Meta-analysis Confirms CR1, CLU, and PICALM as Alzheimer Disease Risk Loci and Reveals Interactions With APOE Genotypes

Alzheimer disease (AD) is the most common form of dementia, affecting 5% of the population older than 65 years and 30% to 50% older than 80 years. Substantial progress was made identifying genes for rare forms of early-onset AD^1-4 and this early success significantly contributed to biologic study of AD mechanisms and, more recently, multiple drug discovery approaches. Late-onset AD, the common form of the disease, has been more difficult to solve, with apolipoprotein E (APOE) being the only confirmed susceptibility locus.⁵ The combination of high-density genotyping methods, large well-characterized AD and control populations, and statistical methods to evaluate population stratification now provide the tools to identify additional genes contributing to AD risk.

Recently, 2 genome-wide association studies (GWAS) reported evidence that variations in CLU (encoding clusterin), PICALM (encoding the phosphatidylinositol binding clathrin assembly protein), and CR1 (encoding complement component [3b/4b] receptor 1) confer genetic risk for AD.^6,7 Evidence for these 3 loci reached genome-wide significance in samples consisting of 5964 cases and 10 188 controls (PICALM and CLU) and 5887 cases and 8508 controls (CRI and CLU). To analyze the role of these genes in AD risk, the Alzheimer's Disease Genetics Consortium (ADGC) performed a meta-analysis using GWAS data for 15 239 subjects from 9 Northern European white cohorts and 5 cohorts that included African American, Israeli-Arab, and Caribbean Hispanic individuals (Table 1). Genotypes for CR1, CLU, and PICALM were analyzed for association with AD using cohorts that are completely independent of those originally used to identify these 3 loci as AD susceptibility factors. The controls used are all elderly (>60 years). We also examined the interaction of APOE with CR1, CLU, and PICALM on AD risk.

All cohorts are described in more detail in the eAppendix and eTables 1, 2, and 3. The National Institute on Aging (NIA) Alzheimer's Disease Center (ADC) subjects were ascertained, evaluated, and sampled by the clinical and neuropathology cores of the 29 NIA-funded ADCs (Table 1). Subject data collection is coordinated by the National Alzheimer's Coordinating Center. DNA from these samples for genotyping was prepared by the National Cell Repository for Alzheimer's Disease. The Alzheimer's Disease Neuroimaging Initiative (ADNI) subjects are AD cases and controls ascertained for neuroimaging, biomarker, and genetic studies. Data used herein were generated as previously described⁸ and obtained from the ADNI database (http://www.loni.ucla.edu/ADNI/). The Collaborative Aging and Memory Project subjects are from the Amish communities of central Ohio and northern Indiana.^9,10 The Columbia University subjects are a Hispanic cohort described in detail elsewhere.¹¹ The Framingham Heart Study is a single-site, community-based, ongoing cohort study described elsewhere.^12-14 Phenotype and GWAS data were from the dbGaP Web site (http://www.ncbi.nlm.nih.gov/gap). The Johns Hopkins University subjects are from the Genetic and Environmental Risk Factors for Alzheimer's Disease Among African Americans (GenerAAtions) Study identified through the electronic claims database of the Henry Ford Health System. The Multi-Institutional Research on Alzheimer’s Genetic Epidemiology (MIRAGE) Study is a family-based genetic epidemiological study of AD in which AD cases and unaffected sibling controls were enrolled at 17 clinical centers in the United States, Canada, Germany, and Greece.¹⁵ The NIA Late-Onset Alzheimer's Disease (NIA-LOAD) Family Study (E. M. Wijsman, PhD, Y Choi, MS, J. H. Rothstein, MS, et al, unpublished data, June 2010) cohort are families with 2 or more affected siblings with late-onset AD and unrelated control subjects without dementia similar in age and ethnic background. One case per family was selected and controls were determined to be cognitively normal after an in-person neurological examination and were not related to a study participant. The Oregon Health and Science University cohort were recruited from aging research cohorts at 10 NIA-funded ADCs and do not overlap with other ADGC samples. The Translational Genomics Research Institute data set is a publicly available sample of AD cases and controls (http://www.tgen.org/research/index.cfm?pageid=1065).¹⁶ The University of Miami/Vanderbilt University/Mount Sinai School of Medicine cohort were new and previously published^17-20 subjects ascertained at the University of Miami, Vanderbilt University, and Mount Sinai School of Medicine. The Wadi Ara data set are from a highly consanguineous, genetically isolated Arab community in northern Israel.^21-24

The cohorts used were genotyped either on Illumina (San Diego, California) or Affymetrix (Santa Clara, California) single-nucleotide polymorphism (SNP) arrays (Table 2). We selected 17 SNPs from CR1, CLU, and PICALM that were recently reported to be significantly associated with AD in 2 large GWAS^6,7 (Table 3). Additional genotypes were obtained using TaqMan assays (Applied Biosystems, Foster City, California) including genotypes for rs7982. Genotyping for the APOE ε2/ε3/ε4 alleles was performed as described in the eAppendix and eTables 1, 2, and 3.

The analysis included only individuals with a censoring age of 60 years or older. The age used for cases was that most closely approximating the age at disease onset. For some cohorts, age at onset was ascertained while for others, only age at ascertainment was available. For some autopsied subjects, only age at death was available and was used as the censoring age. For all studies, the age used for controls was the age at last examination or death (eAppendix and eTables 1, 2, and 3).

We imputed genotypes for all SNPs within 10 kilobases of the 3 genes using the Markov chain haplotyping software²⁵ to obtain a common set of SNPs across all data sets. We imputed SNPs from both HapMap releases 2 and 3 (International HapMap Project, http://snp.cshl.org/) and retained those with pairwise linkage disequilibrium (r² >0.50) for further analysis (see eAppendix and eTables 1, 2, and 3 for more detail and for data cleaning protocols).

To determine if population substructure existed in the different data sets, we used 30 000 to 100 000 SNPs with minor allele frequency more than 0.25 and minimal between-SNP linkage disequilibrium (r² <0.20) sampled at random from the autosomes and analyzed with the STRUCTURE software package.^26,27 To account for population substructure in association analyses, EIGENSTRAT²⁸ was used on each cohort to generate loadings from principal components analysis on the sampled SNPs (eAppendix and eTables 1, 2, and 3).

Genotyped and imputed SNPs were tested for association with AD using a logistic generalized linear model in case-control data sets and a logistic generalized estimating equation in family-based data sets. Genotyped SNPs were coded as 0, 1, or 2 according to the number of minor alleles under the additive genetic model, whereas APOE was coded as 0 or 1 according to the presence or absence of the ε4 allele. For imputed SNPs, a quantitative estimate between 0 and 2 for the dose of the minor allele was used to incorporate the uncertainty of the imputation estimates. Regression models for each SNP without covariates were evaluated for comparison with results from the original reports^6,7 Additional models containing all permutations of covariates for age, sex, and APOE ε4 status were also tested. Formal tests of interaction between the SNPs and APOE were assessed by including the main effects and an interaction term. Regression models were evaluated using the R package.²⁹ Heterogeneity among odds ratios (ORs) was assessed using the Cochran Q, which was calculated as the weighted sum of squared differences between individual study effects and the pooled effect across studies, with the weights being those used in the pooling method. Q was distributed as a χ² with k (number of studies) minus 1 df. The I² statistic^30,31 describes the percentage of variation across studies that is due to heterogeneity rather than chance and was calculated as follows: I² = 100% × (Q − df)/ Q. I² is an intuitive and simple expression of the inconsistency of studies' results. Unlike Q, it does not inherently depend on the number of studies considered. The SNP association results obtained from individual data sets were combined by meta-analysis using the inverse variance method implemented in the software package METAL (http://www.sph.umich.edu/csg/abecasis/Metal/index.html). An additive model was assumed and the association results across data sets were combined by summing the regression coefficients weighted by the inverse variance of the coefficients. The meta-analysis P value of the association was estimated by the summarized test statistic.

To analyze the role of CR1, CLU, and PICALM in AD risk, the ADGC performed a meta-analysis using phenotypes and GWAS data from 12 different cohorts (Table 1). The ADGC is a collaborative network in the United States that includes the 29 NIA-funded ADCs and numerous AD genetics investigators who are working to identify genes responsible for AD. Of 7070 cases with AD examined, 3055 had autopsy documentation of AD. Of the 8169 cognitively normal elderly subjects (>60 years) examined, 1155 had autopsies documenting absence of significant AD neuropathology. The cohorts used included unrelated white cases and controls from the following sources: the NIA-funded ADCs, ADNI,^8,32 University of Miami/Vanderbilt University/Mount Sinai School of Medicine^17-19 (A.C.N, G.W.B, and E.R.M, unpublished data, November 2009), Translational Genomics Research Institute,¹⁶ and Oregon Health and Science University.³³ White cases and controls from the following family-based studies were also included: the MIRAGE Study,¹⁵ Framingham Heart Study,^13,14,34 NIA-LOAD Family Study, and Collaborative Aging and Memory Project.^9,10 Populations not of white descent included African American subjects from several ADCs, a community-based (Detroit, Michigan) study of AD, and the MIRAGE Study¹⁵; Caribbean Hispanic individuals from Manhattan, New York, the Dominican Republic, and Puerto Rico; and members of a genetically isolated Arab community in Wadi Ara, Israel.^21-24

In each data set, we evaluated the association of AD with SNPs in or near CR1, CLU, and PICALM that were genotyped on various platforms or imputed (Table 2). Results were combined across data sets using a meta-analysis approach (Table 3). We analyzed each racial/ethnic group separately. In white individuals, the largest group (5935 cases, 7034 controls), we found significant evidence of association with multiple SNPs at each locus. In the unadjusted analyses, we obtained an OR of 0.91 with a 95% confidence interval (CI) of 0.85 to 0.96 for CLU SNP rs11136000, which is comparable with the effect size reported previously for the same SNP (ORs, 0.86⁷ and 0.91⁶). For the CR1 SNP rs3818361, we obtained an OR of 1.14 (95% CI, 1.07-1.22) compared with the previous report of 1.19.⁷PICALM SNP rs3851179 had an OR of 0.89 (95% CI, 0.84-0.94) compared with 0.86 observed previously.⁶ None of the SNPs were significantly associated with AD in any of the other ethnic groups analyzed together or separately, possibly because of small sizes of these groups (1135 cases and 1135 controls, eTable 1).

We also examined the influence of APOE on the associations of the 3 genes with AD, since APOE is a known AD susceptibility locus in most ethnic groups^5,35 and several APOE genotypes have been reported to modify disease expression in persons with rare mutations in presenilin 1 (PSEN1),³⁶ presenilin 2 (PSEN2),³⁷ and the amyloid precursor protein (APP)^37,38 genes. For the 13 cohorts where APOE genotype data were available, presence of 1 or more APOE ε4 alleles was significantly associated with AD (ORs, 1.80-9.05) in all groups except the Amish and Israeli-Arab individuals (Table 4). We next reevaluated the association of AD with the CR1, CLU, and PICALM SNPs in the white cohorts adjusting for age, sex, and the presence of at least 1 APOE ε4 allele and found greatly reduced evidence for association with PICALM after adjustment (Table 3 and eTable 2), an effect that is attributable primarily to APOE (eTable 2). To explore this effect further, we analyzed the association of CR1, CLU, and PICALM SNPs with AD in subgroups stratified by the presence or absence of the APOE ε4 allele. This analysis revealed that the association with CLU was evident only among subjects without the APOE ε4 allele, whereas the association with PICALM was evident only among subjects with the APOE ε4 allele (Table 5). Analysis of models containing terms for the main effects of each SNP and the presence or absence of the APOE ε4 allele and an interaction term showed significant evidence of interaction for the presence or absence of the APOE ε4 allele and 7 of the 9 PICALM SNPs, with indications of a synergistic effect of these 2 genes on AD risk (Table 5 and eTable 3). Interactions of CR1 and CLU SNPs with the presence or absence of the APOE ε4 allele were not statistically significant.

Using a large multicenter data set of AD cases and controls, we confirm that CR1, CLU, and PICALM are AD susceptibility loci in European ancestry populations. The ORs we get for each are similar to those obtained in the original discovery cohort, suggesting that these estimates of risk are quite accurate for the white AD population, reflecting in part the large size of the cohorts used.^6,7 Clearly, a large data set is required to replicate these small-effect loci. We were unable to replicate the association of these 3 genes in the African-American, Arab, and Hispanic populations. However, further analysis is merited in these racial/ethnic groups using larger cohorts.

While this article was being prepared for publication, a GWAS on AD was reported by Seshadri et al.³⁹ There was some overlap in that study and ours in that the Translational Genomics Research Institute and Framingham Heart Study cohorts are used in both studies. However, whereas Seshadri et al used only prospectively diagnosed AD cases (n = 52) and unrelated controls (n = 2091) from the Framingham Heart Study, we included these subjects as well as prevalent and newly diagnosed cases and related controls, yielding a total sample of 197 AD cases and 2392 controls. Both studies independently confirm that CLU and PICALM are AD susceptibility genes. A primary difference between the 2 studies is that herein we confirm CR1 as an AD locus while Seshadri et al³⁹ obtained only nominal support for CR1.

The cohorts used herein have several features worth mentioning in the context of GWAS for AD. First, the cohorts have a large number of autopsies in the cases (n = 3055). Because the gold standard for diagnosis is neuropathologic confirmation of AD pathology, using autopsied cases reduces etiologic heterogeneity. Second, the controls used herein were elderly, of comparable age to case onset ages, and cognitively normal. Since these subjects lived to a comparable age to cases without developing AD, the case-control contrast should be more robust than if young controls were used. In addition, cases and controls will be comparably censored at other non-AD loci responsible for common diseases of elderly individuals that are unrelated to AD. Third, the cohorts used herein were not involved in the initial discovery of CLU, CR1, and PICALM and thus represent a completely independent replication data set. This is critical in terms of evaluating evidence that these genes are truly AD risk loci. The ideal controls for an AD GWAS would be subjects who were cognitively normal at death, had autopsy documentation that plaque load and tangle distribution did not reach criteria for AD pathology, and were elderly. In autopsy series of older cognitively normal subjects, most have some neurofibrillary tangles and some nonneuritic, and possibly spare neuritic, amyloid deposits but do not reach the accepted threshold for AD, although about a third of these normal subjects do meet neuropathologic criteria for AD.^40-43 In autopsy series of subjects with mild cognitive impairment, up to two-thirds of subjects have AD-level neuropathology.⁴⁴ These findings give rise to the hypothesis that amyloid deposition and tangle formation begin before cognitive decline becomes detectable, an idea strengthened by recent biomarker and amyloid imaging work.⁴⁵ Thus, in persons without dementia, a fraction, mostly those with mild cognitive impairment, will develop AD within a few years and this conversion rate increases with the age of the population, decreasing the contrast between cases and controls and reducing power. To minimize the potential confounding effect of mild cognitive impairment, we excluded them from these analyses and emphasized 1155 controls with autopsy information (Table 1).

When we examined the interaction of the CR1, CLU, and PICALM and APOE genotypes, we detected synergy between APOE and PICALM but not with CR1 or CLU. Our results show that the PICALM association is predominantly in subjects carrying the APOE ε4 allele. Consistent with conclusions from previous studies showing interaction of APOE with PSEN1,³⁶PSEN2,³⁷ and APP,^37,38 our results suggest that the APOE and PICALM gene products participate in a common pathogenic pathway leading to AD. Since PSEN1, PSEN2, and APP are all involved in β-amyloid production, PICALM may also participate in this process, though a more indirect involvement cannot be ruled out and the biology of these interactions remains to be determined. We did not detect an interaction of APOE with CR1 or CLU, though this could be because of sample size, which was not large enough to detect very weak interactions. Also, since the APOE effect on AD risk is much stronger in young case populations,³⁵ the age structure of our study and of others may not be optimal for detecting these interactions.

Our study and those from other consortia^6,7 (E. M. Wijsman, PhD, Y. Choi, MS, J. H. Rothstein, MS, et al, unpublished data, June 2010) show that AD susceptibility loci can be identified by GWAS. Initial AD GWAS had samples sizes that, in comparison with those from the large consortia, were modest and inadequately powered to detect the small effect loci replicated herein.^18,46-51 As sample sizes increase, as in other complex disorders, we expect additional loci to be identified.

Correspondence: Gerard D. Schellenberg, PhD, Department of Pathology and Laboratory Medicine, University of Pennsylvania School of Medicine, Room 609B, Stellar-Chance Laboratories, 422 Curie Blvd, Philadelphia, PA 19104-6100 (gerardsc@mail.med.upenn.edu).

Accepted for Publication: June 24, 2010.

Published Online: August 9, 2010. doi:10.1001/archneurol.2010.201

Author Contributions: All members of the ADGC are authors of this article. Data used in preparation of this article were obtained from the ADNI database (www.loni.ucla.edu/ADNI). As such, Dr Saykin, an investigator within the ADNI, contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this article. A complete listing of the ADNI investigators is available at http://www.loni.ucla.edu/ADNI/Collaboration/ADNI_Manuscript_Citations.pdf. Study concept and design: Jun, Buxbaum, Friedland, Raskind, Shelanski, Bennett, Martin, Montine, Goate, Blacker, Hakonarson, Kukull, Haines, Mayeux, Farrer, Pericak-Vance, and Schellenberg. Acquisition of data: Jun, Wang, Buxbaum, Ertekin-Taner, Fallin, Friedland, Inzelberg, Kramer, Rogaeva, St. George-Hyslop, Arnold, Baldwin, Barber, Beach, Bigio, Bird, Boxer, Burke, Cairns, Carroll, Chui, Clark, Cotman, DeCarli, Diaz-Arrastia, Dick, Dickson, Ellis, Fallon, Farlow, Ferris, Frosch, Galasko, Gearing, Ghetti, Gilman, Glass, Graff-Radford, Green, Growdon, Hamilton, Harrell, Head, Honig, Hulette, Hyman, Jicha, Jin, Johnson, Karlawish, Karydas, Kaye, Kim, Koo, Kowall, Lah, Levey, Lieberman, Lopez, Mack, Markesbery, Marson, Martiniuk, Masliah, McKee, Mesulam, J. W. Miller, B. L. Miller, C. A. Miller, Parisi, Perl, Peskind, Poon, Quinn, Reisberg, Ringman, Roberson, Rosenberg, Sano, J. A. Schneider, L. S. Schneider, Seeley, Smith, Spina, Stern, Tanzi, Troncoso, Van Deerlin, Vinters, Vonsattel, Weintraub, Woltjer, Younkin, Cantwell, Dombroski, Saykin, Reiman, Bennett, Morris, Beekly, Kukull, Foroud, Haines, Mayeux, Farrer, Pericak-Vance, and Schellenberg. Analysis and interpretation of data: Jun, Naj, Beecham, Wang, Buros, Gallins, Friedland, Cummings, Geschwind, Giordani, Jicha, Markesbery, C. A. Miller, Petersen, Trojanowski, Welsh-Bohmer, Reiman, Lunetta, Martin, Blacker, Tsuang, Cupples, Hakonarson, Haines, Farrer, Pericak-Vance, and Schellenberg. Drafting of the manuscript: Jun, Naj, Beecham, Buros, Gallins, Buxbaum, St. George-Hyslop, Cummings, Hamilton, Hulette, Karydas, Martiniuk, Poon, Ringman, Rosenberg, Welsh-Bohmer, Cantwell, Reiman, Tsuang, Haines, Mayeux, Farrer, Pericak-Vance, and Schellenberg. Critical revision of the manuscript for important intellectual content: Naj, Beecham, Wang, Buxbaum, Ertekin-Taner, Fallin, Friedland, Inzelberg, Kramer, Rogaeva, St. George-Hyslop, Arnold, Baldwin, Barber, Beach, Bigio, Bird, Boxer, Burke, Cairns, Carroll, Chui, Clark, Cotman, DeCarli, Diaz-Arrastia, Dick, Dickson, Ellis, Fallon, Farlow, Ferris, Frosch, Galasko, Gearing, Geschwind, Ghetti, Gilman, Giordani, Glass, Graff-Radford, Green, Growdon, Harrell, Head, Honig, Hyman, Jicha, Jin, Johnson, Karlawish, Kaye, Kim, Koo, Kowall, Lah, Levey, Lieberman, Lopez, Mack, Markesbery, Marson, Masliah, McKee, Mesulam, J. W. Miller, B. L. Miller, C. A. Miller, Parisi, Perl, Peskind, Petersen, Quinn, Raskind, Reisberg, Roberson, Sano, J. A. Schneider, L. S. Schneider, Seeley, Shelanski, Smith, Spina, Stern, Tanzi, Trojanowski, Troncoso, Van Deerlin, Vinters, Vonsattel, Weintraub, Woltjer, Younkin, Cantwell, Dombroski, Saykin, Reiman, Bennett, Morris, Lunetta, Martin, Montine, Goate, Blacker, Beekly, Cupples, Hakonarson, Kukull, Foroud, Haines, Mayeux, Farrer, Pericak-Vance, and Schellenberg. Statistical analysis: Jun, Naj, Beecham, Wang, Buros, Gallins, Lunetta, Cupples, Haines, Farrer, and Pericak-Vance. Obtained funding: Buxbaum, Fallin, Friedland, Rogaeva, St. George-Hyslop, Beach, Chui, Hulette, Levey, Lopez, McKee, Petersen, Sano, Spina, Vonsattel, Younkin, Saykin, Bennett, Morris, Martin, Montine, Goate, Kukull, Foroud, Haines, Pericak-Vance, and Schellenberg. Administrative, technical, and material support: Naj, Wang, Buros, Ertekin-Taner, Fallin, Friedland, Inzelberg, Kramer, Arnold, Barber, Beach, Boxer, Burke, Cairns, Carroll, Chui, Cotman, Cummings, Diaz-Arrastia, Ellis, Fallon, Farlow, Ferris, Frosch, Galasko, Gearing, Ghetti, Gilman, Hamilton, Harrell, Head, Honig, Hulette, Hyman, Johnson, Karydas, Kaye, Koo, Kowall, Lah, Levey, Lieberman, Mack, Markesbery, Marson, Martiniuk, Masliah, Mesulam, J. W. Miller, B. L. Miller, C. A. Miller, Parisi, Perl, Peskind, Poon, Reisberg, Ringman, J. A. Schneider, L. S. Schneider, Seeley, Shelanski, Tanzi, Trojanowski, Van Deerlin, Vinters, Welsh-Bohmer, Woltjer, Cantwell, Saykin, Reiman, Bennett, Montine, Blacker, Beekly, Kukull, Farrer, Pericak-Vance, and Schellenberg. Study supervision: Jun, Beecham, Friedland, St. George-Hyslop, Baldwin, Diaz-Arrastia, Ferris, Glass, Growdon, Bennett, Lunetta, Hakonarson, Farrer, Pericak-Vance, and Schellenberg.

Financial Disclosure: Dr Gilman serves on safety monitoring committees for Elan, Pfizer, Janssen, and Allergan pharmaceutical companies and a steering committee for a trial of rasagiline for multiple system atrophy sponsored by Teva Pharmaceuticals. He receives reimbursement only for his time by each of these sponsors. He also consults for Longitude Capital and the Gerson Lehman Group. Dr Reiman has received research grants and contracts from the NIA, state of Arizona, Kronos Life Sciences, GlaxoSmithKline, AstraZeneca, and Avid and has provided consultation and advisory board services to AstraZeneca, Amnestix/Sygnis, Elan, Eli Lilly, and Siemens. Dr Rosenberg is editor of the Archives of Neurology and obtained an independent review and assessment of the manuscript from outside the editorial office prior to its acceptance. The ADNI is funded through generous contributions from the following: Abbott, AstraZeneca AB, Bayer Schering Pharma AG, Bristol-Myers Squibb, Eisai Global Clinical Development, Elan Corporation, Genentech, GE Healthcare, GlaxoSmithKline, Innogenetics, Johnson & Johnson, Eli Lilly and Co, Medpace Inc, Merck and Co Inc, Novartis AG, Pfizer Inc, F. Hoffman-La Roche, Schering-Plough, Synarc Inc, and Wyeth.

Funding/Support: The ADGC is funded by the US National Institutes of Health (NIH), NIA grants U01 AG032984 and RC2 AG036528 and a grant from a private foundation wishing to remain anonymous. The NIH-NIA also provides financial support to National Alzheimer's Coordinating Center (grant U01 AG016976), National Cell Repository for Alzheimer's Disease (grant U24-AG021886), and the ADCs: Banner Alzheimer's Institute (grant P30 AG019610), Boston University (grants P30 AG013846, R01 HG02213, K24 AG027841, U01 AG10483, R01 CA129769, and R01 MH080295), Columbia University (grant P50 AG008702), Duke University (grant P30 AG028377), Emory University (grant AG025688), Indiana University (grant P30 AG10133), Johns Hopkins University (grant P50 AG005146), Massachusetts General Hospital (grant P50 AG005134), Mayo Clinic (grant P50 AG016574), Mount Sinai School of Medicine (grant P50 AG005138), New York University (grants P30 AG08051, UO1 AG16976, MO1 RR00096, and UL1 RR029893), Northwestern University (grant P30 AG013854), Oregon Health and Science University (grant P30 AG008017), Rush University (grant P30 AG010161), University of Alabama at Birmingham (grant P50 AG016582 and grant UL1 RR02777 through the University of Alabama at Birmingham Center for Clinical and Translational Science), University of California, Davis (grant P30 AG010129), University of California, Irvine (grants P50 AG016573, P50 AG016574, P50 AG016575, P50 AG016576, and P50 AG016577), University of California, Los Angeles (grant P50 AG016570), University of California, San Diego (grant P50 AG005131), University of California, San Francisco (grants P50 AG023501 and P01 AG019724), University of Kentucky (grant P30 AG028383), University of Michigan (grant P50 AG008671), University of Pennsylvania (grant P30 AG010124), University of Pittsburgh (grant P50 AG005133), University of Southern California (grant P50 AG005142), University of Texas Southwestern (grant P30 AG012300), University of Washington (grant P50 AG005136), and Washington University (grants P50 AG005681 and P01 AG03991). The work completed by Boston University is also supported by Alzheimer's Association grant IIRG 08-89720 and VA New England Geriatric Research Education and Clinical Center.

This project was also made possible by the many contributions of individual study data sets, supported in part by NIH. These include the NIA-LOAD Family Study (NIH grant U24 AG026395), Columbia University study (NIH grant R37 AG015473), ADNI (grants U01 AG024904 and RC2 AG036535), Framingham Heart Study (grants N01 HC-25195, R01 NS017950, R01 AG08122, R01 AG16495, R01 AG033193, and R01 AG031287), Collaborative Aging and Memory Project (grant R01 AG019085), Johns Hopkins University (grant R01 AG020688), MIRAGE Study (grant R01 AG009029), Wadi Ara study (grant R01 AG017173), and the Multiethnic Genome-wide Association Study (Dr Farrer; grant R01 AG025259). The University of Miami/Vanderbilt University/Mount Sinai School of Medicine work was supported by grants from the NIA-NIH (AG010491, AG002219, AG005138, AG027944, AG021547, AG019757, and R01 AG 027944) and from the Alzheimer's Association (IIRG 05-14147). A subset of these participants was ascertained while Dr Pericak-Vance was a faculty member at Duke University. The study by Oregon Health and Science University was supported by NIA grants R01 AG026916, P30 AG028377, P50 AG005146, P30 AG028383, P50 AG16574, U01 AG06786, P30 AG008017, P30 AG10161, R01 AG17917, P30 AG10129, P50 AG05131, P50 AG08671, P50 AG05681, P01 AG03991, and U01 AG016976 of the NIH and by the Natural Science Foundation of China (project numbers 30730057 and 30700442). The Translational Genomics Research Institute is supported by NIH grant R01 AG031581, Kronos Life Sciences, and the state of Arizona.

The ADNI data collection and sharing for this project was funded by NIH grant U01 AG024904 (principal investigator, Michael W. Weiner, MD). The ADNI is funded by the NIA and the National Institute of Biomedical Imaging and Bioengineering, as well as nonprofit partners the Alzheimer's Association and Alzheimer's Drug Discovery Foundation, with participation from the US Food and Drug Administration. Private sector contributions to ADNI are facilitated by the Foundation for the National Institutes of Health (http://www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer's Disease Cooperative Study at the University of California, San Diego. The ADNI data are disseminated by the Laboratory of Neuro Imaging at the University of California, Los Angeles. This research was also supported by NIH grants P30 AG010129 and K01 AG030514 and the Dana Foundation.

Online-Only Material: The eAppendix and eTables are available at http://www.archneurol.com.

Additional Contributions: We thank Creighton Phelps, PhD, Marcelle Morrison-Bogorad, PhD, and Marilyn Miller, PhD, from the NIA for help in acquiring samples and data; they are ex officio members of the ADGC. Duke University acknowledges John Ervin, BA, from the Brain Bank and Kathleen Hayden, PhD, in the Clinical Core for their respective efforts in the DNA/data pulls required.

This article was corrected for errors on January 5, 2011.