The 315 men in the study were drawn from our previous study [21], including 300 alcohol-dependent (AD) inpatients recruited from psychiatric hospitals in northern China and 15 healthy controls (HCs) from Inner Mongolia Autonomous Region were recruited from December 2009 to December 2012. All patients met the criteria for ND based on the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV). The exclusion criteria were as follows: (1) other drug abuse or dependence, except alcohol; (2) severe cardiovascular disease, liver disease, or kidney disease; (3) participant, or a first-degree relative of the participant, with a severe mental illness.

The participants were asked to complete a series of questionnaires and provide a blood sample for DNA extraction. All staff involved in this study were trained before the study commenced. The Institutional Review Board of the Inner Mongolian Medical University approved the study (Ethic approval number: YKD2015003). All patients provided written informed consent and were told that the blood sample would be subjected to a gene assay.

Genomic DNA was extracted from 5 ml of peripheral blood of each participant using the salting-out method. Primers were designed using Assay Design 3.1 software from Sequenom (San Diego, CA, USA) and synthesized by Gene-Cloud Biotechnology Co., Ltd. (Beijing, China). PCR-specific and single-base extension primer sequences are provided as follows: 5-ACGTTGGATGACACCTTGGACTTTCTTGGC-3, 5-ACGTTGGATGATGTGTGGCTTGCAAGAACC-3, and single-base extension primer: 5-GCTCATGTTTTGTTTTTGAAG-3. Reactions were carried out according to the manufacturer’s protocol [29, 30]. All laboratory procedures were carried out in a manner blind to case-control status. The conditions of PCR were as follows: 50 °C for 2 min, 95 °C for 10 min, followed by 50 cycles of 95 °C for 15 s and 60 °C for 1 min. Ten percent of the DNA samples were duplicated randomly and tested, and no-fault genotyping was found.

First, the Hardy-Weinberg equilibrium for genotype distributions of ADPN rs266729 was tested using the Chi-square test for goodness of fit. Then, Pearson correlations were examined between genetic polymorphisms, age, years of education, ND, and anxiety. Consistent with other studies [31, 32, 33], CG and GG genotypes were collapsed into a G allele group coded as one and the CC genotype coded as 0. Then, traditional linear regression was used to test the interaction between ND and the rs266729 polymorphism. When a significant interaction was found, region of significance (RoS) analysis was used to examine the form of the interaction [34].

Finally, a re-parameterized regression model was fitted to examine the specific pattern of gene $\times$ environment pattern [35] with the following form:

(1)

Diathesis-Stress model is generated by the re-parameterized regression equations. In this regression equation, independent variable (X) is FTND scores, dependent variable (Y) is anxiety, age and years of education are demographic covariates (X2, X3). All variables are continuous variables and model were fitted using the least-squares method [35]. Crossover point C is a point where the slopes of the different groups cross. In differential-susceptibility, crossover point C is estimated by regression equations. What distinguishes the diathesis-stress and differential susceptibility models is the estimate and interval estimate of crossover point C. If the estimate and interval estimate of crossover point C fall within the range of ND, the model is consistent with the differential susceptibility model. Otherwise, if crossover point C is over the maximum of ND, the model is consistent with the diathesis-stress model. Results of the final best-fit models arrived at via model comparison, including F test, Akaike information criterion (AIC) and Bayesian information criterion (BIC).

The diathesis-stress and differential susceptibility models can be further subdivided into strong and weak versions. Strong versions assume that non-risk/non-plasticity allele carriers are not susceptible to the environment. Weak versions assume both allele carriers are susceptible to the environment; however, non-risk/non-plasticity carriers are less susceptible to the environment than risk/plasticity carriers. These models are nested within one another. Therefore, the F-test was used to compare the models and identify a difference in the parameter estimates. For non-nested models, the Akaike information criterion and Bayesian information criterion were compared to evaluate which model was a better fit.

Of the 315 male inpatients, 157 (49.8%) were CC homozygotes, 133 (42.2%) were CG heterozygotes, and 25 (7.9%) were GG homozygotes. The genotype distribution of ADPN rs266729 was consistent with Hardy-Weinberg equilibrium ($\chi$${}^2$ = 0.19, p $>$ 0.05, Table 1). Table 2 displays the demographic and clinical characteristics of the participants. The mean age, academic years, mean Nicotine Dependence (FTND scores), and Anxiety (SAS scores) are all listed separately with genotype CC and CG/GG in Table 3. The ADPN genotypes were dummy-coded into 0 = CC (CC homozygotes) and 1 = CG/GG (G allele carriers). A series of t-tests were then conducted to determine whether participants with and without ND and anxiety symptoms differed regarding the polymorphism ADPN rs266729. We found no significant differences (FTND: t = –0.43; SAS: t = 0.60, both p $>$ 0.05, Table 3).

The descriptive statistics for each research variable are shown in Table 4. FTND scores were positively correlated with SAS scores (r = 0.15, p 0.01). No significant correlation between polymorphisms of rs266729 and FTND (p $>$ 0.05), as well as SAS scores (p $>$ 0.05) were observed.

Next, hierarchical regression models were used to predict anxiety from ND and test the interaction between ND and rs266729 polymorphism on anxiety, with age and years of education as covariates. Table 5 shows that ND had a significant main effect on anxiety symptoms ($\beta$ = 0.14, p 0.05); however, there was no main effect of genotype ($\beta$ = 0.04, p = 0.51). The interaction between ADPN rs266729 and ND was significant ($\beta$ = –0.19, p 0.05). Then, the RoS test was used to examine the interaction effect. As shown in Fig. 1, the slopes for ND on anxiety were as follows: CC homozygotes, $\beta$ = 0.14, t = 4.43, p 0.01; G allele carriers, $\beta$ = –0.05, t = 1.07, p 0.01. Relative to G allele carriers, CC homozygotes were more likely to be affected by ND, causing anxiety.

Fig. 1.

Region of significance (RoS) test on anxiety from nicotine dependence severity in rs266729 allelic groups. Note: Anxiety and nicotine dependence: z scores (unit: standard deviations [SDs]); Simple slope at CC homozygote: 0.14, t = 4.43, p 0.001; Simple slope at G allele: –0.05, t = 1.07, p = 0.287; Lower threshold for RoS with respect to nicotine dependence = –0.115; Upper threshold for RoS with respect to nicotine dependence = 0.610; Crossover point on predictor = 0.211.

In addition, we re-analyzed the data to explain how alcohol use influences these results. As shown in Supplementary Table 1, MAST scores were positively correlated with SAS scores (r = 0.36, p 0.001), suggesting that anxiety was positively correlated with alcohol dependence, as our previous study reported [36]. However, there is no significant correlation between polymorphisms of rs266729 and MAST scores (r = 0.003, p $>$ 0.05). In addition, as shown in Supplementary Table 2, alcohol dependence significantly affected anxiety symptoms ($\beta$ = 0.35, p 0.001). However, there was no significant interaction between alcohol dependence and ADPN rs266729 on anxiety ($\beta$ = –0.22, p $>$ 0.05). Thus, although anxiety was positively correlated with alcohol dependence, there was no significant interactive effect of alcohol dependence and ADPN rs266729 on anxiety.

To improve the accuracy and robustness of the interaction results, a re-parameterized regression model was fitted to examine the specific pattern of G $\times$ E. The fit of the strong differential susceptibility model, Model A, yielded a significant R${}^2$ = 0.05, p 0.001 (Table 6), explaining a significant amount of variance in anxiety, in which the slope for the G allele group was significant (B2 = 0.28, p 0.001). The estimated crossover point C and 95% confidence interval (CI) of C both fell within the range of ND, C = 0.27 (SE = 0.41), 95% CI = [–0.53, 1.07]. We analyzed variance and compared the AIC and BIC values to confirm if Model A is the best model. As shown in Table 6, compared with Model A, Model B had one more parameter, but the explained variance was not significantly increased ($\mathrm{\Delta }$R${}^2$ 0.001, p $>$ 0.05). Model C explained a significantly lower variance ($\mathrm{\Delta }$R${}^2$ = 0.03, p 0.05) than Model A. The AIC and BIC values were smaller than Model D. These findings suggest that Model A provides strong support for the strong differential susceptibility model, indicating that the G allele carriers with ND are less susceptible to anxiety than CC homozygotes.