An update on the role of common genetic variation underlying substance use disorders

Abstract

Purpose of the review:

Sample size increases have resulted in novel and replicable loci for substance use disorders (SUDs). We summarize some of the latest insights into SUD genetics and discuss some next steps in addiction genetics.

Recent findings

Genome-wide association studies have substantiated the role of previously known variants (e.g., rs1229984 in ADH1B for alcohol) and identified several novel loci for alcohol, tobacco, cannabis, opioid and cocaine use disorders. SUDs are genetically correlated with psychiatric outcomes, while liability to substance use is inconsistently associated with these outcomes and more closely associated with lifestyle factors. Specific variant associations appear to differ somewhat across populations, although similar genes and systems are implicated.

Summary:

The next decade of human genetic studies of addiction should focus on expanding to non-European populations, consider pleiotropy across SUD and with other psychiatric disorders, and leverage human and cross-species functional data to elucidate the biological mechanisms underlying SUDs.

Introduction

Excessive substance use imposes considerable personal and societal burden (1–3). Tobacco smoking is amongst the leading direct contributors to preventable death in the United States, and worldwide(4). Heavy alcohol consumption contributes to numerous causes of mortality ranging from vehicular fatalities to cardiovascular and chronic liver disease (1,5). Illicit drugs have impacted the mental and physical health of individuals in developed and developing nations to varying degrees(6). Notably, use of opioids, including prescription opioids, heroin, and fentanyl has attained epidemic proportions in the United States, with steady escalations in overdose deaths in the United States(7–9). In addition to adverse physical health consequences, substance use disorders (SUDs) occur comorbidly with a majority of common psychiatric illnesses, including those with high morbidity (e.g., major depression (10,11)). Therefore, outlining the etiology of use and misuse of alcohol, tobacco and other psychoactive substances is a public health priority.

Neurobiological conceptualizations separate the addiction syndrome into early phases of impulsive use (e.g., binge/intoxication) that are characterized by drug use to elicit reward, followed by repeated use that can result in the emergence of compulsive use, characterized by negative affect during withdrawal and inappropriate investment of resources towards drug seeking behaviors(12–14). Under the Fifth edition of the Diagnostic and Statistical Manual (DSM-5; (15)), a diagnosis of SUD relates to the endorsement of two or more of 11 criteria that document drug tolerance, withdrawal related impairment, heavier and more prolonged use, including in hazardous situations, alongside persistent and failed efforts to curb use, as well as recurrent use despite associated interpersonal, emotional and physical consequences. SUD (or substance abuse, substance dependence) is generally considered a more pathological aspect of the addictive process, preceded by earlier stages of substance use, as indexed by measures of quantity and frequency of use, that for legal, especially less addictive substances, can relate partly to routine lifestyle practices (e.g., how many alcohol containing drinks one consumes in a typical week) but also serve as a harbinger of SUD (e.g., cigarettes smoked per day).

Both substance use and SUD are complex, polygenic and multifactorial traits. Broadly speaking, the estimated heritability of substance use is somewhat lower than that for SUD, which tends to range from 40 to 60% in estimates from twin and family studies (16,17). Individual drug use disorders appear to be co-heritable, with genetic correlations across SUDs > 0.80. For nicotine and opioid dependence, both substances with very high addictive potential, there is also evidence from twin studies for drug-specific genetic factors(18,19). In addition, there is evidence from twin studies for considerable comorbidity with externalizing features such as conduct and antisocial personality disorder and generalized personality traits of impulsivity and risk-taking(20–23).

In the last decade, genome-wide association studies (GWAS) of substance use and SUDs have undergone a seismic shift, moving from sporadic identification of 1–2 loci that were largely unreplicated to multiple loci with robust replication and increases in polygenic risk prediction(24–27). In this review, we discuss recent insights into addiction genetics brought about by GWAS and secondary analyses of genome-wide data. First, we discuss novel locus discovery from association studies conducted in the last five years. Although most large GWAS have been conducted on individuals of European ancestry, SUDs have made unique contributions to our understanding of addictions in non-European populations. Next, we outline differences in genetic architecture between substance use and use disorders that has been revealed by GWAS. Finally, we outline some opportunities and challenges for the field of SUD genetics moving forward in the “post-GWAS” era, including integrating across genetic, transcriptomic, and other ‘omics data, a continued emphasis on the study of non-European populations, and considering SUDs in the context of frequently-comorbid psychiatric disorders.

Recent insights from GWAS

To date, the largest GWAS for any substance use trait has been for quantitative measures of consumption of alcohol and tobacco (e.g., drinks per week N = 941,280, cigarettes per day N = 337,334(27)) or lifetime measures of ever using a substance (e.g., lifetime cannabis use N = 184,765(28), lifetime regular smoking N = 1,232,091(27)). For tobacco-related phenotypes, >100 independent loci have been identified. With respect to validation of loci in well-known genes for these traits (e.g., ADH1B variants for alcohol, CHRNA5 variants for smoking) and discovery of novel loci, this GWAS rivals those for most behavioral traits (e.g., neuroticism(29): N = 329,821; 116 independent genome-wide significant loci), though the number of genome-wide significant loci identified falls behind some traits, such as educational attainment(30) (N = 1,131,881; 1,271 independent genome-wide significant single nucleotide polymorphism (SNPs)), which might suggest a more polygenic architecture for substance use phenotypes. In contrast, GWAS of problematic substance use (particularly for illicit drugs including cocaine and opioids) have been much smaller (Table 1) and consequently, fewer genetic risk loci have been identified for these traits. None of the current large GWAS have included sex-stratified analyses, likely due to low power to delineate sex differences.

Table 1. Summary of findings from GWAS of substance use disorders from the past five years.

Only studies that have undergone peer review are included (i.e., preprints not included).

Drug	Year	PMID	Phenotype	N	Ncases	Number of genome-wide significant loci	Genes or regions implicated	h2 (SE)c
ALCOHOL
Alcohol	2018	30482948	DSM-IV alcohol dependence	52,848b	14,904	1	ADH1B	0.09 (0.02)
Alcohol	2018	30336701	AUDIT-P	121,604	NA	3	DRD2, KLB, CADM2	0.09 (0.005)
Alcohol	2019	31151762	Maximum habitual alcohol use	144,865b	NA	6	ADH1B, CRHR1, FGF14	0.078
Alcohol	2019	30940813	ICD-9 codes 303.X, 305–305.03 and ICD-10 codes F10.1, F10.2	274,391b	55,584	10	ADHIB, ADHIC, ADH4, GCKR, SIX3, SLC39A8, DRD2, chr10q25.1, FTO	0.06 (0.004)
NICOTINE
Nicotine	2015	26440539	FTND scores (categorized into mild, moderate, or severe)	17,074	NA	1	CHRNA4	NA
Nicotine	2018	28972577	FTND scores (categorized into mild, moderate, or severe)	38,602b	NA	2	DNMT3B, CHRNA5-CHRNA3-CHRNB4	NA
CANNABIS
Cannabis	2016	27028160	DSM-IV cannabis dependence criterion counts	14,754b	NA	3	SLC35G, CSMD1, RP11–206M11.7	0.190.25
Cannabis	2018	29112194	DSM-IV cannabis dependence	8,515b	2,080	1	RP11–215A21.1	NA
COCAINE
Cannabis	2019	31209380	ICD-9 codes	51,372	2,387	1	CHRNA2	0.04 (0.01) - 0.09 (0.03)
Cocaine	2014	23958962	DSM-IV cocaine dependence	5,697b	4,291	1	FAM53B	NA
Cocaine	2019	31212010	DSM-IV cocaine dependence	6,378	2,085	0	HIST1H2BD	0.27 (0.03) - 0.30 (0.06)
OPIOIDS
Opioids	2014	24143882	DSM-IV dependence and criterion count	5,697b	2,066	3	PITPNM3, KCNC1, HHLA2 (in AA subset)	NA
Opioids	2016	26239289	DSM-IV dependence vs. nondependent daily injectors	1,328	1,167	1	CNIH3	NA
Opioids	2018	29478698	DSM-IV dependence; criterion counts used for analysis	3,058b	1,290	1	RGMA	NA
Opioids	2017	28115739	Methadone dose	1,410b	-	1	OPRM1	NA

^aNote: this study did not identify any genome-wide significant SNPs but did identify one genome-wide significant gene through gene-based analyses.

^bIndicates studies that included more than one ancestry (i.e., not only European Americans).

^cHeritability is given in raw estimates, rather than percentages; thus, an h²_SNP = 0.09 is equivalent to an h²_SNP = 9%. Only SNP-heritabilities (heritability estimated from all common SNPs assessed) are presented in the table; twin and family estimates for SUDs range from 50%–70. As the heritability estimated only from genome-wide significant SNPs is expected to be negligible and is often not reported, we do not report those statistics here.

Summary of key findings from GWAS:

For alcohol, “disease” cases defined using clinician ratings or self-reports on clinical interviews (e.g., alcohol dependence or alcohol use disorder (AUD)) and questionnaire-based assessments of recent problem drinking (e.g., the Alcohol Use Disorders Identification Test Problem subscale (AUDIT-P)(31)) have solidified the role of variants in the alcohol dehydrogenase genes (ADH1B, ADH1C, ADH4). The strongest finding in this cluster arises for rs1229984, a missense SNP in ADH1B that facilitates conversion of alcohol to acetaldehyde much more efficiently (up to 11 times the rate of the ADH1B*1 isoform), increases facial flushing in Asian populations, and is robustly associated with decreased risk of typical and heavy drinking and alcohol use disorders (32). The association between rs1229984 and AUD is amongst the most well-recognized in psychiatry, similar to the role of variants in apolipoprotein ε4 (APOEε4) in Alzheimer’s Disease(33,34). Another genome-wide significant locus of interest is in the gene encoding dopamine receptor D2 (DRD2). Meta-analyses of historical candidate gene searches that previously examined the association between DRD2 variants, particularly the “Taq1A” variant (rs1800497) and AUD (35), have been largely inconclusive. However, a recent GWAS finding for a variant that is within DRD2, although not in linkage disequilibrium (LD) with rs1800497 (Taq1A), provides hypothesis-free support for the role of variants in the gene in the etiology of problem drinking. Intriguingly, a large study of AUD in a veteran population(36) identified FTO, a gene linked to adiposity and body mass index(37); however, the finding was no longer significant when the confounding effects of body weight on alcohol consumption were accounted for. That same GWAS of AUD also identified a significant association at the Glucokinase Regulator gene (GCKR), which is involved in carbohydrate metabolism in the liver(38,39); GCKR has also been implicated in several studies of alcohol consumption (40,41), including the largest GWAS of drinks per week to date (27). Finally, a GWAS of maximum habitual alcohol intake measure in the Million Veteran Program data(42) reported a genetic correlation between this trait and alcohol dependence(43) to be 0.87 (p = 4.78e-9), and identified variants in ADH1B but also a variant rs77804065 in the region of CRHR1, the gene that encodes the corticotropin releasing hormone receptor which is a key component of hypothalamic-pituitary-adrenal (HPA) axis regulation. Furthermore, this finding actually increased in statistical significance in the trans-ancestral analyses of data from European and African ancestry individuals. While this region on chromosome 17 is gene dense, and the variant is an eQTL for KANSL1, the potential involvement of CRHR1 in drinking to alleviate stress is particularly intriguing as the region also includes genome-wide significant variants associated with re-experiencing post-traumatic stress disorder in the same sample(44).

Similar successes have been noted for nicotine dependence, which is variously assessed by DSM criteria and with the Fagerstrom Test for Nicotine Dependence(45), which includes items assessing cigarettes smoked per day and measures of tolerance. Results have centered around the nicotinic acetylcholine receptors, notably in the CHRNA5-CHRNA3-CHRNB4 cluster on chromosome 15 which includes the missense SNP, rs16969968, which has been shown to alter the function of the nicotinic receptor by modifying the extent to which nicotine can bind to it(46). Another GWAS(47) also identified CHRNA4 – the smoking cessation drug Varenicline is a partial agonist of the alpha 4/beta 2 receptor complex(48). Variants in both of these gene clusters have also been identified in the large GWAS of cigarette smoking. Much like alcohol, the role of variants involved in nicotine metabolism have emerged as strong signals for cigarette smoking but unlike for AUD, these metabolism-related variants have not been identified for nicotine dependence. Notably, variants in the cytochrome P450 (CYP2A6) enzyme-encoding gene were associated with cigarettes per day and smoking cessation (but as expected, not with smoking initiation or age at smoking initiation). The conversion of nicotine to cotinine by this family of enzymes has been previously used to classify individuals based on the speed of metabolism, with studies suggesting higher likelihood of successful cessation in slow versus fast metabolizers(49). The lack of findings with CYP2A6 for measures of nicotine dependence may be due to smaller sample size for the dependence GWAS.

Findings for cannabis, cocaine and opioid use disorder are more limited, due to smaller sample sizes and thus reduced statistical power. For cannabis use disorder (CUD), two findings are notable: first, a recent GWAS implicated a variant (rs56372821) that is an expression quantitative trait locus (eQTL) for CHRNA2, with evidence of a significant association between genetically-regulated expression of CHRNA2 in brain tissue and CUD(50). This particular variant has also been implicated in GWAS of ever smoking cigarettes(27) and schizophrenia(51) both of which are genetically correlated with CUD. Based on a series of sensitivity analyses, the researchers speculated that this variant exerts effects on liability to CUD that are independent of its effect on smoking or schizophrenia risk. In addition, a genome-wide significant variant (rs77378271) in CSMD1 (CUB and Sushi multiple domains 1 gene) was identified in a GWAS of cannabis dependence symptom counts (significant for both the European-American meta-analysis and the trans-ancestral analysis with African ancestry individuals(52).) Variants in CSMD1 have also been implicated in schizophrenia. For cocaine dependence, there has been one genome-wide significant variant identified (rs2629540), located within the FAM53B gene(53). This association, which co-localizes to a prior linkage region(54), was seen in a meta-analysis of European-American and African-American individuals, but appeared to be primarily driven by the African ancestry samples. Another study (which only focused on European ancestry individuals) did not identify any genome-wide significant SNPs, but did find a significant (at 10% FDR) association between cocaine dependence and HIST1H2BD in a gene-based analysis(55). This gene is located within the major histocompatibility complex (MHC) region, which has been robustly associated with schizophrenia (although the strongest signal in that region for schizophrenia has been isolated to the C4 genes (C4A and C4B)(56).) Finally, for opioid use disorder, individual studies have identified interesting signals but few, if any, have been consistently replicated (Table 1). Opioids are amongst a class of drugs with enormous therapeutic value, not only as analgesics but also as treatment for opioid dependence. A recent GWAS(57) utilized a novel approach of studying methadone maintenance dosing to identify a genome-wide significant locus near OPRM1, the gene encoding the mu-opioid receptor where methadone exerts its effect.

It should be noted that for most GWAS of SUD, the identified SNPs that reach genome-wide significance may not necessarily be the “causal” variants; the SNPs which reach the greatest statistical significance in a particular locus may simply be tagging (via linkage disequilibrium) the true causal variant. This is the case for a majority of the significant variants identified by GWAS, but there are also examples in which the true functional variant or gene has been identified (or validated) via GWAS, such as rs1229984 in ADH1B and rs16969968 in CHRNA5. Identifying the true “causal” variants - and, subsequently, the “causal” genes - and unraveling the mechanisms by which genome-wide signals influence disease risk can be challenging. While curated databases integrated within software packages such as FUMA(58), along-side fine-mapping tools, have shown significant promise for gene identification in silico, it is likely that fine-grained molecular analyses that emerge from these discoveries will be necessary. As an illustration, interdisciplinary efforts such as the one that demonstrated that the robust genetic associations with schizophrenia in the major histocompatibility complex region (which has complicated, long-range LD patterns) were driven primarily by the C4 protein (encoded by the complement component 4 genes, C4A and C4B, expression of which varies depending on an assortment of structurally diverse alleles) and its mediation of synaptic pruning(56), may help us elucidate the functional variants and genes contributing to SUD risk.

Use versus use disorder: differing genetic architectures?

Across SUD GWAS, genetic liability to these problematic aspects of substance involvement have been unequivocally linked to liability to other psychiatric outcomes, including robust genetic correlations with schizophrenia (r_g ranging from 0.2 – 0.37), major depressive disorder (r_g = 0.4 – 0.56) and attention deficit hyperactivity disorder (r_g = 0.44 – 0.5). Furthermore, genetic predisposition to SUDs has been shown to relate to susceptibility for greater socio-economic disadvantage and lower educational attainment(41,43,50), and overall, to lower subjective wellbeing (absolute value r_g ranging from 0.26 – 0.64). In contrast, and at least for alcohol and cannabis, the genetics of use (e.g., ever using or quantity-frequency) have markedly deviated from this pattern of genetic correlations(24,36; Figure 1).

Figure 1. Pattern of observed genetic correlations across substance use traits, SUDs, and other relevant trait categories.

Genetic correlations between cannabis use disorder and cardiometabolic traits have not been published, to our knowledge; thus, this cell is marked not applicable (“n/a”).

We illustrate with alcohol. The largest GWAS of alcohol consumption (drinks consumed in a typical drinking week) reported genetic correlations with tobacco smoking initiation, lifetime cannabis use, generalized risk tolerance, as well as several metabolic indices (e.g., negatively with insulin and obesity, and positively with HDL levels)(27). No associations with educational attainment or any psychiatric illnesses were statistically significant. Similar genetic correlations with tobacco smoking, cannabis use and risk-taking were also noted for the current largest GWAS of AUD, but in contrast, that GWAS also implicated psychiatric disorders, notably depression and depressive symptoms (r_g ~ 0.39 – 0.41), as being significantly genetically correlated with AUD, while simultaneously finding no support for a genetic overlap with metabolic traits. Importantly, the pattern of negative genetic correlations with educational attainment and positive genetic correlations with neighborhood disadvantage have been confirmed across multiple GWAS of problematic alcohol use(36,41,43). The divergence in patterns of genetic correlation (Figure 1) when comparing typical consumption of alcohol with problem drinking partly reflects the distinctions between dietary habits that include alcohol intake and maladaptive aspects (e.g., loss of control over drinking) associated with AUD, but further complexity underlies them. Even across different measures of alcohol consumption, we see varying patterns of genetic relationship with traits related to socioeconomic status. Marees et al. (59) examined whether quantity of alcohol consumption and frequency of consumption showed different patterns of association with socioeconomic status, SUD, psychiatric disorders, and personality traits. They found that measures of how much someone drinks (quantity), even during a period of typical use, is genetically more closely related to increased risk for AUD, other psychiatric disorders, and lower socio-economic status than how often one drinks (frequency). Another recent study used a genetic instrumental variable approach and found evidence that genetic instruments associated with increased educational attainment were also associated with increased weekly white and red wine intake; in contrast, these instruments were associated with decreased weekly distilled spirits intake(60), further suggesting heterogeneity as a function of social class. The non-representativeness of the sample that has been used to assess alcohol consumption, the UK Biobank, likely played some role in these findings – notably, this sample consists of older individuals that were skewed towards higher socioeconomic status(61). However, a similar contrast in genetic correlations was noted in the Million Veteran Program data, a sample that is likely partially enriched for problem drinking and psychiatric comorbidity (62). Collectively, these findings suggest that (a) alcohol consumption is more closely genetically related to metabolic characteristics and likelihood of engaging in risky behaviors, including experimenting with substances; (b) despite overlapping loci with consumption, the genetics of AUD in general indexes greater psychopathological vulnerability than GWAS of alcohol consumption.

Use versus use disorder: role of addictive potential

While a similar pattern has been informally observed for cannabis use and use disorder, the relative lack of genetic links between consumption and psychopathological aspects of SUD does not currently appear to hold for tobacco/cigarette smoking, despite a high degree of observed pleiotropy across tobacco smoking and alcohol use. In the largest GWAS to date that included the study of drinks/week, the genetic liability to cigarettes smoked in a typical day was positively correlated with liability to schizophrenia, and similar to AUD, negatively correlated with genetic predisposition to lower educational attainment(27). Similar to alcohol consumption, metabolic traits were also genetically correlated with cigarettes smoked per day, however with divergent effects such that genetic liability to cigarettes per day was positively correlated with obesity and insulin levels, and negatively with HDL levels (while genetic correlations with alcohol consumption were reversed). We speculate that these distinctions between two relatively similar indices of drug consumption (i.e., drinks/week and cigarettes/day) relate to the normativeness of alcohol (but not cigarettes) as an aspect of dietary practices (especially in the populations under study), the inherent increase in accuracy when quantifying cigarettes smoked (e.g., standard pack size) versus variations in drink sizes and ethanol content(63), and the significantly higher addictive potential of nicotine such that quantity (i.e., packs per day) is roughly an index of problem use. In other words, while a majority of heavy smokers are likely to be nicotine dependent, many other psychological and physiological features accompany heavy alcohol use before it corresponds to AUD.

Addictive potential refers to the likelihood of dependence on a drug once its use has been initiated. Drugs with a high addictive potential, such as nicotine and opioids, demonstrate very high conditional likelihood of problem use, while for substances with low addictive potential, such as alcohol and cannabis, distinct stages that involve casual and problem use are evident(64). Interestingly, twin studies support the role of drug-specific genetic influences on drugs of high addictive potential (17). It is presumed that earlier stages of substance use are more closely related to the broader genetic susceptibility to risk-taking(65), while later stages of dependence relate to other serious psychopathology as well as risk-taking; for drugs with high addictive potential, this genetic boundary between earlier and later stages maybe relatively blurred. However, even for such addictive substances, evidence for distinctions between early use and later problems is accumulating. For instance, similar to a recent GWAS (36) that identified a variant in DRD2 for AUD but not for alcohol consumption scores, the most recent GWAS of smoking behaviors also identified variants near DRD2 for cigarettes per day and smoking cessation but not for ever smoking cigarettes (27). We predict that delineating the key genetic differences between substance initiation or casual use versus problem use will continue to reveal important new insights in human genetic studies, of addiction especially as sample sizes increase for more addictive, less studied (in terms of GWAS) drugs such as opioids.

Next steps and opportunities in addiction genetics

GWAS have brought about numerous insights for psychiatric disorders, including SUDs, over the past five years. Combined with the expansion and development of publicly available databases with curated ‘omics data (e.g., GTEx and CommonMind(66–68)), key advances in GWAS annotation (e.g., FUMA(58)), and the integration of the two, we have begun to assign biological context, especially from a regulatory angle, to findings from human genetic studies. Still, there are areas for improvement where SUD research has lagged behind other psychiatric disorders, as well as areas of rich opportunity uniquely for the study of addictions. Here, we discuss some of these challenges and opportunities for the next five years of addiction GWAS:

• Increase SUD GWAS sample sizes: Despite substantial leaps over the past five years, SUD GWAS sample sizes still require significant increases in order to have enough power to detect common variants of relatively small effect size. This need is compounded by some evidence(43) that, even relative to other psychiatric disorders, SUD might have more polygenic genetic architectures; this means that GWAS of SUD will require larger samples to achieve the same success as GWAS of less polygenic disorders. Furthermore, it is clear that substance use and substance use disorders (SUD) have partially overlapping genetics but also some distinct genetic architecture. Thus, it is imperative that future GWAS collection efforts focus not only on easily-accessible measures of consumption, such as drinks per week or cannabis ever-use, but also measures of problem use, whether through short self-administered surveys such as the AUDIT, more stringent clinical diagnoses (e.g., DSM-5 criteria), or electronic health records data (e.g., ICD codes), when available.

• Integration with SUD-specific postmortem data: As the majority of GWAS associations have been found in regulatory regions, rather than directly disrupting protein-coding genes (69), integrating genome-wide findings with data from gene expression and 3D chromatin contact analyses have proven crucial for understanding how these genetic associations influence the phenotype. For example, Demontis et al. (50) identified a genome-wide significant SNP (rs56372821) that was an expression quantitative trait locus (eQTL) for CHRNA2 in brain tissues; in the cerebellum, the risk allele was significantly associated with decreased expression of CHRNA2. This analysis would not have been possible without the use of transcriptomic data publicly available through GTEx (66), as well as improved models for understanding the effects of genetically-regulated gene expression on phenotype through tools such as PrediXcan (70). Tools such as FUMA(58) allow for rapid annotation of GWAS findings using tissuespecific gene expression databases such as GTEx(66) and CommonMind(67). Despite these advances, the addiction field is currently limited by the scant number of postmortem brain samples specifically from SUD cases that can be compared with controls. For example, a recent landmark study(71) examined gene-expression patterns in postmortem prefrontal cortex tissue of individuals from five psychiatric disorders (AUD, N=17; autism, N=50; bipolar disorder, N=94; depression, N=87; and schizophrenia, N=159)) compared to controls (N=293). The authors found patterns of shared gene-expression across disorders that paralleled polygenic overlap among the disorders except for AUD, where transcriptomic correlations were reversed. Whether this finding was related to a true difference in expression profiles, study design differences, or impacted by low sample size requires further analysis. Kapoor et al(72) used a substantially larger sample of AUD prefrontal cortex (N=65) data to identify 129 differential expressed genes and show enrichment of GWAS signals for AUD, and to a lesser extent, for alcohol consumption. If biological discoveries for SUD are to keep pace with other psychiatric disorders, the requirement to increase expression data in various brain regions will be necessary. Notably, chronic substance exposure has wide ranging physiological impact and therefore, expression profiles in other tissue types (e.g., liver for alcohol) will also be valuable.

• Epigenetics: biomarkers vs. indices of change: The epigenetic effects of alcohol across the stages of use - exposure, regular consumption, withdrawal - have been of particular interest (73), but have thus far largely been limited by small sample sizes and limited tissue types (74–76). However, a relatively recent study performed a cross-tissue (brain, blood, and liver) epigenome-wide analysis of individuals with AUD and identified a significant differentially methylated region in the promoter region of the proprotein convertase subtilisin/kexin 9 (PCSK9) gene, which is primarily expressed in the liver and functions as a regulator of LDL cholesterol (77). In a translational mouse model, the researchers further demonstrated that alcohol exposure led to downregulation of PCSK9. Similarly, the epigenetic effects of nicotine have been an area of active interest in animal models (e.g.,(78)) and studies in human tissue continue to emerge(79). While there have been few (if any) published epigenome-wide analyses of CUD to date, the epigenetic effects of cannabis exposure has been an area of keen and growing interest for the field (80). Importantly, longitudinal studies that are able to contrast epigenetic marks as biomarkers (simply cross-sectionally correlated with drug exposure) vs. dynamic indices of environmental injuries over time will help to disentangle the role of substance exposure as, both, a consequence and provocateur of dynamic DNA changes. Furthermore, when integrated with GWAS data, a variant that correlates with methylation subsequent to substance exposure might deserve greater priority than one with global meQTL effects, further underscoring the need for drug-specific epigenetic profiles.

• Discovery in non-European samples: A deficiency in most genetic studies of behavioral traits is that they have, thus far, relied heavily on data from participants of European ancestry(81,82). While GWAS of SUD have typically included more non-European ancestry individuals than other psychiatric disorders, even the largest representative non-European ancestral group of African Americans (~56,000 (36)) is only about six percent of the largest sample size for discovery in Europeans (~900,000 (27)). Studying non-European ancestry samples is important for a variety of reasons: first, if GWAS and secondary analyses (including polygenic risk score analyses) are conducted only in individuals of European descent, it has been shown that Europeanancestry individuals are also the ones most likely to benefit from any clinically-useful findings (82), which may become an even more pertinent limitation in the future as GWAS become more powerful and produce more accurate polygenic prediction. There are important genetic distinctions in allele frequencies and the LD architecture across ancestral groups, which severely restricts the use of summary statistics from one ancestral group in polygenic risk prediction in another group(83,84). For instance, one study found that even a substantially smaller GWAS of alcohol dependence conducted in individuals of African-American descent was a superior predictor of alcohol dependence in an independent sample of the same ancestry than was a larger GWAS in individuals of European descent (Nagelkerke’s R² ~ 1.61% vs. 0.36%) (43). A second motivation for studying genetic associations in other populations is that this practice actually improves fine-mapping efforts (81). For example, while AUD associated signals in ADH1B are amongst the strongest in psychiatry, the actual genetic variants of effect are distinct in Asian and European (lead SNP rs1229984) and African-ancestry (lead SNP rs2066702) samples (43). Undoubtedly, buried in this region of chromosome 4, as well as genome-wide, there are likely to be novel causal variants that could be identified in an ancestry-specific manner. This relative lack of studies of non-European populations is particularly problematic for SUD, which have varied patterns of onset, severity, remission and relapse, across populations, cultures and communities; in the context of also being genetically understudied, this only furthers clinical disparities(82). Furthermore, a heightened awareness of genetic admixture in other populations (e.g., LatinX(85)) presents opportunities for continuing to study the development of these disorders in the changing landscape of world-wide communities. Even the study of populations where variants of very strong effect are near fixation in the population could result in promising new clues. In certain East Asian populations, a variant in ALDH2 limits the efficiency with which acetaldehyde, generated from the metabolism of alcohol, is converted to acetate. The accumulation of acetaldehyde has been implicated in flushing, a reportedly aversive physiological response (facial reddening, nausea, dizziness) that impedes the likelihood of progression to problem drinking. The effect size associated with rs671 in ALDH2 is so large that it is easily detected at genome-wide significant levels even in small samples of less than 1000 individuals. In larger Asian samples (N=512,715 (86)), combining the effects of rs1229984 (ADH1B) and rs671 (ALDH2) has provided unprecedented precision in being able to quantify the effect of genotype in sex-limited and regionally variable patterns of alcohol intake in China, but also opens up the possibility of discovery of additional loci.

• Cross substance and cross disorder genetics: Another important insight from GWAS has been the pervasive nature of pleiotropy(69); in other words, genetic associations identified for SUD are unlikely to solely influence the trait under study. Genetic variants that contribute to AUD, for example, are likely to influence other SUD too, as well as other complex traits (e.g., educational attainment) and psychiatric disorders, cementing their categorization as psychopathology with genetic variation relating to neurobiological dysfunction(12). A recent cross-disorder GWAS of eight psychiatric disorders identified 109 pleiotropic loci (which, on average, showed increased expression in the brain beginning in the second gestational trimester and throughout the lifespan) and leveraged genetic correlations to identify three meaningful subgroups of inter-related disorders(87). This cross-disorder effort did not include any SUD data; thus, next steps for the field of addiction include two broad cross-disorder efforts: amongst different SUDs themselves (i.e., polydrug use disorders), and between SUDs and other psychiatric disorders.

• Cross-species analyses - increase collaboration among human and animal geneticists: Among psychiatric disorders, addictions are probably some of the most well-studied in terms of non-human animal models. Well-validated animal paradigms (e.g., self-administration, conditioned place preference) can thus afford an added source of functional validation for GWAS findings. The schism between human and animal genetics occurs at several levels, including prevailing research philosophies (e.g., candidate focus vs. unbiased approaches) and lack of perfect behavioral consilience across species (e.g., limited overlap between common alcohol consumption phenotypes studied in animals vs. light or “social” drinking in humans(88)). However, databases such as GeneWeaver(89) - curated gene-sets from functional genomic studies in model organisms - provide a starting point for such integration. Of course, there will be challenges in collaborating across human and animal genetics - for example, cross-species comparisons are typically done at the level of the gene (although variant-level analyses are currently being developed, and there is already some evidence of convergence across human and rodent GWAS(90)), but more advanced future methods will need to account for imperfect gene homology across the species.

Conclusion

It is likely that additional loci will be discovered for all SUDs in the next five years, as large-scale efforts to aggregate electronic health record data, and national repositories such as All of Us (https://allofus.nih.gov) begin to come to fruition. AUD are amongst the most common diagnoses in medical records, however the extent of unmet healthcare need for SUDs in general, and especially for illicit drugs, raises concerns about high rates of false negatives. Coordinated efforts that leverage genetic approaches to combine various types of data, such as the Psychiatric Genomics Consortium’s Substance Use Disorders Working Group, can offer one possible pathway to further discovery. In this review, we have outlined a few next steps for the field of addiction genetics research. One aspect that this review does not delve into is the translation of genetic findings to the clinic, which is a key goal of psychiatric genetics research. Notably, clinical utility extends beyond our confidence in a particular genetic finding (i.e., the strength of its statistical association) to how that finding might be communicated to patients, including the evaluation of the overall practical benefit relative to potential risks associated with knowledge regarding these findings. These latter and critical aspects of genetic literacy involve the contributions of many lines of expertise (e.g., genetic counsellors, patient advocates). From a statistical perspective, prediction models (typically in the form of polygenic scores) derived from GWAS of SUD are not yet ready for clinician use, but there is hope that this will change with larger (and more diverse) samples. In addition, there have been recent efforts to harness GWAS data for drug repurposing, using tools such as Drug Targetor (91); this approach holds promise for bringing findings from genetic studies of SUDs into actual clinical implementation. In this “post-GWAS” era(92), the strategy of multi-disciplinary, collaborative research - extending across modalities and among interdisciplinary groups of geneticists, bioinformaticists, and clinicians - will be essential to shaping our understanding of the genetics of SUD and the biological mechanisms underlying these complex disorders.