In the present study, we identified several variants in CNR1 showing significant associations with various dimensions of personality in an AA sample, especially for extraversion, for which four SNPs were identified. Importantly, SNPs associated with the other four dimensions of personality were all associated with extraversion. Considering that variants in CNR1 have been reported to be associated with an increased risk of mental health conditions, these data indicate the presence of genetic influences shared by personality traits and psychiatric disorders (Hettema et al., 2006; Kendler and Myers, 2010).

Our study might benefit greatly from the AA samples used in this study compared with those reported GWAS studies on various personality traits (de Moor et al., 2012; Smith et al., 2016), where samples with multiple ethnic origins, locations, or studies designed for different purposes were included. By using two approaches to analyze the same imputed dataset, we found that the frequentist test performed better than the regression model. When compared with PLINK in the context of genotype uncertainty, the frequentist tests, including the Score and Expected tests, improved statistical significance. One of the possible explanations for this result is that SNPTEST fully accounts for the uncertainty in imputed genotypes by weighting the imputation probability (Score test) or by calculating the allele dosage (Expected test). Conversely, the PLINK-based regression analysis uses only imputed genotypes that have a posterior probability above some threshold, which might lead to both false-positive findings and loss of power (Marchini and Howie, 2010). By implementing a gene-only model-based power analysis, we believe that the current genetic study is able to detect significant effects of CNR1 variants on personality traits in this valuable AA sample.

The relations between CNR1 and personality traits as well as other psychiatric disorders, such as depression and anxiety, have been established by Juhasz et al. (2009a), where 10 SNPs in CNR1 were genotyped, but they failed to detect any significant association with any personality trait at the individual SNP level in 1,269 participants. Interestingly, the haplotype trend regression analysis revealed significant associations with neuroticism and agreeableness after correction for multiple testing. Furthermore, those investigators detected SNP-by-SNP interactions for rs806379 with several other SNPs on both neuroticism and agreeableness. Another genetic association study revealed the association of rs7766029 with neuroticism (Aleksandrova et al., 2012).

Our individual SNP-based association analysis indicated significant association of CNR1 variants with extraversion in AAs. Rather than selecting haplotype-tagged SNPs, we imputed SNPs within the region of interest according to a high-density reference panel. An established statistical method was used in this study, which takes the genotype uncertainty into account and appeared to be more sensitive than PLINK in terms of our imputed data. Although almost all the investigated SNPs in the study reported by Juhasz et al. (2009a) were included in our analysis (except for rs7766029 because of the lack of reliable imputation quality), comparable results were not obtained in the two studies. We detected a positive association only between rs806368 and openness regardless of the genetic model. Such divergent results might be explained by the following factors: (Costa et al., 1992) different genetic models were used in these two studies. We detected single marker associations with the additive model, whereas Juhasz et al. (2009a) reported their results from three models (additive, dominant, and recessive); (McCrae et al., 2005) different questionnaires were used in these two studies. Juhasz et al. (2009a) used the Big Five Inventory (BFI-44) to assess personality, whereas we used the NEO Personality Inventory. Although there exist excellent correlations between the BFI-44 Inventory and NEO, a slight discrepancy might arise in phenotypic assessments; and (DeYoung et al., 2010) population stratification: we observed differential allele frequencies between AAs and European-American (EA) populations (Supplementary Table S8). For example, the allele frequency of rs1049359 is rarer in AAs (MAF = 0.029) than in EAs (MAF = 0.258).

The extraversion-associated SNP rs806368 remained significant after Bonferroni correction. In addition, this SNP showed a marginal association with both agreeableness and openness (Table 2). Rs806368 is located in the 3′-UTR region of CNR1. A strong association between rs306368 and impulsivity has been reported for the Southwest California Indians (also known as Mission Indians) (Ehlers et al., 2007). Impulsivity is considered as a facet of personality and acts as a major component of various psychiatric diseases such as ADHD (Nigg, 2001), bipolar disorder (Henry et al., 2001), and antisocial personality disorder (Horn et al., 2003). The study reported by Eysenck (1967) indicates that extraversion is a combination of impulsiveness and sociability, which is consistent with the genetic findings that rs806368 is associated with extraversion and impulsivity.

A pharmacogenomics study highlighted the significant role of rs806371 in both MDD etiology and the clinical response to citalopram treatment (Mitjans et al., 2013). Mitjans et al. (2013) reported a higher frequency of rs806371 G carriers in MDD patients with melancholia and psychotic symptoms than in controls. Moreover, the C allele carriers of rs806371 demonstrate increased remission status of MD symptoms after 12 weeks of treatment with citalopram (Mitjans et al., 2013).

Another interesting SNP revealed by this study is rs2180619, located in the promoter region of CNR1, which is not only significantly associated with both extraversion and conscientiousness, but also nominally with agreeableness (Table 2). By genotyping 706 individuals for the 5′-HTTLPR in the SLC6A4 and four SNPs in CNR1, Lazary et al. (2009) showed that the risk of anxiety was 4.6-fold greater in individuals with genotypes G/G for rs2180619 and S/S for 5-HTTLPR compared with other genotypes. Furthermore, Heitland et al. (2012) found that individuals with genotypes G/G or G/A for rs2180619 showed robust extinction of fear compared with individuals with the A/A genotype. Interestingly, Mroczkowski et al. (2016) found that a high neuroticism score significantly increases the risk of three anxiety disorders (agoraphobia, panic disorder, and social phobia), and extraversion and conscientiousness were negatively associated with separation anxiety disorder and social phobia. Although these relationships were constructed in individuals with obsessive compulsive disorder, the significances survived after adjusting for age at onset of this disorder (Mroczkowski et al., 2016).

Relationships among personality traits are worthy of consideration. For example, extraversion and openness shared genetic overlap in the context of rs806368, rs806370, rs806371, rs806372, and rs9444584; extraversion, conscientiousness, and agreeableness were highly related for rs9353526, rs9362466, and rs2180619 (Table 2). We further found the direction of effects for associated SNPs were opposite between neuroticism and the other four personality traits. It was reported recently that neuroticism shows a negative genetic correlation with the other four personality traits (Lo et al., 2017). Taken together, the multiple dimensions of personality-associated CNR1 variants support the concept that the presence of shared genetic factors contributes to the correlation among personalities (Lo et al., 2017), suggesting a fundamental role of CNR1 in personality formation.

Dysfunction of CNR1 may result in neuropsychological disturbances, leading to psychiatric disorders. Martin et al. (2002) first used CB1 knockout mice to evaluate the potential role of CNR1 in emotional behavior (Ledent et al., 1999). By using a light/dark box, they observed a significant decrease in the number of entries and in the time spent in the lit compartment for CB1 knockout mice, indicating that these mice present an anxiogenic-like response. By using a chronic unpredictable mild stress model (CMS) to simulate anhedonia, the CB1 knockout mice exhibited greater anhedonia, a core symptom of depression and defining feature of melancholia (Klonsky, 2008; Battle, 2013). More recently, Bowers and Ressler (2016) administered Cnr1 antagonist SR141716A to C57BL/6J mice, after which both male and female adult mice were sensitive to Cnr1 antagonist-mediated increases in anxiety-like behavior, thus supporting the protective role of CNR1 against the development of anxiety disorder. These studies confirmed that dysfunction of the CB1 receptor changes several behavioral responses and increases susceptibility to emotional disorders, such as anxiety, depression, and aggressiveness.

There are limitations associated with this study. First, the SNPs investigated were generated primarily by imputation. Although we have great confidence in the accuracy of our imputed genotypes, as demonstrated in the literature, including our own work (Marchini and Howie, 2010; Howie et al., 2011), individual genotyping of variants in CNR1 is desirable in future studies to support the conclusions about the association of CNR1 variants with personality in AAs. Second, our imputation-based SNP selections provide a fine-mapping of CNR1, but the SNPs examined could not accurately tag the causal genetic variants. Last but not least, the detected associations of variants in CNR1 with various personality traits require to be replicated with independent AA samples. However, such type of AA samples with appropriate phenotypes is unavailable to us.