Association between dopaminergic polymorphisms and borderline personality traits among at-risk young adults and psychiatric inpatients

In the US sample 99 young adults from low-to-moderate income families (59.6% female; age range 18-22) participated in a study of adolescent-parent relationships [13]. Forty one of the families had been participating in a longitudinal study of attachment relationships since infancy, and were recruited at child age 0-18 months. All 41 families were at or below federal poverty level at intake, and half of the families were referred to clinical services during the young adult's infancy for concerns about the quality of care provided (the other half were socioeconomically-matched controls). An additional 48 crossectional families were recruited as part of the young adult follow-up study and matched to the longitudinal sample on socioeconomic status. Hence, the combined study sample of 99 young adults is regarded as an at-risk sample (53.5% of the families had an income of $40,000 per year or less; 43.4% of mothers had a high school education or less; 37.4% were single parents). The study was conducted in compliance with the Code of Ethics of the World Medical Association and with the requirements of the Hospital Institutional Review Board and the National Institute of Mental Health. All young adults and their mothers provided written informed consent for their participation.

Ethnicity/race data was obtained from the mother of the young adult during the interview by asking her to classify the ethnicity and race of both herself and the father (possible options for ethnicity: Hispanic or Latino; Not Hispanic or Latino; possible alternatives for race: American Indian or Alaska Native; Black or African American; White; Asian; Native Hawaiian or other Pacific Islander; more than one race). Of the 99 young adults, 66 had two parents reported as White, 26 subjects had one or two African American parents, and 7 subjects had one or two parents who were Hispanic. The genetic analyses yielding significant association for the total sample (N = 99) were repeated for the non-Hispanic Caucasian subgroup (N = 66).

Young adult psychopathology was assessed using the Structured Clinical Interview for Diagnosis (SCID) for DSM-IV Axis I and II [33, 34], which was administered in the laboratory by trained clinicians (see clinical characteristics for Axis I disorders in Table 1). Among the 99 young adults 2 were diagnosed with both BPD and APD, 2 with BPD, 5 with APD, and 14 with other personality disorder(s). The pattern of antisocial or borderline personality disorder traits among the young adults were the following: 4% met criteria for a diagnosis of BPD, however, 20% endorsed two or more borderline symptoms; 7% of the sample met diagnosis of APD (i.e. 3 or more APD symptoms and 2 or more conduct disorder symptoms before age 15 years), and 26% showed two or more antisocial symptoms. In the categorical analyses (chi-square and binary logistic regression) borderline and antisocial traits were coded as present if two or more features of the disorder were endorsed.

In the analyses a severity of abuse variable was also used: An overall rating for severity of abuse from birth to age 18 was assigned by reviewing the Conflict Tactics Scale second version [35], the Traumatic Stress Schedule [36], and the Childhood Traumatic Experiences Scales-Revised coded from one-hour semi-structured Adult Attachment Interview [37], and any history of state protective services involvement. The severity of abuse scale was coded as 0 - no occurrence of violence; 1 - harsh punishment or witnessed violence; 2 - emotional/verbal abuse; 3 - one type of physical or sexual abuse (using state guidelines for abuse), or protective services/foster care involvement; 4 - two or more forms of abuse/trauma out of the three listed previously. Reliability of the severity of abuse scale was assessed by having two coders independently rate 37 of the 99 participants on severity of abuse by reviewing the above set of assessments; reliability between the two coders assessed by intraclass correlation was high, r_i = 0.99.

The independent replication sample consisted of psychiatric inpatients diagnosed with mood disorders, recruited from the Department of Psychiatry, Kútvölgyi Clinical Centre (Budapest, Hungary) during their depressive periods requiring treatment. Unrelated participants of Caucasian (Hungarian) origin were included in the present study, thus creating an ethnically homogenous population [38]. The study was conducted in compliance with the Helsinki Declaration, and was approved by the Local Ethical Committee (TUKEB), all participants provided written informed consent. After screening out organic causes of psychiatric disorders, DSM-IV psychiatric diagnoses [39] were obtained by a clinical psychiatrist, the Axis I disorder frequencies are presented in Table 1. Among the 136 patients (female 76.5%; mean age: 46.3 ± 10.5, age range: 21-64 years) 14 (10.3%) were diagnosed with personality disorders. For the genetic analyses borderline symptoms were assessed from the self-report SCID Screen questionnaire, which showed good correlations with SCID-II interviews [40]. The number of BPD symptoms (0-9) was used as the borderline scale; the internal consistency of the scale was found to be satisfactory (Cronbach's alpha: 0.729).

Non-invasive DNA sampling was applied, and DNA was isolated from buccal cells using the DNA-purification kit obtained from Gentra (Minneapolis, US). Genotyping procedures for the DRD2 TaqIA (rs1800497), TaqIB (rs1079597), TaqID (rs1800498), and DAT1 40 bp VNTR in the 3' untranslated region were carried out using published protocols [41‐44]. Whereas VNTR genotyping methods of the DRD4 48 bp VNTR in exon III and 120 bp duplication in the promoter region, and the allele-specific amplifications of the DRD4 -616 C/G (rs747302), -521 C/T (rs1800955) and COMT Val158Met (rs4680) were developed and optimized in our laboratory [45‐47]. The genotyping accuracy was checked by parallel genotyping of two independent DNA samples per person, and by calculating chi-square tests for deviation from the Hardy-Weinberg equilibrium using Knud Christensen's program [48]. Except for the -616 C/G SNP in the Hungarian psychiatric sample, no significant deviations from the Hardy-Weinberg equilibrium were detected in either populations. We assume that the slight deviation from the equilibrium (χ² = 4.099, df = 1, p = 0.043) in the Hungarian patient sample originates from possible association of the -616 C/G SNP with psychiatric disorder(s). In order to support this assumption, control subjects (with no psychiatric history) were recruited at the Institute of Psychology, Eotvos Lorand University (N = 178, age: 23.2 ± 5.0; 72.5% female) and genotyped for the studied dopaminergic polymorphism (see additional file 1). Our data demonstrated that the non-clinical Hungarian sample was indeed in Hardy-Weinberg equilibrium for -616 C/G SNP (χ² = 0.104, df = 1, p = 0.747).

In the genetic analyses the VNTRs were analyzed according to the number of the risk allele, i.e. the DRD4 48 bp VNTR 7-repeat allele and the DAT1 40 bp VNTR 9-repeat allele (see Table 2 and 3). In the second set of analyses the genotypes were grouped according to the presence of the risk allele, resulting in DRD4 7-present (7+) and 7-absent (7-) groups. At the DAT1 VNTR the grouping was based on the presence of the 9-repeat allele, resulting in 9-present (9+) and 9-absent (9-) groups; the rare 3/10, 6/10, 10/11, and 10/12 genotype were grouped together with the 10/10 genotype in the 9- group, whereas the rare 3/9 and 8/9 genotype were regarded as having one 9-repeat allele, and grouped to the 9+ group. At the SNP markers, homozygotes for the minor allele were grouped together with heterozygotes in the second round of analyses, i.e. DRD2 A1+ group consisted of A1/A1 (TT) and A1/A2 (CT) genotypes, DRD2 B+ group consisted of B1/B1 (AA) and B1/B2 (AG) genotypes. The DRD4 -521 C/T SNP was analyzed in CC + CT vs. TT setting, and at the -616 C/G SNP CC vs. CG + GG groups were compared.

The SPSS program for Windows (17.0 version) was used for the statistical analyses. In the US young adult sample Pearson chi-square analyses were conducted using the symptom cut-off scores at 2 to assess the effect of dopaminergic polymorphisms. In the binary logistic regression analyses gender and severity of abuse were also entered beside the dichotomous genetic variable in the same step (block 1). In the exploratory interaction analyses of DRD4 48 bp VNTR, dichotomous genetic variables (DRD4 7+ vs 7-, DRD2 A1+ vs A1-, DAT1 9+ vs 9-), gender, and the 5-point scale of severity of abuse were entered at block 1, whereas the created interaction variable (DRD4 × DRD2, DRD4 × DAT1, or DRD4 × abuse) was entered in block 2. Genetic association analyses within the Hungarian clinical sample were conducted by univariate analysis of variance using the borderline scale as the dependent variable and the dopaminergic genotypes as the grouping factor, with sex and age used as covariates. To control for multiple comparisons, the False Discovery Rate adjustment of significance levels was used at α = 0.05 level [49].

Haplotype analyses were conducted by likelihood-based association analysis using the Unphased program [50], also direct molecular haplotyping was applied for the DRD4 promoter -616 C/G and -521 C/T SNPs [46]. Linkage disequilibrium between the DRD2 SNPs was calculated by the Haploview program [51] using the Hungarian control sample genotype data: the DRD2 TaqIB SNP was in linkage with TaqIA (D'= 0.947, LOD = 29.77, r² = 0.701) and with TaqID (D'= 1.0, LOD = 12.17, r² = 0.212). Linkage disequilibrium for the DRD4 promoter polymorphisms has been previously reported [46].

The previously indicated polymorphisms in the DAT1 and COMT gene were not associated with either borderline or antisocial traits in the US young adult sample (Table 2). However, both the DRD2 B1-allele and A1-allele were significantly associated with borderline traits (TaqIB 3 genotypes: χ² = 14.935, df = 2, p = 0.001, B1+ vs B1- categories: χ² = 12.977, df = 1, p = 0.0003; TaqIA 3 genotypes: χ² = 10.568, df = 2, p = 0.005, A1+ vs A1- categories: χ² = 9.477, df = 1, p = 0.002). These associations remained significant after correcting for multiple comparisons (p < 0.0056). Subjects carrying at least one B1-allele were six times more likely to develop borderline symptoms, OR = 6.2 (2.15-17.85), whereas A1+ subjects were five times more likely to have borderline symptoms, OR = 5.17 (1.7-15.7). Similar results were found in the Caucasian subgroup (N = 66): TaqIB 3 genotypes: χ² = 15.415, df = 2, p = 0.0004; B1+ vs B1- categories: χ² = 14.139, df = 1, p = 0.0002, OR = 10.1 (2.7-37.6). TaqIA 3 genotypes: χ² = 11.253, df = 2, p = 0.004; A1+ vs A1- categories: χ² = 11.215, df = 1, p = 0.001, OR = 7.7 (2.13-27.99). With gender and severity of abuse accounted for in the model, the effect of DRD2 SNPs was still significant (Table 4). Similar results appeared in the Caucasian subgroup (Table 4). For the APD traits, only the DRD2 TaqID SNP showed a nominally significant association (p = 0.046) which did not remain significant after correcting for multiple comparisons.

Among the DRD4 polymorphisms, the -616 C/G and -521 C/T promoter SNPs showed potentially interesting associations with borderline and antisocial traits (Table 2). The -616 CC and -521 TT genotypes were overrepresented among those who had 2 or more borderline or antisocial symptoms. In the second round of analyses homozygotes for the minor allele were grouped together with heterozygotes. Using the CC vs CG + GG grouping system at the -616 C/G SNP, the difference in genotype distribution was nominally significant for borderline symptoms (p = 0.010), whereas at the -521 C/T SNP the CC + CT vs TT grouping showed nominally significant results for both borderline (p = 0.029) and antisocial (p = 0.011) symptoms. In the Caucasian subgroup similar finding emerged for the -616 C/G SNP when testing borderline symptoms (χ² = 6.732, df = 1, p = 0.009). Whereas at the -521 C/T SNP the analyses showed only marginal effect (borderline: χ² = 3.609, df = 1, p = 0.057, antisocial: χ² = 3.493, df = 1, p = 0.062). Neither of the nominally significant associations (p < 0.05) remained significant after correcting for multiple comparisons (p < 0.0056). However, in the logistic regression analyses (entering gender and severity of abuse in the model), the -616 C/G SNP had a significant effect on BPD symptoms at p < 0.0056 level (Table 4). Similar results emerged in the Caucasian subgroup: severity of abuse and the number of -616 C allele were significant predictors of BPD symptoms (p = 0.004, and p = 0.005, respectively, Table 4). The DRD4 48 bp VNTR did not show any association with either BPD or APD traits (Table 2).

Since dopaminergic polymorphisms have been associated with alcohol and drug abuse, we checked for possible confounding genetic association with substance abuse in our sample. Among the young adults 38.4% were diagnosed with substance abuse (Table 1). At the DRD2 SNPs the association of the TaqIA polymorphism with substance abuse was marginally significant using the 3 genotype categories (χ² = 5.34, df = 2, p = 0.069) and nominally significant for the A1+ vs A1- categories (χ² = 4.52, df = 1, p = 0.034), with 57.9% of the substance abuse present group carrying the A1-allele vs 36.1% of the substance abuse absent group. Similar findings emerged at the TaqIB 3 genotype categories: χ² = 5.07, df = 2, p = 0.079, and at the B1+ vs B1- categories: χ² = 4.21, df = 1, p = 0.04, with 39.5% of the substance abuse present group carrying the B1-allele vs 20.3% of the substance abuse absent group. These results did not remain significant after correcting for multiple testing. None of the DRD4 promoter SNPs showed significant association with substance abuse using either the 3 or 2 genotype categories.

According to previously reported findings, exploratory interaction analyses were conducted with the DRD4 48 bp VNTR (for genotype grouping and analysis description see Methods section). The DRD4 7-repeat allele × severity of abuse interaction analysis did not yield a significant result (p = 0.43). Nor did the DRD4 7+ × DRD2 A1+ interaction analysis result in a significant finding (p = 0.725). There was a nominally significant finding at the DRD4 7+ × DAT1 9+ interaction (p = 0.045), but this result did not remain significant after correcting for multiple testing.

For confirmation of the findings regarding borderline traits, univariate ANOVA was conducted in an independent Caucasian population, using the SCID-II Screen questionnaire borderline scale in a Hungarian sample of depressive patients (Table 3). Neither of the DRD2 polymorphisms showed association with BPD traits. On the other hand, the DRD4 -616 C/G SNP showed similar association with BPD traits as observed among the at-risk US young adults. The -616 CC genotype was associated with more severe borderline symptoms using the three genotype groups (p = 0.007). Results were also significant for the combined CC vs CG + GG genotype grouping (p = 0.005). Other dopaminergic polymorphisms did not show any significant associations with the number of borderline symptoms in this clinical sample.

Since both the DRD2 TaqIB and TaqIA SNPs were associated with BPD traits, haplotype analyses were conducted using the Unphased program. Again, only the US young adult sample analysis yielded significant results (likelihood ratio χ² = 12.49, df = 2, p = 0.002), with the B1 ~ A1 (A ~ T) haplotype conferring risk for BPD traits: OR = 4.89 (2.05 - 11.66). Furthermore, the haplotype constructed from the three DRD2 SNPs was also significantly associated with BPD traits: likelihood ratio χ² = 13.91, df = 4, p = 0.008, OR = 7.35 (2.38 - 22.7) for the A ~ C ~ T haplotype. The estimated haplotype analyses for the DRD4 -616 C and -521 T alleles showed association with BPD traits in both samples: US young adult sample: likelihood ratio χ² = 10.41, df = 3, p = 0.015; Hungarian patient sample: likelihood ratio χ² = 9.81, df = 3, p = 0.02. In addition, the exact chromosomal localization of the two DRD4 SNPs could be determined by direct haplotyping methods [46]. Genetic association analyses were conducted using the number of the -616C ~ -521T (C ~ T) haplotype. In the US young adult sample the frequency of the DRD4 -616C ~ -521T haplotype was increased among those who had 2 or more borderline symptoms (55% had one C ~ T and 25% had two C ~ T haplotype vs 47.4% and 5.1%, respectively, χ² = 9.89, df = 2, p = 0.007). In the logistic regression analyses, with gender and severity of abuse accounted for in the model, the genetic effect was still significant: OR = 4.76 (1.77 - 12.79), p = 0.002. In the Hungarian patient sample univariate ANOVA showed a similar effect: those who had -616C ~ -521T haplotype on one or two chromosomes displayed a higher number of borderline symptoms (for three groups: p = 0.03, for two groups: p = 0.009, Table 3).