Genotype-Phenotype Studies in Bipolar Disorder Showing Association Between the DAOA/G30 Locus and Persecutory Delusions: A First Step Toward a Molecular Genetic Classification of Psychiatric Phenotypes

We defined lifetime psychotic symptoms as the occurrence of delusional or hallucinatory symptoms. Patients with bipolar affective disorder were classified as having a positive history of psychosis if a “yes” rating was given on at least one of the 21 OPCRIT symptoms of delusions or hallucinations (Table 2). Of the 300 patients in the exploratory study, we identified 173 individuals with a definite positive history and 83 with a definite negative history. The latter assignment was only made when all 21 OPCRIT symptoms of psychosis were rated negative with no missing values. Thus, for 44 patients, no clear assignment could be made. Of the 294 patients in the replication study, 121 patients with a definite positive history of psychosis and 163 patients with a definite negative history were identified (10 patients could not be classified).

The DAOA/G30 single nucleotide polymorphisms (SNPs) genotyped in the exploratory study were determined by Masscode Technology (QIAGEN Genomics). The DAOA/G30 SNPs analyzed in the replication study were genotyped using the MassARRAY system (Sequenom, San Diego) according to Ding and Cantor (25) with minor modifications. The genotyping quality between both techniques is comparable: two SNPs (rs7320 and rs2185740) were genotyped using both techniques, the Masscode and the Mass ARRAY technology, and no discrepancies in genotypes were observed. Genotype frequencies were in Hardy-Weinberg equilibrium for cases and comparison subjects in both study groups.

We performed both single-marker and haplotype analyses with the program COCAPHASE 2.40 (http://www.mrc-bsu.cam.ac. uk/personal/frank/software/unphased/) (26). Using a standard unconditional logistic regression, this software package performs likelihood ratio tests under a log-linear model of the probability that an allele or a haplotype belongs to a case rather than a noncase; the expectation-maximization algorithm is used to resolve uncertain haplotypes and provide maximum-likelihood estimates of frequencies.

For the single-marker analyses, a permutation procedure was used to estimate the significance of the best results, correcting for all loci tested. For haplotype analyses, the global null hypotheses that all odds ratios are equal were tested by permutation, owing to the fact that estimated haplotype frequencies cannot be treated as observed data. The permutation method implemented in COCAPHASE randomly reassigns the “patient” and “comparison” labels in the actual data. Ten thousand permutations were performed.

Comparisons were performed across clinical subgroups by stratifying the patients and comparing allele and haplotype frequencies against the same comparison population.

Statistical genotype-phenotype association analysis was performed by using logistic regression analysis as implemented in the SAS procedure PROC LOGISTIC. For this analysis, we chose the genotype of the DAOA/G30 marker M23, which had yielded the most marked result in our previous analysis (3), as the dependent variable with three categories (i.e., the genotypes TT, TC, CC). The 21 OPCRIT symptoms of psychosis (Table 2) were entered as explanatory variables in the model, as well as the variables age, gender, and age at onset. The logistic regression analysis was performed on a complete case basis including only those patients for whom information on all items was available (N=225). The logistic model was used in its generalized version allowing for dependent variables that take their values in a set of more than two ordered categories. This approach is based on the proportional odds assumption according to which the odds ratios related to all possible dichotomized versions of the dependent variable, except for the intercept terms, are the same function of the covariables for any choice of the cutoff. Generalized logistic regression was carried out for the 21 OPCRIT symptoms of psychosis both as a full set and in a stepwise manner selecting the symptoms that showed a significant contribution to the goodness of fit of the model.

It has been suggested that, given a population with significant association, an improvement in p value may exist in random subsets of the population (27). Therefore, we established the empirical significance of any asymptomatically significant finding by simulation. In 10,000 replicates, random subsets of the size of the actual, clinically defined subset were drawn from the overall group and analyzed.

Power calculations were performed using the Genetic Power Calculator (http://statgen.iop.kcl.ac.uk/gpc/cc2.html) (28) and were based on the odds ratio of 1.33 for the best associated marker (M23) in the German sample. We used both a recessive and a dominant model, assuming equal allele frequencies for the disease and marker alleles, and a D′ of 0.8 between disease and marker locus.

In the German group of 300 patients and 300 comparison subjects, the association between bipolar affective disorder and markers and haplotypes at the DAOA/G30 locus yielded odds ratios of around 1.3 (Table 3 and Table 4). We investigated whether this finding was due to an association with the overall group or with the subset of patients displaying psychotic features.

For the analysis of the subgroup of patients with psychotic bipolar disorder, we chose the four markers that made up the risk haplotype for both schizophrenia and bipolar affective disorder in the entire group of patients with bipolar affective disorder, i.e., M12, M15, M23, and M24.

In the single marker analysis, only M23 showed a significant association (p=0.005) in bipolar disorder patients with psychotic symptoms, as was the case for the entire group of patients with bipolar affective disorder (p<0.02). The robust permutation procedure in COCAPHASE, which we applied to correct for the four markers tested, yielded a p value of 0.012. By simulation analysis, a p value of 0.005 or smaller was observed 1,162 times out of 10,000 replicates, suggesting that the observed change in p value obtained through phenotype refinement bears a 12% probability of being a chance finding (empirical p value=0.12).

For the haplotype analyses (Table 4), significant associations were seen for the global four-marker haplotype (p=0.04) and the individual haplotype G-G-T-A (p<0.03), with the G-G-C-T haplotype approaching significance (p<0.09), respectively.

The 83 patients with bipolar affective disorder and no lifetime history of delusional or hallucinatory symptoms showed no association, either in the single marker or in the haplotype analyses (Table 3 and Table 4).

Given that the aforementioned stratification on psychotic features might still be too crude a procedure for refining phenotypes, we were interested in whether the association between bipolar affective disorder and DAOA/G30 was explained instead by an even more specific subset, defined by specific psychotic features. We performed a logistic regression in a model that included age, gender, age at onset, and the 21 OPCRIT symptoms of psychosis (Table 2). This analysis identified the variable “lifetime history of persecutory delusions” as the only significant (p<0.005) explanatory variable for the M23 genotype. Individuals with a history of persecutory delusions were more likely to have genotypes containing the C allele than those with no such history. Generalized logistic regression was carried with the 21 OPCRIT symptoms of psychosis both as a full set and in a stepwise manner. Since both approaches led to the same qualitative conclusions, only the results of analyzing the full multivariate model are presented.

We thus divided our group of bipolar affective disorder patients for whom detailed information regarding the OPCRIT item “persecutory delusion” was available into those with (N=90) and those without (N=165) a history of persecutory delusions in order to perform two separate comparisons with the comparison group. When comparing the subset of 90 bipolar disorder patients with persecutory delusions versus comparison subjects, markers M12 and M15 did not yield significant results. As seen in Table 3, we observed significant association for markers M23 (p<0.0004) and M24 (p<0.007).

In a simulation analysis, p values <0.0004 and <0.007 were observed for, respectively, 24 and 67 out of 10,000 random subsets of 90 individuals, indicating that the association findings at these significance levels are very unlikely to occur in equally sized random subsets (p=0.0024 and p=0.0067).

In the haplotype analysis (Table 4), significant associations were seen for the global four-marker haplotype (p<0.03) and the individual haplotypes G-G-T-A (odds ratio=0.54, p<0.006) and G-G-C-T (odds ratio=1.54, p<0.04). We also evaluated the empirical significance of these haplotype-based results: a p value smaller or equal to 0.006 would be seen in 1,360 out of 10,000 random subsets of 90 individuals (p=0.136). Comparing the larger subsample of patients with bipolar disorder and no history of persecutory delusions (N=165) with the comparison group (N=300) yielded no association (odds ratios ~1) (Table 3, Table 4). We also compared the comparison subjects with those 83 patients that had a history of psychotic symptoms but not of persecutory delusions (i.e., the 173 patients with psychosis minus the 90 patients with a history of persecutory delusions [see Table 1]). For all four markers, again, there was no association. Allele and haplotype frequencies were similar between this particular subset of cases and comparison subjects (odds ratios ranged from 0.86 to 1.14, p values ranged from 0.50 to 0.97 [data not shown]).

The subsample of patients with bipolar disorder with persecutory delusions did not differ from the entire sample of patients with bipolar disorder regarding gender (χ²=0.37, p=0.54), age (t=0.26, p=0.79), or age at onset (t=0.62, p=0.536).

Given that only markers M23 and M24 showed association with bipolar affective disorder, we analyzed these in our replication study with the Polish group. As with the analysis of the German subjects, we performed single-marker and haplotype-based analyses, using three case definitions: all patients with bipolar affective disorder, a subgroup of those with a history of psychotic symptoms, and a subgroup of those with a history of persecutory delusions. In the single-marker analysis, neither M23 nor M24 showed significant association with bipolar affective disorder in any of the three sets (Table 5). However, for M23, we observed a nearly significant association (odds ratio=1.48, p<0.08) in the subset of bipolar disorder patients with persecutory delusions. Similar to the German group, the bipolar disorder patients with persecutory delusions were enriched with the C allele relative to the entire group of bipolar affective disorder patients (57% versus 49%). With regard to the haplotype analysis, no association was observed in the bipolar affective disorder patients and the bipolar affective disorder subgroup with a history of psychotic symptoms. When case definition was confined to those with persecutory delusions, however, we observed a significant association (Table 6): The C-T haplotype, which is part of the associated G-G-C-T haplotype in the German subjects, was similarly more frequent in patients than in comparison subjects (odds ratio=1.61, p<0.04). We also evaluated the empirical significance of this haplotype-based result: a p value ≤0.04 would be seen in 220 out of 10,000 random subsets of 55 individuals (p=0.022). Thus, while there was no association when using the total group of patients with bipolar affective disorder or the subgroup with a history of psychotic symptoms, there was a significant one when cases were restricted to patients with persecutory delusions. The “protective” haplotype T-A did not show a significant association. It is interesting to note, however, that it showed the same frequency in the total group of bipolar affective disorder patients, the subgroup of bipolar affective disorder patients with psychotic symptoms, and comparison subjects while being less frequent in bipolar disorder patients with persecutory delusions, as observed in the German subjects.