None of the authors has any involvement, financial or otherwise, that might bias this work.
TRPM2 variants and bipolar disorder risk: confirmation in a family-based association study
Version of Record online: 9 JAN 2009
© 2009 The Authors. Journal compilation © 2009 Blackwell Munksgaard
Volume 11, Issue 1, pages 1–10, February 2009
How to Cite
Xu, C., Li, P. P., Cooke, R. G., Parikh, S. V., Wang, K., Kennedy, J. L. and Warsh, J. J. (2009), TRPM2 variants and bipolar disorder risk: confirmation in a family-based association study. Bipolar Disorders, 11: 1–10. doi: 10.1111/j.1399-5618.2008.00655.x
- Issue online: 9 JAN 2009
- Version of Record online: 9 JAN 2009
- Received 22 November 2007, revised and accepted for publication 14 February 2008
- bipolar disorder;
- case-control design;
- family study design;
- intracellular calcium homeostasis;
Objective: Recent case-control studies implicate the transient receptor potential melastatin 2 (TRPM2) channel in conferring risk for bipolar disorder (BD), though the risk variants differed. As confounding effects of population structure could not be unequivocally ruled out as the basis for the discordance, we tested the association of TRPM2 with BD in a family design, which is immune to population stratification, for those TRPM2 single nucleotide polymorphisms (SNPs) previously reported as associated with BD.
Methods: The exon 11 SNP (rs1556314) and four informative intronic SNPs (rs1785437, rs1618355, rs933151, and rs749909) were genotyped in 300 BD families by TaqMan allelic discrimination and results were analyzed using χ2 test, transmission disequilibrium test, and pedigree-based association. SNP rs1556314 was also genotyped in our case-control sample set comprised of 184 BD and 195 healthy Caucasian subjects.
Results: The SNP rs1556314 in exon 11 was significantly associated with bipolar disorder type I (BD-I) (p = 0.011, ppermutation = 0.015) in the case-control dataset and in the family design (p = 0.018, ppermutation = 0.052, TDTPHASE). Interestingly, the C-T-A haplotype of SNPs rs1618355, rs933151, and rs749909 was significantly associated with early age at onset in BD-I families.
Conclusion: Significant association of TRPM2 genetic variants with BD in case-control and family datasets further supports a role for TRPM2 in the pathogenesis of this disorder. Overtransmission of the G allele of rs1556314 at exon 11 of TRPM2 in BD-I but not bipolar disorder type II (BD-II) further supports different genetic contributions to the pathogenesis of these bipolar phenotypes.
Although genome scan studies of bipolar disorder (BD) have illuminated several chromosomal loci which satisfy consensus criteria as harboring genes involved in BD (1, 2), evidence is sparse regarding specific genes within these regions that contribute to this complex trait. Guided by observations implicating altered intracellular calcium (Ca2+) signaling dynamics in the pathophysiology of BD (3), we focused on a priority candidate, the nonselective cation permeable transient receptor potential melastatin type 2 (TRPM2) channel, based on its involvement in processes regulating intracellular Ca2+ homeostasis (4) and regulation by oxidative stress (5), its location on a confirmed chromosomal region (21q22.3) harboring a BD susceptibility gene(s) [reviewed briefly in (6)], and its decreased expression in B lymphoblast cell lines (BLCLs) from BD type I (BD-I) patients with elevated basal intracellular Ca2+ concentrations ([Ca2+]B) compared with those with normal BLCL [Ca2+]B and with BLCLs from healthy subjects (7). Significant association was found between BD and several single nucleotide polymorphisms [(SNPs) rs1785437, rs1618355, rs1612472, and rs933151 in introns 16, 18, 19, and 20, respectively] of the TRPM2 gene in our recent case-control study design (6). An independent fine-mapping study of a susceptibility locus for BD on 21q22.3 (8) also revealed significant association of TRPM2 with BD, but the risk variant, SNP rs1556314 in exon 11, was different from those observed in the region between intron 16 and 20 SNPs in our recent report (6). In contrast, Roche et al. (9) failed to find evidence of association between SNPs rs1556314 and rs933151 of the TRPM2 gene and BD in a family-based sample set; one possible explanation may be the small sample size, consisting of only 125 BD-I families.
In view of these somewhat discordant findings, we sought to confirm the association of TRPM2 variation with BD in a larger family study design (n = 300 BD families), since such a design would be immune to confounding effects of population stratification (10), which could not be unequivocally excluded as a factor contributing to potentially spurious findings in previous case-control studies (6, 8). In the present study, we focused on those SNPs previously reported in case-control designs to be significantly associated with BD, including exon 11 SNP rs1556314 (8), rs1785437, rs1618355, and rs933151 in introns 16, 18, and 20, respectively, and an SNP, rs749909 in intron 27, which showed nominal statistically significant association with BD as a whole and with BD-I in particular (6). The exon 11 SNP rs1556314 (Asp543Glu), which was reported to be associated with BD in the case-control study of McQuillin et al. (8), is the only exonic SNP of high minor allele frequency identified to date in TRPM2. The SNP rs749909 in intron 27 is located proximally to the exon encoding an important structural region of the TRPM2 protein, the C-terminal coil-coiled region and NUDT9-H domain. The latter region exhibits intrinsic adenosine diphosphate (ADP)-ribose hydrolase activity (4) and is responsive to stimulation by ADP-ribose (11). The TRPM2 channel gates Ca2+ entry into cells in response to oxidative stress signaling mediated by ADP-ribose, to which it is sensitive (5). Disturbances of oxidative stress signaling (12), intracellular Ca2+ dynamics (3), and endoplasmic reticulum stress response (13) have been implicated in the pathophysiology of BD. Deletion of a partial sequence of exon 11, encompassing SNP rs1556314, ablates the response of TRPM2 to H2O2 or ADP-ribose stimulation (11), raising the possibility that this BD-associated SNP may be functionally relevant.
Here we report confirmatory evidence of an association between BD and the genetic variant of TRPM2 SNP rs1556314 in exon 11. Significant 2-locus-haplotype association (consisting of exon 11 and intron 16 SNPs) and 2-locus, 3-locus associations (consisting of rs1618355, rs933151, and rs749909 in intron 18, 20, and 27 SNPs) with early age at onset (AAO) in BD-I families were also observed.
Material and methods
As previously described (14), the family dataset was comprised of probands from 300 BD families with a diagnosis of BD-I, bipolar disorder type II (BD-II), or schizoaffective disorder–bipolar type, along with their living parents and, in some cases other relatives. Probands and family members were assessed by trained interviewers using the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I), patient edition (15), and diagnoses were made according to a ‘best estimate’ procedure (14). Subjects ranged in age from 16 to 67 years (mean ± SD, 35 ± 10 years). Female subjects comprised 52% of the sample and 63% of BD probands. Age at onset of illness, defined as the first distinct episode meeting DSM-IV criteria for depression or hypomania/mania, was extracted from the SCID-I. BD families were also categorized as those with (43.7%) or without (56.3%) psychotic probands based on lifetime history of psychosis within the context of an episode of mania or depression as extracted from the SCID-I. Almost all probands recruited were outpatients at the time of their participation in the study (14, 16). The study was approved by the Research Ethics Board of the Centre for Addiction and Mental Health, and following a complete description of the study, participating subjects provided written informed consent. Among the 300 families, 96.8% were of European, Caucasian origin and included 252 trios (84% of the family sample), 11 extended families, 14 affected sib-pair families, and 23 discordant sib-pair families. In total, five SNPs were selected for genotyping: four intronic SNPs (rs1785437, rs1618355, rs933151, and rs749909) which were previously found to be significantly associated with BD at different significance levels in case-control study designs (6, 8), and one putative functional SNP (rs1556314 in exon 11), which also showed an association with BD in the study by McQuillin and coworkers (8). An intron 19 SNP (rs1612472) also showed an association with BD in our previous report (6); however, it was not examined in the current family study design due to its strong linkage disequilibrium (LD) (D′ = 0.98) with intron 20 SNP (rs933151) and a lower genotype-calling rate that required multiple independent assays to ensure genotype concordance.
The SNP rs1556314 in exon 11 was also genotyped in our case-control dataset as it had not been previously examined in this sample. The case-control dataset was comprised of 184 Caucasian BD patients (131 BD-I and 53 BD-II) and 195 age-, sex-, and ethnicity-matched healthy subjects. Psychiatric diagnoses were confirmed using the SCID-I as previously described (6). Healthy subjects had no personal or family (first-degree relative) history of psychiatric disorder, based on structured interview with the SCID-I nonpatient version (17).
DNA extraction and SNP genotyping
For the family study samples, genomic DNA was extracted from EDTA anticoagulated blood samples using a high-salt method (18) as previously described (14). For case-control subjects, genomic DNA was extracted from subjects’ BLCLs using a QIAamp DNA Mini Kit (Qiagen, Mississauga, Ontario, Canada) as previously described (6). Genotyping was performed using the 5′ nuclease allelic discrimination TaqMan assay (Applied Biosystems, Inc, Foster City, CA, USA) in a 96-well format (6). Briefly, 5 μl of PCR master mix (TaqMan® Universal PCR Master Mix), 5–10 ng of genomic DNA, and 0.5 μl of 20× Assay-on-Demand or 40× Assay-by-Design SNP Genotyping Assay Mix (sequence-specific primers and probes) were added to each well. Amplification reactions were carried out with endpoint detection on an ABI PRISM® 7300 Sequence Detection System (Applied Biosystems, Inc., Foster City, CA, USA). SNP genotyping was performed blind to subject diagnosis, characteristics, and sample replication.
Pedigree, genotyping errors, and data handling
The family-based association test [(FBAT) version 1.7.3 (19)] was used to check for Mendelian inconsistencies. Those showing inconsistencies were subjected to repeat genotyping and the families were excluded from further data analyses when the discrepancies remained. To monitor genotyping accuracy, negative control samples without genomic DNA, the same positive controls with known genotypes (1/1, 1/2, and 2/2 genotypes), and a random selection of a proportion of the samples with known genotypes were included on each 96-well plate. The genotype concordance rates were 100% for the respective positive controls and 98.5% for duplicate samples. However, increasing the concentrations or further purification of the genomic DNA of those showing duplicate sample errors improved the genotype concordance rates to 100%. Missing data rates were 0.67% for the case-control study and 0.85% for the family study design. Genotype data were independently checked by two research staff who were blind to the diagnosis. Genotype distributions for the five tested SNPs were in Hardy-Weinberg equilibrium (p > 0.05).
In the family design, the transmission disequilibrium test (TDT) from the TDTPHASE program of the UNPHASED software, version 2.403 (20), was used to examine individual SNP association with BD and all two-, three-, four-, and five-marker haplotype associations, excluding rare haplotypes (<1%), in a sliding window fashion. For phase-certain haplotypes a conditional logistic regression model was used, corresponding to the probability of the offspring haplotype conditional upon the parents’ haplotypes. When phase was uncertain, unconditional logistic regression on the full likelihood of the haplotype of parents and offspring was used (21). In order to maximize power in the TDTPHASE statistical analyses, permutation procedures were implemented to calculate the empirical p-values derived from 10,000 permutations. This handles multiple testing directly and avoids correction bias. Moreover, single-marker and haplotype analyses with AAO as a quantitative trait were performed using the FBAT-Wilcoxon test from the Pedigree-Based Association Test [(PBAT), version 3.2 (22)]. The latter is a multivariate extension of the FBAT with the advantage that it does not bias subsequently computed FBAT statistics and provides nonparametric screening for non-normally distributed quantitative trait variables (22) using the FBAT-Wilcoxon test, which has been shown in simulation testing to provide greater power than other non-parametric tests (23). In addition, the FBAT/PBAT approach is a generalized version of the TDT, which is also able to analyze extended pedigrees maximizing phenotype and genotype information used in analyses.
In the case-control study, the allele or genotype frequencies of the exon 11 SNP rs1556314 were compared in BD, BD-I, and BD-II with healthy individuals using two-tailed χ2 tests (6). COCAPHASE (20) was used to calculate the empirical p-values derived from 10,000 permutations.
K-means cluster analysis [SPSS 14.0 (SPSS, Inc., Chicago, IL, USA] was used to identify relatively homogeneous subgroups of AAO of 255 affected offspring from families with BD-I probands by using the initial mean and standard deviation for admixture analysis. Four BD families had no AAO information and were excluded from this analysis. The 41 BD-II families were not subjected to separate K-means cluster analysis due to small sample size. In admixture analysis using the NOCOM program (http://linkage.rockefeller.edu/ott/nocom.htm), maximum likelihood analyses were performed to determine the best-fit model (e.g., unimodal, bimodal, and trimodal normal distributions) of AAO. The Kolmogorov-Smirnov Z-test was used to test the normality of each distribution.
In testing of the hypothesis that the inheritance pattern of TRPM2 variants might be influenced by parent-of-origin and gender-specific effects, parental and offspring sex were selected as modifiers in the TDTPHASE analysis.
Power analyses based on the current family study design (n = 300 BD families genotyped) yielded an estimate of 0.93 at α = 0.05 and 0.80 at α = 0.01 for a gene of moderate effect [odds ratio (OR) = 2.0], as estimated using the TDT for discrete traits of the Genetic Power Calculator (http://pngu.mgh.harvard.edu/~purcell/gpc/). The calculated power for the case-control study design (n = 379) was 0.81 at α = 0.05 for a gene of moderate effect.
In agreement with the haplotype block structure observed in our earlier Caucasian case-control study (6), one haplotype block was identified (Fig. 1). The three SNPs rs1785437, rs1618355 and rs933151 in introns 16, 18, and 20 showed strong LD using unrelated founders in this family cohort (pairwise D′ ranged from 0.91 to 0.95 and pairwise r2 ranged from 0.67 to 0.91; see Fig. 1).
Nominally statistically significant increased G allele (minor allele) frequency was observed for rs1556314 in exon 11 in BD as compared with healthy controls (27.2% versus 21.0%, OR = 1.40, p = 0.047; see Table 1). Moreover, the G allele frequency of this putative functional SNP was also significantly increased in the BD-I subtype patient group as compared with controls (29.8% versus 21.0%, OR = 1.59, p = 0.011; see Table 1). After permutation analyses to correct for multiple testing using COCAPHASE, allelic association of exon 11 SNP with BD-I remained significant (ppermutation = 0.015) and showed borderline significance (ppermutation = 0.056; see Table 1) with the BD patient group as a whole. This suggests that the G allele of exon 11 SNP had a stronger effect on the BD-I subtype than on BD as a whole.
|Diagnosis||Allele T, n (%)||Allele G, n (%)||OR (95% CI)||χ2||p-value||ppermutation valuea|
|Bipolar disorder (N = 184)||268 (72.8)||100 (27.2)||1.40 (1.00–1.96)||3.92||0.047||0.056|
|BD-I (n = 131)||184 (70.2)||78 (29.8)||1.59 (1.11–2.28)||6.47||0.011||0.015|
|BD-II (n = 53)||84 (79.2)||22 (20.8)||1.02 (0.60–1.73)||0.00||0.952||0.096|
|Control (n = 195)||308 (79.0)||82 (21.0)|
In addition to rs1556314, SNPs rs1785437, rs1618355, rs933151, and rs749909 were genotyped to determine whether those variants in TRPM2, which showed association with BD in our previously reported case-control sample set (6), were also associated with BD phenotypes in this family design. No statistically significant associations were detected between any of the SNPs and BD (p > 0.05) when analyzed individually. However, when BD families were subgrouped based on proband BD subtype, statistically significant overtransmission of the G allele of the rs1556314 SNP was found in those families with BD-I probands [transmitted/not transmitted (T/NT) = 127/94, p = 0.018, ppermutation = 0.052; see Table 2]. An apparent undertransmission of the G allele of SNP rs1556314 was observed in families with BD-II probands, but did not reach statistical significance (T/NT = 21/33, p = 0.097; see Table 2).
|SNPs or haplotypes||Frequency||T/NT||χ2||p-values|
|rs1556314 (allele G in exon 11)||0.22||148/127||1.87||0.172|
|rs1785437 (allele C in intron 16)||0.29||185/181||0.06||0.812|
|rs1618355 (allele C in intron 18)||0.25||158/166||0.20||0.653|
|rs933151 (allele C in intron 20)||0.26||165/168||0.04||0.846|
|rs749909 (allele G in intron 27)||0.24||158/154||0.06||0.803|
|rs1556314 (allele G in exon 11)||0.25||127/94||5.58||0.018a|
|rs1785437 (allele C in intron 16)||0.30||150/142||0.22||0.638|
|rs1618355 (allele C in intron 18)||0.27||129/124||0.08||0.782|
|rs933151 (allele C in intron 20)||0.27||133/132||0.01||0.935|
|rs749909 (allele G in intron 27)||0.24||124/115||0.38||0.536|
|rs1556314 (allele G in exon 11)||0.14||21/33||2.76||0.097|
|rs1785437 (allele C in intron 16)||0.24||35/39||0.23||0.629|
|rs1618355 (allele C in intron 18)||0.19||29/41||2.33||0.127|
|rs933151 (allele C in intron 20)||0.21||32/37||0.46||0.499|
|rs749909 (allele G in intron 27)||0.23||34/39||0.38||0.537|
|BD-I probands (AAO ≤ 24 years)|
|rs1556314 (allele G in exon 11)||0.18||101/73||5.04||0.024|
For the haplotype analysis, we analyzed two-, three-, four-, and five-marker haplotypes from adjacent SNPs in a sliding-window fashion using TDTPHASE. No statistically significant associations (p’s > 0.05) were found between any of these SNPs and BD, BD-I or BD-II, however.
Potential effects of AAO on the association of the examined TRPM2 variants with BD were also tested. The AAO of the 255 affected offspring in families with BD-I probands only (mean ± SD = 20.2 ± 7.52) showed significant deviation from a single Gaussian distribution (Kolomgorov-Smirnov Z-test, p = 0.001). Based on the NOCOM analysis, the likelihood ratio criterion between the models with uni- and bimodal distributions of AAO was 38.40 (df = 2, p < 0.001), while the likelihood ratio criterion between the models with bi- and trimodal distributions was 5.7 (p > 0.05). Further, the two components were normally distributed (Table 3) in BD-I families. Thus, the model that best fit the observed AAO distribution of BD-I was a mixture of Gaussian distributions, consonant with those previously reported (24, 25). The threshold best separating patients with early- and adult-onset BD was 24 years of age (Table 3). Thus defined, the 296 families were divided into early- and adult-onset subgroups of BD, and the association with BD was examined. First, AAO was analyzed as a quantitative trait using the FBAT-Wilcoxon test from the PBAT package; no significant association was detected between AAO as a quantitative variable and individual TRPM2 variants for five tested SNPs. However, using binary trait analysis with BD-I probands from BD-I families subdivided based on the empirically derived threshold into early- (≤ 24 yr) and adult-onset (>24 yr) groups, haplotype analyses revealed statistically significant associations between BD-I probands with early AAO and the C-T haplotype of rs1618355 and rs933151 in introns 18 and 20 (p = 0.032, PBAT) and with the C-T-A haplotype of rs1618355, rs933151, and rs749909 in introns 18, 20, and 27 (p = 0.045, PBAT). Overtransmission of the G allele of the exon 11 SNP was also found using TDTPHASE (T/NT = 101/73, χ2 = 5.04, p = 0.024; see Table 2) for early AAO in BD-I families. There was no significant association of the currently tested TRPM2 variants with adult-onset BD-I, nor with early AAO and adult AAO in BD as a whole. Although higher G allele frequency of the exon 11 SNP was also observed in the case-control study in BD-I patients with early AAO (27%) as compared with healthy controls (20%), the difference did not reach statistical significance (p = 0.11), likely due to the small sample size of this BD-I subgroup.
|Parameter||Early onset||Adult onset|
|No. of affected offspring (n)||198||57|
|Proportion of affected offspring||77.6%||22.4%|
|Mean of AAO ± SD (years)||16.9 ± 4.15||31.5 ± 5.29|
|p-value of Z-test||0.05||0.16|
As parent of origin and sex effects may modify risk for BD in a gene-specific manner (26–29), we also examined whether the potential TRPM2 risk variants are transmitted with a parent-of-origin effect or show gender-dependent over-representation, as the family study design permits. No statistically significant effects of parent of origin or gender were detected (data not shown; p’s > 0.05) in BD and BD-I patients, however. Finally, we also examined whether the risk alleles were over- or under-represented in BD and BD-I patients with a history of psychosis but found no statistically significant associations (p’s > 0.05)
The major finding of this study is the association of the putative functional SNP rs1556314 in exon 11 of TRPM2 with BD-I, found in both Caucasian case-control and family study designs. Confirmation that specific TRPM2 variants are associated with BD in the current family dataset is particularly important given the insensitivity of this study design to population stratification, a factor that can lead to spurious associations (30). Of note, rs1556314, which results in the amino acid Asp543Glu substitution, was previously found to be associated with BD by McQuillin and coworkers (8) in a case-control dataset. This SNP is located in a region that in deletion studies was shown to impair response to oxidative stress signals H2O2 and ADP ribose, which might be related to the effect of such deletion mutants on channel gating, channel assembly, or surface trafficking (11).
Interestingly, overtransmission of the rs1556314 G allele was observed essentially in BD-I families, whereas a trend towards undertransmission of this allele was seen in BD-II families (Table 2). A similar result was also observed in the current case-control analysis, in which the G allele frequency was significantly increased in the BD-I group (27.2%, p = 0.011) but was not different in the BD-II (20.8%) group as compared with that in healthy controls (21.0%). This result supports the possibility of differences in genetic contributions to the pathogenesis of these BD phenotypes, at least with respect to TRPM2 variants. While the distinction of BD-I and BD-II subtypes is supported by differences in phenomenology, course of illness, and response to pharmacotherapy [reviewed in (31, 32)], which underlie the basis for considering these subgroups to be relevant subphenotypes of the BD, the pathogenetic basis for this subgrouping is still obscure. Molecular genetic findings specific to BD-I or BD-II are not unique, however. For example, the GABA A receptor α5 subunit gene was found to be associated with BD-I but not with BD-II (33). In another example, DNA hypomethylation in the promoter region of peptidylpropyl isomerase E-like gene was found in BD-II, but not BD-I patients, compared with healthy controls (34). The current result should be interpreted cautiously, however, since the sample size of BD-II families was small.
Age at onset and history of psychosis also represent heritable subphenotypes which may define more homogeneous subgroups of BD in which the associations of risk genotypes/alleles are more readily revealed (24, 35, 36). The finding of a significant relationship between several risk TRPM2 haplotypes in addition to exon 11 SNP rs1556314 and early AAO in BD-I families is interesting in light of reports suggesting that AAO distinguishes genetically more homogeneous subtypes of BD which may differ in aspects of clinical phenotype, such as in the severity of illness and response to treatment (37). The resolution of AAO into the two normally distributed populations found in the current study, with mean onset and range similar to those previously reported (24, 25), further supports that AAO may be a valid clinical indicator of more homogenous subtypes of BD (24). Thus, when taken into consideration in the analyses, early AAO reveals potentially meaningful genetic association otherwise obscured by underlying clinical and, in turn, genetic heterogeneity, in the BD sample. Clearly, further studies with larger, independently recruited samples are warranted to validate the current finding of an effect of early AAO on the association of TRPM2 variants with BD-I and to determine whether it is attributable to specific biological and clinical characteristics in this BD subgroup.
The results of the current study with respect to the lack of strong association of the four tested SNPs rs1785437, rs1618355, rs933151, and rs749909 in introns 16, 18, 20, and 27 are discordant with the significant association found in our previous case-control cohort (6). Moreover, the haplotype analyses did not reveal any significant associations between TRPM2 variants rs1618355, rs933151, and rs749909 and BD, BD-I, or BD-II in the current family design, except in the group of BD-I families with early AAO. Although the specific reason(s) for the discrepancies between the two study designs with respect to this region are uncertain, several factors may have contributed to the discordant findings. First, selection of healthy controls (or unaffected members in the family study) and clinical heterogeneity of the BD patient group in both designs may be factors that underlie inconsistency in replication or confirmation for those three SNPs of the gene: phenotypic heterogeneity such as reflected by AAO, for example, is well recognized as a factor that can contribute to discordant findings between family and case-control designs (38, 39). Second, because of the challenges to recruitment that result in smaller datasets than can be obtained in case-control studies, family study designs are relatively underpowered to detect genetic effects of modest or small magnitude (40). Third, lack of control for confounding factors, selection bias, and genetic heterogeneity have also been noted to contribute to differences in findings between case-control and family study designs (38, 39). That said, the possibility that the lack of allelic and genotypic associations in intron 18, 20, and 27 SNPs in the current family dataset is a true negative result cannot be discounted.
No detectable relationship was found between history of psychosis and TRPM2 variants. While this may be consequent to limitations of the sample size, it is also possible that variants of this gene contribute primarily to the general risk of illness per se. Furthermore, no effects of parent of origin or gender were observed between BD phenotypes and TRPM2 variants, arguing against confounding effects of imprinting on the evinced associations.
While our findings concur with those of McQuillin and coworkers (8) in regard to association of the TRPM2 exon 11 SNP rs1556314 with BD in both family and case-control datasets in this study, they are at odds in regard to the risk allele: McQuillin et al. (8) reported a frequency for the G allele of 18% in BD as compared with 23% in healthy controls. This contrasts with our observation of a higher frequency of the G allele found in the case-control dataset (29.8% in BD-I versus 21.0% in healthy controls) and overtransmission of the G allele with BD in the family design. Such differences in which allele, major or minor, is associated with a disorder, recently referred to as the ‘flip-flop’ phenomenon, have been noted in other studies of complex traits (41). Although this situation may be taken to suggest the results are spurious, this is unlikely to be the case given the consistency of our findings in both case-control and family study designs. Importantly, such flip-flop differences may be explained by multilocus effects and variation in interlocus correlations, by the variant being noncausal and in LD with a true causal one, by variation in LD architecture across populations, and/or by sampling variation (41).
Our findings also contrast with those of Roche et al. (9), who reported no evidence of association of examined TRPM2 genetic variants with BD. However, the sample size reported on by the latter authors was quite small, one-third of the current family sample size, and clearly of limited power to adequately test for association. It is more difficult, however, to reconcile the discrepancy between the studies of McQuillin et al. (8) and Roche et al. (9) and the current study with those of our earlier case-control report (6) in regard to the relationship, or lack thereof, between SNPs rs1785437, rs1618355, or rs933151 and BD phenotypes. Differences in clinical subphenotype of the BD patients and healthy control selection (or unaffected family members) among the study designs cannot be discounted as contributing to the discordant results among studies of TRPM2. In this regard, neither of the former studies examined associations with BD-I and AAO, as done in our studies. Second, the haplotype block structure we reported previously (6) was consistent with the Caucasian haplotype block structure described in Applied Biosystems SNPbrowser Software (version 3.5). However, haplotype block structure has been shown to vary between as well as within populations (42). Unfortunately, it is not possible to assess block structure of the British cohorts reported by McQuillin et al. (8), and no details were given by Roche et al. (9) regarding the structure of haplotype blocks in their family study design to assess whether this could have accounted for the different observations among the three studies. Any discrepancy in findings between our family study dataset and other reports must be ascribed to possibilities other than population stratification, however, as the family design is resistant to population effects, as noted.
Among other possibilities that may explain the discordant findings are allelic heterogeneity, in which different alleles at the same locus could be responsible for increased disease risk in ethnically similar populations, and locus heterogeneity, which may exist in the TRPM2 gene in association with BD phenotype(s). As multiple loci acting in concert are thought to cause complex diseases, such as BD, a single-locus association could be confounded by effects of other loci (43). Furthermore, the association findings among the different studies may reflect correlative effects of a causal variant located in a different region of the gene, as raised above to explain the differences in the risk allele. Finally, epigenetic (gene-environmental interactions) and epistatic effects (gene-gene interactions) may also operate to confound the systematic detection of associations with these SNPs; these have not been addressed in any of the three published studies of TRPM2 genetic variants in BD.
The pathophysiological relevance of the exon 11 SNP in BD is at present unknown. The amino acid encoded, Asp543, is not conserved among species, and thus may not impair TRPM2 protein function in the absence of any changes in its messenger ribonucleic acid and protein levels. In light of the recent finding showing random monoallelic expression of a TRPM2 intron 2 SNP (rs13313851) in monoclonal B lymphoblast clones (44), further examination of whether this BD-associated exon 11 SNP displays monoallelic expression and alters levels of TRPM2 expression is warranted.
Our results should be interpreted in light of several potential limitations. The possibility of spurious association, a potential risk in many genetic association studies of complex traits, cannot be entirely dismissed. However, two features of this study, together with the positive findings of our previous case-control study (6), reduce the likelihood of our findings being falsely positive. These include the use of the family design, which is immune to population stratification, and the finding of statistically significant association not only with individual SNPs but also with their haplotype, which supports the inference that a true signal is being detected. That said, it remains to be established whether any of the tested intronic SNPs and/or unidentified variants in the region between intron 16 and 27 are likely to be causal or, alternatively, in LD with variants which are causally related to development of the BD phenotype. While intronic SNP variation can affect transcriptional regulation in some instances (45), and we have reported decreased TRPM2 transcript abundance in BLCLs showing elevated [Ca2+]B from BD-I patients (7), it is unknown whether this applies to those BD-associated intronic variants studied here. The modest association between BD and TRPM2 variants from among the five tested SNPs in the current family study and rs1556314 in the family and case-control designs may also have been influenced by factors such as the relatively high minor allele frequency and strong LD among the five tested SNPs, low population attributable fraction, and unobserved confounders, such as environmental factors shared among the BD subjects (46, 47). In addition, p-values of results for the tested SNPs evaluated in this study are nominal and not corrected for multiple testing, except for rs1556314 SNP in exon 11, for which the p-values retained significance on permutation testing. That said, this study was predicated on an a priori hypothesis of association between formerly tested TRPM2 variants based on previous positive findings, justifying less stringency and conservatism in correcting for multiple testing. Finally, the current sample size is relatively modest and limited for rigorous statistical analyses of interactive effects of multiple potential modifiers on disease association, such as the effect of parent of origin, gender, and other subphenotypes of BD. Clearly, careful replication with appropriate designs in independent sample sets is warranted as advocated for stringent confirmation of genetic associations (48), as well as further LD fine-mapping, and mutation screening in the current BD-associated regions (e.g., exon 11 and between introns 16 and 27).
In conclusion, the evidence for association of TRPM2 variants with BD found in the current family-based and case-control designs adds further support to the two independent Caucasian case-control studies strengthening the notion that genetic variation of TRPM2 influences the susceptibility to BD. In particular, overtransmission was evinced for the G allele of rs1556314 in exon 11 to BD cases in the current family study design. Further investigations of the TRPM2 gene are thus warranted to increase our understanding of its role in the pathogenesis of BD.
This study was supported in part by Canadian Institutes of Health Research grant MOP 12851 (JJW). The assistance of Bronwen Hughes, Wendy Hiscox, Hester Tims, Nicole King, and Sajid Shaikh is gratefully acknowledged. Michelle Liu offered helpful advice regarding statistical analyses.
- 15Structured Clinical Interview for DSM-IV Axis I Disorders-Patient Edition (SCID-I/P) Version 2.0. New York, N.Y.: Biometrics Research Department, New York Psychiatric Institute, 1995., , , .
- 17Structured Clinical Interview for DSM-IV Axis I Disorders - Nonpatient Edition (SCID-I/NP), Version 2.0. New York, N.Y.: Biometrics Research Department, New York Psychiatric Institute, 1995., , , .
- 20UNPHASED user guide. Technical Report 2006/5. Cambridge, UK: MRC Biostatistics Unit, 2006..