Background
Illicit and hazardous drugs, such as cocaine, methamphetamine, inhalants, stimulants, and opioids, pose significant health risks, contributing to injuries, psychiatric and physical diseases, and an increased risk of death (Afonso, Mohammad, & Thatai, Reference Afonso, Mohammad and Thatai2007; Isoardi et al., Reference Isoardi, Mudge, Harris, Dimeski and Buckley2020; Jones & Rayner, Reference Jones and Rayner2015; Kaye, McKetin, Duflou, & Darke, Reference Kaye, McKetin, Duflou and Darke2007; Maraj, Figueredo, & Morris, Reference Maraj, Figueredo and Morris2010; Mick, McManus, & Goldberg, Reference Mick, McManus and Goldberg2013; Valente et al., Reference Valente, Carvalho, Bastos, de Pinho and Carvalho2012; van der Woude, Reference van der Woude2000; Verna, Schluger, & Brown, Reference Verna, Schluger and Brown2019). Nevertheless, drug use is a widespread phenomenon with increasing prevalence (Barocas et al., Reference Barocas, White, Wang, Walley, LaRochelle, Bernson and Linas2018; Han et al., Reference Han, Compton, Jones, Einstein and Volkow2021; John & Wu, Reference John and Wu2017; Wu & Ringwalt, Reference Wu and Ringwalt2006; Yockey, King, & Vidourek, Reference Yockey, King and Vidourek2020). Therefore, both environmental and genetic factors influencing the use of these drugs are major public health concerns.
In 2023, the past-year prevalence of cocaine use in the United States was 1.8%, inhalant use was 0.9%, and methamphetamine use was 0.9%. In addition, there was a past-year 1.4% frequency of prescription stimulant use and 3.1% use of legal and illegal opioids (SAMHSA, 2023). The lifetime prevalence of opioid use and misuse was between 11.9% and 37.8% (Han et al., Reference Han, Compton, Blanco, Crane, Lee and Jones2017; Zajacova et al., Reference Zajacova, Grol-Prokopczyk, Limani, Schwarz and Gilron2023), while the prevalence of illegal street opioid use, such as heroin, was between 1.6% and 1.84% (Ihongbe & Masho, Reference Ihongbe and Masho2016; Martins et al., Reference Martins, Sarvet, Santaella-Tenorio, Saha, Grant and Hasin2017). The lifetime prevalence of prescription stimulant use is 9.5% (McCabe & West, Reference McCabe and West2013).
Transition rates from initial use to developing a use disorder or addiction vary between different drugs. For example, between 0.3% and 0.4% of the population reportedly develops cocaine use disorder (CocUD) (John & Wu, Reference John and Wu2017; Kerridge et al., Reference Kerridge, Chou, Pickering, Ruan, Huang, Jung and Hasin2019) (suggesting a transition rate of up to ~2.5%), whereas 0.8–4.6% develop opioid use disorder (OUD) (Barocas et al., Reference Barocas, White, Wang, Walley, LaRochelle, Bernson and Linas2018; Han et al., Reference Han, Compton, Blanco, Crane, Lee and Jones2017) (suggesting a transition rate of up to ~39%). There is also high comorbidity between the use of different types of drugs; for example, cocaine use is prevalent in 11.8% of heavy alcohol users, with 2.10% of this population developing CocUD (John & Wu, Reference John and Wu2017). In addition, 73% of individuals with OUD report co-use of other substances (Mahoney, Marshalek, Haut, Hodder, et al., Reference Mahoney, Winstanley, Lander, Berry, Marshalek, Haut and Hodder2021), and 27.5% of them develop alcohol use disorder (Baskerville et al., Reference Baskerville, Grodin, Lin, Miotto, Mooney and Ray2023). Tobacco smoking is strongly associated with opioid use and OUD (Rajabi et al., Reference Rajabi, Dehghani, Shojaei, Farjam and Motevalian2019).
There is also high comorbidity between substance use and various psychiatric disorders. The point prevalence of depression is 24% among cocaine-dependent users (López & Becoña, Reference López and Becoña2007) and 27–61% among opioid users (Rogers et al., Reference Rogers, Zvolensky, Ditre, Buckner and Asmundson2021), compared to 5% in the general population (WHO, 2023). The odds of depression among methamphetamine users are 18–66% higher than in the general population (Leung et al., Reference Leung, Mekonen, Wang, Arunogiri, Degenhardt and McKetin2023) and 32% higher among inhalant users (Gentile, Bianco, Nordstrom & Nordstrom, Reference Gentile, Bianco, Nordstrom and Nordstrom2021). Other disorders, such as attention-deficit/hyperactivity disorder (ADHD) (Obermeit et al., Reference Obermeit, Cattie, Bolden, Marquine, Morgan, Franklin and AIDS2013; Oliva et al., Reference Oliva, Mangiapane, Nibbio, Berchialla, Colombi and Vigna-Taglianti2021; Wilens et al., Reference Wilens, Adler, Adams, Sgambati, Rotrosen, Sawtelle and Fusillo2008), schizophrenia (Hunt et al., Reference Hunt, Large, Cleary, Lai and Saunders2018), post-traumatic stress disorder (PTSD) (Dahlby & Kerr, Reference Dahlby and Kerr2020), and bipolar disorder (BD) (Cerullo & Strakowski, Reference Cerullo and Strakowski2007), are also more common in people with histories of substance use. Drugs, such as cocaine, opioids, and stimulants, pose health risks other than psychiatric effects, which can deeply impair the quality of life and may be life-threatening; these include hypertension, elevated heart rate, myocardial infarction and heart failure (Afonso et al., Reference Afonso, Mohammad and Thatai2007; Kaye et al., Reference Kaye, McKetin, Duflou and Darke2007; Maraj et al., Reference Maraj, Figueredo and Morris2010; Mick et al., Reference Mick, McManus and Goldberg2013), liver damage (Valente et al., Reference Valente, Carvalho, Bastos, de Pinho and Carvalho2012; Verna et al., Reference Verna, Schluger and Brown2019; Zhao et al., Reference Zhao, Zhai, Song, Wu, Ge, Zhang and Jiao2020), and kidney injury (Isoardi et al., Reference Isoardi, Mudge, Harris, Dimeski and Buckley2020; Jones & Rayner, Reference Jones and Rayner2015; Valente et al., Reference Valente, Carvalho, Bastos, de Pinho and Carvalho2012; van der Woude, Reference van der Woude2000). Out of 600,000 drug-related deaths in a 37-year period, the vast majority were caused by cocaine, prescription opioids, and street opioid overdose (Jalal et al., Reference Jalal, Buchanich, Roberts, Balmert, Zhang and Burke2018). However, this study did not include alcohol- and tobacco-related deaths, which are estimated to be even higher (Karaye, Maleki, Hassan, & Yunusa, Reference Karaye, Maleki, Hassan and Yunusa2023; Siegel et al., Reference Siegel, Jacobs, Newton, Feskanich, Freedman, Prentice and Jemal2015). Opioid abuse has been considered a national epidemic in the United States (Volkow & Blanco, Reference Volkow and Blanco2021), with annual opioid-associated mortality rates rising to 73,838 in 2022 (NIH, 2022). In total, ~150,000 people die every year in the United States due to drug overdose (CDC, 2024).
Most genome-wide association studies (GWASs) of substance use traits have dealt with substance use frequency, substance dependence, and substance use disorders (SUDs) (Dao et al., Reference Dao, Zhou, Small, Gordon, Li, Kember and Justice2021; Deak et al., Reference Deak, Zhou, Galimberti, Levey, Wendt, Sanchez-Roige and Gelernter2022; Levey et al., Reference Levey, Galimberti, Deak, Wendt, Bhattacharya, Koller and Program2023; Sanchez-Roige et al., Reference Sanchez-Roige, Palmer, Fontanillas, Elson, Adams, Howard, Edenberg, Davies, Crist, Deary, McIntosh and Clarke2019; Zhou et al., Reference Zhou, Kember, Deak, Xu, Toikumo, Yuan and Gelernter2023), and less so with substance lifetime use (LU) (with the exception of tobacco smoking initiation, which is similar to an LU trait; e.g. see Saunders et al., Reference Saunders, Wang, Chen, Jang, Liu, Wang and Vrieze2022). Although it is clear why SUDs are of great interest, the liability to use drugs should be investigated too, considering the high transition rates from use to use disorder and possible hazards that arise from nondependent use. In addition, studies conducted so far regarding cannabis (Levey et al., Reference Levey, Galimberti, Deak, Wendt, Bhattacharya, Koller and Program2023; Pasman et al., Reference Pasman, Verweij, Gerring, Stringer, Sanchez-Roige, Treur and Vink2018) and alcohol (Saunders et al., Reference Saunders, Wang, Chen, Jang, Liu, Wang and Vrieze2022; Zhou et al., Reference Zhou, Kember, Deak, Xu, Toikumo, Yuan and Gelernter2023), for example, indicate that substance LU (SubLU) and SUD traits are somewhat different from one another genetically. Here, we conducted GWAS analyses to investigate the genetic basis of LU of cocaine, methamphetamine, inhalants, illegal opioids, prescription opioids, and prescription stimulants in European (EUR), African (AFR), and Latin American (AMR) ancestry research participants. We examined the genetic and phenotypic similarities and differences between these traits and their global and local genetic correlations with traits of interest, including chronic pain, SUDs, depression, and other psychiatric, medical, and socioeconomic traits. We also evaluated the genetic underpinnings of the inclination to use illicit drugs of any kind, as well as the amount of drugs a person uses over a lifetime.
Methods
See also Supplementary Methods.
Cohorts
We included subjects of EUR, AFR, and AMR ancestries from the All of Us (AoU; v7) biobank. Genotyping and quality control procedures were described previously (Bick et al., Reference Bick, Metcall, Mayo, Lichtenstein, Rura, Carroll and Genomics2024). LU of cocaine, methamphetamine, inhalants, street opioids, prescription opioids, and prescription stimulants phenotypes were defined using a lifestyle survey. The ‘any substance’ (SubLU) and quantitative ‘number of substances used’ (nSubLU) phenotypes were defined using the same survey.
GWAS analysis and meta-analysis
GWAS analyses were conducted using PLINK 2.0, with sex, age, and the first 10 genetic PCs as covariates. We removed subjects due to relatedness, and variants were excluded due to minor allele frequency <0.1% and Hardy–Weinberg equilibrium p < 1 × 10−6. Cross-ancestry meta-analyses were performed using METAL (Willer, Li, & Abecasis, Reference Willer, Li and Abecasis2010).
Genetic correlations and SNP-based heritability
We used linkage disequilibrium score regression (LDSC) (Bulik-Sullivan et al., Reference Bulik-Sullivan, Loh, Finucane, Ripke, Yang and Neale2015) to calculate single-nucleotide polymorphism (SNP)-based heritability (h 2) for all the traits and inter-trait genetic correlations between all six individual SubLU traits. Then, we calculated the genetic correlation between each trait, including the composite traits SubLU and nSubLU, and 12 selected traits (Deak et al., Reference Deak, Zhou, Galimberti, Levey, Wendt, Sanchez-Roige and Gelernter2022; Demontis et al., Reference Demontis, Walters, Athanasiadis, Walters, Therrien, Nielsen and Borglum2023; Docherty et al., Reference Docherty, Mullins, Ashley-Koch, Qin, Coleman, Shabalin and Ruderfer2023; Johnston et al., Reference Johnston, Adams, Nicholl, Ward, Strawbridge, Ferguson and Smith2019; Levey et al., Reference Levey, Stein, Wendt, Pathak, Zhou, Aslan and Gelernter2021; Levey et al., Reference Levey, Galimberti, Deak, Wendt, Bhattacharya, Koller and Program2023; Nievergelt et al., Reference Nievergelt, Maihofer, Atkinson, Chen, Choi, Coleman and Koenen2024; O’Connell et al., Reference O’Connell, Koromina, van der Veen, Boltz, David and Yang2025; Trubetskoy et al., Reference Trubetskoy, Pardinas, Qi, Panagiotaropoulou, Awasthi and Bigdeli2022; Watanabe et al., Reference Watanabe, Jansen, Savage, Nandakumar, Wang, Research and Posthuma2022; Zhou et al., Reference Zhou, Kember, Deak, Xu, Toikumo, Yuan and Gelernter2023) (Supplementary Table S1). For this analysis, we selected substance use traits with well-powered GWAS available. Psychiatric traits were included if they were previously phenotypically associated with substance use traits, such as schizophrenia, depression, BD, and PTSD (Dahlby & Kerr, Reference Dahlby and Kerr2020; Hunt et al., Reference Hunt, Large, Cleary, Lai and Saunders2018; López & Becoña, Reference López and Becoña2007; Rogers et al., Reference Rogers, Zvolensky, Ditre, Buckner and Asmundson2021), or if they are commonly associated with the use of, or treatment via, specific substances (e.g. ADHD, which is associated with stimulant use [Wilens et al., Reference Wilens, Adler, Adams, Sgambati, Rotrosen, Sawtelle and Fusillo2008], and chronic pain, which is associated with opioid use [Weiss et al., Reference Weiss, Potter, Griffin, McHugh, Haller, Jacobs and Rosen2014]). The trait ‘academic degree’ was selected to represent cognitive function (Braatveit, Torsheim, & Hove, Reference Braatveit, Torsheim and Hove2018).
Phenotypic correlations
We estimated inter-trait phenotypic correlations using χ 2- and φ-coefficient (r φ).
Local genetic correlations
We used local analysis of covariant association (LAVA) (Werme, van der Sluis, Posthuma, & de Leeuw, Reference Werme, van der Sluis, Posthuma and de Leeuw2022) to calculate inter-trait local genetic correlations between all six SubLU traits (a total of 15 pairs). Then, we calculated the local genetic correlations between each trait – including SubLU and nSubLU – and 12 selected traits of interest (Supplementary Table S1).
Cross-ancestry genetic correlations
We used Popcorn (Brown et al., Reference Brown, Ye, Price, Zaitlen and Network2016) to calculate the cross-ancestry genetic correlations between the SubLU trait in AFR and AMR populations and a selected list of traits in EUR. We used the same set of traits that were measured for genetic correlation among EUR using LDSC (see above). We applied the Benjamini–Hochberg procedure for correction of false discovery rate (FDR).
Genomic structural equation modeling (gSEM)
We utilized gSEM (Grotzinger et al., Reference Grotzinger, Rhemtulla, de Vlaming, Ritchie, Mallard, Hill and Tucker-Drob2019) to examine the underlying latent factor structure of the six individual SubLU traits, along with nine other psychiatric and health-related traits associated with substance use specifically.
Results
GWAS analyses
The average number of participants for the analysis of the six SubLU traits (cocaine, methamphetamine, inhalants, street opioids, prescription opioids, and prescription stimulants) was 115,578 in EUR, 47,039 in AFR, and 36,407 in AMR (for details, see Table 1). Demographics for the nSubLU trait are presented in Supplementary Table S2. For each trait, we also conducted a cross-ancestry meta-analysis. All lead SNPs were significant in the range of 5 × 10−8 and 1 × 10−9, except as noted (Table 2). Fifteen analyses yielded at least one significant SNP (Manhattan plots and regional plots in Supplementary Figures S1–15).
Table 1. Sample size for each trait
Table 2. Lead SNPs of substance lifetime use traits
a Noncoding RNA.
For Cocaine LU, we found two genome-wide significant (GWS) variants in EUR (CADM2*rs146999751 and GBF1*rs60331671) and one in AFR (MTNR1A*rs78664860). A cross-ancestry meta-analysis revealed four significant loci: LINC02008*rs9869718, CADM2*rs4856591, TCF4*rs72930769, and MTNR1A*rs78664860. For Inhalants LU, we found two significant hits: CDH12*rs4235547 in EUR and CYP24A1*rs11699278 in AMR. A cross-ancestry meta-analysis revealed two intergenic hits. For Methamphetamine LU, we found one significant variant in AMR in a noncoding region (rs141493660, p = 8.6 × 10−10) and a different one in a cross-ancestry meta-analysis (EXT1*rs17431748). For Street Opioids LU, we found a significant hit only in AFR (MAGI1*rs113325386), which was preserved in the cross-ancestry meta-analysis. For Prescription Opioids LU, we found one significant protein-coding locus in AMR (CNTN5*rs7298982), and one in a cross-ancestry meta-analysis (GRAMD1A*rs1052457445). For Prescription Stimulants LU, we found significant hits in AMR (TBCD*rs55775765), and in a cross-ancestry meta-analysis (MEF2C-AS2*rs116758901).
We then conducted a GWAS of composite (cumulative) traits: an ‘any substance’ LU (SubLU) trait, in which cases were defined as subjects that used any of the substances included for the analyses described above; and a quantitative trait of ‘number of substances used’ (nSubLU), defined as the number of substances subjects listed as ‘ever used’ in their lifetime (a range of 0–6 substances). The number of participants used for the analysis was 115,635 in EUR, 47,086 in AFR, and 36,439 in AMR (Supplementary Table S2). All lead SNPs were significant in the range of 5 × 10−8 and 10−9, unless stated otherwise (Table 2).
For SubLU, the ‘any substance’ trait, we found three significant variants in EUR (CADM2*rs1821351, p = 1.29 × 10−10; LINC02758*rs7931884, p = 1.55 × 10−10; and rs4813097, which is located in an overlapping area of two genes: SDCBP2-AS1 and FKBP1A-SDCBP2) (Figure 1a) and one in AFR (DPRX*rs1600561869) (Supplementary Figure S16). A cross-ancestry meta-analysis revealed two significant loci that were also significant for EUR (CADM2*rs1821351, p = 7.09 × 10−10, and LINC02758*rs7931884) and three new significant SNPs (EBF1*rs116350745 and the intergenic rs76518228 and rs12489967) (Figure 1b and Supplementary Figure S19–20).
Figure 1. Manhattan plot of substance lifetime use (LU) (LU of one or more of the drugs discussed in this study) in (a) EUR (nCases = 34,017, nControls = 81,618); (b) cross-ancestry meta-analysis (nCases = 54,237, nConrols = 144,923), and number of lifetime substances used (the number of different drugs, of the drugs discussed in this study, and a subject used in his or her lifetime; a quantitative trait with a range of 0–6) in (c) EUR (n = 115,635); (d) cross-ancestry meta-analysis (n = 199,190).
For nSubLU, the quantitative trait, we found two hits in EUR (GBF1*rs60331671 and LINC02758*rs7931884) (Figure 1c and Supplementary Figure S21): one in AFR (the intergenic rs141793374) (Supplementary Figure S17), and two in AMR (LOC124900725*rs17465728 and the intergenic rs149549558) (Supplementary Figure S18). A cross-ancestry meta-analysis revealed six SNPs that were significantly associated with nSubLU: the AFR-bound rs141793374 and five that were not significant for the individual ancestries: ACTN2*rs2275399, CD47*rs327133, LOC105372130*rs28758902, and the intergenic rs35149938 and rs58899690 (Figure 1d and Supplementary Figure S22).
Inter-trait genetic correlations
LDSC was used to calculate SNP-based heritability (h 2) and inter-trait genetic correlations for the individual SubLU traits in EUR. Heritability estimates ranged between 8.08% (for Street Opioids LU; SE = 0.009) and 13.9% (for Prescription Opioids LU; SE = 0.019). Intercept measures ranged between 0.992 and 1.012, and attenuation ratios were of 0.085 or lower (Supplementary Table S3). Inter-trait genetic correlations were calculated between all six individual SubLU traits – a total of 15 pairs. A significant positive moderate-to-strong correlation was found between all pairs. The strongest correlations were between Street Opioids LU and Prescription Opioids LU (r g = 0.968, p = 5.87 × 10−68), Street Opioids LU and Methamphetamine LU (r g = 0.965, p = 9.96 × 10−129), and Cocaine LU and Methamphetamine LU (r g = 0.95, p = 1.66 × 10−133). In total, 12 pairs of traits had a strong genetic correlation of r g > 0.7 (Figure 2a and Supplementary Table S4).
Figure 2. (a) Inter-trait genetic correlations between all six individual substance lifetime use (LU) traits in EUR. (b) Inter-trait phenotypic correlations between all six individual substance LU traits in EUR. (c) Genetic correlations between all six individual and cumulative substance LU traits in EUR and a selected list of traits. Statistically nonsignificant values are in dark gray Note: ADHD, ‘attention-deficit/hyperactivity disorder’; BD, ‘bipolar disorder’; CanUD, ‘cannabis use disorder’; ns, ‘nonsignificant’; OUD, ‘opioid use disorder’; PAU, ‘problematic alcohol use’; PTSD, ‘post-traumatic stress disorder’ [Deak et al., Reference Deak, Zhou, Galimberti, Levey, Wendt, Sanchez-Roige and Gelernter2022; Demontis et al., Reference Demontis, Walters, Athanasiadis, Walters, Therrien, Nielsen and Borglum2023; Doherty et al., Reference Doherty, Smith-Byrne, Ferreira, Holmes, Holmes, Pulit and Lindgren2018; Johnston et al., Reference Johnston, Adams, Nicholl, Ward, Strawbridge, Ferguson and Smith2019; Levey et al., Reference Levey, Galimberti, Deak, Wendt, Bhattacharya, Koller and Program2023; Levey et al., Reference Levey, Stein, Wendt, Pathak, Zhou, Aslan and Gelernter2021; Nievergelt et al., Reference Nievergelt, Maihofer, Atkinson, Chen, Choi, Coleman and Koenen2024; O’Connell et al., Reference O’Connell, Koromina, van der Veen, Boltz, David and Yang2025; Trubetskoy et al., Reference Trubetskoy, Pardinas, Qi, Panagiotaropoulou, Awasthi and Bigdeli2022; Watanabe et al., Reference Watanabe, Jansen, Savage, Nandakumar, Wang, Research and Posthuma2022; Zhou et al., Reference Zhou, Kember, Deak, Xu, Toikumo, Yuan and Gelernter2023].
Inter-trait phenotypic correlations
Inter-trait phenotypic correlations were calculated for individual SubLU traits using χ 2-estimates and φ-coefficients. The analyses were conducted separately for EUR, AFR, and AMR, and revealed positive weak-to-moderate phenotypic correlations between all the traits in all the ancestries (Supplementary Table S5). In EUR (Figure 2b) and AMR (Supplementary Figure S23), the strongest correlation was observed between Cocaine LU and Methamphetamine LU (r φ = 0.5 and r φ = 0.47, respectively). In AFR (Supplementary Figure S24), the strongest correlation was between Cocaine LU and Street Opioids LU (r φ = 0.31).
Inter-trait local genetic correlations
Inter-trait local genetic correlations were calculated for individual SubLU traits in EUR using LAVA (Werme et al., Reference Werme, van der Sluis, Posthuma and de Leeuw2022). For every trait, only regions that reached the significance threshold of p < 0.05 were used to calculate genetic correlations with the other traits (a total of 3,242 regions in 15 pairs). After Bonferroni correction for 3,242 tests, the statistically significant threshold was set on p = 1.54 × 10−5. All 15 pairs had genetic correlations in at least three regions, and all the correlations were positive. The most highly correlated traits were Street Opioids LU and Methamphetamine LU (40 significantly correlated genetic regions), Street Opioids LU and Prescription Opioids LU (37 regions), and Street Opioids LU and Inhalants LU (37 regions). The most prominent region was region 1,225 (chr7:152253295–153241228), which genetically correlated eight pairs (the full data are presented in Supplementary Table S6).
Correlations between SubLU traits and other traits of interest
LDSC was used to calculate genetic correlations between the individual SubLU traits in EUR and 12 selected traits related to substance use, substance dependence, and common psychiatric illnesses (Figure 2c and Supplementary Table S7). In total, 72 pairs were analyzed (6 substance use traits × 12 ‘other’ traits); there were only 4 instances of nonsignificant (but nominally correlated) relationships, all of them involving Inhalants LU. For all other pairs, there were significant genetic correlations with a wide spectrum of statistical significance: the weakest effect was for Prescription Opioids LU and schizophrenia (r g = 0.193, p = 4.64 × 10−6) and the strongest was for Street Opioids LU and cannabis use disorder (CanUD) (r g = 0.843, p = 1.82 × 10−66). For 10 of the 11 traits tested, all the correlations with SubLU traits were positive, while for academic degree, all the correlations were negative. The two strongest correlations of OUD were with opioid use traits (Street Opioids LU: r g = 0.789, p = 2.4 × 10−24; Prescription Opioids LU: r g = 0.749, p = 4.31 × 10−14), and the strongest correlation of chronic pain was with Prescription Opioids LU (r g = 0.487, p = 4.05 × 10−25). Even though prescription stimulants are prescribed mostly to treat ADHD (Piper et al., Reference Piper, Ogden, Simoyan, Chung, Caggiano, Nichols and McCall2018), Methamphetamine LU (r g = 0.665, p = 5.10 × 10−45) and Cocaine LU (r g = 0.6212, p = 8.10 × 10−43) had higher genetic correlation with ADHD compared to Prescription Stimulants LU (r g = 0.528, p = 4.02 × 10−26). We also calculated the genetic correlations between SubLU and nSubLU and the same 12 selected traits. All the tests had significant results. For all the traits, the r g values with SubLU and nSubLU were highly similar. Nevertheless, in all cases, the correlation with nSubLU had much lower p-values.
Local genetic correlations between SubLU traits and other traits of interest
Local genetic correlations between individual SubLU traits in EUR and other traits of interest were calculated using LAVA (Werme et al., Reference Werme, van der Sluis, Posthuma and de Leeuw2022). For every trait, only regions that reached the significance threshold of p < 0.05 were used to calculate genetic correlations with the other traits (a total of 15,841 regions in 96 pairs). After Bonferroni correction for 15,841 tests, the statistical significance threshold was set at p = 3.16 × 10−6. In total, 25 significant correlations between individual and cumulative SubLU traits and the other traits of interest were found. Methamphetamine LU was the SubLU trait with the greatest number of shared regions, with six different associations, two of them with CanUD. One region – 1,292 (chr8: 55275355–56346878) – was associated with problematic alcohol use (PAU) as well as two individual LU traits: Inhalants LU and Methamphetamine LU. Region 1,966 (chr14:33591114–34695195) associated schizophrenia with both Prescription Stimulants LU and nSubLU. Region 727 (chr4:139553761–141087047) associated chronic pain with Prescription Stimulants LU and BD with Street Opioids. Region 267 (chr2:59251997–60775066) associated CanUD with Methamphetamine LU, BD with Cocaine LU, academic degree with SubLU, and BD with SubLU. All the other significant regions were associated with one pair of traits (Supplementary Table S8).
Cross-ancestry genetic correlations
Using Popcorn (Brown et al., Reference Brown, Ye, Price, Zaitlen and Network2016), genetic correlations were calculated for SubLU in AMR and AFR against a set of traits of interest in EUR. After FDR correction, there were three significant correlations in AFR – for chronic pain, suicidality, and academic degree – and eight in AMR – for OUD, PAU, CanUD, chronic pain, ADHD, depression, BD, and suicidality. In all the cases, the correlation with SubLU was in the same direction as in EUR (as calculated by LDSC) (Supplementary Table S9 and Supplementary Figure S25).
gSEM modeling
Heritability estimates and genetic correlations were calculated across traits using LDSC (Supplementary Table S10). Parallel analysis indicated that a two-factor model best fit the data, although a third factor was relatively close to reaching the threshold for consideration (Supplementary Figure S26). Therefore, we performed two exploratory factor analyses (two- and three-factor solutions). Comparisons across the two- and three-factor model EFA results suggested that the two-factor model best fit the data due to the minimal amount of variance accounted for by inclusion of the third factor (<0.06). For the two-factor EFA, cocaine, inhalants, methamphetamines, street opioids, prescription opioids, prescription stimulants, CanUD, PAU, and OUD (i.e. substance-related traits) loaded onto factor 1, while suicidality, PTSD, schizophrenia, depression, and chronic pain (i.e. nonsubstance psychiatric traits) loaded onto factor 2. ADHD cross-loaded onto both factors. A confirmatory factor analysis (CFA) was performed, inputting traits on the relevant factor as indicated by EFA loading results, and the CFA results indicated adequate fit with comparative fit index (CFI) = 0.9476 and standardized root mean square residual (SRMR) = 0.0784 (Figure 3 and Supplementary Table S11).
Figure 3. Path diagram for genomic structural equation modeling for confirmatory factor analysis (CFA) results of the two-factor model. The diagram presents the results of the correlated two-factor CFA model of 15 substance use, psychiatric, and chronic pain traits for European ancestry participants. Standardized estimates are provided for each path with standard errors included in parentheses.
Discussion
Substance use and use disorders, even for the same substance, have often been seen to differ genetically (Levey et al., Reference Levey, Galimberti, Deak, Wendt, Bhattacharya, Koller and Program2023; Pasman et al., Reference Pasman, Verweij, Gerring, Stringer, Sanchez-Roige, Treur and Consortium2019; Sanchez-Roige et al., Reference Sanchez-Roige, Palmer, Fontanillas, Elson, Adams, Howard, Edenberg, Davies, Crist, Deary, McIntosh and Clarke2019; Zhou et al., Reference Zhou, Sealock, Sanchez-Roige, Clarke, Levey, Cheng and Gelernter2020) and, therefore, need to be studied individually. While there have been numerous studies of alcohol and tobacco use traits (Buchwald et al., Reference Buchwald, Chenoweth, Palviainen, Zhu, Benner, Gordon and Tyndale2021; Sanchez-Roige et al., Reference Sanchez-Roige, Palmer, Fontanillas, Elson, Adams, Howard, Edenberg, Davies, Crist, Deary, McIntosh and Clarke2019; Xu et al., Reference Xu, Li, McGinnis, Vickers-Smith, Dao, Sun and Program2020; Zhou et al., Reference Zhou, Kember, Deak, Xu, Toikumo, Yuan and Gelernter2023) and several for cannabis (Levey et al., Reference Levey, Galimberti, Deak, Wendt, Bhattacharya, Koller and Program2023; Pasman et al., Reference Pasman, Verweij, Gerring, Stringer, Sanchez-Roige, Treur and Vink2018), many substance use traits have not yet been subject to genetic study. We present here the first genetic investigations for several of these traits. We found several genes associated with different SubLU traits, with no overlap among the significant loci between traits: although there were very high genetic correlations between these traits (in EUR), different lead SNPs – located within different genes – were significantly associated with each of these phenotypes. This may, however, reflect power limitations, that is, as more cases become available for each substance and more loci are identified, it is plausible that more loci common to multiple traits may be identified.
For the individual SubLU traits, there were a total of three hits in AFR, two of them within protein-coding genes. For Cocaine LU, the MTNR1A gene had a significant effect (all significant SNPs mentioned in this section had a p-value in the range of 5 × 10−8–1 × 10−10). This gene encodes one of the two main melatonin receptors, known for its major role in circadian rhythms and the sleep–wake cycle (Hardeland, Pandi-Perumal, & Cardinali, Reference Hardeland, Pandi-Perumal and Cardinali2006). There is a strong association between sleep disorders and SUDs (Conroy & Arnedt, Reference Conroy and Arnedt2014), and specifically cocaine (Schierenbeck, Riemann, Berger, & Hornyak, Reference Schierenbeck, Riemann, Berger and Hornyak2008), and melatonin was used as an experimental treatment for SUDs, with mixed results (Das, Prithviraj, & Mohanraj, Reference Das, Prithviraj and Mohanraj2022). The effect of MTNR1A on Cocaine LU may provide a window into the interaction between sleep disorders and cocaine use, especially considering that it was also significant in a cross-ancestry meta-analysis. For Street Opioids LU, there was a hit in an intronic variant of MAGI1. Its encoded protein participates in inter-cell interactions and has been associated with neuroticism (Genetics of Personality et al., Reference Personality, de Moor, van den Berg, Verweij, Krueger, Luciano and Boomsma2015), BD, and schizophrenia (Karlsson et al., Reference Karlsson, Graae, Lekman, Wang, Favis, Axelsson and Paddock2012).
To our knowledge, there have been no previous genetic studies of inhalant use or dependence. In AMR, rs11699278 – an intronic variant of CYP24A1, a gene involved in the regulation of vitamin D and calcium homeostasis (Jones, Prosser, & Kaufmann, Reference Jones, Prosser and Kaufmann2012) – was associated with Inhalants LU; it did not have high LD with any significant (or nearly significant) SNPs (Supplementary Figure S4), so this finding might be a false-positive. There was a significant effect of CNTN5, a member of the contactin gene family, on Prescription Opioids LU. This gene has a role in cell-surface interactions during the developmental phase of the nervous system. Variants are associated with Alzheimer’s disease (Dauar et al., Reference Dauar, Labonte, Picard, Miron, Rosa-Neto, Zetterberg and Poirier2023), autism (van Daalen et al., Reference van Daalen, Kemner, Verbeek, van der Zwaag, Dijkhuizen, Rump and Poot2011), and ADHD (Lionel et al., Reference Lionel, Crosbie, Barbosa, Goodale, Thiruvahindrapuram, Rickaby and Scherer2011).
In EUR, four loci were associated with individual SubLU traits. GBF1 had a significant effect on Cocaine LU. The protein encoded by this gene has a role in diseases that affect muscular function, such as Charcot–Marie–Tooth (Mendoza-Ferreira et al., Reference Mendoza-Ferreira, Karakaya, Cengiz, Beijer, Brigatti, Gonzaga-Jauregui and Wirth2020) and Parkinson’s disease (PD) (Zhao et al., Reference Zhao, Li, Niu, Li, Luo, Zhou and Liu2020). The latter might suggest its involvement in dopaminergic processes: PD is related to dopaminergic cell death (Latif et al., Reference Latif, Jahangeer, Maknoon Razia, Ashiq, Ghaffar, Akram and Azam Ansari2021), while substance use elevates dopamine release in the synapses (Trifilieff & Martinez, Reference Trifilieff and Martinez2014). Cocaine acts via dopamine transporter inhibition (Gether, Andersen, Larsson, & Schousboe, Reference Gether, Andersen, Larsson and Schousboe2006). In a multi-trait analysis of GWAS, GBF1 had a significant effect on CanUD (Xu et al., Reference Xu, Toikumo, Crist, Glogowska, Jinwala, Deak and Kember2023). CADM2 was also associated with Cocaine LU in EUR, as well as cross-ancestry, in line with numerous studies suggesting that CADM2 is involved in substance use traits and risk-taking behavior (Arends et al., Reference Arends, Pasman, Verweij, Derks, Gordon, Hickie and Vink2021; Koller et al., Reference Koller, Mitjans, Kouakou, Friligkou, Cabrera-Mendoza, Deak and Polimanti2024; Pasman et al., Reference Pasman, Chen, Smit, Vink, Van Den Oever, Pattij and Verweij2022; Sanchez-Roige et al., Reference Sanchez-Roige, Jennings, Thorpe, Mallari, van der Werf, Bianchi and Palmer2023; Zhou et al., Reference Zhou, Sealock, Sanchez-Roige, Clarke, Levey, Cheng and Gelernter2020). For Inhalants LU, there was a lead SNP located in the CDH12 gene, encoding a protein that was associated with infertility and endometriosis in females (Golawski et al., Reference Golawski, Soczewica, Kacperczyk-Bartnik, Manka, Kiecka, Lipa and Laudanski2022). It has also been previously associated with schizophrenia (Hawi et al., Reference Hawi, Tong, Dark, Yates, Johnson and Bellgrove2018).
Cross-ancestry meta-analyses revealed additional novel associations: TCF4 had a significant effect on Cocaine LU; this gene is associated with Pitt–Hopkins syndrome, a disorder of severe developmental delay and intellectual disability (Sweatt, Reference Sweatt2013), and also with PTSD reexperiencing (Gelernter et al., Reference Gelernter, Sun, Polimanti, Pietrzak, Levey, Bryois and Million Veteran2019), corneal endothelial dystrophy, schizophrenia (Forrest et al., Reference Forrest, Hill, Quantock, Martin-Rendon and Blake2014), and depression (Mossakowska-Wojcik et al., Reference Mossakowska-Wojcik, Orzechowska, Talarowska, Szemraj and Galecki2018). LINC02008, a gene associated with blood pressure (Chen et al., Reference Chen, Wang, Li, Xu, Tian and Zhang2021), affected Cocaine LU too. For Methamphetamine LU, there was one GWS variant that maps to EXT1. Variants at this locus are associated with exostoses (Ludecke et al., Reference Ludecke, Ahn, Lin, Hill, Wagner, Schomburg and Wells1997) and were previously associated with other substance use traits such as smoking initiation (Saunders et al., Reference Saunders, Wang, Chen, Jang, Liu, Wang and Vrieze2022). As for Prescription Opioids LU, there was a cross-ancestry effect for a variant located within the GRAMD1A gene, encoding a protein responsible for cholesterol transfer. Studies link cholesterol levels to substance use (Buydens-Branchey & Branchey, Reference Buydens-Branchey and Branchey2003; Lin et al., Reference Lin, Yang, Lee, Hsieh, Chen, Lu and Chen2012), suggesting that low plasma cholesterol may be associated with increased craving for drugs (Lin et al., Reference Lin, Yang, Lee, Hsieh, Chen, Lu and Chen2012). For Prescription Stimulants, we found an effect for the noncoding RNA MEF2C-AS2, previously associated with sleep duration, physical activity (Doherty et al., Reference Doherty, Smith-Byrne, Ferreira, Holmes, Holmes, Pulit and Lindgren2018), and educational attainment (Okbay et al., Reference Okbay, Wu, Wang, Jayashankar, Bennett, Nehzati and Young2022).
Genes that were significantly associated with the individual SubLU traits cover a broad range of functionality, yet the inter-trait genetic correlations were very high. Looking at the local inter-trait genetic correlations, two of the three most highly correlated pairs (Street Opioids LU–Methamphetamine LU and Street Opioids LU–Prescription Opioids LU, with 40 and 37 correlated regions, respectively) also had the highest r g between them. The pair with the lowest observed r g, that is, Inhalants LU–Street Opioids LU, also had 37 common regions, an indication of the complexity of genetic correlations and of the hidden information that might be exposed using LAVA. In comparison, the phenotypic correlations between the traits were low to moderate, suggesting that the differences between the traits (the inclination to prefer one substance over another) may be based mostly on environmental factors. In other words, genetically, SubLU traits are quite similar to one another, yet factors such as substance availability (Kiang, Basu, Chen, & Alexander, Reference Kiang, Basu, Chen and Alexander2019), prescription patterns (Schieber et al., Reference Schieber, Guy, Seth, Young, Mattson, Mikosz and Schieber2019), income, socioeconomic status (Han et al., Reference Han, Compton, Jones, Einstein and Volkow2021; John & Wu, Reference John and Wu2017; Patrick, Wightman, Schoeni, & Schulenberg, Reference Patrick, Wightman, Schoeni and Schulenberg2012; Skoog et al., Reference Skoog, Midlov, Beckman, Sundquist and Halling2014; Wu & Ringwalt, Reference Wu and Ringwalt2006), and education (Han et al., Reference Han, Compton, Jones, Einstein and Volkow2021; Skoog et al., Reference Skoog, Midlov, Beckman, Sundquist and Halling2014; Wu & Ringwalt, Reference Wu and Ringwalt2006) might lead people to choose one drug over the other.
We also calculated the genetic correlations between individual SubLU traits and 12 traits of interest. Generally, the prominent results came from the two opioid LU traits. As expected, Prescription and Street Opioids LU had the highest genetic correlations with OUD. Prescription Opioids LU had the highest genetic correlation with depression, which was less expected; a clinical study showed that the comorbidity of depression with OUD is actually lower than with CocUD or methamphetamine use disorder (MetUD) (Calarco & Lobo, Reference Calarco and Lobo2021). The fact that the present findings with the same substances differ may reflect the differences between substance use and use disorders, as well as differences between genetic and phenotypic relationships. While initiation of cocaine use is mainly recreational (van der Poel et al., Reference van der Poel, Rodenburg, Dijkstra, Stoele and van de Mheen2009), opioids are usually prescribed to treat pain (Weiss et al., Reference Weiss, Potter, Griffin, McHugh, Haller, Jacobs and Rosen2014, which occurs in high comorbidity with depression (IsHak et al., Reference IsHak, Wen, Naghdechi, Vanle, Dang, Knosp and Louy2018). Therefore, the fact that Prescription Opioids LU also had the highest genetic correlation with chronic pain was expected. Prescription Opioids LU also had the strongest negative correlation with having an academic degree, in accordance with findings regarding impairing effects of opioids on academic performance (Darolia, Owens, & Tyler, Reference Darolia, Owens and Tyler2022; Ellis, Kasper, & Cicero, Reference Ellis, Kasper and Cicero2020). Moderate negative correlations between academic degree and methamphetamine and cocaine LU are also backed by literature (Dean, Morales, Hellemann, & London, Reference Dean, Morales, Hellemann and London2018; Jeynes, Reference Jeynes2022). A negative genetic correlation between Prescription Stimulants LU and academic degree is in line with its positive correlation with ADHD. The relatively low (compared to other pairs) correlation between Prescription Stimulants LU and ADHD, even though stimulants are mainly prescribed to treat the latter, was surprising. Generally, the trait that was most strongly associated with SubLU traits was CanUD, with a mean r g of 0.76 (±0.073) among individual traits and >0.82 for the cumulative SubLU traits. This can be partially explained by the fact that cannabis is the most commonly used intoxicating drug in daily or near-daily use in the United States, even more than alcohol (5.21%) (Caulkins, Reference Caulkins2024). Yet, daily or near-daily tobacco smoking was more common, with 11.6% prevalence (2022).
Given the high correlations indicating moderate to strong phenotypic and genetic associations across the substance use and psychiatric traits, we utilized gSEM to examine the genetic architecture encompassed by these phenotypes. This investigation revealed two latent genetic factors with substance use traits loading more strongly onto factor 1 and other traits (psychiatric disorders and chronic pain) loading strongly onto factor 2, with the exception of ADHD, which adequately cross-loaded onto both factors. These results align with existing data regarding common genetic architecture between different SUDs (Abdellaoui et al., Reference Abdellaoui, Smit, van den Brink, Denys and Verweij2021; Deak & Johnson, Reference Deak and Johnson2021; Hatoum et al., Reference Hatoum, Colbert, Johnson, Huggett, Deak, Pathak and Agrawal2023) and are similar to previous findings that showed that substance use traits tend to align under the same latent factor, while other psychiatric traits usually fit under a separate psychopathology-oriented factor (Abdellaoui et al., Reference Abdellaoui, Smit, van den Brink, Denys and Verweij2021; Levey et al., Reference Levey, Galimberti, Deak, Wendt, Bhattacharya, Koller and Program2023).
We conducted two further GWAS analyses of composite traits: (i) counting any substance use trait (of the substances included in this study) as an indicator of any-SubLU; and (ii) considering substance use as a quantitative trait (nSubLU), measuring the total number of substances used by each subject (a range of 0–6 substances), which can be viewed as a severity measure. By this, we examined whether the genetic factors that are associated with substance use in general are different, to a degree, from those that influence the severity of substance use, or the tendency to use multiple drugs. We found different loci associated with each of these traits, suggesting that different factors affect any-substance use and substance use severity as judged by the comorbidity of the same substances. In AFR, the only hit for nSubLU was in a noncoding region, while for SubLU, there was a significant variant within the DPRX gene, encoding a DNA-binding protein thought to be involved in embryonic development (Madissoon et al., Reference Madissoon, Jouhilahti, Vesterlund, Tohonen, Krjutskov, Petropoulos and Kere2016). In AMR, there were two GWS findings for nSubLU, though one has the appearance of a false positive, and the second maps to a noncoding RNA (Supplementary Figure S18).
In EUR, there was one mutual finding between SubLU and nSubLU, a variant that maps to LINC02758, which was previously associated with smoking initiation (Saunders et al., Reference Saunders, Wang, Chen, Jang, Liu, Wang and Vrieze2022), suicidal thoughts (Kimbrel et al., Reference Kimbrel, Ashley-Koch, Qin, Lindquist, Garrett and Dennis2023), depression (Zhang et al., Reference Zhang, Zheng, Yu, Yu, Luo, Yin and Ai2024), ADHD (Chen et al., Reference Chen, Liu, Yi, Fan, Zhao and Liu2024), and risk-taking behavior (Baselmans et al., Reference Baselmans, Hammerschlag, Noordijk, Ip, van der Zee, de Geus and van’t Ent2022). For SubLU, its significance (p = 1.55 × 10−10) was higher compared to nSubLU (p = 1.14 × 10−8) by nearly two orders of magnitude, and only in SubLU it was significant for the cross-ancestry meta-analysis.
CADM2 was significantly associated with SubLU in EUR and cross-ancestry, but not with nSubLU. As already mentioned, CADM2 is one of the most strongly associated genes with substance use traits (Arends et al., Reference Arends, Pasman, Verweij, Derks, Gordon, Hickie and Vink2021; Koller et al., Reference Koller, Mitjans, Kouakou, Friligkou, Cabrera-Mendoza, Deak and Polimanti2024; Pasman et al., Reference Pasman, Chen, Smit, Vink, Van Den Oever, Pattij and Verweij2022; Sanchez-Roige et al., Reference Sanchez-Roige, Jennings, Thorpe, Mallari, van der Werf, Bianchi and Palmer2023; Zhou et al., Reference Zhou, Sealock, Sanchez-Roige, Clarke, Levey, Cheng and Gelernter2020). These results support that it could be more important in the inclination to use substances, but less so for severity. We can also suggest the opposite regarding GBF1, which was not significantly associated with SubLU, but was with nSubLU, suggesting its possible involvement in increased risk taking. As with other differential associations, this could also reflect power differences between analyses and random variation. This effect of GBF1 remained only in EUR, and did not appear cross-ancestry.
There was a significant association between nSubLU and ACTN2, previously associated with heart failure (Arvanitis et al., Reference Arvanitis, Tampakakis, Zhang, Wang, Auton and Battle2020), smoking initiation (Saunders et al., Reference Saunders, Wang, Chen, Jang, Liu, Wang and Vrieze2022), externalizing behavior (Karlsson Linner et al., Reference Karlsson Linner, Mallard, Barr, Sanchez-Roige, Madole, Driver and Dick2024), and educational attainment (Okbay et al., Reference Okbay, Wu, Wang, Jayashankar, Bennett, Nehzati and Young2022). CD47, associated with smoking initiation (Saunders et al., Reference Saunders, Wang, Chen, Jang, Liu, Wang and Vrieze2022) and BD (Li et al., Reference Li, Zhang, Hui, Zhou, Li, Zhang and Gese2021), was GWS for nSubLU too.
The genetic correlations of SubLU and nSubLU with traits of interest were almost identical. Nevertheless, some differences in local genetic correlation were revealed: nSubLU was locally correlated with schizophrenia and PAU (the latter association in a region near CADM2, suggesting a possible mediation of this association by CADM2), which showed more severe traits that suggest a greater association between nSubLU and pathologies. In Cross-ancestry, we found similar genetic correlations between SubLU and chronic pain across all three ancestries. In AMR, SubLU also had a significant genetic correlation with OUD, PAU, CanUD, ADHD, and depression, while in AFR, there was a genetic correlation between SubLU and academic degree. Nevertheless, these correlations were not as strikingly similar to those found in EUR, which may reflect the lower power in AFR analyses compared to EUR. These findings indicate that a similar underlying genomic architecture drives the genetic correlations between SubLU and chronic pain across different ancestries, whereas other traits exhibit greater dissimilarities.
To our knowledge, this is the first study to examine the genetic mechanisms behind a broad range of SubLU traits. Therefore, nearly all the results presented here are novel. We found genes that suggest a possible association between SubLU and some neurological disorders, including AD, PD, and autism. We also pointed to a genetic association with psychiatric phenomena like psychosis, depression, ADHD, and SUDs, observed through several mutual genes, such as CADM2, TCF4, LINC02758, and CD47, and via genetic correlations. The genetic correlations between SubLU traits were very high, but no gene had a significant effect on more than one individual trait. Most findings in the cumulative traits were novel too (i.e. did not appear in the individual substances). Our findings offer insights into the genetic predisposition to substance use. This knowledge could facilitate future risk assessments for substance use and provide improved understanding of the biology that underlies these traits.
This study has limitations. First, all the data used was based on the AoU cohort. AoU is a high-quality US sample with good representation from a range of US populations, although it is still important to explore these traits in other cohorts to confirm our results and be able to generalize them to other populations. Better powered studies might also reveal mutual (pleiotropic) effects of specific SNPs or loci between SubLU traits, which we were not able to detect in the current study due to limited sample sizes. Future availability of well-powered datasets for a variety of additional phenotypically associated traits like CocUD and MetUD may provide important insights regarding a broader range of genetic correlations between these phenotypes and SubLU traits. Second, all substance lifetime users were considered as a singular group, with no differentiation between occasional and high-frequency users, including people with SUD. Third, substance use traits were defined by self-report; this could impair the nature of assigning participants to the right group (i.e. case or control), considering the possible inaccuracies that may appear in a self-reported trait, and especially considering the sensitivity of reporting the use of an illegal substance. Fourth, for some of the traits, the sample sizes for the AFR and AMR populations were relatively small and, therefore, lacked the power to detect an effect that might exist. Future studies with larger sample sizes for those ancestries may reveal significant effects that did not appear in the current study.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291725101293.
Acknowledgments
The authors would like to acknowledge All of Us participants for their contributions, without whom this research would not have been possible. The authors also thank the National Institutes of Health’s All of Us Research Program for making available the participant data examined in this study (researchallofus.org).
Funding statement
This research was supported by NIH grants R01DA058862 (JG) and R01DA054869 (JG) and by funding from the Department of Veterans Affairs Office of Research and Development, USVA, grant I01CX001849 (JG), and the VA Cooperative Studies Program study, no. 575B, the VA National Center for PTSD Research, the West Haven VA Mental Illness Research, Education and Clinical Center, and a Career Development Award CDA-2 from the Veterans Affairs Office of Research and Development (DFL; IK2BX005058).
Competing interests
JG is paid for editorial work by the journal Complex Psychiatry.
All other authors report no competing interests.
Open access
