Human Genetics of Addiction: New Insights and Future Directions

Abstract

Purpose of review

With the advent of the genome-wide association study (GWAS), our understanding of the genetics of addiction has made significant strides forward. Here, we summarize genetic loci containing variants identified at genome-wide statistical significance (P<5×10⁻⁸) and independently replicated, review evidence of functional or regulatory effects for GWAS-identified variants, and outline multi-omics approaches to enhance discovery and characterize addiction loci.

Recent findings

Replicable GWAS findings span 11 genetic loci for smoking, eight loci for alcohol, and two loci for illicit drugs combined and include missense functional variants and noncoding variants with regulatory effects in human brain tissues traditionally viewed as addiction-relevant (e.g., prefrontal cortex [PFC]) and, more recently, tissues often overlooked (e.g., cerebellum).

Summary

GWAS have discovered several novel, replicable variants contributing to addiction. Using larger samples sizes from harmonized datasets and new approaches to integrate GWAS with multiple ‘omics data across human brain tissues holds great promise to significantly advance our understanding of the biology underlying addiction.

Introduction

Addiction is a chronic, relapsing disease that alters the brain’s reward circuitry and consequently leads to compulsive drug seeking and other behavioral changes. The long-lasting biological effects of drug exposure cause a multitude of adverse effects throughout the body. Despite these well-known health consequences and widespread public health campaigns to curb use of addictive drugs, prevalence remains high. Among individuals aged 12 and older in the U.S. in 2015, an estimated 30.2 million (11.3%) smoked cigarettes daily in the past month; 15.7 million (5.9%) had an alcohol use disorder and 7.7 million (2.9%) had an illicit drug use disorder in the past year [1]. Individuals with addiction often have strong desires to quit, but rates of successful treatment and recovery are low. For example, among adult U.S. smokers during 2015, an estimated 68% wanted to quit, 55% had made a quit attempt in the past year, but only 7% had recently quit [2].

Addiction to drugs of abuse follows a recurring cycle, with each of the three stages being driven by a major neurobiological circuit: basal ganglia (including ventral tegmental area and nucleus accumbens) for the binge/intoxication stage; extended amygdala and habenula for the withdrawal/negative affect stage; and prefrontal cortex (PFC), insula, and allocortex for the preoccupation/anticipation (craving) stage [3]. This framework has expanded as knowledge of the complex neurocircuitry of addiction has evolved [3, 4].

Inter-individual differences in neurobiological circuits, due to genetic variation and their downstream effects, alter susceptibility to developing addiction. Although addiction is multifactorial, heritability estimates have indicated that around 40%–60% of the population variability in becoming addicted to nicotine, alcohol, or illicit drugs is attributable to genetic factors [3, 5]. Some genetic factors may influence an overarching susceptibility to developing addiction; thus, their effects are shared across different drugs of abuse. In contrast, other genetic factors may underlie susceptibility to developing specific drug addictions. To pinpoint the genetic factors underlying addiction, variants selected from biologically plausible candidate genes were long-studied in hypothesis-driven studies, but genetic variant associations were not firmly established until the advent of the agnostic, hypothesis-generating genome-wide association study (GWAS) approach. As seen across the field of complex human disease genetics, the potential of GWAS to make novel discoveries continues today with ever increasing sample sizes and statistical power, denser genomic coverage, and phenotype harmonization. This review outlines GWAS reports to date for nicotine, alcohol, or other drug addictions and focuses on common variants (mostly, single nucleotide polymorphisms [SNPs]) with robust evidence for association (i.e., identified at the standard genome-wide significance threshold [P<5×10⁻⁸] and replicated in an independent dataset).

GWAS for cigarette smoking

The earliest GWAS in the addiction field was conducted for nicotine dependence [6, 7], implicating nicotinic acetylcholine receptor genes on chromosomes 15q25 (CHRNA5-CHRNA3-CHRNB4) and 8p11 (CHRNB3-CHRNA6), which have since been firmly established in other studies of nicotine dependence and related smoking phenotypes (Table 1). To date, at least 26 GWAS analyses (largest N=74,035 [8]) have been conducted for self-reported phenotypes [6, 9–11, 8, 12–27], including ever vs. never smoking, former vs. current smoking, cigarettes per day (CPD), nicotine dependence defined by Fagerström Test for Nicotine Dependence (FTND) or Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) [6, 9–11, 8, 12–27], nicotine withdrawal defined by DSM-IV, as well as smoking biomarkers [28–32]. The FTND, a validated, expert-recommended, low-burden questionnaire of six items used to assess severity of physiological nicotine dependence symptoms among cigarette smokers [33], is the most widely used measure of nicotine dependence.

Table 1.

Evidence supporting 11 genetic loci identified at genome-wide statistical significance (P<5×10⁻⁸) in GWAS of cigarette smoking phenotypes/biomarkers and independently replicated or extended to a related phenotype/biomarker. For each locus, replication (or extension) evidence is presented for GWAS-identified lead SNP/indel or proxy variants. P values shown in bold correspond to the first GWAS demonstrating genome-wide significant evidence for SNP/indel association. Loci are sorted by chromosome, and studies supporting each locus are presented chronologically.

Genetic locus	Chr	Lead GWAS SNP/indela	SNP/indel annotation	Smoking phenotype or biomarker	N, by ancestry				P
					European	African	Asian or Pacific Islander	Latin o/Hispanic
UGT2B10-UGT2A3b	4q13	rs115765562c	Upstream	Cotinine glucuronidation [30]	437	364	985	453	1.6×10−155
		rs114612145c (r2=0.31, D′=0.94)	Downstream	Cotinine levels [32]	4,548	0	0	0	5.9×10−10
					755	0	0	0	0.020
PDE1C	7p14	rs215605	Intronic	CPD [11]	77,012	0	0	0	5.4×10−9
				Heavy vs. never [22]	48,931	0	0	0	2.0×10−3
CHRNB3-CHRNA6	8p11	rs6474412	Upstream	CPD [11]	84,956	0	0	0	1.4×10−8
		rs1451240 (r2=0.99, D′=1)	Upstream	Nicotine dependence [15]	3,201	999	0	0	6.7×10−16
DBH	9q34	rs3025343	Intronic	Current vs. former [8]	64,924	0	0	0	3.6×10−8
				Current vs. former [14]	1,164	0	0	0	0.002
				Current vs. former [16]	0	11,644	0	0	0.03
				FTND score [142]	1,430	0	0	0	0.023
				Heavy vs. never [22]	48,931	0	0	0	1.2×10−5
				Nicotine dependence [26]	28,677	0	0	0	1.7×10−5
BDNF	11p 14	rs6265	Missense	Ever vs. never [8]	143,023	0	0	0	1.8×10−8
				Heavy vs. never [22]	48,931	0	0	0	2.4×10−4
CHRNA5-CHRNA3-CHRNB4	15q 25	rs16969968	Missense	Nicotine dependence [6, 7]	1,929	0	0	0	6.4×10−4
				CPD [11, 12, 8]	73,853	0	0	0	5.6×10−72
				CPD [14]	3,440	0	0	0	1.5×10−4
				CPD [16]	0	11,480	0	0	0.027
				CPD [143]	28,772	0	0	0	3.2×10−25
				Years of smoking [143]					1.1×10−6
				Pack-years [143]					3.0×10−23
				Current vs. former [143]					6.9×10−10
				CPD [144]	12,364	0	0	0	5.2×10−6
				CPD [145]	1,9420	0	0	0	3×10−4
				FTND score [146]	815	0	0	0	0.0068
					1,121	0	0	0	0.0028
				Heavy vs. light [25]	0	0	0	5,085	2.2×10−7
				Heavy vs. light [147]	14,786	10,912	6,889	0	1.1×10−17
				Heavy vs. light [148]	140	0	0	0	0.007
				Low vs. high dependence [148]					0.01
				Age of cessation [46]	29,072	0	0	0	0.0042
				Exhaled CO [28]	1,521	247	0	0	1.7×10−8
				Cotinine levels [32]	4,548	0	0	0	6.9×10−17
				Cotinine levels [144]	12,364	0	0	0	2.7×10−11
		rs1051730 (r2=0.87, D′=1)	Synonymous	CPD [9]	15,771	0	0	0	6×10−20
				CPD [11, 12, 8]	73,853	0	0	0	2.8×10−73
				CPD [10]	4,342	0	0	0	6.1×10−4
				CPD [14]	3,440	0	0	0	1.1×10−4
				CPD [16]	0	15,552	0	0	0.0079
				CPD [149]	788	0	0	0	0.004
				Heavy vs. light [22]	48,931	0	0	0	4.4×10−15
				Heavy vs. light [150]	1,264	0	0	0	0.0057
				Adherence to NRT [151]	633	0	0	0	0.044
				NRT dose [151]					0.026
		rs2036527 (r2=0.67, D′=0.97)	Intergenic	CPD [16]	0	15,554	0	0	1.8×10−8
				Nicotine dependence [152]	1,428	0	0	0	9×10−6
				Abstinence [153]	0	1,295	0	0	0.004
		rs34684276 (r2=0.88, D′=0.95)	Intronic	Nicotine dependence [21]	17,074	0	0	0	3.5×10−17
EGLN2c	19q 13	rs3733829	Intronic	CPD [8]	73,853	0	0	0	1.0×10−8
				CPD [154]	1,395	0	0	0	0.026
				exhaled CO [154]					3.8×10−5
CYP2A6-CYP2B6d	19q 13	rs4105144	Upstream	CPD [11]	83,317	0	0	0	2.2×10−12
				CPD [14]	3,440	0	0	0	3.9×10−5
				Cotinine levels [155]	845	0	0	0	1×10−4
		rs8102683 (r2=0.21, D′=0.95)	Upstream	CPD [17]	0	0	11,696	0	4.3×10−26
					0	0	5,462	0	1.5×10−17
		rs56113850 (r2=0.22, D′=0.53)	Intronic	NMR [29]	1,518	0	0	0	5.8×10−86
				NMR [31]	212	49	51	0	6.6×10−18
DNMT3Bd	20q 11	rs910083	Intronic	Nicotine dependence [26]	28,677	9,925	0	0	3.7×10−8
				Heavy vs. never [22, 26]	48,931	0	0	0	3.6×10−4
NOL4L d	20q 11	rs57342388b	Intronic	Heavy vs. never [22]	48,931	0	0	0	4.7×10−9
				Nicotine dependence [26]	28,677	9,925	0	0	0.0017
CHRNA4	20q 13	rs2273500	Splice acceptor	Nicotine dependence [21]	24,543	0	0	0	8.0×10−9
				Heavy vs. never [22, 26]	48,931	0	0	0	1.4×10−6

CPD, cigarettes per day; CO, carbon monoxide; FTND, Fagerström Test for Nicotine Dependence; NMR, nicotine metabolite ratio; NRT, nicotine replacement therapy

^aFor loci with more than one GWAS lead or proxy SNP indicated, r² and D′ values are shown in reference to all 1000 Genomes phase 3 panels as calculated using LDlink [117].

^bOf the 11 loci, UGT2B10-UGT2A3 is the only one implicated for biomarkers but not extended to nicotine dependence or related smoking phenotypes.

^crs115765562, as originally reported, has since merged into rs34100980. rs57342388 has the same alignment as rs143125561. rs114612145, as originally reported, has since merged into rs77107237.

^dLoci located on the same chromosomal arm, but presented separately, have been reported as independent signals based on weak linkage disequilibrium and conditional association results: (1) EGLN2 and CYP2A6 [154] (2) DNMT3B and NOL4L [26].

Most GWAS analyses have been conducted in European ancestry populations [6, 9–11, 8, 12–14, 18, 19, 21, 22, 29, 32], although one focused on Hispanics [25], one focused on Japanese [17], eight included or focused exclusively on African Americans [15, 16, 20, 23, 24, 26, 28, 27], and two others included multiple ancestries [30, 31]. Two GWAS analyses examined copy number variants [17, 23], while all others focused on SNP and insertion/deletion (indel) variants.

Table 1 presents the 11 genetic loci with common SNPs/indels implicated at genome-wide significance (P<5×10⁻⁸) and replicated in one or more independent datasets for the same phenotype/biomarker or extended to a related smoking phenotype/biomarker. The UGT2B10-UGT2A3 locus was identified in cotinine biomarker GWAS, but this finding has not been extended to smoking phenotypes [32]; the 10 other GWAS-identified loci influence susceptibility of smoking phenotypes. Lead SNPs with the smallest P value from each distinct genetic locus are presented, although several loci include many significantly associated SNPs, often strongly correlated with the lead SNPs. For example, 360 SNPs in the CHRNA5-CHRNA3-CHRNB4 locus were significantly associated in GWAS meta-analysis of FTND-defined nicotine dependence [21], and 719 SNPs in the CYP2A6-CYP2B6 locus were significantly associated in GWAS meta-analysis of the nicotine metabolite ratio [29]. SNPs in some GWAS-identified loci also influence susceptibility of developing smoking-related health outcomes, including: CHRNA5 with lung cancer (most recently [34]), lung function [35], airflow obstruction [36], COPD (most recently [37]), and peripheral arterial disease [9]; the nicotine metabolizing gene CYP2A6 with lung cancer [11], lung function [35], and emphysema [38]; and CHRNB3-CHRNA6, CHRNA4, and the DNA methyltransferase gene DNMT3B with lung cancer [11, 21, 26]. Beyond common SNPs, the CHRNA5-CHRNA3-CHRNB4 [39–42] and CHRNA4 [40, 43, 44] loci also harbor smoking-associated rare variants, albeit some do not have replication evidence or occur in an isolated population. Genome-wide significant SNP associations in other loci have been reported but not yet independently replicated [8, 13, 20, 22, 24].

The GWAS-identified SNPs each explain a small percent of the variance in nicotine dependence or related smoking phenotypes, but their discoveries have revealed important neurobiology that impacts susceptibility to developing addiction and exerts clinically important effects downstream, as best illustrated by CHRNA5. Following the discovery of its association with nicotine dependence (Table 1), the CHRNA5 missense SNP rs16969968 was found to alter α5 receptor function [45] and later its high-risk rs16969968-AA genotype was associated with a 4-year median delay in quitting smoking and a 4-year earlier median age of lung cancer diagnosis, even when adjusted for CPD [46]. Moreover, the haplotype carrying the rs16969968-A risk allele interacts with cessation treatment, with smokers at highest nicotine dependence risk being less likely to quit smoking overall but responding most effectively to pharmacologic treatment [47], demonstrating the potential for personalized cessation treatment based on genetic risk variants for nicotine dependence. Independent of rs16969968, noncoding CHRNA5 SNPs have been found to tag cis-expression quantitative trait loci (cis-eQTL) and cis-methylation QTL (cis-meQTL) variants that regulate CHRNA5 RNA expression (rs588765 [48] and rs880395 [49]) and DNA methylation (rs11636753 [50]), respectively, in postmortem human PFC. Consistent with these findings, mouse models have indicated that genetically altered CHRNA5 RNA expression profoundly influences behavioral traits characteristic of nicotine dependence [51–53].

While the regulatory effects of CHRNA5 SNPs have mainly centered on PFC, widely recognized for its involvement in addiction [3, 54], noncoding SNPs nearby genes identified in more recent GWAS analyses have highlighted gene regulatory effects in unexpected brain tissues. The CHRNA4 splice site acceptor SNP rs2273500 (Table 1) was indicated as a cis-eQTL SNP for CHRNA4 in intralobular white matter [21]. The DNMT3B intronic SNP rs910083 (Table 1) [26] and CHRNA2 upstream SNP rs117804171, discovered for its association with lung cancer but then extended to smoking [34], were each indicated as cis-eQTL SNPs for their proximal gene in cerebellum, which has been often overlooked despite evidence for its involvement in the neurobiology of addiction [55–57]. Follow-up of these and other GWAS discoveries for functional or regulatory effects, in normal physiological vs. smoking-exposed states, is needed across the wide array of human brain tissues to expand our neurobiological understanding of initiating smoking, becoming a regular smoker, developing nicotine dependence, and ultimately improving cessation treatment strategies.

GWAS for alcohol

At least 34 GWAS analyses (largest N=112,117 [58]) have been reported for alcohol use disorder, related alcohol phenotypes, or alcohol biomarkers from studies focused on European [13, 18, 59–74, 58, 75, 76], European and African [77–83], or Asian [84–88] ancestries, or multiple ancestries combined [89]. Table 2 presents the genome-wide significant and replicable SNP associations for self-reported phenotypes in or near seven loci—serpin family C member 1 (SERPINC1), glucokinase regulator (GCKR,) shugoshin 1 (SGOL1), β-klotho (KLB), alcohol dehydrogenase (ADH) cluster, autism susceptibility candidate 2 (AUTS2), and aldehyde dehydrogenase 2 (ALDH2); an eighth locus—transferrin (TF)—was implicated for biomarkers of excessive alcohol intake but has not been associated with alcohol use disorder or related phenotypes. Emerging evidence suggests that the ADH gene cluster contains more than one independent association signal [58]. Other genome-wide significant loci have been reported, but independent replication has not been attained [13, 59, 60, 64, 68, 70, 73, 74, 58].

Table 2.

Evidence supporting eight genetic loci identified at genome-wide statistical significance (P<5×10⁻⁸) in GWAS of alcohol phenotypes/biomarkers and independently replicated or extended to a related phenotype/biomarker. For each locus, replication (or extension) evidence is presented for GWAS-identified lead SNP/indel or proxy variants. P values shown in bold correspond to the first GWAS demonstrating genome-wide significant evidence for SNP association. Loci are sorted by chromosome, and studies supporting each locus are presented chronologically.

Genetic locus	Chr	Lead GWAS SNPa	SNP annotation	Alcohol phenotype or biomarker	N, by ancestry				P
					European	African	Asia n	Latino/His panic
SERPINC1	1q25	rs1799876	Intronic	Maximum drinks in 24-hour period [82]	2,328	0	0	0	4.0×10−8
				Weekly alcohol intake [89]	47,967	0	0	0	0.031
GCKR	2p23	rs780094	Intronic	Daily alcohol intake [73]	98,477	0	0	0	3.6×10−9
		rs4665985 (r2=0.41, D′=0.68)	Downstream	Drinker vs. non-drinker [89]	71,071	2,475	6,034	7,047	5.0×10−4
				Weekly alcohol intake [89]	47,967	1,310	2,746	4,374	2.3×10−8
SGOL1	3p24	rs11128951	Upstream	Maximum drinks in 24-hour period [69]	2,687	0	0	0	4.3×10−8
				Drinker vs. non-drinker [89]	71,071	0	0	0	0.019
TF-SRPRBb	3q22	rs1799899	Missense	CDT% [64]	5,181	0	0	0	3.4×10−35
					1,509	0	0	0	9.7×10−6
		rs3811647 (r2=0.02, D′=0.96)	Intronic	Total transferrin [64]	5,181	0	0	0	1.3×10−10
					1,509	0	0	0	5.3×10−19
					745	0	0	0	8.8×10−9
		rs1534166 (r2=0.07, D′=0.86)	Intronic	CDT concentration [64]	1,509	0	0	0	5.8×10−12
					5,181	0	0	0	3.0×10−7
KLB	4p14	rs11940694	Intronic	Daily alcohol intake [73]	98,477	0	0	0	9.2×10−12
				Weekly alcohol intake [58]	112,117	0	0	0	8.4×10−19
		rs7686419 (r2=0.47, D′=0.78)	Upstream	Drinker vs. non-drinker [89]	71,071	2,475	6,034	7,047	4.1×10−5
				Weekly alcohol intake [89]	47,967	1,310	2,746	4,374	3.4×10−10
ADH1B-ADH1C	4q23	rs1789891	Intronic	DSM-IV-defined cases vs. controls [65]	3,501	0	0	0	1.3×10−8
				DSM-IV-defined cases vs. controls [156]	2,103	0	0	0	7.2×10−5
		rs1229984 (r2=0.02, D′=0.89)	Missense	Cases vs. controls with varied definitions [91]	5,850	0	12,266	0	1.0×10−36
				Ever vs. never [85]	0	0	6,720	0	3.6×10−4
				DSM-IV-defined cases vs. controls [157]	3,789	1,721	0	0	6.6×10−10
				Maximum drinks in 24-hour period [157]	3,777	1,698	0	0	3.2×10−13
				DSM-IV-defined cases vs. controls [86]	0	0	1,371	0	2.6×10−21
				DSM-IV symptom count [81]	7,001	0	0	0	2.9×10−18
				Maximum drinks in a 24-hour period [82]	2,587	0	0	0	6.0×10−15
				DSM-IV-defined cases vs. controls [156]	2,103	0	0	0	8.4×10−6
				Drinker vs. non-drinker [89]	71,071	2,475	6,034	7,047	8.0×10−4
				Weekly alcohol intake [89]	47,967	1,310	2,746	4,374	3.0×10−4
		rs2066702 (r2=0.01, D′=0.96)	Missense	DSM-IV symptom count [81]	0	4,638	0	0	2.2×10−13
				Maximum drinks in 24-hour period [82]	0	5,064	0	0	2.5×10−10
		rs145452708 (r2<0.01, D′=1)	Intronic	Weekly alcohol intake [58]	112,117	0	0	0	1.2×10−30
		rs141973904 (r2<0.01, D′=1)	Intronic	Alcohol use disorder identification test (AUDIT) score [76]	20,328	0	0	0	4.4×10−7
AUTS2	7	rs6943555	Intronic	Average consumption (grams/day/kg body weight) [60]	26,316	0	0	0	4.1×10−9
				Weekly alcohol intake [89]	47,967	0	0	0	0.0050
ALDH2	12q24	rs2074356	Intronic	Average consumption (grams alcohol/d ay) [84]	0	0	1,721	0	5.8×10−46
					0	0	1,113	0	1.3×10−16
		rs671 (r2=0.49, D′=0.83)	Missense	Ever vs. never [85]	0	0	6,720	0	3.6×10−211
				DSM-IV-defined cases vs. controls [86]	0	0	396	0	8.4×10−8
				Past year drinkers vs. nondrinkers [87]	0	0	23,199	0	5.2×10−215
				Alcohol dependence [88]	0	0	595	0	4.7×10−8
				Maximum drinks in a 24-hour period [88]	0	0	595	0	1.5×10−16
				Flush response [88]	0	0	595	0	4.8×10−26
				Drinker vs. non-drinker [89]	0	0	6,034	0	2.3×10−72
				Weekly alcohol intake [89]	0	0	2,746	0	5.4×10−4
		rs11066280 (r2=0.44, D′=0.88)	5′ untranslated	Past year drinkers vs. nondrinkers [87]	0	0	23,199	0	3.3×10−215
				Daily alcohol intake [87]	0	0	6,211	0	4.0×10−21

CDT, carbohydrate-deficient transferrin (biomarker of excessive alcohol intake)

^aFor loci with more than one GWAS lead SNP indicated, r² and D′ values are shown in reference to all 1000 Genomes phase 3 panels as calculated using LDlink [117].

^bOf the 8 loci, TF is the only one implicated for biomarkers but not extended to alcohol use disorder or related alcohol phenotypes.

The GWAS-identified variants include missense functional SNPs that alter the alcohol metabolism pathway converting alcohol to acetaldehyde via ADH enzymes and then to acetate via ALDH enzymes. For example, the ADH1B SNP allele rs1229984-T is a missense variant that alters enzymatic activity resulting in up to 100-fold higher rate of alcohol to toxic acetaldehyde metabolism, causing the alcohol flush reaction, and reducing susceptibility to developing alcohol drinking problems [90]. Moreover, the rs671-A allele in ALDH2, another missense variant, alters the latter step of the alcohol metabolism pathway by greatly reducing ALDH2 activity and rendering high acetaldehyde concentrations upon alcohol exposure [85]. Both the ADH1B rs1229984-T and ALDH2 rs671-A alleles occur frequently in East Asian populations but infrequently in other world populations, and their highly significant associations with reduced susceptibility to drinking alcohol and developing an addiction were studied as candidate SNPs based on their functional consequences on enzymatic activity before they were highlighted via agnostic GWAS [85, 91]. Associations of these SNPs or their proxies on alcohol-related disease outcomes include coronary heart disease [85], various cancers [92, 93], renal function [94], blood pressure [95], and cirrhosis [96]. Beyond these functional missense SNPs, the intronic AUTS2 SNP rs6943555 was implicated as a cis-eQTL for AUTS2 in PFC [60], and the intronic KLB SNP rs11940694 was implicated as a cis-eQTL for the replication factor RCF1 gene in cerebellum [58]. Little else is known about the biological relevance of other noncoding SNPs that may exert independent effects on regulation of genes that underlie alcohol metabolism in liver or the neurobiology of developing an alcohol use disorder.

GWAS for other specific drugs

Relatively few GWAS have been reported for cannabis, stimulants, or opioids. Unlike GWAS of cigarette smoking and alcohol phenotypes, GWAS of these specific drugs have had limited success at identifying and replicating variant associations.

For cannabis, four of the five reported GWAS had no genome-wide significant SNP associations [97–100], most notably including a large GWAS meta-analysis (total N=32,330 of European ancestry) [99]. The most recent GWAS meta-analysis (total N=14,754 European Americans and African Americans) identified three loci that attained genome-wide significance, but these findings await independent replication [101].

For stimulants, four GWAS have been reported. Two focused on methamphetamine and found no genome-wide significant associations [102, 103]. The other two GWAS focused on amphetamine [104] and cocaine [105], and each reported a single genome-wide significant SNP association, but these findings also await replication.

For opioids, six GWAS have been reported. Three SNP-based GWAS were small (N<1,000), and no SNPs attained genome-wide significance [106–108]. The three larger GWAS, including two SNP-based [109, 110] and one copy number variant-based [111], each found genome-wide significant associations in different genetic loci, but these association signals all await independent replication. GWAS analyses have not yielded genome-wide significant evidence for any of the biologically plausible opioid receptor genes. However, such evidence emerged from a targeted study of cis-eQTL SNPs of the μ-opioid receptor gene (OPRM1); smallest P=4.3×10⁻⁸ for intronic SNP rs3778150 with total N=16,729 of European or African American ancestry [112]. Much attention has focused on OPRM1 SNP associations but with inconclusive findings from mostly studies of single cohorts. The OPRM1 SNP rs1799971, which results in an amino acid change in the μ-opioid receptor, has been the most extensively studied. In an accumulation of evidence for rs1799971, meta-analysis of >28,000 European ancestry participants from 25 studies supported a modest association with general substance dependence, rather than opioids or any substance specifically [113]. As we previously reported, the cis-eQTL SNP rs3778150 may explain the prior inconsistent associations, as the rs1799971-A allele conferred increased heroin addiction risk only in the presence of the rs3778150-C allele that lowers OPRM1 expression levels and plausibly disrupts the opioid system of a key compensatory response to exogenous opioid exposure [112]. The association pattern observed between heroin addiction and the haplotype carrying rs3778150-C and rs1799971-A has been extended to subjective alcohol response [114].

Relatedly for opioid use disorder phenotypes, two GWAS of clinical opioid dosing have been reported [115, 116]. Most recently, a genome-wide significant association of methadone dosing among 383 opioid-dependent African Americans was observed for the SNP rs73568641 (P=2.8×10⁻⁸), located upstream of OPRM1 [116]. The rs73568641 association was extended to morphine dosing for post-operative pain in an independent dataset of 241 African Americans (P=0.039) [116]. Rs73568641 is in weak linkage disequilibrium (r²<0.01 and D′=0.54 across 1000 Genomes African ancestry reference panels [117]) with the cis-eQTL SNP rs3778150, suggesting that variants marking more than one signal across the OPRM1 gene region may be driving different opioid phenotypes. However, association analyses of these SNPs conditioned on one another are needed to formally assess independence of the signals.

GWAS for illicit drugs combined

Given the high levels of co-morbidity across drugs of abuse and the plausible shared heritability underlying susceptibility to developing addiction in general, three GWAS have combined cases addicted to any illicit drug [18, 118, 119]. Two of these GWAS reported genome-wide significant and replicable associations. First, a GWAS of general substance dependence liability, based on factor analysis of DSM-IV symptoms, identified the intergenic SNP rs2952621 (P=1.8×10⁻⁸, N=2,322 European Americans) and replicated the finding in an independent dataset (P=0.02, N=2,647 European Americans) [118]. Rs2952561 is a SNP of unknown biological consequence that is located upstream of the uncharacterized gene LOC151121.

Next, GWAS of people who inject drugs vs. controls, revealed an African American-specific association for the intronic SNP rs9829896 in the lysine acetyltransferase gene (KAT2B; P=4.6×10⁻⁸, N=3,742), which replicated in an independent dataset (P=0.0016, N=755) [119]. Follow-up analyses in postmortem human PFC implicated rs9829896 as a cis-eQTL SNP, specifically in African Americans, for KAT2B and other genes in KAT2B-containing pathways, including OPRM1 and the CREB binding protein (CREBBP) genes [119] that have been highlighted in prior SNP- or pathway-based analyses of opioid phenotypes [109, 112, 116].

Potential for future GWAS to broaden the known genetic etiology unique to and shared among addiction phenotypes

Prior GWAS have utilized varied addiction phenotypes and biomarkers as well as study designs. A major distinction among the case-control studies involves the assessment of substance use and misuse for controls. Some studies have used population-based, unassessed controls when study controls were not available, for example [119], or when intending to increase sample size and statistical power with minimal offset from misclassification due to the relatively low prevalence of addiction, particularly for illicit drugs in healthy cohorts [112]. Other studies have focused on assessed controls with prior exposure or history of misuse but without symptoms of addiction, for example [110], aimed at detecting genetic variants associated with progression from misuse to later stages of addiction. Relatedly, studies with assessed controls have either adjusted for co-morbid exposure and/or addiction to other substances as covariates, under the premise of identifying genetic variants associated with specific substances, for example [105, 109]. Other studies have not adjusted for other substance co-morbidities [110], which may enhance detection of generalizable genetic variant associations. To our knowledge, there has been no formal testing to outline the best scenarios for future addiction studies to balance statistical power with the use of unassessed vs. assessed controls. Ultimately, study design decisions will depend on the primary research question and expected misclassification rate. For strategizing adjustment for co-morbid addictions, simulation testing of genetically correlated outcomes and covariates suggests that the impact of covariate adjustment depends on the direction and magnitude of correlation with the outcome [120].

We fully expect that future GWAS using larger sample sizes from expanded meta-analyses and large-scale biobanks with data on cigarette smoking, alcohol, and other drugs will identify and replicate additional loci with robust statistical evidence. These loci will likely include some variants that underlie specific addictions, while others are shared across different drugs of abuse and across psychiatric diseases that are highly co-morbid with addiction. This notion of shared genetic susceptibility is supported by schizophrenia-associated common genetic variation having significant correlations with smoking-associated genetic variation: for example, r_g=0.14 for nicotine dependence, r_g=0.12–0.14 for CPD, and r_g=0.10 for ever/never smoking [121, 122]. Polygenic risk scores (weighted summations of top ranking subsets of genetic variant associations) have provided additional evidence for shared genetic etiologies. Polygenic risk scores based on GWAS-identified variants for schizophrenia or bipolar disorder have shown significant associations with smoking, alcohol, and substance use disorder phenotypes [123–125]. Early examples of genetic variants that demonstrate such pleiotropy, i.e., exert effects on more than one outcome, across addiction and other psychiatric diseases include: ZNF804A SNP rs1344706, the first genome-wide significant finding for schizophrenia [126] and later extended to heroin addiction [127, 128]; and the BDNF SNP rs6265 and CHRNA5 SNPs rs16969968 and rs1051730 implicated at genome-wide significance for smoking (Table 1) and extended to schizophrenia/bipolar disorder outcomes [129, 130]. As the sample sizes for addiction GWAS expand to sizes comparable with other psychiatric disease GWAS, polygenic risk scores and top loci that comprise these scores are likely to reveal even more extensive sharing. Once established, polygenic risk scores may have clinical utility for predicting individuals at highest risk for developing addiction or guiding treatment strategies.

Ongoing sequential integration of multiple ‘omics data

Establishing robust statistical evidence for a GWAS-identified variant marks the starting point in the path to elucidating its biological consequences as related to disease outcomes. Its annotation in the context of DNA, RNA, and protein sequences often provides the first clue. Coding SNPs that alter the protein’s amino acid sequence can be undoubtedly important for risk of developing complex human diseases, including addiction (e.g., CHRNA5 SNP rs16969968 for smoking [Table 1] and ADH1B SNP rs1229984 and ALDH2 SNP rs671 for alcohol [Table 2]). However, >95% of GWAS discoveries come from noncoding SNPs [131]. Noncoding SNPs can alter gene activity by influencing DNA methylation, RNA expression, protein expression, or metabolite levels and are thus well-recognized for their disease risk potential. Indeed, the GWAS catalog is heavily enriched for regulatory variants, such as eQTLs [131–133], and variants residing in sequences that are highly conserved, sensitive to cleavage by DNase I, or mark enhancer, promoter, or protein binding sites [133].

Integration of GWAS with other ‘omics data types (such as, DNA methylation or RNA expression) commonly happens in a sequential fashion to infer functional or regulatory effects of top GWAS findings. Each ‘omics type is analyzed separately, and comparisons are made across results. For example, rs910083 was identified in the largest GWAS meta-analysis of nicotine dependence, and its association was extended to heavy vs. never smoking in the UK Biobank (Table 1); rs910083 was then followed-up via a cis-eQTL analysis using RNA expression data measured across multiple postmortem human brain tissues in the Genotype-Tissue Expression (GTEx) and Brain eQTL Almanac projects, leading to the implication of rs910083 as a DNMT3B cis-eQTL in cerebellum [26]. This type of sequential integration in moving from GWAS to functional or regulatory characterization has also yielded other important discoveries in addiction [21, 48, 119].

A reversal in the direction of sequential integration is an alternative, but not as widely utilized, approach that first identifies functional or regulatory variants to narrow the search space, thereby reducing the multiple testing burden of GWAS to uncover disease associated variants that may have otherwise been missed. The OPRM1 cis-eQTL SNP rs3778150 finding for heroin addiction serves as an example of moving from gene regulation into GWAS [112].

In taking integration to the next level, PrediXcan [134] is a promising new gene-based method that utilizes genome-wide SNP genotypes and RNA-sequencing data in GTEx to build genetically regulated gene expression models in a diverse set of tissues and applying those models in GWAS with disease phenotypes of interest. The PrediXcan framework is applicable to other sources of ‘omics data (e.g., DNA methylation) in GTEx or elsewhere.

Importance of brain tissue specificity for multi ‘omics analyses

These sequential integration examples highlight a challenge for the field of addiction, and psychiatric disease more broadly, because functional and regulatory effects can be highly tissue-specific [135] and brain is the most relevant tissue for studying the neurobiology of addiction. GWAS genotypes, other ‘omics data in brain, and addiction phenotypes are seldom available in the same dataset. Other ‘omics data derived in blood or other peripheral tissues that are easily accessible in living participants may offer informative biomarkers of addiction but overall provide poor indicators of neurobiology. Poor correlations in RNA expression levels between blood and brain has been repeatedly shown [136, 137, 135]. Most recently, pilot GTEx analyses with RNA expression measured in 43 tissues showed that blood vs. brain had the most distinct expression profiles among the tissue comparisons [135]. There is also limited overlap of meQTL and eQTL [136, 137, 135] variants mapped in blood vs. brain. Thus, when analyzing gene regulatory potential to study the neurobiology of addiction, it is critical to use disease-relevant brain tissue. Reliance on blood-specific regulatory effects could even lead to erroneous conclusions, as illustrated by the cis-eQTL SNP rs880395 having opposing directions of association with CHRNA5 expression in lymphoblastoid cell lines, as compared to frontal cortex [49]. Moreover, gene regulatory effects can differ across the different brain tissues, so comprehensive functional and regulatory assessment will require the availability of ‘omics data across multiple brain tissues, including ones traditionally viewed as being addiction-relevant (e.g., PFC, nucleus accumbens) and others often overlooked (e.g., cerebellum). GTEx, Brain eQTL Almanac, and others provide an unprecedented opportunity to carry out comprehensive analyses in normal brains. Future studies are needed for more thorough and well-powered assessment of brain tissues in participants with addiction phenotype data.

Upcoming concurrent integration of multi ‘omics data

Where once a single ‘omics type was used, studies that capture multiple ‘omics data types in the same dataset are emerging. As these datasets become available, concurrent integration that jointly assess all data may unveil relationships not evident when each data type is analyzed separately by virtue of increased statistical power and their explicit biological relationships (Figure 1). It is expected that genetic variants with large effect sizes are identifiable in sequential analyses but that concurrent integration will enable the identification of genetic variants with moderate-sized, but multi-faceted, functional or regulatory effects. This approach can operate in a bidirectional fashion, with datasets rich in ‘omics informing follow-up in large-scale GWAS of addiction phenotypes/biomarkers and vice versa (Figure 1), and accelerate understanding of the biology underlying statistical associations with addiction phenotypes.

Figure 1.

Concurrent integrative analysis approach with multiple types of ‘omics data used to inform discovery or characterize top findings from large-scale addiction GWAS. Genomics encompasses SNP/indel genotypes, rare variants, and structural variants including copy number variants; epigenomics includes DNA methylation and histone modification; and transcriptomics refers to expression of all RNA types. ‘Omics data may pertain to endogenous factors along the flow of information according to the Central Dogma of Biology or exogenous factors such as environmental exposures.

New methods are being developed that will allow researchers to analyze data concurrently, as applied in plant [138] and mouse model [139] studies. In an early adaptation of this approach in humans [140], the Two-Way Orthogonal Partial Least Squares (O2PLS) method to RNA expression and metabolomics data in 466 Finnish participants recapitulated signals detected in sequential analyses of the two data types and detected new signals by modeling the data types together. Colocalization (COLOC) [141] also enables concurrent integrative analyses. Comparisons of these and other methods will be needed to establish standard analytic practices.

Conclusions

GWAS analyses have identified several genetic loci with statistically robust and reproducible SNP/indel associations, cementing the polygenic nature of addiction. However, with only a fraction of the heritability explained (for example, 15% of the variance in nicotine dependence [121] and 13% of the variance in alcohol consumption [58] explained by common SNPs together) and limited knowledge of the neurobiological pathways leading to addiction, much remains to be discovered. We fully expect that GWAS analyses conducted with sample sizes into the hundreds of thousands and millions will replicate previously suggested, but currently unreplicated, genetic variants and will also unveil novel variants. As evidenced by the GWAS-identified variants identified to date (mainly SNPs), novel variants will likely exert small effect sizes on developing addiction but potentially uncover previously unrecognized neurobiological pathways. Additional heritability may also be explained by complex interactions of genetic variants with one another and with prominent environmental exposures, which will likely require very large sample sizes with harmonized exposures and addiction phenotypes across multiple datasets. In the future expansion of GWAS sample sizes, we emphasize the importance of including diverse ancestry groups that rival the current representation of European ancestry to address important health disparities, identifying variants that manifest under certain genetic, cultural, or environmental backgrounds and narrowing the resolution of association signals that arise from the varied linkage disequilibrium patterns across diverse populations.

New genetic discoveries are also likely to arise from leveraging novel analytic approaches such as integrating GWAS with multiple ‘omics data generated across addiction-relevant brain tissues, even at the resolution of individual cell types and single cells. With the ever-expanding postmortem human brain resources being made available to the scientific community, the field has begun to unravel the biology of top GWAS-identified variants, including a few functional missense SNPs that alter susceptibility to developing nicotine dependence or alcohol use disorder and several noncoding SNPs that perturb gene regulation and influence nicotine dependence and substance use disorders. The recent availability of ‘omics data across multiple brain tissues has offered unparalleled opportunities to study the neurobiological underpinnings of addiction and led to somewhat unexpected connections, such as smoking- and alcohol-associated cis-eQTL SNPs specific to cerebellum [26, 34, 58]. In this era of precision medicine, genetic variant discoveries hold great promise in tailoring prevention and treatment strategies for addiction. Discovery of more genetic loci and biological characterization of the associated variants is needed to enhance our understanding of the neurobiology underlying addiction and lessen its widespread health consequences on individuals and the public health burden.