Analyses of the National Institute on Aging Late-Onset Alzheimer's Disease Family Study: Title and subTitle BreakImplication of Additional Loci FREE

Although the apolipoprotein E ε4 allele (APOE ε4) (OMIM 107741) is the most consistently replicated genetic variant influencing the risk of late-onset Alzheimer disease (LOAD),¹ it explains only 20% of the attributable genetic risk.² Daw et al³ reported that there may be 4 additional genes influencing LOAD risk. Although a number of susceptibility genes have been reported (http://www.alzgene.org/), the number of genes that have been replicated across multiple studies remains small. Whereas some genes (eg, sortilin-related receptor 1 [SORL1]⁴ and angiotensin-converting enzyme [ACE]⁵ ) are strongly supported by several studies across multiple ethnic groups, other genes need further evaluation (eg, low-density lipoprotein receptor-related protein 6 [LRP6],⁶ GRB-2–associated binding protein [GAB2],⁷ and cholesterol 25-hydroxylase [CH25H]⁸ ).

In early 2007, 968 association studies of 398 candidate genes were compiled by AlzGene (http://www.alzgene.org/), but most have not been replicated. Ertekin-Taner⁹ has described several reasons for the lack of replication, but progress in identifying and confirming genetic variants related to LOAD may also be limited because of the paucity of data sets and samples available to the scientific community. In 2002, the National Institute on Aging (NIA) launched the NIA Genetics Initiative for Late-Onset Alzheimer's Disease (NIA-LOAD) to expand resources needed to identify the remaining genes contributing to the risk for LOAD. The NIA-LOAD Family Study, a major component of the initiative, has as its goals to identify and recruit families with 2 or more affected siblings with LOAD and unrelated, nondemented control subjects similar in age and ethnic background. The clinical data, DNA, genotyping results, and preliminary analyses will be made available to investigators worldwide. Herein we describe the families and the results of linkage, family-based association, and case-control analyses from a genomewide scan using approximately 6000 single-nucleotide polymorphic (SNP) markers.

Recruitment took place throughout the United States at 18 participating AD centers (ADCs), each of which had received approval by their institutional review board. A collaborative effort by each ADC, the NIA, the Alzheimer's Disease Education and Referral Center, and the Alzheimer's Association led to national media coverage, which facilitated recruitment. A toll-free number at the National Cell Repository for Alzheimer's Disease (http://ncrad.iu.edu) was made available. When qualifying families contacted the National Cell Repository, research staff referred the family to the geographically closest participating ADC for evaluation.

The recruitment criteria included a family with multiple members affected with LOAD that could provide clinical information and a biological sample for DNA extraction. The proband had to have a diagnosis of definite or probable LOAD¹⁰ with onset after 60 years of age and a full sibling with definite, probable, or possible LOAD with onset after 60 years of age. A third biologically related family member was required, who could have been a first-, second-, or third-degree relative of the affected sibling pairs and 60 years or older if unaffected or 50 years or older if diagnosed as having LOAD or mild cognitive impairment.¹¹ Unaffected persons were required to have had documented cognitive testing and clinical examination results to verify the clinical designation.

A minimal data set included demographic variables, diagnosis, age at onset, method of diagnosis, Clinical Dementia Rating Scale score,¹² and the presence of other relevant health problems. Each ADC was required to use standard research criteria for the diagnosis of LOAD.¹⁰ Participants with advanced disease or those living in a remote location who could not complete a detailed in-person evaluation contributed blood samples, and the site investigator conducted a detailed review of medical records to document the presence or absence of LOAD.

The age at onset for patients with LOAD was the age at which the family first observed memory problems, but, if this information was not available, the age at first examination was used. For controls, we used their age at the time of their examination confirming the absence of dementia. For 137 deceased family members who had undergone a postmortem brain evaluation, neuropathologic results were used to document the diagnosis. The clinical diagnosis of LOAD agreed with the autopsy diagnosis for 95% of the case patients who had both diagnoses.

Before SNP genotyping, we verified the reported family relationships using 9 microsatellite markers (7 autosome markers and 1 X and 1 Y chromosome marker). Subsequently, we used 3 sets of more than 570 SNPs by selecting every 10th SNP to evaluate relationships among family members with the Pedigree Relationship Statistical Test.^{13 - 14} Based on the results, we corrected family relationships in 40 families (detailed information is available from the authors upon request), and we excluded individuals who were not biologically related to any family (n = 3 individuals). Four families were excluded because problems with reported relationships could not be resolved adequately, and we excluded 1 family with a presenilin 1 gene (PSEN1) mutation. We then checked for inconsistencies in mendelian transmission using PedCheck¹⁵ and corrected them. We considered erroneous genotypes from these individuals as missing.

We assessed Hardy-Weinberg equilibrium using the Haploview software (http://www.broad.mit.edu/mpg/haploview/)¹⁶ and excluded SNPs that deviated from Hardy-Weinberg equilibrium with a P value of less than .001. To identify regions with high linkage disequilibrium (LD) we computed pairwise LD coefficients and created 95% confidence bounds on D′ to define SNP pairs in strong LD.¹⁷ For multipoint linkage analysis, we used 1 SNP from each haplotype block to ensure that the D′ between adjacent markers remained low; as a result, we dropped 255 SNPs that were in strong LD with adjacent SNPs.

Single-nucleotide polymorphisms were genotyped at the Center for Inherited Disease Research using a marker panel (Illumina Linkage-IVb Marker Panel; http://www.cidr.jhmi.edu). From this panel, 5954 SNP markers were originally genotyped. After eliminating SNP genotypes with uncertain calls, excess missing data, or mendelian errors, a total of 5616 SNPs were available for statistical analysis at an intermarker distance of 0.65 cM (519 kilobase [kb]); the average marker heterozygosity was 0.43. Missing data rate among the released genotype data was 0.21% (32 581 of 15 450 676 total genotypes).

Genotyping of APOE polymorphisms (based on SNPs rs7412 and rs429358) was performed at PreventionGenetics (http://www.preventiongenetics.com). Genotyping was performed in array tape using allele-specific polymerase chain reaction analysis with universal molecular beacons. The DNA sequencing of positive control DNA samples was completed to ensure correct assignment of alleles.

Unless stated otherwise, analyses were conducted using the following definitions of LOAD based on standard research criteria: (1) broad, which included definite, probable, or possible LOAD and (2) narrow, which included as affected only those individuals who met criteria for definite or probable LOAD. We classified the affection status of family members with other forms of dementia or with mild cognitive impairment as unknown for the purposes of genetic analyses. For the linkage and family-based analyses using the narrow definition, we also classified patients with possible LOAD as unknown.

Single-point and multipoint nonparametric linkage analyses based on the algorithm of Kong and Cox¹⁸ were implemented using a multipoint engine for rapid likelihood inference (MERLIN),^{19 - 20} and we calculated nonparametric logarithm of odds (LOD) scores based on an established algorithm.²¹ We computed allele frequencies using all genotyped subjects. Given the important role of APOE ε4 in LOAD, we performed a conditional linkage analysis to test for a 2-locus model in which a polymorphism or variant at a given locus has an influence on LOAD only in the presence of the APOE ε4 allele.

We conducted single-point family-based association test (FBAT) analysis as implemented in version 1.7.3 of the FBAT software.^{22 - 23} We tested the hypothesis of no linkage and no association under an additive model, rather than the hypothesis of no association in the presence of linkage, because the primary goal of the analysis was to identify a novel candidate region rather than to fine map previously identified loci from the linkage analysis. We estimated allele frequencies for FBAT using parental genotype data, which we estimated from the offspring genotype database using the expectation-maximization algorithm. We also used the FBAT software to confirm the relation between APOE and LOAD. For the case-control data set, we first performed the χ² test to assess the allelic association between LOAD and SNPs.

For the case-control analysis, we selected 1 affected individual from each family with definite or probable LOAD. The unrelated, unaffected individuals served as controls. For the case-control analysis, we used the χ² test to assess the allelic association between LOAD and SNPs. We assessed population stratification using the Structure program, version 2.2 (http://pritch.bsd.uchicago.edu/structure.html),^{24 - 25} by using 103 unlinked SNPs to measure population substructure. We chose 103 SNPs that were present in both the Illumina-IVb linkage panel and the HapMap data set (http://www.hapmap.org). This was necessary because the present study participants were predominantly white and we lacked genotype data for nonwhite subjects. Thus we used the genotype data from the NIA-LOAD samples for white subjects and used the genotype data from the HapMap data set for nonwhite subjects. The allele frequencies for white subjects with LOAD in the NIA-LOAD data set were similar to those for white subjects in the HapMap data set. The results from the Structure analysis were used in an association analysis implemented in the STRAT program, version 1.0.^{24 - 25}

To determine the consistency of our findings, we examined allelic associations in an independent, publicly available data set from the Translational Genomics Research Institute (TGen) that included 859 patients and 552 controls, for a total of 1411 individuals (http://www.tgen.org/neurogenomics/data).⁷ We restricted our evaluation of the TGen data to SNPs that were only significant in the case-control analysis discussed in the “Results” section at P < .005. However, the TGen data set was genotyped using a microarray platform (Affymetrix platform; Affymetrix, Inc, Santa Clara, California) that included approximately 500 000 SNPs. Because SNPs were not identical, we included 5 SNPs on either side of the candidate SNP location derived from the current analysis. Imputation was not possible owing to the sparse genotyping in the NIA-LOAD families. Single-point allelic association was performed using Haploview software.¹⁶ Haplotype analysis was not performed.

The linkage and association analyses were restricted to 328 white families (1902 individuals) because more than 90% of the cohort were of European or North American ancestry (Table 1). The mean (SD) age at onset of symptoms for affected individuals was 73.9 (7.5) years, and the mean (SD) age at diagnosis was 77.2 (7.5) years. Of the 1902 individuals, 40.8% were affected and 45.9% were considered unaffected. The remaining 13.3% had other forms of dementia or mild cognitive impairment. All data, including pedigree structure, affection status, and genotype data used in the analysis, are available at the Web site at the NIA Genetics of Alzheimer Disease Data Storage Site (http://www.niageneticsdata.org).

For the case-control analysis, we studied 328 patients and 236 unrelated controls. The mean age at onset of dementia was 73.3 (range, 60-92) years, and the mean age of the last evaluation for controls was 78.1 (range, 60-99) years. Women constituted 61.9% of the participating family members and 58.5% of the controls. The APOE ε4 allele was present in 43.1% of the cases and 9.5% of the controls.

Using the broad definition of LOAD, 15 SNPs had LOD scores exceeding 2, including the following 2 SNPs with LOD scores of greater than 3: rs798485 (7p22.2; LOD score, 3.77) and rs1482258 (16q21; LOD score, 3.32) (Table 2). The SNP rs1482258 and 3 adjacent markers within a 6-cM region showed strong support for linkage with LOD scores exceeding 2. In addition, rs719423 (7p21.3) showed evidence of suggestive linkage (LOD score, 2.89). At 19q13.32, SNP rs2341000 similarly showed a strong support for linkage (LOD score, 2.49), most likely due to its proximity to APOE. Three markers with evidence of suggestive linkage, rs2036256 (6q22.31), rs720974 (9p21.3), and rs1537626 (10p14), have been previously reported as statistically significant in other studies (http://www.alzgene.org).

Using the narrow definition, the following 3 SNPs achieved LOD scores of 3.0 or greater: rs719423 at 7p21.3 (7.28 cM from rs798485 at 7p22.2), rs735144 at 16q13 (4.25 cM from rs1482258 at 16q21), and rs1482258 at 16q21. For the 3 most significant SNPs under the broad definition, LOD scores for rs1482258 at 16q21 and rs719423 at 7p21.3 increased under the narrow definition, whereas the LOD score for rs798485 at 7p22.2 decreased slightly. At or near 16q21 within a 27-cM region, a total of 9 SNPs had LOD scores exceeding 2.

Using the broad definition, the strongest evidence of linkage in the multipoint analysis was for 2 SNPs near APOE at 19q13.31-2 (LOD scores, 3.10 and 3.19) (Figure 1A). In fact, 23 SNPs within a 14.8-cM region near APOE had LOD scores greater than 2.0 in the region extending from 19q13.12 to 19q13.32. Three additional SNP clusters at chromosomes 7p22.1, 8p21.3, and 18q12.2 also had LOD scores of 2.0 or greater. The LOD scores decreased slightly using the narrow definition of LOAD in the multipoint analysis for the SNPs clustering at 19q13.31-32 (Figure 1B). Findings for 7 SNPs near 7p22.1-3, 2 SNPs at 8p21.3, and 4 SNPs in a 1.5-cM region around 16q21 remained suggestive of linkage.

Using the broad definition, 6 SNPs showed association with LOAD, with P values of less than .001 (range, .000063 to <.000968). The SNP rs174345 at 22q11.21 showed the strongest association (P = .000063). Of interest, rs744281 proximal to 17q21.31 was also strongly associated with LOAD, a marker near the gene encoding tau at 17q21.1. The SNPs at 2p14, 3q13.31, 8p21.3, and 11p14.3 were also associated with LOAD. Under the narrow definition, 9 SNPs had P values of less than .001 (range, .000174 to <.000815). At 8p21.3, SNP rs4427168 showed the most significant association with LOAD (P = .000174). This SNP, along with rs174345 at 22q11.21, was associated with LOAD under both disease definitions. These associations are illustrated in Figure 2.

In addition to the 2 coding SNPs for APOE, the most significant association was observed with rs762883 at 22q12.3 (P = .000069) (Table 3). We found that other loci showed evidence of association (defined as −log [P] > 2.5) included SNPs at 1p34.3, 1q41, 2p21, 2q24.3, 3q26.1, 7q21.3, 7q31.1, 8q23.3, 11q24.3, 14q13.1, 15q15.1, 20q13.33, and 22q12.3. All SNPs that were significant in the χ² analysis remained significant in the STRAT analysis.^{24 - 25}

Using the TGen LOAD data set,⁷ we found concordance with allelic associations at P < .05 for 3 SNPs that were significant in this case-control analysis (results available from the authors upon request). Single-nucleotide polymorphism A-2236481 (rs41377151) (located 10.9 kb away from rs7412, 1 of the coding SNPs for APOE) was significantly associated with LOAD (P = 3.29 × 10⁻³⁶) in the TGen data set. In addition, SNP A-1968867 (rs6027452) on chromosome 20 (located 4.5 kb away from the candidate SNP rs1381100 [20q13.33; P = .04]) and SNP A-4212589 (rs728273) on chromosome 7 (located 14.2 kb away from rs43077 [7q31.1; P = .047]) were also associated with LOAD in the TGen data set. The candidate SNPs identified from the TGen study were in strong LD with those from the NIA-LOAD study (D′ range, 0.97-1.00). (Details are available from the authors upon request.)

The FBAT analysis indicated that the 2 SNPs within APOE designating the ε4 allele were significantly associated with Alzheimer disease (Z = 8.68; P = 1.98 × 10⁻¹⁸; data not shown), as did the case-control analysis (χ² = 150.46; P = 1.4 × 10⁻³⁴). In the APOE conditional linkage analysis, many loci that were significant in the unadjusted analysis remained significant; however, some loci were significant only in the presence of the APOE ε4 allele. Five SNPs within a 5-cM region surrounding APOE provided LOD scores ranging from 3.06 to 4.84. Outside the APOE region, rs1482258 (located at 16q21), rs798485 (7p22.3), and rs985942 (8q12.1) had LOD scores suggestive of linkage. Some SNPs were found to be significant in the APOE ε4 conditional linkage analysis only, including rs2034222 (located at 5p; 31.31cM), rs1349710 (6q; 144.87 cM), rs189811 (7q; 185.95 cM), and rs337663 (12q; 101.97 cM); 19q13.32 showed the strongest multipoint LOD score (11.5). (Additional figures are available from the authors upon request.)

Using the NIA-LOAD family data set, we identified loci that may contain genetic variants related to the risk of LOAD. Not surprisingly, the APOE locus was identified and remains the most consistently replicated genetic risk factor for LOAD. Multiple candidate loci were identified from linkage and association analyses, but the results obtained from the family-based linkage and association analyses shared little overlap with the results from the case-control analysis (Table 4). The sample size for the case-control set provided 68% power to detect an allelic association at a significance level of .001, assuming an allele frequency of 0.3, risk ratio of 1.5, and genotyping error of 0.1%. Thus, it is possible that reduced power was a factor contributing to the observed differences.

Linkage analysis tests the cosegregation of the disease and genetic markers within families, without reference to a specific allele, and this method is powerful in diseases that conform to mendelian inheritance. In contrast, association analysis determines the excess transmission of a specific allele to affected individuals within families or co-occurrence of the specific allele among cases compared with unrelated individuals without disease.²⁷ It is possible to observe an allelic association in the absence of linkage when the allele frequency is high. Without dense SNP coverage, it is possible to miss allelic associations, even in the presence of linkage. Gene identification for common diseases is difficult when a single method is applied^{28 - 30} ; thus, it is optimal to apply linkage as well as association analyses when the data are available.

There continues to be support for linkage for LOAD at 6p, 9q, 10q, 12p, and 19q, but it has been extraordinarily difficult to identify the specific genes at each locus (see Kamboh³¹ for review). Moreover, there is little concordance between case-control and family-based linkage or association studies suggesting clinical and genetic heterogeneity. For example, variants in the alpha-2-macroglobulin gene (A2M); catenin alpha 3 gene (CTNNA3); plasminogen activator, urokinase gene (PLAU); insulin-degrading enzyme gene (IDE); glutathione S-transferase omega 1 and 2 genes (GSTO1 and GSTO2); and glyceraldehyde-3-phosphate dehydrogenase gene (GAPDH) have been identified in family-based or case-control studies but lack consistent replication. Loci in a broad region of 12p11 to 12q13 may contain genetic variants for LOAD, including GAPDH³² and LRP6 at 12p13.31,⁶ but both remain unconfirmed. In addition, several other genetic variants surrounding a locus at 10q24 have been related to LOAD.^{33 - 36} Linkage to LOAD and plasma amyloid β at 10q24 were reported.^{37 - 39} Li et al^{40 - 41} found support for an association with GSTO1 and GSTO2 at 10q25.1 with LOAD, but neither finding has been confirmed,⁴² and a new variant in the ribosomal protein S3A gene (RPS3A; located at 4q31.3) has been reported.⁴³ The locus on 9p21-22 has also eluded identification, but an association between LOAD and variants in the ubiquilin 1 gene (UBQLN1) at 9q22 have been described^{44 - 45} and confirmed in at least one study.⁴⁶ However, other studies do not support this finding.^{47 - 48}

Putative loci in our report overlap with some of those compiled by Bertram and colleagues⁴⁹ (http://www.alzgene.org), but many will remain unconfirmed. For independent confirmation without performing additional genotyping, we compared the findings from the publicly available LOAD data set from TGen against the findings from the present study. We found similar associations proximal to 7q31.1 and 20q13.33 in the NIA-LOAD data set. Second, we investigated candidate genes associated with LOAD using SNPs that were near or within these genes. As recommended, markers from previously implicated regions were treated differently from markers for which there was no prior evidence of an association.⁵⁰ We investigated 18 such candidates and found that at least 1 SNP at 12p13.2 (LRP6), 11q23.3 (SORL1), 17q23.2 (ACE), and 14q24.2 (PSEN1) was modestly associated with AD, with P values ranging from .004 to more than .05 (Table 5). Although none of these findings would survive a conservative correction for multiple testing, this exercise demonstrates the consistency in our findings and the value of this well-characterized data set for discovery or confirmation of genetic variants predisposing to LOAD.

Correspondence: Richard Mayeux, MD, MSc, Gertrude H. Sergievsky Center, 630 W 168th St, Columbia University, New York, NY 10032 (rpm2@columbia.edu).

Accepted for Publication: March 14, 2008.

Author Contributions:Study concept and design: Lee, Foroud, and Mayeux. Acquisition of data: Graff-Radford, Foroud, and Mayeux. Analysis and interpretation of data: Lee, Cheng, Graff-Radford, and Mayeux. Drafting of the manuscript: Lee, Foroud, and Mayeux. Critical revision of the manuscript for important intellectual content: Lee, Cheng, Graff-Radford, Foroud, and Mayeux. Statistical analysis: Lee, Cheng, and Mayeux. Obtained funding: Graff-Radford, Foroud, and Mayeux. Study supervision: Lee.

Investigators from the Consortium of Alzheimer’s Disease Centers in the NIA-LOAD Family Study Group: Robert Green, MD, Neil Kowal, MD, and Lindsay Farrer, PhD (Boston University, Boston, Massachusetts); Jennifer Williamson, MS, and Vincent Santana, MBA (Columbia University, New York, New York); Donald Schmechel, MD, and Peter Gaskel, BS (Duke University, Durham, North Carolina); Bernardino Ghetti, MD, Martin R. Farlow, MD, and Kelly Horner (Indiana University, Indianapolis); John H. Growdon, MD, Deborah Blacker, MD, ScD, Rudolph E. Tanzi, PhD, and Bradley T. Hyman, MD (Massachusetts General Hospital, Boston); Bradley Boeve, MD, Karen Kuntz, RN, Lindsay Norgaard, BS, and Nathan Larson, BS (Mayo Clinic, Rochester, Minnesota); Dana Kistler, BSH, Francine Parfitt, MS, and Jenny Haddow, BS (Mayo Clinic, Jacksonville, Florida); Jeremy Silverman, PhD, Michal Schnaider Beeri, PhD, Mary Sano, PhD, Joy Wang, BA, and Rachel Lally, BA (Mount Sinai School of Medicine, New York); Nancy Johnson, PhD, Marcel Mesulam, PhD, Sandra Weintraub, PhD, and Eileen Bigio, MD (Northwestern University, Chicago, Illinois); Jeffery Kaye, MD, Patricia Kramer, PhD, and Jessica Payne-Murphy, BA (Oregon Health and Science University, Portland); David Bennett, MD, Holli Jacobs, BA, Jeen-Soo Chang, MD, and Danielle Arends, RN (Rush University, Chicago); Lindy Harrell, MD, PhD (University of Alabama, Birmingham); George Bartzokis, MD, Jeffery Cummings, MD, Po H. Lu, PsyD, and Usha Toland, MS (University of California, Los Angeles); William Markesberry, MD, Charles Smith, MD, and Alise Brickhouse, BA (University of Kentucky, Lexington); John Trojanowski, MD, PhD, Vivianna Van Deerlin, MD, PhD, and Elisabeth McCarty Wood, MS (University of Pennsylvania, Philadelphia); Steven DeKosky, MD, Robert Sweet, MD, and Elise Weamer, MPH (University of Pittsburgh, Pittsburgh, Pennsylvania); I. Helena Chui, MD, and Arousiak Varpetian, MD (University of Southern California, Los Angeles); Ramon Diaz-Arrastia, MD, PhD, Roger Rosenberg, MD, and Barbara Davis, MA (The University of Texas Southwestern Medical Center, Dallas); Thomas Bird, MD, Malia Rumbaugh, MS, Gerard D. Schellenberg, PhD, and Murray Raskind, MD (University of Washington, Seattle); and Alison Goate, DPhil, John Morris, MD, Joanne Norton, MSN, RN, Denise Levitch, RN, Betsy Grant, MSW, PhD, and Mary Coats, MSN, RN (Washington University, St Louis, Missouri).

Financial Disclosure: None reported.

Funding/Support: This study was supported by the following federal grants: U24AG026395 (NIA-LOAD Family Study); U24AG021886 (National Cell Repository for Alzheimer's Disease); P50AG08702 (Boston University and Columbia University); P30AG028377 (Duke University); P30AG010133 (Indiana University); PO1 AG05138 (Massachusetts General Hospital; Mayo Clinic, Rochester; Mayo Clinic, Jacksonville; and Mount Sinai School of Medicine); and P30AG010124 (Northwestern University Medical School; Oregon Health and Science University; Rush University Medical Center; University of Alabama at Birmingham; David Geffen School of Medicine, University of California, Los Angeles; University of Kentucky, Lexington; University of Pennsylvania; University of Pittsburgh; University of Southern California; The University of Texas Southwestern Medical Center; University of Washington; and Washington University School of Medicine). Genotyping services were provided by the Center for Inherited Disease Research, which is fully funded through federal contract N01-HG-65403 from the National Institutes of Health (The Johns Hopkins University).

Additional Contributions: Susan LaRusse Eckert, MS, and Stephanie Doan, MPH (Columbia University), and Michele Goodman and Kelley Farber, MS (Indiana University), helped coordinate the project across the United States. Creighton H. Phelps, PhD, Marcelle Morrison-Bogorod, PhD, and Marilyn Miller, PhD, at the NIA provided guidance.

Additional Information: The following Web resources were used in this study: AlzGene (http://www.alzforum.org/res/com/gen/alzgene/); the Center for Inherited Disease Research (http://www.cidr.jhmi.edu/human_snp.html); the National Cell Repository for Alzheimer Disease (http://ncrad.iu.edu); and the NIA Genetics of Alzheimer's Disease data storage site (http://www.niageneticsdata.org).