whole blood (z = 5.88, P = 4.17 × 10
–9
). Of note, although CADM2 is
expressed more widely in brain than in other tissues (Supplementary
Fig. 7), the top SNP, rs2875907, regulates the expression of CADM2
only in nonbrain tissues (Supplementary Fig. 8). Exploration of
S-PrediXcan results in UK Biobank data (https://imlab.shinyapps.
io/gene2pheno_ukb_neale/) showed that CADM2 expression is
significantly associated with multiple traits, including increased
risk-taking, body mass index and reduced feelings of anxiety. Like
the SNP- and gene-based tests of association, the S-PrediXcan
analysis detected a strong signal in a high-LD region at 16p11.2.
Supplementary Table 3 provides an overview of all genes that were
identified in the gene-based test of association and the S-PrediXcan
analyses, along with information about the gene product and previ
-
ously identified associations with the gene.
Genetic correlations with other traits. Using our GWAS results
and those of other GWASs, we estimated the genetic correlation of
lifetime cannabis use with 25 traits of interest, including substance
use, personality and mental health phenotypes. Fourteen traits were
significantly genetically correlated with lifetime cannabis use after
correction for multiple testing (Fig. 4 and Supplementary Table 6).
Positive genetic correlations were found with substance use pheno
-
types, including smoking and alcohol use and dependence, as well
as with mental health phenotypes, including ADHD and schizo
-
phrenia. Furthermore, positive genetic correlations were found
with risk-taking behavior, openness to experience, and educational
attainment, as well as a negative correlation with conscientiousness.
Causal association between cannabis use and schizophrenia:
two-sample Mendelian randomization. A positive genetic correla
-
tion was found between genetic risk factors for cannabis use and
schizophrenia (r
g
= 0.24, s.e. = 0.03, P < 0.01). To examine whether
there was evidence for a causal effect of cannabis use on schizo
-
phrenia risk and vice versa, we performed bidirectional two-sample
Mendelian randomization analysis
22
. In our main analysis, inverse-
variance-weighted (IVW) regression analysis, we found some weak
(nonsignificant) evidence for a causal influence of lifetime canna
-
bis use on schizophrenia risk, but only for the genetic instrument
containing SNPs associated with cannabis use under the P-value
threshold 1 × 10
–5
. The IVW regression odds ratio was 1.10 (95%
confidence interval (CI) 0.99–1.21, P = 0.074). We found stronger
evidence for a causal positive influence of schizophrenia risk on life
-
time cannabis use, the IVW regression odds ratio being 1.16 (95%
CI 1.06–1.27, P = 0.001; see Table 3, Supplementary Figs. 9 and 10,
and Supplementary Tables 7–9).
To determine the robustness of these findings, we performed
four sensitivity analyses that rely on distinct assumptions regard
-
ing instrument validity. The sensitivity analyses showed a consistent
pattern supporting weak evidence for a causal effect of cannabis use
on schizophrenia and strong evidence for a causal effect of schizo
-
phrenia on cannabis use (Table 3). As an exception, the evidence
provided by MR-Egger SIMEX (Mendelian randomization Egger
simulation extrapolation) for a causal relation from schizophrenia
risk to cannabis use was very weak. However, since the Egger intercept
NATURE NEUROSCIENCE | VOL 21 | SEPTEMBER 2018 | 1161–1170 | www.nature.com/natureneuroscience
1162
Articles
Nature NeuroscieNce
was not significantly different from 0 (Supplementary Table 10),
indicating no pleiotropic effects for the SNPs included in the genetic
instruments
27
, it is likely that this method simply lacked power to be
able to reject the null hypothesis of no causal effect
28
.
Discussion
SNP- and gene-based tests revealed several SNPs and genes strongly
associated with lifetime cannabis use. Overall, 11% of the varia
-
tion in the phenotype was explained by the combined effect of
SNPs, which amounts to approximately 25% of twin-based heri
-
tability estimates
4
. CADM2 and NCAM1, both identified in the
original ICC meta-analysis
10
, were among the strongest findings in
the SNP-based and gene-based tests. The CADM2 gene (cell adhe
-
sion molecule 2) is a synaptic cell adhesion molecule and is part of
the immunoglobulin superfamily. Notably, CADM2 has previously
been identified in GWASs of other behavioral phenotypes, includ
-
ing alcohol consumption
14
, processing speed
29
, and number of
offspring and risk-taking behavior
16
. A large-scale phenome-wide scan
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Chromosome
rs2875907
rs10085617
rs9773390
rs9919557
rs10499
rs17761723
–log
10
(P )
0
5
10
15
b
λ = 1.11
Expected –log
10
(P )
0246
0
5
10
15
Observed –log
10
(P )
a
Fig. 1 | Q–Q and Manhattan plot of the GWAS meta-analysis. a, Q–Q plot of the distribution of the –log
10
(P) observed for the SNP associations
with lifetime cannabis use against those expected under the null hypothesis. Expected –log
10
(P) values under the null hypothesis are indicated
by the red line. Genomic inflation is indicated by λ in the plot. There was no evidence for population stratification (LD score regression b
0
= 1.00,
s.e. = 0.007). b, Manhattan plot for the SNP-based GWAS meta-analysis. Results are based on N = 184,765 individuals and N
SNPs
= 11,733,371 SNPs. The
SNP with the lowest P-value for each independent (R
2
< 0.1, window size 250 kb) genome-wide significant locus is annotated by a red circle with rs-
number. The red line represents the conventional genome-wide significance threshold of P < 5 × 10
–8
. The statistical test comprised linear regression;
significance was tested two-sided.
Table 1 | Association results of eight independent SNPs that are significantly associated with lifetime cannabis use
SNP Chr Gene BP A1 A2 Freq A1 N β SE P-value Direction
rs2875907 3p12.1 CADM2 85,518,580 A G 0.352 181,675 0.070 0.009 9.38 × 10
–17
+ + +
rs1448602 3p12.1 CADM2 85,780,454 A G 0.756 184,765 –0.062 0.010 6.55 × 10
–11
– – –
rs7651996 3p12.1 CADM2 85,057,349 T G 0.477 184,765 0.049 0.008 2.37 × 10
–9
+ + +
rs10085617 7p22.2 SDK1 3,634,711 A T 0.416 184,765 0.046 0.008 2.93 × 10
–8
+ + +
rs9773390 8q21.13 ZNF704 81,565,692 T C 0.933 44,595 –0.171 0.029 5.66 × 10
–9
– –?
rs9919557 11q23.2 NCAM1 112,877,408 T C 0.614 180,428 –0.055 0.009 9.94 × 10
–11
– – –
rs10499 16p11.2 RABEP2, ATP2A1 28,915,527 A G 0.651 179,767 0.053 0.009 1.13 × 10
–9
+ + +
rs17761723 17p13.3 SMG6 2,107,090 T C 0.346 184,765 0.047 0.009 3.24 × 10
–8
+ + +
Independent hits were defined as R
2
< 0.01, window size 250 kb. The threshold was set at P < 5 × 10
–8
(conventional genome-wide significant threshold; significance was tested two-sided). Table gives
chromosomal region (Chr), gene the SNP is located in or the nearest gene (within 500 kb), base pair (BP) location SNP on Hg19, allele 1 (A1), allele 2 (A2), frequency of allele 1 (Freq A1), number of
individuals for which variant was included (N), β coefficient of the effect allele A1, standard error (SE) of the β coefficient, and direction for each sample: allele A1 increases (+ ) or decreases (–) liability
for cannabis use, or sample did not contribute to this SNP (?). Order of samples within the Direction column, from left to right: ICC, 23andMe, UK Biobank. Independent SNPs were selected as SNPs
with linkage disequilibrium R
2
< 0.1 using a window size of 250 kb. SNP rs9773390 was not present in the UK Biobank sample and its effect is rather isolated (see Figs. 1b and 2); it might not represent a
robust association.
NATURE NEUROSCIENCE | VOL 21 | SEPTEMBER 2018 | 1161–1170 | www.nature.com/natureneuroscience
1163
Articles
Nature NeuroscieNce
showed that CADM2 was associated with various personality traits,
with the risk variant being associated with reduced anxiety, neuroticism
and conscientiousness and with increased risk-taking
15
. Taken together,
these findings suggest that risk variants in CADM2 are associated with
a broad profile of a risk-taking, optimistic and care-free personality
15
.
Cannabis use has previously been associated with related personality
traits, including high levels of impulsivity and novelty seeking
30,31
.
NCAM1 (neural cell adhesion molecule 1) also encodes a cell
adhesion protein and is member of the immunoglobulin superfam
-
ily. The encoded protein is involved in cell–matrix interactions and
cell differentiation during development
32
. NCAM1 is located in the
NCAM1–TTC12–ANKK1–DRD2 gene cluster, which is related to
neurogenesis and dopaminergic neurotransmission. This gene cluster
has been associated with smoking, alcohol use and illicit drug
use
12,33–35
and has been implicated in psychiatric disorders, such as
schizophrenia and mood disorders
36,37
.
A putatively novel finding comprises the 16p11.2 region
(identified in the SNP and gene-based tests of association and in
S-PrediXcan analysis). Deletions and duplications in this region
have previously been reported to be associated with autism and
schizophrenia
38,39
, while a common 16p11.2 inversion underlies
susceptibility to asthma and obesity
40
. The inversion explains a
substantial proportion of variability in expression of multiple
genes in this region, including TUFM and SH2B1
40
. Given the
high LD in this region and high levels of coexpression of the dif
-
ferentially expressed genes, follow-up studies will be needed to
determine which genes are functionally driving the association
with cannabis use.
Several of the top genes from the gene-based and/or S-PrediXcan
analyses have previously shown an association with other traits,
including schizophrenia (for example, TUFM, NCAM1), body
mass index or obesity (for example, SH2B1, APOBR, ATXN2L),
CADM2
CADM2-AS2
MIR5688
Plotted SNPs
85.2 85.4 85.6 85.8 86
–log
10
(P )
Position on chr3 (Mb)
rs2875907
15
10
5
0
100
80
60
40
20
0
r
2
0.8
0.6
0.4
0.2
Plotted SNPs
–log
10
(P )
Position on chr8 (Mb)
rs9773390
100
80
60
40
20
0
81.4 81.5 81.6 81.7 81.8 81.9 82
10
8
6
4
2
0
ZBTB10
ZNF704
PAG1
r
2
0.8
0.6
0.4
0.2
r
2
0.8
0.6
0.4
0.2
ZPO6
NPIPB6
SBK1
CLN3
SULT1A2
EIF3CL
EIF3C
MIR6862-1
MIR6862-2
APOBR
IL27
SULT1A1
NUPR1
CCDC101
NPIPB9
ATXN2L
ATP2A1
TUFM
MIR4721
SH2B1
LOC100289092
RABEP2
CD19
NFATC2IP
MIR4517
SPNS1
LAT
RRN3P2
SNX29P2
Plotted SNPs
–log
10
(P )
Position on chr16 (Mb)
rs10499
28.2 28.4 28.6 28.8 29 29.2 29.4
10
100
80
60
40
20
0
8
6
4
2
0
10
8
6
4
2
0
r
2
0.8
0.6
0.4
0.2
Position on chr7 (Mb)
Recombination rate (cM/Mb)
rs10085617
3.23.4 3.6 3.84
SDK1
100
80
60
40
20
0
r
2
0.8
0.6
0.4
0.2
Position on chr11 (Mb)
Recombination rate (cM/Mb)
rs9919557
112.7 112.8 112.9 113 113.1 113.2
100
80
60
40
20
0
LOC101928847
NCAM1
NCAM1-AS1
TTC12
10
8
6
4
2
0
RPA1
RTN4RL1
DPH1
HIC1
OVCA2
MIR132
MIR212
SMG6
LOC101927839
SRR
TSR1
SNORD91B
SNORD91A
SGSM2
LOC284009
MNT
METTL16
Position on chr17 (Mb)
Recombination rate (cM/Mb)
rs17761723
1.81.9 22.1 2.22.3 2.4
100
80
60
40
20
0
r
2
0.8
0.6
0.4
0.2
10
8
6
4
2
0
Fig. 2 | Regional plots of the genome-wide significant SNPs. Underlined in yellow are the genes that were significant in the gene-based test (tested
two-sided; P < 2.74 × 10
–6
, Bonferroni corrected threshold of P < 0.05 adjusted for 18,293 tests); underlined in green are the genes that were identified
in the S-PrediXcan analysis only (P < 1.92 × 10
–7
, Bonferroni corrected threshold of P < 0.05 adjusted for 259,825 tests). Colors of the dots indicate
the level of LD (blue for low and red for high LD) with the lead SNP (purple; independent defined as R
2
< 0.1, window size 250 kb).
NATURE NEUROSCIENCE | VOL 21 | SEPTEMBER 2018 | 1161–1170 | www.nature.com/natureneuroscience
1164
