A prior genome-wide association study of DSM-IV cannabis dependence, conducted in the sample used in this study, failed to identify genetic variants at a statistically significant level . This has resulted in speculation regarding the biological underpinnings of cannabis use disorders; in particular, the question of whether common variation available in commercially available genome-wide arrays captures it . Aggregating the effects of all single nucleotide polymorphisms on commercial arrays might quantify the overall role of common SNPs as well as causal variants in linkage disequilibrium Table 1 Prevalence of individual DSM-IV and proposed DSM-5 criteria for cannabis use disorder in 3053 lifetime cannabis users of European-American and African American ancestry. When significant, this would indicate that heritable variation in the trait is at least partially captured by these SNPs in a highly polygenic manner. Applying this methodology, investigators have successfully attributed 23–51% of the variation in current smoking, major depression, schizophrenia and human intelligence to genetic influences . The present study uses a multi-pronged phenotypic and genomic approach to evaluate, respectively, the architecture and genetic underpinnings of DSM-5 cannabis use disorders, defined as a quantitative phenotype. Instead of relying on a diagnostic measure, we first utilize item response models to construct a factor representing liability to DSM-5 cannabis use disorders, while accounting for sex and ethnic differences. Second, we use this psychometrically constructed factor score in a genome-wide association analysis. Finally, we evaluate whether genome-wide SNPs and putative causal variants in linkage disequilibrium with them explain a significant proportion of the heritable variation in DSM-5 cannabis use disorders.The genotyping and quality control procedures applied to these data are explained in detail in earlier publications . In brief, DNA samples from 3988 individuals were genotyped on the Illumina Human 1 M bead chip by the Center for Inherited Diseases Research at Johns Hopkins University. As described earlier, 948,658 SNPs passed data cleaning protocols. No imputed data were used for these analyses. HapMap genotyping controls, duplicates, related subjects, and outliers were removed. For the current analyses, data on 3053 individuals reporting a lifetime history of cannabis use were used. Self-identified ethnicity was 2018 European Americans and 1035 African Americans.
We used MPlus to conduct exploratory and confirmatory factor analyses of the 12 DSM-IV/DSM-5 criteria in the same sample. Exploratory analyses were conducted in the full sample, while subsequent confirmatory factor analyses were conducted in African-American and European-Americans ,trim tray for weed separately by sex, using a multi-group framework. Initially, factor loadings and thresholds were constrained across the ethnic groups and across sexes. Individual sub-models were tested to determine whether allowing the factor loading and threshold for each criterion to vary across the groups resulted in a significant improvement in model fit. The model that accommodated all statistically significant differences was used to generate factor scores that were subsequently used for genome-wide association analysis.The sample used for analyses was restricted to those who reported at least one lifetime use of cannabis . These individuals are characterized with respect to the 12 individual DSM-IV/DSM- 5 criteria in Table 1. Prevalence of each criterion was higher in males than females for both ethnic groups, and males, regardless of ethnicity, were more likely than females to meet criteria for DSM-IV and DSM-5 diagnoses. However, several intriguing ethnic differences emerged. For both sexes, hazardous use, use of larger amounts or for a longer period of time and desire to quit or multiple failed quit attempts were differentially endorsed by EA and AA. EA men and women were more likely to endorse hazardous use and less likely to endorse using larger amounts or for longer than intended and failed quit attempts than their AA counterparts. In addition, tolerance, time spent using cannabis and the DSM-5 criteria of withdrawal and craving were more commonly reported by AA women than their EA counterparts—similar differences were not noted for men. The prevalence of DSM-IV cannabis abuse/dependence was higher in men compared with women, but no within-sex ethnic differences were noted. For DSM-5, cannabis use disorder was again more common in men than women, and there were no ethnic differences in men. However, AA women were more likely to meet criteria compared with their EA counterparts . Comparing the prevalence of DSM-IV vs. DSM-5 cannabis use disorders—within each group, very modest changes were observed. Decrease in overall prevalence was noted for EA, while AA women showed a slight increase and AA men remained unchanged. Examining the [95%] confidence limits for the point estimates, only the decrease in prevalence in the EA was statistically significant while the estimates in AA subjects could be equated across diagnostic classification scheme .An exploratory factor analysis of the full sample revealed that a single factor solution provided a reasonable fit to the data : 0.996, root mean square error of approximation : 0.054. While a 2-factor exploratory solution modestly improved these fit indices,the inter-factor correlation was 0.90. Hence, we proceeded with the more parsimonious single factor confirmatory analysis, which readily approximates item response parameters.
Confirmatory factor analysis of the 4 DSM-IV abuse, 6 DSM-IV dependence and the DSM-5 withdrawal and craving criteria revealed high factor loadings for all criteria except legal problems , which was excluded from further analyses comparing factor loadings and thresholds for each individual criterion across EA and AA males and females. The factor loadings and thresholds from the model allowing for statistically significant differences across individual items are shown in Table 2. Factor loadings and thresholds could not be constrained across the groups for hazardous use,interpersonal problems, withdrawal,using more than intended , repeated/failed quit attempts,time spent and physical/psychological problems . Factor scores that accommodated these differing thresholds and factor loadings were created for each of the four subgroups and used for genomic analyses.Individual signals did not surpass the Bonferroni corrected genome-wide significance threshold of p < 5 × 10−8. The results for the top 20 SNPs are presented in Table 3 . For the EA sub-sample, 11 SNPs on 17q23-24 appeared to be associated at nominal levels of significance although none surpassed the genome-wide threshold of 5 × 10−8.The top SNP, rs6504555, was an intronic variant in the bromodomain PHD finger transcription factor gene—a regional association plot for this region of chromosome 17 is shown inFig. 1, indicating a high degree of linkage disequilibrium across the associated SNPs. With the exception of rs11870068, the remaining chromosome 17 SNPs were in moderate to high linkage disequilibrium . In the AA sub-sample, results did not aggregate in any particular chromosomal region. The most significant SNP, rs4364205, on chromosome 3, was intergenic. Meta-analysis of the results from the EA and AA sub-samples did not yield a boost in statistical significance . This was evident from a comparison of results in the EA and AA sub-samples. Of all SNPs with p-values < 0.05 in EA sub-sample, only 5% had corresponding p-values < 0.05 in AA sub-sample. However, particularly for the SNPs for the EA sub-sample shown in Table 2, the direction of effect in the AA sub-sample predominantly concurred with the EA sub-sample.We sought to examine the phenotypic and genomic architecture of a continuously distributed cannabis use disorders factor, psychometrically derived from DSM-5 criteria, in samples ascertained for alcohol, nicotine and cocaine dependence.Analysis of ethnic differences indicated a modest reduction in the prevalence of DSM-5 cannabis use disorders, relative to DSM-IV, in EA. Genomic analyses, using a genome-wide scan, failed to identify SNPs that satisfied statistical thresholds for significance; however, gene-based association implicated genes on the q-arm of chromosome 17. A genome wide variance calculation revealed that 21% of the phenotypic variance in cannabis use disorders was captured by the available common variation on the genome-wide array, but this estimate had a large standard error and was not significant. We used the factor score as our phenotype for genomic analyses.
Incorporating withdrawal and craving, excluding legal problems and combining across DSM-IV abuse and dependence criteria, this factor embodies the ‘spirit’ of the new DSM-5 diagnostic scheme while not being encumbered by concerns that the threshold of 2 or more criteria for diagnosis of disorder is too lax . From a psychometric perspective, our results are consistent with the extant literature . For instance, despite our sample being ascertained for alcohol, nicotine and cocaine dependence, which inflated endorsement rates of individual criteria , our high rates of hazardous use were comparable with those reported for lifetime cannabis users from the general population as reflected in data from the National Epidemiological Survey of Alcohol and Related Conditions . Likewise, broadly consistent with numerous other studies, the DSM-IV abuse criterion of legal problems was infrequently endorsed and had a weak factor loading, affirming its proposed exclusion from DSM- 5. The overall prevalence of the remaining criteria, although much higher than in general population cohorts, supports the presence of a unidimensional construct across sexes and ethnicities. Craving and withdrawal, both of which have been added to DSM-5, performed well, with high factor loadings supporting their inclusion. Overall, trimming tray weed rates of diagnostic DSM-5 cannabis use disorders appear to be modestly lower than those for DSM-IV abuse/dependence, but only in EA, particularly men. This finding is highly comparable with epidemiological analyses of alcohol symptomatology in U.S. and with results from the 2007 Australian National Survey of Mental Health and Well being, which reported a decrease in the lifetime rate of cannabis use disorder from 6.2% to 5.4% when transitioning from DSM-IV to DSM-5 . In our sample, this decrease was uniformly attributable to individuals who endorsed hazardous use alone, which results in a DSM-IV diagnosis of cannabis abuse but not a DSM-5 diagnosis of cannabis use disorder, because it falls below the latter’s minimum two-symptom threshold. No differences were noted in AA men , and this is also not surprising. Individuals endorsing this criterion alone tend to be of higher socio-economic standing and tend to, overwhelmingly, endorse this criterion due to a history of drinking and driving . That socio-economic status may correlate with ethnicity is expected—in our data, 45.9% of AA participants reported a gross annual income of less than $20,000, vs. 15.4% of their EA counterparts. Upon examining gender and ethnic differences within classification version , the only significant variation was noted for DSM-5 diagnoses in AA women who were more likely to receive a diagnosis of DSM-5, but not DSM-IV cannabis use disorder, relative to their EA female counterparts. Intriguingly, also relative to their EA counterparts, they were less likely to endorse hazardous use but more likely to endorse numerous other criteria, with the exception of giving up important activities and use despite physical/psychological problems. This finding may be attributable to the larger number of AA women that were ascertained from the cocaine dependence study vs. other studies. Although this observation holds true for the men as well, and the prevalence did not vary across AA and EA women, it is possible that AA women from the FSCD study represent a high-risk group. For instance, when compared to the alcohol and nicotine dependence studies, AA women from the cocaine study were more likely to report lower household income and a greater likelihood of less than a high school education . Thus, this vulnerability might reflect environmental adversity rather than increased genetic susceptibility, and in any case, is accounted for in the genomic analyses by incorporating study sample and gender as covariates. From a genetic perspective, the single SNP analyses did not reveal any genome wide significant signals. This is likely because our sample is under powered, even with a quantitative trait, to detect single variants of modest effect size. Using GWA Power , we estimated power available in our dataset to identify SNPs of varying effect size. Power was 80% when an effect size of 0.01 was anticipated . Increasing efforts to amass larger samples with comparable cannabis-related data would afford greater power to detect variants of more modest effect size via meta- and mega-analyses. However, few current studies have DSM-5 criteria data. In this regard, factor scores such as ours may prove to be useful phenotypes as they can accommodate DSM-IV and DSM-5 based assessments of vulnerability to cannabis use disorders. In contrast,the gene-based analyses conducted with the EA sub-sample identified a cluster of genes, of varied function, on the q-arm of chromosome 17 that appeared to contain an aggregation of variants associated with DSM-5 cannabis use disorders.