We recently demonstrated the feasibility of high-density genome-wide association studies in our neuropathologically characterized cases and controls, providing empirical support for the suggestion that the APOE
locus is unparalleled in its contribution to LOAD risk (Coon et al., 2007
). With the exception of an SNP only 14 kb pairs distal to and in linkage disequilibrium (LD) with the APOE
ε4 variant on chromosome 19, no other SNP distinguished LOAD cases from controls after Bonferroni correction for multiple comparisons (Figure S1A at http://www.tgen.org/neurogenomics/data
). For the previously noted reasons, we divided each cohort into two subgroups: allelic APOE ε4 carriers (Figure S1B) and APOE ε4 noncarriers (Figure S1C). We now report associations between a common gene and LOAD in APOE
ε4 carriers in our three cohorts; we show that the implicated gene is associated with AD neuropathology in neuronal microarray and immunohistochemical studies; and we consider a possible mechanism by which GAB2
modifies AD risk in a small-interfering RNA (siRNA) study. Finally, we deposit all of the data into the public domain for use by the community (http://www.tgen.org/neurogenomics/data
High-Density Genome-Wide Association Studies
Genome-wide genotyping was performed on each individual sample from a “neuropathological discovery cohort” of 736 brain donors, a “neuropathological replication cohort” of 311 brain donors, and an additional “clinical replication cohort” of 364 living subjects who were at least 65 years old at the time of their death or last clinical assessment and who were independently assessed for their APOE genotype. For the two neuropathological cohorts, brain tissue for DNA extraction, neuropathological diagnoses, and data were supplied by investigators from 20 of the National Institute on Aging (NIA)-sponsored Alzheimer’s Disease Centers (ADCs) (in accordance with agreements with the NIA, the ADCs, and the National Alzheimer’s Coordinating Center) and from the Netherlands Brain Bank. For the hypothesis-testing clinical replication cohort, DNA extracted from blood, clinical diagnoses, and data from subjects assessed in Rochester, MN were supplied by investigators from the Mayo Clinic.
The neuropathological discovery cohort included 446 LOAD cases (299 ε4 carriers and 147 ε4 noncarriers) and 290 controls (61 ε4 carriers and 229 ε4 noncarriers); the neuropathological replication cohort included 197 LOAD cases (113 ε4 carriers and 84 ε4 noncarriers) and 114 controls (27 ε4 carriers and 87 ε4 noncarriers); and the clinical replication cohort included 218 LOAD cases (115 ε4 carriers and 103 ε4 noncarriers) and 146 controls (29 ε4 carriers and 117 ε4 noncarriers). Brain donor cases satisfied clinical and neuropathological criteria for LOAD, and were age 73.5 ± 6.2 at death. Brain donor controls did not have significant cognitive impairment or significant neuropathological features of AD, and were age 75.8 ± 7.5 at death. Clinical cases satisfied criteria for probable AD, and were age 78.9 ± 7.8 at last clinical visit. Clinical controls did not have clinically significant cognitive impairment and were age 81.7 ± 6.6 at last clinical assessment.
We initially surveyed SNPs in the neuropathological discovery cohort to explore LOAD associations in the ε4 carrier and noncarrier subgroups. Within the discovery subgroup of APOE
ε4 carriers, 10 of the 25 SNPs with the most significant LOAD-association significance levels (contingency test p = 9 × 10−5
to 1 × 10−7
; uncorrected for multiple comparisons) were located in the GRB
-associated binding protein 2 (GAB2
) gene on chromosome 11q14.1 (). LOAD associations in six of these SNPs were then confirmed in both the neuropathological replication and clinical replication cohorts (). These ten SNPs were not significantly associated with LOAD in the APOE
ε4 noncarrier group (contingency test p = 0.08 to 0.97) (Table S1A). Combining data from all 644 APOE
ε4-carrying cases and controls, we found highly significant associations between LOAD and all ten GAB2
SNPs (contingency test p = 1.19 × 10−5
to 9.66 × 10−11
), with five of the six consistently implicated alleles surviving the highly conservative Bonferroni correction for 312,316 independent comparisons (p = 1.55 × 10−7
) (). When data from the ε4 carriers and noncarriers were analyzed together, as in our previous report (Coon et al., 2007
), the ten GAB2
SNPs were still associated with LOAD (contingency test p = 0.013 to 2.7 × 10−6
, Table S1B), but these associations did not survive Bonferroni correction.
GAB2 SNPs Implicated in the Neuropathological Discovery, Neuropathological Replication, and Clinical Replication Cohorts
The PLINK analysis toolset (http://pngu.mgh.harvard.edu/~purcell/plink/index.shtml
) was used for whole-genome analysis. Haploview v3.32 was used to determine the LD structure of the chromosome 11q14.1 region surrounding GAB2
in each of the three APOE ε4-stratified cohorts (). Three haplotype blocks are present in this region: one block upstream of GAB2
, roughly corresponding to the ALG8
locus; one 189 kb-pair block encompassing most of the GAB2
locus; and one downstream block corresponding to the NARS2
locus. These blocks were consistent with the LD structure of the HapMap CEPH populations. The GAB2
gene is completely encompassed by a single haplotype block extending from rs901104 to rs2373115 (SNPs 5–22 in ), which has three major haplotypes: an extremely common “GAB2
risk haplotype,” a common “GAB2
protective haplotype,” and a relatively uncommon GAB2
haplotype unrelated to LOAD risk in APOE
ε4 carriers (). In all three cohorts, the GAB2
CT-AAG-CAGATCAGACG haplotype was associated with higher
LOAD risk, the GAB2
TC-GCA-TGAGGTGTCTT haplotype was associated with a lower
LOAD risk, and the CT-AAG-CAGAGCA GCCG was unrelated to LOAD risk in the APOE
ε4 carriers ().
Linkage Disequilibrium Structure and Haplotype Significance Levels for the Region Encompassing GAB2
Data from the 1411 subjects (including 644 APOE ε4 carriers and 767 noncarriers) in all three cohorts were combined to characterize odds ratios (ORs) and 95% confidence intervals (CIs) for rs2373115, the most significant SNP in our screen (). In ε4 carriers, LOAD cases had a risk genotype frequency of 0.88 and controls had a frequency of 0.71. In comparison with the other ε4 carriers, those with rs2373115 genotype GG had a significantly higher risk of LOAD (OR 2.36, 95% CI 1.55–3.58) than those with genotype GT and TT. In contrast, APOE ε4 noncarriers with rs2373115 genotype GG did not have a higher LOAD risk than the other ε4 noncarriers (OR 1.01, 95% CI 0.74–1.38).
GAB2 LOAD Odds Ratios in APOE ε4 Carriers and Noncarriers
Whereas we confirmed a younger age at dementia onset in the APOE ε4 carriers than in noncarriers (age 75.5 ± 7.2 versus 77.8 ± 7.9, p = 2.4 × 10−4, two-tailed unpaired t test; unequal variance is statistical test used for all tests), there was no significant effect of rs2373115 genotype on age at dementia onset in either the ε4 carriers (t test, p = 0.32) or noncarriers (t test, p = 0.84).
Neuronal Microarray Studies
To provide converging evidence that GAB2 is biologically relevant to AD neuropathology, expression profiling using the Affymetrix Human Genome U133 Plus 2.0 array was used to characterize and compare GAB2 expression in laser-capture microdissected non-tangle-bearing neurons of cases and controls in six brain regions differentially affected by AD. LOAD cases had significantly greater neuronal GAB2 expression in the posterior cingulate cortex (9 cases, 13 controls, 4.50-fold change, t test, p = 0.00039) and hippocampus (9 cases, 13 controls, 2.94-fold change, t test, p = 0.00085), and no significant expression differences in the entorhinal cortex (10 cases, 13 controls, 1.20-fold change, t test, p = 0.46), middle temporal gyrus (13 cases, 12 controls, 1.44-fold change, t test, p = 0.14), superior frontal gyrus (22 cases, 11 controls, 1.25-fold change, t test, p = 0.47), or primary visual cortex (17 cases, 12 controls, 1.53-fold change, t test, p = 0.14).
The hippocampus is known to be especially vulnerable to AD-related neurofibrillary tangles (Braak and Braak, 1991
), neuronal loss, and brain atrophy (Bobinski et al., 2000
). It is preferentially involved in AD-related memory impairment (Jack et al., 1999
) and is associated with the highest cerebral Gab2 expression in the rodent brain (Lein et al., 2007
). The posterior cingulate cortex is known to be preferentially vulnerable to AD-related hypometabolic abnormalities and fibrillar amyloid deposition, and is also involved in AD-related memory impairment (Reiman et al., 1996
; Johnson et al., 2006
; Buckner et al., 2005
; Mintun et al., 2006
). While the entorhinal cortex and temporal and prefrontal regions are also affected by AD neuropathology, the visual cortex is relatively spared. Using a repeated measures analysis of variance to analyze neuronal GAB2
gene expression data from the same eight LOAD cases and ten controls, there was a significant group-by-region interaction (p = 0.011), with LOADrelated increases in neuronal GAB2
gene expression that were greater in the posterior cingulate cortex and hippocampus than in the visual cortex.
Tau Phosphorylation siRNA Study
In addition to its other properties, GAB2
is the principal activator of the phosphatidylinositol 3-kinase (PI3K) signaling pathway (Pratt et al., 2001
). PI3K activates Akt, which in turn promotes glycogen synthase kinase-3 (Gsk3) phosphorylation/inactivation. This mechanism suppresses Gsk3-dependent phosphorylation of tau at AD-related hyperphorylated tau residues, the principal component of neurofibrillary tangles, and prevents apoptosis of confluent cells (Baki et al., 2004
; Kang et al., 2005
). Based on these findings, we hypothesized that Gab2 might function to protect cells from neuronal tangle formation and cell death and that a loss-of-function GAB2
haplotype would diminish such protection. We thus postulated that interference with GAB2
expression using siRNA treatment would increase tau phosphorylation at the serine-262 residue known to be hyperphosphorylated in AD. As shown in , GAB2
siRNA treatment was associated with a 1.70-fold increase in serine-262 phosphorylated tau. This increase was not attributable to a concomitant increase in total tau levels. Additional siRNA and protein validation studies are now being performed to determine the extent to which GAB2
affects tau phosphorylation at additional relevant epitopes.
siRNA Knockdown of GAB2 Increases Tau Phosphorylation
Gab2 immunohistochemistry was assessed in hippocampus and posterior cingulate cortex in LOAD cases. In hippocampus, Gab2 immunoreactivity was observed in structures with the morphology of dystrophic neurites or neuropil threads, neurons, and corpora amylacea. The putative neurons were almost entirely dystrophic in appearance () or had cytoplasmic inclusions resembling neurofibrillary tangles (). Dystrophic neurons and neurites () and neurofibrillary tangle-bearing cells () were also revealed by the Gab2 antibody in posterior cingulate. Here, however, many relatively normal neurons were observed as well, with long stretches of immunoreactive apical dendrites ascending through the cortical layers ().
Gab2 Colocalizes with Dystrophic Neurons in the AD Brain
In order to characterize and confirm associations between the GAB2 gene and LOAD risk in APOE ε4 carriers, our studies capitalized on the genome-wide survey of more than 300,000 SNPs, two clinically characterized and neuropathologically verified cohorts of AD cases and controls, a third cohort of clinically well characterized subjects, and stratification of the samples with respect to carriers and noncarriers of a major LOAD susceptibility gene, APOE. Six SNPs that are part of a common haplotype block encompassing the entire GAB2 gene were implicated in three independent cohorts. Maximal significance of the association was at SNP rs2373115 (p = 9 × 10−11) with an odds ratio of 4.06 (CI 2.81–14.69). An odds ratio of 24.64 (CI 7.44–116.79) for overall genetic risk is achieved when both ε4 and the GAB2 rs2373115 risk alleles are present. Data from a microarray study of laser-capture microdissected neurons in LOAD cases and controls, immunohistochemistry, and an siRNA study provided converging evidence for the relevance of GAB2 to the neuropathology of LOAD, raising testable hypotheses about the mechanisms by which GAB2 could modify LOAD risk in ε4 carriers and provide targets at which to aim new treatments.
Although only one genotyping platform was used in all three cohorts, our findings are unlikely to be attributable to any platform-related bias in genotyping calls because the observed association was not limited to a single SNP but was related to a large haplotype block in agreement with the LD structure of the HapMap CEPH population. Furthermore, all six of the implicated SNPs had high-quality SNiPer-HD scores (greater than 0.45), indicating that the data for each SNP clustered well (see Figure S2 for cluster diagrams).
Individual genotype data for all samples across >300,000 high-quality SNiPer-HD calls (see Experimental Procedures for QC metrics) is made available to the community as a .ped file at http://www.tgen.org/neurogenomics/data
. This genome-wide scan data will enable replication of putative common LOAD risk alleles, and also enable further discovery of both independent and combinatorial genetic associations.
haplotype block spans 189 kb and includes at least 614 known SNPs. Four of the six hundred and fourteen known SNPs in this locus are nonsynonymous coding SNPs, which are generally considered to be the best candidates for functional variation. However, all four of these SNPs are reported to have minor allele frequencies of 0.0% in the CEPH population (The International HapMap Consortium, 2005
), and therefore are not candidates for the common functional variant on the GAB2
is a scaffolding protein involved in multiple signaling pathways, which could affect AD-related tau, amyloid, metabolic, or other aspects of AD pathology and cell survival in different ways (Koncz et al., 2002
; Gu et al., 2001
; Zompi, Gu and Colucci, 2004
), and it has been found to be coexpressed with other putative AD-related genes (Li et al., 2004a
). Discovery of this LOAD susceptibility gene, if further replicated, provides new opportunities to investigate LOAD pathogenesis, predisposition, treatment, and prevention. Genome-wide studies using even higher density platforms and compound genetic analyses in sufficiently large samples of well-characterized cases and controls promise to play increasingly important roles in the scientific understanding, evaluation, treatment, and prevention of AD and other common and genetically complex disorders. In the interim, public access to the raw genotyping data from our series will provide valuable information to assess the contribution of other putative risk loci to this devastating disease.