Introduction

It is well known that the development of Alzheimer’s disease (AD) is consequence of complex interactions between multiple genetic and environmental factors. To date, only the ɛ4 allele of the apolipoprotein E gene (APOE) is universally recognized as a genetic risk factor for late-onset AD (LOAD) in a variety of populations,1, 2, 3 but not in elderly Nigerians.4 As the presence of risk genes other than APOE is speculated,5 many studies have been performed to identify them.

Genome-wide association studies (GWAS) involving high-density single-nucleotide polymorphism (SNP) genotyping technologies have led to great success in the identification of risk genes for various common diseases.6, 7 With regard to LOAD, the GRB2-associated binding protein 2 gene (GAB2) on chromosome 11q was recently identified in Caucasians through GWAS: 10 SNPs of this gene have been shown to be associated with LOAD in APOE-ɛ4 carriers.8 It is noteworthy that the most significant SNP, rs2373115, exhibits an odds ratio (OR) of 4.1 (95% confidence intervals (Cis), 2.8–14.7), which is almost equal to the strong risk effect exerted by the APOE-ɛ4 allele (ɛ3 vs ɛ4, OR=3.2–4.1).2 Furthermore, the following findings with relation to AD neuropathology have been made:8 in LOAD brains GAB2 is detected in highly dystrophic neurons, including neurofibrillary tangle (NFT)-bearing neurons, and interference with GAB2 expression increases TAU phosphorylation, which leads to NFT formation. With this genetic and biological evidence, GAB2 is considered to be a promising candidate for LOAD, although a recent replication study revealed a lack of association of this gene with LOAD in Caucasians.9 Therefore, we here assessed whether or not the genetic association of GAB2 with LOAD can be reproduced in Japanese.

Subjects and methods

Subjects

Blood samples were collected by the Japanese Genetic Study Consortium for AD (JGSCAD): the members are listed in our recent publications.10, 11 The LOAD patients were clinically validated, and satisfied the criteria of the National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer's Disease and Related Disorders Association for a diagnosis of probable AD.12 Non-demented controls living in an unassisted manner in the local community were recruited from among elderly subjects. The Mini-mental State Examination (MMSE) and Clinical Dementia Rating and/or the Function Assessment Staging were used to assess severity of the cognitive impairment. Basic information on the sample sets used is presented in Table 1. The total sample size is 1656 LOAD patients (female, 71.6%) and 1656 controls (female, 58.7%), which is large enough to detect risk alleles assuming OR>1.3 (range of risk allele frequency=0.1–0.9, α=0.05, power=80%). This subject group is referred to as overall sample set All in this study (Table 1). A large proportion (79.4%) of the subjects are the same as in our previous overall sample set.10, 11 To construct two sub-sample sets, the All set was stratified as to the APOE-ɛ4 carrier status: Negative-ɛ4 (LOAD, 790; control, 1378) and Positive-ɛ4 (LOAD, 866; control, 278) (Table 1).

Table 1 Information on sample sets
Full size table

This study was approved by the Institutional Review Board of Niigata University, and by all participating institutes. Informed consent was obtained from all controls and appropriate proxies for patients, and all subjects were anonymously subjected to SNP genotyping.

Genotyping

Genomic DNA preparation and genotyping were described previously.10. We did not genotype additional SNPs to the 10 ones reported by Reiman et al,8 as strong LD (∣D′∣>0.8) was observed across the GAB2 (see Supplementary Figure 1).

Statistical analysis

We carried out a Hardy–Weinberg equilibrium (HWE) test based on an exact test, single SNP and haplotype-based case–control studies, haplotype inference, and computation of LD measures (D′). As an estimate of the relative risk of disease, OR with 95% CIs of each marker or haplotype was calculated from a 2 × 2 contingency table. For all statistical analyses mentioned above, we used SNPAlyze® software version 6.0.1 (DYNACOM): the analytical methods were described in detail elsewhere.11 For evaluation of the LD block structure in and around GAB2, Haploview software version 3.32 was used. We considered P<0.05 statistically significant.

Results

To determine whether the GAB2 association can be replicated in Japanese or not, we analyzed the 10 SNPs using a total of 3312 clinical subjects for genotyping (see Supplementary Figure 2). These SNPs are encompassed by GAB2 (Table 2), which consists of 10 exons and spans about 202.4 kb on chromosome 11. To examine population differences in the allele frequencies of the SNPs, we first assessed the HapMap genotype data (http://www.hapmap.org/index.html) for four populations: Japanese in Tokyo (JPT), US Utah residents with northern and western European ancestry (CEU), Han Chinese in Beijing (CHB) and Yoruba in Ibadan, Nigeria. These 10 SNPs for JPT exhibited similar allelic frequencies to CHB, but not to CEU: for example, the frequencies for allele G of SNP rs2373115 were 0.47 for JPT and 0.89 for CEU (see Supplementary Figure 3). HWE exact tests were performed to detect genotyping errors. SNP rs7101429 slightly deviated from the HWE in the All (P=0.0497) and Negative-ɛ4 (P=0.0416) sample sets in LOAD. Remaining nine SNPs were in HWE (P≥0.05) (Table 2).

Table 2 SNP information on 10 GAB2 SNPs
Full size table

A single SNP case–control study (χ2 test) was then carried out. We did not observe any significant association of the SNPs with LOAD in not only the All set but also the two subsets (Negative-ɛ4 and Positive-ɛ4) (Table 3). Multiple logistic regression analysis, with adjustment for the carrier status of the APOE-ɛ4 allele, age and gender as covariates, did not reveal any significant evidence of association (data not shown).

Table 3 Genotypic and allelic associations
Full size table

Pairwise LD measures, D′, of the SNPs are given in Supplementary Table 1. We found a strong correlation (∣D′∣>0.93) between the 10 SNPs in each of the three sample sets. No difference in the LD block structure was observed between LOAD and control subjects. Using the HapMap genotype data for JPT and CEU, we further performed in silico LD mapping of a genomic region spanning about 500 kb. It was found that GAB2 was completely encompassed by a highly structured single LD block in both JPT and CEU (see Supplementary Figure 1). However, there was an evident difference in the LD block boundary in the 5′ region of GAB2: in JPT, we observed a definitive break point in the block, but not in CEU (see Supplementary Figure 1).

In the LD block, including the whole GAB2, three common haplotypes (frequency >1%), H1, H2 and H3, were inferred in all sample sets (see Supplementary Table 2). Haplotype H2 consisted of all major alleles of the 10 SNPs. In every sample set, no haplotypes exhibited significant differences between LOAD and controls (see Supplementary Table 2).

Discussion

Recently, it was shown that GAB2, encoding a scaffolding adaptor protein involved in several signal-transduction pathways, is associated with LOAD in Caucasians.8 At SNP rs2373115 located within this gene, a noticeable significance in allelic association (Pallele=9.7 × 10−11) has been observed.8 Interestingly, the disease risk of this gene is increased by the APOE-ɛ4 allele: maximum OR of 24.6 (95% CIs, 7.4–116.8) was computed in carriers with both the APOE-ɛ4 and GAB2 SNP rs2373115 risk (G) alleles,8 suggesting a genetic interaction between these two genes. On the basis of these findings, we attempted here to replicate the genetic association of GAB2 with LOAD in Japanese. In Reiman et al's study,8 neuropathologically well-characterized brains of Caucasians were largely used (LOAD, 643; control, 404), whereas we utilized only clinically confirmed subjects (LOAD, 1656; control, 1656). However, no evidence of association of this gene was obtained in the All, Negative-ɛ4 and Positive-ɛ4 sets (Table 3). GAB2 may not be a disease susceptibility gene for LOAD in Japanese.

As a possible explanation for the discrepancy between our results and the initial study,8 we consider an ethnic difference (Japanese vs Caucasian), genotyping technology (TaqMan® vs GeneChip® genotyping) and subject selection (clinically vs neuropathologically verified subjects) described above. With regard to the ethnic difference, Wright's FST statistic has been proposed for clarifying the level of between-population differentiation.13 FST is 0.145 (estimated from 3845 SNPs) among Asian (Japanese and Chinese), African-American and European-American, and 0.013 (estimated from 8801 SNPs) between Japanese and Chinese.14 Across the 10 GAB2 SNPs, we calculated FST using HapMap genotype data of JPT, CHB and CEU. The mean FST of these SNPs was 0.012 (standard deviation (SD), 0.006; range, 0.000–0.025) between JPT and CHB, and 0.219 (SD, 0.077; range, 0.164–0.435) between JPT and CEU. These data indicate that a higher level of genetic differentiation exists between JPT and CEU for GAB2. Recently, Chapuis et al9 could not replicate the initial finding8 even in European-Caucasian subjects (N>3000), suggesting that GAB2 is at best a minor disease susceptibility gene for LOAD. A meta-analysis is needed to confirm the association of GAB2 with LOAD.