Multi‐ancestry meta‐analysis identifies genetic modifiers of age‐at‐onset of Alzheimer's disease at known and novel loci

Abstract

INTRODUCTION

Much of Alzheimer's disease (AD) risk is explained by age, apolipoprotein E (APOE) genotype, and sex. We sought to identify genetic modifiers of age at onset (AAO) of AD while probing the influence of sex and APOE among those with diverse ancestry.

METHODS

We performed genome‐wide association studies (GWASs) of AAO in two diverse samples followed by meta‐analysis, contrasting results with and without adjustment for sex and APOE. Genome‐wide significance was set to p < 5×10⁻⁸.

RESULTS

GWASs adjusting for sex, APOE, population structure, and relatedness revealed 17 significant loci including independent associations at AD risk loci and four novel signals. APOE adjustment influenced GWAS effect sizes across the genome while sex adjustment had minimal effect.

DISCUSSION

We identified association signals within a diverse but relatively small sample, replicating loci recently discovered in large European ancestry‐only GWASs, and illustrated the power of using a quantitative trait like AAO over a binary diagnosis trait.

Highlights

• Survival analysis approach identified known and novel genetic modifiers of Alzheimer's disease (AD).

• Multi‐ancestry analyses revealed independent signals at known AD loci.

• Apolipoprotein E adjustment influenced variant effects across the genome.

1. BACKGROUND

Age, sex, and apolipoprotein E (APOE) genotype are major risk factors for Alzheimer's disease (AD). The annual incidence of AD increases 19‐fold between the ages of 65 to 70 and 85+ years; the prevalence of AD is one third higher in women than men; ¹ and APOE genotype, specifically elevated AD risk with ε4 and the protective effect of ε2, can explain nearly a quarter of AD heritability. ² Despite this, genome‐wide association studies (GWASs) for AD do not consistently address the effects of age, sex, and APOE genotype in their modeling approaches. This inconsistency complicates the interpretation of meta‐analyses combining results across outcomes and models ³ , ⁴ and reduces statistical power to detect novel AD susceptibility loci.

While age at onset (AAO) of AD is highly heritable (57%–78%), ⁵ , ⁶ relatively few GWASs have considered it as an outcome. ² , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ Analyzing AAO as a censored trait offers improved statistical power relative to a typical case/control design by leveraging age information not only from those affected by the disease but also those who appear protected from AD at older ages. ¹⁵ , ¹⁶ , ¹⁷ Despite their smaller sample sizes, AAO GWASs have replicated 20 AD risk loci and nominated three novel signals near genes associated with molecular or neuropathological changes in AD. ² , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ This suggests that AAO of AD and AD risk share an overlapping, but not necessarily identical, genetic architecture.

Most of the > 150 GWASs of AD risk or AAO represent non‐Hispanic White participants ¹⁸ , ¹⁹ while other populations remain underrepresented, with underpowered sample sizes unable to detect association signals with comparable effect sizes. ²⁰ Many AD risk loci identified in European ancestry–‐focused GWASs are nominally associated with AD in Black Americans (APOE, ABCA7, BIN1, CASS4, CD33, CELF, CR1, BIN1, EPHA1, NME8), ²¹ , ²² Caribbean Hispanics (BIN1, CLU, Picalm), ²³ and East Asians (ABCA7, BIN1, CD33, CNTNAP2, PICALM, SORL1). ²² , ²⁴ , ²⁵ Many variants associated with AD vary substantially in frequency across continental populations ²⁶ due to the history of human migration (e.g., APOE ²⁷ ). Haplotypic variation at those loci influences effect size estimates ²⁸ and hampers our ability to generalize GWAS results and estimate polygenic risk of disease. ²⁹ , ³⁰ , ³¹ Local ancestry differences capturing this haplotypic variation have been significantly associated with AD risk near ABCA7, CD33, and GRINB3 in Blacks ³² and at APOE in Hispanics. ³³ , ³⁴ Multi‐ancestry analyses offer an alternative strategy to identify genetic variation associated with AD phenotypes while better capturing human diversity ²² , ³⁵ , ³⁶ and offer results that improve the generalizability of polygenic risk scores for AD. ³⁷

Here, we perform a GWAS for AAO of AD in a diverse set of AD cases and controls in both a discovery and replication data set, then meta‐analyze results to detect shared signals. Given the relationships among APOE, ancestry, and AD risk, ¹⁹ , ³³ , ³⁴ , ³⁸ we explore the effects of APOE adjustment on association signals. We find shared genetic architecture between AD risk and AAO loci, novel loci associated with AAO of AD, and evidence that APOE adjustment alters the association between AAO of AD and variants across the genome.

2. METHODS

2.1. The discovery data set

We gathered controlled‐access genotype array and phenotype data through approved applications to the database of Genotypes and Phenotypes (dbGaP; study accession: phs00496) and the National Institute on Aging Genetics of Alzheimer's Disease Data Storage Site (NIAGADS; study accessions NG00020, NG00022‐NG00024, NG00026, NG00028‐NG00031, NG00034, NG00047, NG00068‐NG00071). Study‐specific sample quality control (QC) required missing < 5% of genotypes as well as pedigree ³⁹ and sex checks ⁴⁰ to exclude sample swaps. We restricted analysis to those with AD case or control status (excluding mild cognitive impairment), AAO for cases and age ‐at ‐last ‐evaluation (AAE) data for controls, and both sex and APOE genotype data. Where data were available, outliers were excluded where age at death was > 10 years greater than AAO. APOE covariate adjustment used imputed ε2 and ε4 allele dosages to account for uncertainty in genotyping. Samples with discordant observed versus imputed sex or APOE genotype were excluded. When selecting which duplicate sample to exclude, we prioritized those missing phenotype data, then by genotype missingness rate, and finally selected one at random if they otherwise did not differ. Study‐specific autosomal variant QC included genotype completeness ≥ 95%, minor allele frequency (MAF) > 5%, Hardy–Weinberg equilibrium ⁴¹ p < 0.01 using the Robust Unified Hardy–Weinberg Equilibrium Test to accommodate population structure within a study, and unambiguous genotypes. Genotypes were aligned to hg19, phased by Eagle v2.4, ⁴² then imputed ⁴³ to the Trans‐Omics for Precision Medicine (TOPMed) ⁴⁴ reference panel aligned to hg38 using minimac4 ⁴⁵ as implemented in the TOPMed Imputation Server. ⁴³ Study‐specific imputation QC included average call rate ≥ 0.95 and r ² ≥ 0.3 for common variants with MAF ≥ 0.05 or r ² ≥ 0.5 for uncommon variants with MAF < 0.05. Study‐specific imputed genotypes and phenotype data were combined into a single data set. Imputed variants with MAF < 1% or missing ≥ 5% of genotypes in the combined data set were excluded. Samples with missing or discordant observed and imputed ε2/ε3/ε4 genotypes were excluded from APOE‐adjusted GWAS.

RESEARCH IN CONTEXT

1. Systematic review: The authors performed a literature review of published articles describing genome‐wide association studies of Alzheimer's disease (AD) and age at onset (AAO). Previous studies offer evidence that the association between genotype and AD can vary dramatically in strength, direction of effect, or both when studies vary by major risk factors including age, sex, and ancestry.

2. Interpretation: Our multi‐ancestry genome‐wide association study of AAO of AD identified shared genetic architecture with AD risk, including novel association signals and independent signals at established AD risk loci. Apolipoprotein E adjustment influenced variant effects genome wide, consistent with a strong relationship with both AD risk and human population structure.

3. Future directions: Future studies of genome sequence data in independent data sets representing similar levels of population diversity are needed to validate and fine‐map these associations. Functional studies in appropriate cell types are needed to test causal relationships among prioritized variants, genes, and AD.

2.2. The replication data set

The Alzheimer's Disease Genetics Consortium (ADGC) data were imputed to the TOPMed reference as previously described. ³⁵ Data were shared as study‐specific cohorts defined by reported race and ethnicity. We selected a subset of this ADGC data to minimize overlap with the discovery data (ACT‐AA, ACT2, ACT3, ACT3‐AA, ACT3‐Asian, ACT3‐Hispanic, ADC1‐2‐AA, ADC10, ADC10‐AA, ADC10‐Asian, ADC10‐Hispanic, ADC11, ADC11‐AA, ADC11‐Asian, ADC11‐Hispanic, ADC12, ADC12‐AA, ADC12‐Asian, ADC12‐Hispanic, ADC3‐AA, ADC8, ADC8‐AA, ADC8‐Hispanic, ADC9, ADC9‐AA, ADC9‐Hispanic, ADNI, BIOCARD, CHAP‐AA, CHAP2, CHOP‐AA, EAS, GSK, MAYO, MIRAGE300‐AA, MIRAGE600‐AA, NBB, NIALOAD‐NCRAD‐AA, OHSU, PRADI‐Hispanic, REAAADI‐AA, RMayo, TARCC1, TARCC3, TARCC3‐AA, TARCC3‐Hispanic, TARCC4‐Hispanic, UKS, UMVUTARC2, WASHU2, WHICAP). Study‐specific imputation QC included average call rate ≥ 0.95 and r ² ≥ 0.3 for common variants with MAF ≥ 0.05 or r ² ≥ 0.5 for uncommon variants with MAF < 0.05. Study‐specific imputed genotypes and phenotype data were combined into a single data set. Sample QC included missing < 5% of genotypes and restricted analysis to those with AD case or control status (excluding mild cognitive impairment), AAO for cases or AAE for controls, and both sex and APOE genotype data. APOE covariate adjustment used imputed allele dosages to account for uncertainty in genotyping. Samples with discordant observed versus imputed sex or APOE genotype were excluded. Imputed variants with MAF < 0.5% or missing ≥ 5% genotypes in the combined data set were excluded.

2.3. Human subjects

This study was approved by the University of Washington Human Subjects Division (STUDY00000240) and was performed in accordance with ethical standards consistent with the 1964 Declaration of Helsinki. This study included data from participants across demographic boundaries including reported sex, race, and ethnicity. As described below, analyses adjusted for relatedness and population structure within the data, improving the generalizability of our results.

2.4. Relatedness and population structure

Relatedness and principal components (PC) analysis was performed on imputed genotypes after QC, filtering variants for MAF > 5%. For each data set, relatedness was initially estimated by KING‐robust ³⁹ after pruning variants for linkage disequilibrium (r ² < √0.01). Duplicate samples were excluded, prioritizing first those missing phenotype data, then higher genotyping rate, and finally selecting one at random if they otherwise did not differ. PCs accounting for relatedness were first estimated using PC‐AiR ⁴⁶ then relatedness estimates were recalibrated by PC‐Relate ⁴⁷ and then these estimates were used for another round of PC‐AiR/PC‐Relate to orthogonally partition PCs and the genetic relatedness matrix (GRM). The most informative PCs were selected using the inflection point of scree plots as well as clustering within pairwise PC plots. To speed GWAS, we used a sparse GRM, setting < fourth‐degree relatedness to zero.

2.5. Association testing

GWAS analyses were restricted to imputed genotypes with MAF ≥ 1% and AAO or AAE. The phenotype was defined as the Martingale residuals from a Cox proportional hazards analysis performed using the survival package in R (v3.5), using AAO of AD for cases and AAE for controls, adjusted for sex, imputed ε2 dosage, imputed ε4 dosage, and the most informative PCs. Using the LMM‐OPS framework, ⁴⁸ we tested the association between imputed genotypes and this phenotype, adjusting for the most informative PCs with fixed effects and the GRM with random effects.

We performed two secondary GWASs to evaluate the sensitivity to covariate adjustment. The first alternative GWAS mimicked the original GWAS but removed APOE covariates, while the second removed both APOE and sex covariates.

Genomic inflation ⁴⁹ in the discovery and replication GWAS was measured using λ and linkage disequilibrium (LD) score regression intercepts. ⁵⁰ Results from the discovery and replication GWAS were meta‐analyzed using Han and Eskin's Random Effects model (RE2) implemented in Metasoft ⁵¹ and correcting for genomic inflation of mean effects. Genome‐wide significance was defined as p < 5×10⁻⁸, and Manhattan plots were drawn in R. ⁵²

2.6. Interpreting association signals

Hazard ratios for individual markers were estimated by including the marker as a covariate in the same Cox proportional hazards test used to define the associated outcome. The genomic context of GWAS signals was plotted using the locuszoomr (v0.3.5) R package, annotated with LD information from 1000 Genomes Europeans ⁵³ , ⁵⁴ provided by the LDlinkR R package (v1.4.0) ⁵⁵ and gene annotations from Ensembl v113 ⁵⁶ provided by the AnnotationHub R package (v3.14.0). Correlation and differences in effect sizes were calculated and illustrated in R.

We sought evidence of replication for loci previously associated with AD risk or AAO of AD. We extracted the lead marker from genome‐wide significant associations with AD risk across 13 major GWASs for AD or proxy AD phenotypes ⁵⁷ and study‐wide significant associations with AAO ² , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ outside the APOE region (chr19:43,905,796‐45,909,393; hg38). We selected one marker per locus, prioritizing first by strength of evidence of replication, then by effect size, and finally by p value. Because APOC1 is near APOE, we performed a separate search of the GWAS catalog ¹⁸ for “APOC1” and reviewed all listings with the reported trait “Alzheimer's disease.”

Variants were connected to target genes using Open Targets Genetics, ⁵⁸ , ⁵⁹ https://genetics.opentargets.org/, last updated October 2022. If a variant fell at a known AD GWAS locus, annotation was restricted to the nominated gene for that locus; otherwise, the top candidate was selected based on variant to function (V2F) support. Phenome‐wide association study (PheWAS) data were extracted from the NHGRI/EMBL‐EBI GWAS Catalog, ¹⁸ restricted to those with study‐specific p < 5×10⁻⁸. Genes were connected to AD biological domains ⁶⁰ and therapeutic targets using Agora, https://agora.adknowledgeportal.org/, site v3.4.0 and data version syn13363290‐v68. A subset of variants was annotated with allele frequencies in 1000 Genomes/Human Genome Diversity Project (HGDP) reference populations. ⁵³

3. RESULTS

3.1. Diversity within discovery and replication data sets

After QC, the discovery and replication data sets were similarly balanced between cases and controls by sample sizes, age distributions, sex, and APOE genotype frequencies (Table 1). APOE genotype mismatch rates were similar in the discovery and replication sets for ε2 (1.13% vs. 1.87%) while the ε4 mismatch rate was greater in the replication data set (2.29% vs. 5.23%). Sample‐level QC for missing data in the replication data demonstrated a disproportionate effect on cohorts reporting Asian or Black race. Both the discovery and replication data sets are enriched for non‐Hispanic White participants (72% and 63%, respectively), with higher proportions of Hispanic ethnicity and lower proportions of Black race in the discovery data. The discovery data showed relative enrichment of non‐Hispanic White cases and Black controls, but this pattern was less evident in the replication data. PC analysis (PCA) within each data set reveals complex patterns of genetic variation (Figure 1), corresponding to human population structure rather than technical artifacts (Figures S1 and S2 in supporting information).

TABLE 1.

Sample description.

	Discovery (n = 22,044)		Replication (n = 19,483)
n	Controls	Cases	Controls	Cases
Males	4197	4484	4417	2930
Females	6963	6400	7938	4198
Age (years)
Mean	76.3	73.3	74.5	72.5
Min	32	30	43	33
Max	103	102	108	107
Median	77	74	74	73
APOE frequencies
ε2	7.2%	3.5%	7.5%	3.9%
ε4	15.2%	36.4%	15.6%	34.8%
Reported race/ethnicity
Non‐Hispanic White	63.5%	78.7%	61.4%	66.7%
Black	13.5%	3.0%	28.2%	22.9%
Hispanic	22.9%	17.9%	10.4%	10.5%
Other values	0.1%	0.4%	0.0%	0.0%

Notes: Age: age ‐at onset of Alzheimer's disease for cases, age at last evaluation or visit for controls. APOE frequencies and reported race/ethnicity proportions represent only non‐missing values. “Other” values include American Indian/Alaskan Native, Asian/Pacific Islander, and Other demographic categories.

Abbreviation: APOE, apolipoprotein E.

FIGURE 1.

Genetic diversity within the discovery and replication data sets. X axis: the first principal component (PC). Y axis: the second PC. Each PC is labeled by the amount of genetic variance explained.

3.2. GWAS for AAO of AD with both sex and APOE adjustment

The discovery GWAS of 9,111,557 variants and 21,736 participants adjusted for sex and APOE ε2 and ε4 dosages identified significant evidence of association between AAO of AD and seven known AD risk loci near APOE, BIN1, CASS4, CR1, HAVCR2, MS4A4A, and PICALM (λ = 1.03, LD score intercept = 1.04; Figure 2A, Table S1 in supporting information). The replication analysis of 8,634,132 variants and 19,483 participants supported each of these signals while no other loci reached genome‐wide significance (λ = 1.03, LD score intercept = 1.04; Figure 2B, Table S1). Meta‐analysis of the discovery and replication GWAS identified 16 genome‐wide significant associations (λ = 1), including 14 known AD risk loci near APOE (APOC1), BIN1, CASS4, CD2AP, CR1, ECHDC3 (LOC105376412), EPHA1, HAVCR2 (ADRA1B), MS4A4A, MYO15A (FBXW10), PICALM, pILRA (NYAP1), SHARPIN, and UMAD1 while adding two new signals: chromosomes 11q13.1 near CATSPERZ and 18p11.21 near LDLRAD4 (Figure 2C, Table 2). Most of these signals had robust support from both the discovery and the replication GWASs (2q14.3 BIN1, 6p12.3 CD2AP, 7p21.3 UMAD1, 7q22.1 NYAP1, 7q34‐q35 EPHA1, 8q24.3 SHARPIN, 10p14 LOC105376412, 11q12.2 MS4A4A, 11q14.2 PICALM, 20q13.31 CASS4) though two showed some attenuation of effect in the replication study (1q32.2 CR1, 19q13.32 APOC1; Table S2 in supporting information). Four signals had both nominally significant evidence of heterogeneity and opposing directions of effect (5q33.3 ADRA1B, 11q13.1 CATSPERZ, 17p11.2 FBXW10, 18p11.21 LDLRAD4).

FIGURE 2.

GWAS for AAO of AD, adjusting for sex, ε2, ε4, population structure, and relatedness. A, GWAS in the discovery data. B, GWAS in the replication data. C, Meta‐analysis of discovery and replication GWASs, adjusted for genomic inflation. AAO, age at onset; AD, Alzheimer's disease; GWAS, genome‐wide association study; LDSC, linkage disequilibrium score regression.

TABLE 2.

Genome‐wide significant loci in meta‐analysis of GWAS for AAO of AD, adjusting for sex, APOE genotypes, population structure, and relatedness.

			Discovery				Replication				Meta‐analysis
Locus	Gene	Variant	AAF	β	SE	p	AAF	β	SE	p	β	SE	p	I 2 (%)	Q p
1q32.2	CR1	1:207623552:A:T	0.817	−0.078	0.009	6.68E‐17	0.846	−0.051	0.009	1.69E‐08	−0.064	0.014	1.35E‐21	77.30	3.58E‐02
2q14.3	BIN1	2:127135234:C:T	0.404	0.073	0.007	3.16E‐23	0.405	0.058	0.007	3.04E‐19	0.065	0.007	1.52E‐37	52.95	1.45E‐01
5q33.3	HAVCR2	5:159951499:G:A	0.024	−0.130	0.023	1.85E‐08	0.031	0.044	0.018	1.53E‐02	−0.042	0.087	4.43E‐12	97.15	3.21E‐09
6p12.3	CD2AP	6:47586441:G:A	0.255	0.035	0.008	1.67E‐05	0.242	0.031	0.007	2.77E‐05	0.033	0.005	8.70E‐09	0.00	7.32E‐01
7p21.3	UMAD1	7:7816063:G:A	0.555	−0.035	0.007	9.92E‐07	0.538	−0.022	0.006	4.74E‐04	−0.028	0.006	1.87E‐08	41.97	1.89E‐01
7q22.1	pILRA	7:100494172:T:C	0.729	0.039	0.008	1.34E‐06	0.748	0.029	0.007	1.25E‐04	0.033	0.005	4.81E‐09	0.00	3.59E‐01
7q34‐q35	EPHA1	7:143435090:G:A	0.567	0.027	0.007	1.66E‐04	0.524	0.034	0.007	2.49E‐07	0.031	0.005	1.11E‐09	0.00	5.13E‐01
8q24.3	SHARPIN	8:144103704:G:A	0.080	0.067	0.014	2.49E‐06	0.065	0.054	0.014	1.51E‐04	0.061	0.010	8.24E‐09	0.00	5.13E‐01
10p14	ECHDC3	10:11678309:A:G	0.358	0.032	0.008	2.19E‐05	0.331	0.030	0.007	2.21E‐05	0.031	0.005	8.78E‐09	0.00	8.26E‐01
11q12.2	MS4A6A	11:60251788:G:GTA	0.334	−0.042	0.008	5.39E‐08	0.293	−0.043	0.007	3.59E‐09	−0.042	0.005	1.35E‐14	0.00	9.43E‐01
11q13.1	CATSPERZ	11:64301284:T:C	0.933	0.034	0.017	4.75E‐02	0.856	−0.060	0.011	1.20E‐07	−0.014	0.047	7.18E‐09	95.15	5.57E‐06
11q14.2	PICALM	11:86120648:A:G	0.679	0.039	0.008	3.79E‐07	0.704	0.035	0.007	4.23E‐07	0.037	0.005	6.44E‐12	0.00	7.67E‐01
17p11.2	MYO15A	17:18737894:G:A	0.029	−0.065	0.024	5.94E‐03	0.024	0.104	0.024	1.30E‐05	0.019	0.085	1.88E‐08	96.05	4.93E‐07
18p11.21	LDLRAD4	18:13585011:C:T	0.911	0.043	0.012	4.47E‐04	0.916	−0.043	0.012	2.46E‐04	0.000	0.043	2.34E‐08	96.11	3.94E‐07
19q13.32	APOC1	19:44916825:A:C	0.184	0.095	0.015	1.31E‐10	0.137	0.030	0.013	2.43E‐02	0.062	0.033	1.02E‐10	90.76	1.01E‐03
20q13.31	CASS4	20:56414777:T:A	0.095	−0.064	0.012	1.35E‐07	0.112	−0.050	0.010	8.88E‐07	−0.056	0.008	6.77E‐12	0.00	3.82E‐01

Notes: Variant: Chromosome: Position on hg38: Reference allele: Alternate allele; β: Effect size for the alternate allele; Q statistic p value: Evidence for heterogeneity between our discovery and replication studies.

Abbreviations: AAF, alternative allele frequency; AAO, age at onset; AD, Alzheimer's disease; APOE, apolipoprotein E; GWAS, genome‐wide association study; SE, standard error of effect size.

3.2.1. Evidence supporting loci with significant heterogeneity

We investigated the support for each genome‐wide significant locus in the meta‐analysis with strong evidence for heterogeneity and effect size estimates in opposite directions between the discovery and replication data sets. These variants were typically uncommon in both data sets (MAF < 10%), though the lead single nucleotide polymorphism (SNP) at 11q13.1 reached nearly 15% in the replication data. Two of these signals are driven by a strong signal in a single data set (5q33.3 and 11q13.1) with only nominally significant support for an opposing direction of effect in the other data set (p < 0.05). The evidence supporting the other two heterogeneous signals (17p11.2 and 18p11.21) was more modest and more evenly distributed between data sets. LocusZoom plots reveal that each signal, when evident, is supported by multiple variants in a haplotype (Figure 3A). Variation in sample demographics and variation in MAFs across reference populations suggest that the underlying haplotypic variation may vary between our discovery and replication data sets (Figure 3B).

FIGURE 3.

Support for heterogeneous meta‐analysis signals. A, Lead variant from meta‐analysis is shown as a purple diamond. Linkage disequilibrium information is specific to 1000 Genomes European samples. Gene labels were restricted to protein coding genes for 11q13.1. B, Minor allele frequency (MAF) variation across 1000 Genomes reference populations. Variants are presented as chromosome:position:reference allele:alternate allele. Reference populations are color coded as gold: admixed Americans; orange: African; magenta: East Asian; purple: European; blue: South Asian.

3.2.2. Relationship with AD risk alleles

To better understand the relationship between our association signals and prior AD GWAS, we compared our evidence of association between AAO of AD and the lead SNP for an AD GWAS signal from the literature ⁵⁷ (Table 3). Three of our lead SNPs are the most frequently reported lead SNP at the locus across recent AD GWASs: BIN1, ECHDC3, and SHARPIN1 (Figures S3–S5 in supporting information). While our lead SNPs at 1q32.2 (CR1), 7p21.3 (UMAD1), and 11q12.2 (MS4A6A) differed, they were in high LD with the reported SNPs at those loci (r ² > 0.8) which reached genome‐wide significance in our study (p < 5 × 10⁻⁸, Figures S6–S8 in supporting information). Our GWAS refines the known haplotypes associated with AD on 6p12.3 (CD2AP), 7q22.1 (pILRA), 11q14.2 (PICALM), and 20q13.31 (CASS4), with lead SNPs in moderate LD with the reported lead SNPs with attenuated p values in our study (Figures S9–S12 in supporting information). However, our analysis does not support the association between the reported SNPs near EPHA1, HAVCR2, and MYO15A (p > 0.05, Figures 3 and S13 in supporting information), suggesting that our signals at 5q33.3, 7q34‐q35, and 17p11.2 are independent. As ε4 dosage is included in the GWAS model it is not significantly associated with AAO of AD in this study, though the APOC1 variant on 19q13.32 is on the same haplotype in both the discovery and replication data sets (Figure S14 in supporting information). While APOC1 SNPs have been associated with AD, our literature review found no prior genome‐wide significant signal at APOC1 that was independent of the APOE ε2 and ε4 alleles.

TABLE 3.

Comparison of lead variants at known AD GWAS loci.

										Discovery		Replication
Locus	Gene	Lead variant	HR	95% CI	p	Known SNP	HR	95% CI	p	r 2	D'	r 2	D'
1q32.2	CR1	rs10863417	0.94	0.91‐0.96	1.35E‐21	rs679515	0.94	0.91‐0.97	2.57E‐21	0.96	0.99	0.65	0.99
2q14.3	BIN1	rs6733839	1.07	1.05‐1.08	1.52E‐37	rs6733839	–	–	–	–	–	–	–
5q33.3	HAVCR2	rs62377696	0.96	0.81‐1.14	4.43E‐12	rs6891966	1.01	0.99‐1.02	4.02E‐01	0	0.03	0	0.15
6p12.3	CD2AP	rs9395285	1.03	1.02‐1.04	8.70E‐09	rs9381563	0.98	0.97‐0.99	2.49E‐04	0.28	0.99	0.13	0.99
7p21.3	UMAD1	rs10276423	0.97	0.96‐0.98	1.87E‐08	rs6943429	0.97	0.96‐0.99	1.87E‐08	0.89	0.96	0.87	0.95
7q22.1	PILRA	rs12539172	1.03	1.02‐1.05	4.81E‐09	rs1859788	1.03	1.02‐1.04	1.18E‐08	0.32	0.92	0.34	0.88
7q34‐q35	EPHA1	rs9640386	1.03	1.02‐1.04	1.11E‐09	rs10808026	0.99	0.97‐1.01	1.17E‐01	0	0.02	0	0.09
8q24.3	SHARPIN	rs34173062	1.06	1.04‐1.08	8.24E‐09	rs34173062	–	–	–	–	–	–	–
10p14	ECHDC3	rs7920721	1.03	1.02‐1.04	8.78E‐09	rs7920721	–	–	–	–	–	–	–
11q12.2	MS4A6A	rs3041800	0.96	0.95‐0.97	1.35E‐14	rs1582763	0.96	0.95‐0.97	1.59E‐14	0.66	1	0.69	1
11q14.2	PICALM	rs472486	1.04	1.03‐1.05	6.44E‐12	rs3851179	1.03	1.02‐1.05	1.78E‐10	0.42	0.85	0.42	0.82
17p11.2	MYO15A	rs149428166	1.02	0.86‐1.20	1.88E‐08	rs2242595	0.99	0.97‐1.01	6.38E‐02	0	0.37	0	0.89
19q13.32	APOC1	rs73052335	1.06	1.00‐1.13	1.02E‐10	rs429358	1.06	0.68‐2.56	4.51E‐01	0.27	0.93	0.36	0.91
20q13.31	CASS4	rs6014722	0.95	0.93‐0.96	6.77E‐12	rs6014724	0.96	0.93‐0.98	1.25E‐07	0.54	0.87	0.53	0.75

Notes: Our meta‐analysis results for both the lead variant in this study and the lead SNP from prior AD GWASs are compared, along with their pairwise estimates of linkage disequilibrium within our discovery and replication data sets. Known SNP, Lead variant in prior AD GWAS; r ², D': Measures of linkage disequilibrium between the lead variants in this study and prior AD GWAS.

Abbreviations: AD, Alzheimer's disease; CI, confidence interval; GWAS, genome‐wide association study; gene: Nominated gene at AD GWAS locus; HR, hazard ratio; SNP, single nucleotide polymorphism.

3.3. Covariate effects

We explored whether differences in model specification might explain the differences between this study and the recent AD GWAS literature focused on case–control status and non‐Hispanic White samples. ³ , ⁴ , ⁶¹ , ⁶² , ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁶⁸ Prior GWASs of AD cohorts with clinical/pathological diagnoses typically included age and sex as covariates, ⁶¹ , ⁶² whereas GWASs including proxy AD phenotypes (e.g., parental history of dementia) in biobank datasets typically did not adjust for age ³ , ⁶⁴ , ⁶⁷ or sex ³ , ⁶⁴ , ⁶⁷ or did so inconsistently. ⁴ , ⁶⁵ , ⁶⁸ These proxy‐GWAS also typically meta‐analyzed their results with GWAS summary statistics from studies that included age and/or sex as covariates. None included APOE genotype as a covariate.

3.3.1. Main covariate effects

Each covariate was strongly associated with AAO of AD after adjustment for the first nine PCs, with both data sets offering similar effect size estimates. Female sex was associated with reduced hazard of AD (discovery hazard ratio [HR]: 0.88 [95% confidence interval (CI): 0.84–0.91], p = 2.75 × 10⁻¹¹; replication HR: 0.87 [95% CI: 0.83–0.91], p = 1.97 × 10⁻⁰⁸). Dose of the APOE ε2 allele was associated with a strong protective effect (discovery HR: 0.66 [95% CI: 0.61–0.71], p < 2 × 10⁻¹⁶; replication HR: 0.68 [95% CI: 0.63–0.74]; p < 2 × 10⁻¹⁶). Dose of the APOE ε4 allele was associated with a strong deleterious effect on the hazard of AD (discovery HR: 2.50 [95% CI: 2.43–2.57], p < 2 × 10⁻¹⁶; replication HR: 2.50 [95% CI: 2.41–2.59], p < 2 × 10⁻¹⁶).

3.3.2. GWAS sensitivity to covariates

We performed two additional GWASs to evaluate the sensitivity of our results to covariate adjustment. The first alternative GWAS mimicked our original GWAS but removed APOE covariates while still adjusting for sex, population structure, and relatedness (Figure S15, Tables S2 and S3 in supporting information). The second alternative GWAS further simplified the model by removing sex as a covariate while still adjusting for population structure and relatedness (Figure S16, Tables S4 and S5 in supporting information). Relative to the original GWAS, the two alternative GWASs identified similar numbers of loci reaching significance and shared similar evidence for genomic inflation, but our original model recognized the greatest number of significant loci. Nine loci failed to reach genome‐wide significance across all three meta‐analyses models: 5q33.3, 7p21.3, 7q34‐q35, 12q24.31, 16q23.2, 17p13.2, 17p11.2, 18p11.21, and 19q13.32 (Table S6 in supporting information). The change at 19q13.32, the APOE locus, is to be expected. Most other loci had only modest changes in effect size estimates, but many failed to reach a suggestive threshold in at least one analysis. The signals at 5q33.3, 7q34‐q35, 17p11.2, and 18p11.21 needed APOE covariate adjustment to reach significance, while the signal at 12q24.31 needed sex as a covariate. This suggests that APOE genotype and sex covariates act as precision variables in GWAS for AD‐related traits.

3.3.3. Effect across the genome

Because we saw changes in both effect size and p value across models for some loci outside the APOE region, we explored the extent to which our covariate selection influenced our results across the genome. The inconsistent handling of covariates could possibly explain some of the variable support for previously reported AD GWAS SNPs at shared loci. A comparison of our original GWAS to the first alternative GWAS without APOE adjustment for all autosomes (excluding chromosome 19 where APOE is located) revealed correlated β estimates (r ² = 0.75), which fluctuated by as much as the magnitude of a genome‐wide significant signal (Figure 4A, B). In contrast, comparison of the two alternative GWASs with and without sex adjustment revealed highly correlated β estimates (r ² = 0.99) with fluctuations more comparable in magnitude to the standard errors at significant loci (Figure 4C, D). The stronger fluctuations due to APOE adjustment are likely explained by population structure, as the frequency distribution of APOE ε2 and ε4 are correlated with geography. In neither comparison do we see strong outliers.

FIGURE 4.

Change in effect size (β) estimates across GWAS models. Top: Contrast effect size estimates from the original GWAS with the first alternative GWAS adjusted for sex but not APOE in panel (A), and the difference between the two values in panel (B). Bottom: Contrast effect size estimates from the first alternative GWAS and the second that did not adjust for sex or APOE in panel (C), and the difference between the two in panel (D). All models adjust for population structure and relatedness. Results on chr19 are excluded to avoid cis effects due to APOE genotype. Dashed line: effect size comparable to the lead variant at 2q14.3 near BIN1. Dotted line: effect size comparable to the lead variant at 6p12.3 near CD2AP. APOE, apolipoprotein E; GWAS, genome‐wide association study.

3.4. Connections between GWAS signals and AD biology

The lead variant at each original AAO GWAS signal (Table 2) was linked to a prioritized gene through its genetic and epigenetic context, then each prioritized gene was linked to AD biology through multi‐omics and systems biology (Table S7 in supporting information). Most variants (13/16, 81%) were linked to their prioritized gene through gene regulation, measured by direct association with transcript abundance (expression quantitative trait loci, eQTLs) or promoter‐gene interactions (promoter capture Hi‐C; Figure 5A). Similarly, nearly all prioritized genes were significantly differentially expressed in the AD brain (15/16, 94%; Figure 5B). Most prioritized genes (11/16, 69%) are robustly associated with late‐onset AD, while the 10p14 locus prioritizing ECHDC3 has also long been associated with AD and family history of AD. ³ , ⁴ , ²² , ³⁶ , ⁶¹ , ⁶² , ⁶⁴ , ⁶⁵ , ⁶⁹ , ⁷⁰ The heterogenous signals at 5q33.3 and 17p11.2 do not implicate the AD risk gene at the same locus (Table 3), while their implicated genes have relatively modest links to AD biology: ADRA1B at 5q33.3 is differentially expressed in the AD brain and belongs to the vasculature AD domain, while LGALS9C at 17p11.2 is not associated with gene or protein expression in the AD brain. In contrast, the two novel AAO GWAS signals at 11q13.1 and 18p11.21 implicate genes differentially expressed in the AD brain and involved in multiple AD biological domains: ESRRA is involved in structural stabilization, metal binding and homeostasis, and epigenetic AD domains, while RNMT is involved in the immune response and synapse AD domains. The prioritized genes represent most AD biological domains (16/19, 84%), excepting autophagy, DNA repair, and the RNA spliceosome. While structural stabilization, immune response, and synapse have the strongest support, 7/19 (37%) AD biological domains are implicated by at least a quarter of the prioritized genes (Figure 5C). Half of the prioritized genes have been independently nominated as AD therapeutic targets in Agora.

FIGURE 5.

Connections between AAO GWAS hits and AD biology. Detailed annotations are provided in Table S7 in supporting information. A, Lead variants from the AAO GWAS are linked to genes based on their position (“within gene”) and their association with gene expression (“eQTL”), protein expression (“pQTL”), gene splicing (“sQTL”), and gene‐promoter interaction (promoter capture Hi‐C, “PC‐HiC”) as reported in Open Targets Genetics. B, Genes are linked to AD biology based on the strength of their association with AD by the Alzheimer's Disease Sequencing Project's Gene Verification Committee (“AD GWAS”), evidence the gene is differentially expressed (“DEG in AD brain”) or the protein is differentially expressed (“DEP in AD brain”) based on the Accelerating Medicines Partnership Program for Alzheimer's Disease consortium work, and their nomination as a therapeutic target as reported in Agora. C, Distribution of prioritized genes within biological domains linked to AD as reported in Agora, restricted to those supported by 4+ genes. AAO, age at onset; AD, Alzheimer's disease; GWAS, genome‐wide association study.

4. DISCUSSION

4.1. Overview

Despite a relatively modest sample size, our multi‐ancestry GWAS of a powerful survival‐based AAO of AD phenotype revealed new insights into known GWAS loci and nominated new genetic modifiers of AD. Variants at known AD risk loci were significantly associated with AAO of AD while analyzing ≈ 5% (5q33.3, 17p11.2, 8q24.3, 7p21.3) to 50% (20q13.31, 10p14, 7q22.1) the original AD GWAS sample size. We found that APOE adjustment exposed an independent signal on 19q13.32 and influenced variant effect sizes across the genome while sex adjustment did not. Despite the diversity within and between the discovery and replication data sets, our GWAS showed less evidence of polygenicity and genomic inflation than AD GWAS meta‐analyses restricted to non‐Hispanic Whites (smaller λ, comparable LD score regression intercepts) ³ , ⁴ , ⁶² , ⁶⁵ , ⁶⁶ , ⁶⁷ while variant associations showed similar levels of heterogeneity. ⁵⁷

4.2. Shared genetic architecture with AD

Most of the AAO GWAS signals intersected AD risk loci (14/16, 88%). Three of these signals share the same lead variant as AD GWAS restricted to non‐Hispanic Whites (3/14, 21%; rs6733839 near BIN1, rs7920721 near ECHDC3, and rs34173062 in SHARPIN). This suggests these signals are transferable across populations and phenotypic measures. In contrast, six others lead variants are on the same haplotype (high D’) but with different allele frequencies (r ² < 0.8) than their corresponding AD GWAS lead variant: 6p12.3, 7q22.1, 11q12.2, 11q14.2, 17p11.2, and 19q13.32. In half the cases, the AD GWAS SNP at these loci didn't reach genome‐wide significance, which suggests that fine mapping these loci in diverse populations might identify different drivers of association than those identified in strictly non‐Hispanic White samples. This multi‐ancestry AAO GWAS also uncovered independent signals at three AD GWAS loci: 5q33.3, 7q34‐q35, and 17p11.2. Together, this may help explain the attenuated portability of polygenic risk scores for AD across populations.

4.3. Novel loci associated with AAO of AD

This study identified two loci not previously associated with AD risk and variation at the APOE locus independently associated with AAO of AD. Preliminary research reveals that each novel locus implicates genes and biological processes with established links to AD. The AAO GWAS signal on 11q13.1 implicated ESRRA, the gene encoding estrogen‐related receptor alpha. Agora reports that ESRRA is differentially expressed in the AD brain and involved in the structural stabilization, metal binding and homeostasis, and epigenetic AD domains. ESRRA is also implicated in the pathogenesis of AD by studies of AD microglia, in vitro neuronal cells, and in vivo mouse models of AD. ⁷¹ , ⁷² , ⁷³ The AAO GWAS signal on 18p11.2 implicates RNMT, the gene encoding RNA guanine‐7 methyltransferase. Agora reports both the gene and protein are differentially expressed in AD and involved in both the immune response and synapse AD domains. ⁶⁰ RNMT plays an important role in mRNA modification with m7G, which has been implicated in aging and AD. ⁷⁴ Finally, the AAO GWAS signal on 19q33.3 independent of ε2 and ε4 genotypes is driven by rs73052335, an intronic APOC1 variant associated with APOE regulation (Figure 5). This APOC1 variant is most common in European and most rare in African reference populations and associated with an increased hazard of AD; this suggests it may explain some of the protective effect associated with African local ancestry at APOE. ³³ , ³⁴

4.4. APOE and sex effects on AAO of AD and GWAS signals

Both female sex and ε2 allele dosage were consistently associated with reduced hazard of AD, while ε4 dosage was associated with a strong and consistent increased hazard of AD. ε2 and ε4 genotypes acted as precision variables in this study, pulling suggestive signals up to significance and shining a light on SNPs within the shadow of APOE. The impact of including precision variables or non‐confounding covariates differs across regression strategies: while adjustment increases precision in linear regression, non‐confounding covariate adjustment can reduce power in logistic regression analyses of case/control status. We therefore recommend, when possible, researchers compare GWAS results from analyses with and without APOE adjustment. In this study, APOE covariate adjustment influenced effect size estimates across the genome. This is consistent with prior AD GWAS, which revealed significant associations that were sharply attenuated in a secondary analysis either adjusting for or stratifying by ε4. ²² , ²⁸ , ⁶⁹ Rather than a biological interaction, this phenomenon is likely explained by the strong association between ε2 and ε4 with fine‐scale population structure in Europe and across the globe. ⁷⁵ In contrast, adjusting for sex as a covariate had minimal effect on variant effect sizes across the genome. This indicates that the observed changes in effect size with APOE adjustment are not driven by changes in the number of covariates alone.

4.5. Limitations and future directions

This study has several limitations. Our reliance on imputed data with strict QC eliminated a substantial number of variants (e.g., 19% in the discovery data harmonization). This prevented our study from investigating some known association signals, such as the ABCA7 locus, which was significant in the discovery data but was missing too many genotypes to pass QC in the replication data. This missing data problem severely limits our ability to fine‐map the variant or variants driving the observed association signals. Our meta‐analysis strategy relied on the RE2 model to allow for heterogeneity expected when analyzing imputed markers and diverse data sets, but it can lead to association signals with opposing directions of effect across studies. Age at diagnosis is difficult to precisely measure, and age at last evaluation was not always captured in older data sets. Further analysis of sequence‐level variation in large, diverse data sets ascertained for AD are necessary to determine whether these meta‐analysis signals represent true disease associations rather than background noise. Together, these limitations underscore the need for large, diverse cohorts such as the Alzheimer's Disease Sequencing Project (ADSP) and All of Us, which pair genome sequence and robust phenotypic data to adequately power discovery and validation efforts. ²⁰

CONFLICT OF INTEREST STATEMENT

The authors report no disclosures relevant to the manuscript. Author disclosures are available in the supporting information.

Supporting information

Supporting information

Supporting information

Supporting information

ADGC Collaborators

1.

Farid Rajabli^1,2, Penelope Benchek³, Giuseppe Tosto^4,5, Gyungah R. Jun^6,7,8, Christiane Reitz^9,10,5, Nicholas Kushch², Jin Sha¹¹, Katrina Bazemore¹¹, Congcong Zhu⁶, Wan‐Ping Lee¹², Jacob Haut¹¹, Kara L. Hamilton‐Nelson², Nicholas R. Wheeler^3,13, Yi Zhao¹², John J. Farrell⁶, Michelle A. Grunin³, Yuk Yee Leung¹², Pavel P. Kuksa¹², Donghe Li⁶, Eder Lucio da Fonseca², Jesse B. Mez¹⁴, Ellen L. Palmer³, Jagan Pillai¹⁵, Richard M. Sherva¹⁶, Yeunjoo E. Song^13,3, Xiaoling Zhang^6,7, Takeshi Ikeuchi¹⁷, Taha Iqbal¹¹, Omkar Pathak¹², Otto Valladares¹², Dolly Reyes‐Dumeyer^4,5, Amanda B. Kuzma¹², Erin Abner¹⁸, Larry D. Adams¹, Perrie M. Adams¹⁹, Alyssa Aguirre²⁰, Marilyn S. Albert²¹, Roger L. Albin^22,23,24, Mariet Allen²⁵, Lisa Alvarez²⁶, Liana G. Apostolova^27,28, Steven E. Arnold²⁹, Sanjay Asthana^30,31,32, Craig S. Atwood^30,31,32, Stanford Auerbach¹⁴, Gayle Ayres²⁰, Clinton T. Baldwin⁶, Robert C. Barber²⁶, Lisa L. Barnes^33,34,35, Sandra Barral^4,10,5, Thomas G. Beach³⁶, James T. Becker³⁷, Gary W. Beecham², Duane Beekly³⁸, Bruno A. Benitez³⁹, David Bennett^33,35, John Bertelson⁴⁰, Thomas D. Bird^41,42, Deborah Blacker^43,44, Bradley F. Boeve⁴⁵, James D. Bowen⁴⁶, Adam Boxer⁴⁷, James Brewer⁴⁸, James R. Burke⁴⁹, Jeffrey M. Burns⁵⁰, Joseph D. Buxbaum^51,52,53, Nigel J. Cairns⁵⁴, Laura B. Cantwell¹², Chuanhai Cao⁵⁵, Christopher S. Carlson⁵⁶, Cynthia M. Carlsson^32,31, Regina M. Carney⁵⁷, Minerva M. Carrasquillo²⁵, Scott Chasse⁵⁸, Marie‐Francoise Chesselet⁵⁹, Nathaniel A. Chin^30,31, Helena C. Chui⁶⁰, Jaeyoon Chung⁶, Suzanne Craft⁶¹, Paul K. Crane⁶², David H. Cribbs⁶³, Elizabeth A. Crocco⁶⁴, Carlos Cruchaga^65,66, Michael L. Cuccaro^2,1, Munro Cullum¹⁹, Eveleen Darby⁶⁷, Barbara Davis⁶⁸, Philip L. De Jager⁶⁹, Charles DeCarli⁷⁰, John DeToledo⁷¹, Malcolm Dick⁷², Dennis W. Dickson²⁵, Beth A. Dombroski¹², Rachelle S. Doody⁶⁷, Ranjan Duara⁷³, NIlüfer Ertekin‐Taner^25,74, Denis A. Evans⁷⁵, Kelley M. Faber⁷⁶, Thomas J. Fairchild⁷⁷, Kenneth B. Fallon⁷⁸, David W. Fardo⁷⁹, Martin R. Farlow⁸⁰, Victoria Fernandez‐Hernandez³⁹, Steven Ferris⁸¹, Robert P. Friedland⁸², Tatiana M. Foroud⁷⁶, Matthew P. Frosch⁸³, Brian Fulton‐Howard⁸⁴, Douglas R. Galasko⁴⁸, Adriana Gamboa^85,86, Marla Gearing^87,88, Daniel H. Geschwind⁵⁹, Bernardino Ghetti⁸⁹, John R. Gilbert^2,1, Rodney C.P. Go⁷⁸, Alison M. Goate⁵¹, Thomas J. Grabowski^42,90, Neill R. Graff‐Radford^25,74, Robert C. Green⁹¹, John H. Growdon⁹², Hakon Hakonarson^93,94, James Hall²⁶, Ronald L. Hamilton⁹⁵, Oscar Harari⁶⁶, John Hardy^96,97, Lindy E. Harrell⁹⁸, Elizabeth Head⁹⁹, Victor W. Henderson^100,101, Michelle Hernandez⁷¹, Timothy Hohman^102,103, Lawrence S. Honig⁴, Ryan M. Huebinger¹⁰⁴, Matthew J. Huentelman¹⁰⁵, Christine M. Hulette¹⁰⁶, Bradley T. Hyman⁹², Linda S. Hynan^19,107,108, Laura Ibanez^109,110, Gail P. Jarvik^111,112, Suman Jayadev⁴², Lee‐Way Jin¹¹³, Kim Johnson⁷¹, Leigh Johnson⁸⁵, M. Ilyas Kamboh^114,115,116, Anna M. Karydas⁴⁷, Mindy J. Katz¹¹⁷, John S. Kauwe^118,119, Jeffrey A. Kaye^120,121, C. Dirk Keene¹²², Aisha Khaleeq⁶⁷, Masataka Kikuchi¹⁷, Ronald Kim⁹⁹, Janice Knebl⁸⁵, Neil W. Kowall^14,123, Joel H. Kramer¹²⁴, Walter A. Kukull¹²⁵, Frank M. LaFerla¹²⁶, James J. Lah¹²⁷, Eric B. Larson¹²⁸, Alan Lerner³, James B. Leverenz¹⁵, Allan I. Levey¹²⁷, Andrew P. Lieberman¹²⁹, Richard B. Lipton¹¹⁷, Mark Logue^6,130,131, Oscar L. Lopez³⁷, Kathryn L. Lunetta⁷, Constantine G. Lyketsos¹³², Douglas Mains^85,86, Flanagan E. Margaret^133,134, Daniel C. Marson⁹⁸, Eden R R. Martin^2,1, Frank Martiniuk¹³⁵, Deborah C. Mash¹³⁶, Eliezer Masliah^48,137, Paul Massman⁶⁷, Arjun Masurkar⁸¹, Wayne C. McCormick⁶², Susan M. McCurry¹³⁸, Andrew N. McDavid⁵⁶, Stefan McDonough¹³⁹, Ann C. McKee^14,140, Marsel Mesulam^133,134, Bruce L. Miller¹⁴¹, Carol A. Miller¹⁴², Joshua W. Miller¹¹³, Thomas J. Montine¹⁴³, Edwin S. Monuki¹⁴⁴, John C. Morris^{54,145,146,110}, Shubhabrata Mukherjee⁶², Amanda J. Myers⁶⁴, Trung Nguyen¹⁰⁷, Obisesan Thomas Thomas Obisesan ADGC¹⁴⁷, Sid O'Bryant¹⁴⁸, John M. Olichney¹⁴⁹, Marcia Ory¹⁵⁰, Raymond Palmer¹⁵¹, Joseph E. Parisi¹⁵², Henry L. Paulson^22,24, Valory Pavlik⁶⁷, David Paydarfar²⁰, Victoria Perez⁷¹, Elaine Peskind¹⁵³, Ronald C. Petersen⁴⁵, Helen Petrovitch¹⁵⁴, Aimee Pierce⁶³, Marsha Polk¹⁵¹, Wayne W. Poon⁷², Huntington Potter¹⁵⁵, Liming Qu¹², Mary Quiceno^156,157, Joseph F. Quinn^120,121, Ashok Raj⁵⁵, Murray Raskind¹⁵³, Eric M. Reiman^{105,158,159,160}, Barry Reisberg^161,81, Joan S. Reisch⁶⁸, John M. Ringman¹⁶², Erik D. Roberson⁹⁸, Monica Rodriguear⁶⁷, Ekaterina Rogaeva¹⁶³, Howard J. Rosen⁴⁷, Roger N. Rosenberg¹⁰⁷, Donald R. Royall¹⁶⁴, Marwan Sabbagh¹⁶⁵, A. Dessa Sadovnick¹⁶⁶, Mark A. Sager³¹, Mary Sano⁵³, Andrew J. Saykin^76,167, Julie A. Schneider^33,35,168, Lon S. Schneider^169,60, William W. Seeley⁴⁷, Susan H. Slifer², Scott Small^4,5, Amanda G. Smith⁵⁵, Janet P. Smith⁶⁸, Joshua A. Sonnen¹²², Salvatore Spina⁸⁹, Peter St George‐Hyslop^170,171, Takiyah D. Starks^172,173, Robert A. Stern¹⁴, Alan B. Stevens^174,175,176, Stephen M. Strittmatter¹⁷⁷, David Sultzer¹⁷⁸, Russell H. Swerdlow⁵⁰, Rudolph E. Tanzi⁹², Jeffrey L. Tilson¹⁷⁹, John Q. Trojanowski¹², Juan C. Troncoso¹⁸⁰, Magda Tsolaki¹⁸¹, Debby W. Tsuang^41,153, Vivianna M. Van Deerlin¹², Linda J. van Eldik¹⁸², Jeffery M. Vance^1,2, Badri N. Vardarajan⁴, Robert Vassar^133,134, Harry V. Vinters^183,162, Jean‐Paul Vonsattel⁴, Sandra Weintraub¹⁸⁴, Kathleen A. Welsh‐Bohmer^49,185, Patrice L. Whitehead², Kirk C. Wilhelmsen⁵⁸, Benjamin Williams¹⁸⁶, Jennifer Williamson⁴, Henrik Wilms⁷¹, Thomas S. Wingo¹²⁷, Thomas Wisniewski^187,188, Randall L. Woltjer¹⁸⁹, Martin Woon⁴⁰, Clinton B. Wright¹⁹⁰, Chuang‐Kuo Wu⁷¹, Steven G. Younkin^25,74, Chang‐En Yu⁶², Lei Yu^33,35, Xiongwei Zhu¹⁹¹, Brian W. Kunkle^1,2, William S. Bush^3,13, Miyashita Akinori¹⁷, Byrd Goldie¹⁹², Li‐San Wang¹², Lindsay A. Farrer^{16,14,8,193,7}, Jonathan L. Haines^3,13, Richard Mayeux⁴, Margaret A. Pericak‐Vance^1,2, Gerard D. Schellenberg¹²

¹Dr. John T. Macdonald Foundation Department of Human Genetics, Miller School of Medicine, University of Miami, Miami, Florida, USA; ²The John P. Hussman Institute for Human Genomics, University of Miami, Miami, Florida, USA; ³Department of Population and Quantitative Health Sciences, Case Western Reserve University, Cleveland, Ohio, USA; ⁴Taub Institute for Research in Alzheimer's Disease and the Aging Brain, The Gertrude H. Sergievsky Center, Department of Neurology, Columbia University, New York, New York, USA; ⁵Department of Neurology, Columbia University, New York, New York, USA; ⁶Department of Medicine (Biomedical Genetics), Boston University Chobanian & Avedisian School of Medicine, Boston, Massachusetts, USA; ⁷Department of Biostatistics, Boston University School of Public Health, Boston, Massachusetts, USA; ⁸Department of Ophthalmology, Boston University Chobanian & Avedisian School of Medicine, Boston, Massachusetts, USA; ⁹Department of Epidemiology, Columbia University, New York, New York, USA; ¹⁰Gertrude H. Sergievsky Center, Columbia University, New York, New York, USA; ¹¹Department of Biostatistics, Epidemiology, and Informatics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, USA; ¹²Penn Neurodegeneration Genomics Center, Department of Pathology and Laboratory Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, USA; ¹³Cleveland Institute for Computational Biology, Case Western Reserve University, Cleveland, Ohio, USA; ¹⁴Department of Neurology, Boston University Chobanian & Avedisian School of Medicine, Boston, Massachusetts, USA; ¹⁵Cleveland Clinic Lou Ruvo Center for Brain Health, Cleveland Clinic, Cleveland, Ohio, USA; ¹⁶Department of Medicine (Biomedical Genetics), Boston University Chobanian & Avedisian School of Medicine, Boston, Massachusetts, USA; ¹⁷Molecular Genetics Division, Brain Research Institute, Niigata University, Niigata, Japan; ¹⁸Sanders‐Brown Center on Aging, Department of Epidemiology, College of Public Health, University of Kentucky, Lexington, Kentucky, USA; ¹⁹Department of Psychiatry, University of Texas Southwestern Medical Center, Dallas, Texas, USA; ²⁰Department of Neurology, Dell Medical School, University of Texas at Austin, Austin, Texas, USA; ²¹Department of Neurology, Johns Hopkins University, Baltimore, Maryland, USA; ²²Department of Neurology, University of Michigan, Ann Arbor, Michigan, USA; ²³Geriatric Research, Education and Clinical Center (GRECC), VA Ann Arbor Healthcare System (VAAAHS), Ann Arbor, Michigan, USA; ²⁴Michigan Alzheimer's Disease Center, University of Michigan, Ann Arbor, Michigan, USA; ²⁵Department of Neuroscience, Mayo Clinic, Jacksonville, Florida, USA; ²⁶Department of Pharmacology and Neuroscience, University of North Texas Health Science Center, Fort Worth, Texas, USA; ²⁷Departments of Neurology, Radiology, and Medical and Molecular Genetics, Indiana University School of Medicine, Indianapolis, Indiana, USA; ²⁸Indiana Alzheimer's Disease Research Center, Indiana University School of Medicine, Indianapolis, Indiana, USA; ²⁹Department of Psychiatry, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, USA; ³⁰Geriatric Research, Education and Clinical Center (GRECC), University of Wisconsin, Madison, Wisconsin, USA; ³¹Department of Medicine, University of Wisconsin, Madison, Wisconsin, USA; ³²Wisconsin Alzheimer's Disease Research Center, Madison, Wisconsin, USA; ³³Department of Neurological Sciences, Rush University Medical Center, Chicago, Illinois, USA; ³⁴Department of Behavioral Sciences, Rush University Medical Center, Chicago, Illinois, USA; ³⁵Rush Alzheimer's Disease Center, Rush University Medical Center, Chicago, Illinois, USA; ³⁶Civin Laboratory for Neuropathology, Banner Sun Health Research Institute, Phoenix, Arizona, USA; ³⁷Departments of Psychiatry, Neurology, and Psychology, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania, USA; ³⁸National Alzheimer's Coordinating Center, University of Washington, Seattle, Washington, USA; ³⁹Department of Psychiatry and Hope Center Program on Protein Aggregation and Neurodegeneration, Washington University School of Medicine, St. Louis, Missouri, USA; ⁴⁰Department of Psychiatry, University of Texas at Austin/Dell Medical School, Austin, Texas, USA; ⁴¹VA Puget Sound Health Care System/GRECC, Seattle, Washington, USA; ⁴²Department of Neurology, University of Washington, Seattle, Washington, USA; ⁴³Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts, USA; ⁴⁴Department of Psychiatry, Massachusetts General Hospital/Harvard Medical School, Boston, Massachusetts, USA; ⁴⁵Department of Neurology, Mayo Clinic, Rochester, Minnesota, USA; ⁴⁶Swedish Medical Center, Seattle, Washington, USA; ⁴⁷Department of Neurology, University of California San Francisco, San Francisco, California, USA; ⁴⁸Department of Neurosciences, University of California San Diego, La Jolla, California, USA; ⁴⁹Department of Medicine, Duke University, Durham, North Carolina, USA; ⁵⁰University of Kansas Alzheimer's Disease Center, University of Kansas Medical Center, Kansas City, Kansas, USA; ⁵¹Department of Genetics and Genomic Sciences, Ronald M. Loeb Center for Alzheimer's Disease, Icahn School of Medicine at Mount Sinai, New York, New York, USA; ⁵²Department of Neuroscience, Icahn School of Medicine at Mount Sinai, New York, New York, USA; ⁵³Department of Psychiatry, Mount Sinai School of Medicine, New York, New York, USA; ⁵⁴Department of Pathology and Immunology, Washington University, St. Louis, Missouri, USA; ⁵⁵USF Health Byrd Alzheimer's Institute, University of South Florida, Tampa, Florida, USA; ⁵⁶Fred Hutchinson Cancer Research Center, Seattle, Washington, USA; ⁵⁷Mental Health and Behavioral Science Service, Bruce W. Carter VA Medical Center, Miami, Florida, USA; ⁵⁸Department of Genetics, University of North Carolina Chapel Hill, Chapel Hill, North Carolina, USA; ⁵⁹Neurogenetics Program, University of California Los Angeles, Los Angeles, California, USA; ⁶⁰Department of Neurology, University of Southern California, Los Angeles, California, USA; ⁶¹Section of Gerontology and Geriatric Medicine Research, Wake Forest School of Medicine, Winston‐Salem, North Carolina, USA; ⁶²Department of Medicine, University of Washington, Seattle, Washington, USA; ⁶³Department of Neurology, University of California Irvine, Irvine, California, USA; ⁶⁴Department of Psychiatry and Behavioral Sciences, Miller School of Medicine, University of Miami, Miami, Florida, USA; ⁶⁵NeuroGenomics and Informatics, Washington University, St. Louis, Missouri, USA; ⁶⁶Department of Psychiatry, Washington University in St. Louis, St. Louis, Missouri, USA; ⁶⁷Alzheimer's Disease and Memory Disorders Center, Baylor College of Medicine, Houston, Texas, USA; ⁶⁸Department of Population and Data Sciences, University of Texas Southwestern Medical Center, Dallas, Texas, USA; ⁶⁹Center for Translational and Computational Neuroimmunology, Department of Neurology, Columbia University Medical Center, New York, New York, USA; ⁷⁰Department of Neurology, University of California Davis, Sacramento, California, USA; ⁷¹Departments of Neurology, Pharmacology and Neuroscience, Texas Tech University Health Science Center, Lubbock, Texas, USA; ⁷²Institute for Memory Impairments and Neurological Disorders, University of California Irvine, Irvine, California, USA; ⁷³Wien Center for Alzheimer's Disease and Memory Disorders, Mount Sinai Medical Center, Miami Beach, Florida, USA; ⁷⁴Department of Neurology, Mayo Clinic, Jacksonville, Florida, USA; ⁷⁵Rush Institute for Healthy Aging, Department of Internal Medicine, Rush University Medical Center, Chicago, Illinois, USA; ⁷⁶Department of Medical and Molecular Genetics, Indiana University, Indianapolis, Indiana, USA; ⁷⁷Office of Strategy and Measurement, University of North Texas Health Science Center, Fort Worth, Texas, USA; ⁷⁸Department of Pathology, University of Alabama at Birmingham, Birmingham, Alabama, USA; ⁷⁹Sanders‐Brown Center on Aging, Department of Biostatistics, College of Public Health, University of Kentucky, Lexington, Kentucky, USA; ⁸⁰Department of Neurology, Indiana University, Indianapolis, Indiana, USA; ⁸¹Department of Psychiatry, New York University, New York, New York, USA; ⁸²Department of Neurology, University of Louisville School of Medicine, Louisville, KY, USA; ⁸³C.S. Kubik Laboratory for Neuropathology, Massachusetts General Hospital, Charlestown, Massachusetts, USA; ⁸⁴Department of Neuroscience, Ronald M. Loeb Center for Alzheimer's Disease, Icahn School of Medicine at Mount Sinai, New York, New York, USA; ⁸⁵Department of Health Behavior and Health Systems, University of North Texas Health Science Center, Fort Worth, Texas, USA; ⁸⁶Department of Health Management and Policy, School of Public Health, University of North Texas Health Science Center, Fort Worth, Texas, USA; ⁸⁷Department of Pathology and Laboratory Medicine, Emory University, Atlanta, Georgia, USA; ⁸⁸Emory Alzheimer's Disease Center, Emory University, Atlanta, Georgia, USA; ⁸⁹Department of Pathology and Laboratory Medicine, Indiana University, Indianapolis, Indiana, USA; ⁹⁰Department of Radiology, University of Washington, Seattle, Washington, USA; ⁹¹Division of Genetics, Department of Medicine and Partners Center for Personalized Genetic Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts, USA; ⁹²Department of Neurology, Massachusetts General Hospital/Harvard Medical School, Boston, Massachusetts, USA; ⁹³Center for Applied Genomics, Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, USA; ⁹⁴Division of Human Genetics, Department of Pediatrics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, USA; ⁹⁵Department of Pathology (Neuropathology), University of Pittsburgh, Pittsburgh, Pennsylvania, USA; ⁹⁶UCL Institute of Neurology, University College London, London, UK; ⁹⁷Department of Molecular Neuroscience, UCL Institute of Neurology, University College London, London, UK; ⁹⁸Department of Neurology, University of Alabama at Birmingham, Birmingham, Alabama, USA; ⁹⁹Department of Pathology and Laboratory Medicine, University of California Irvine, Irvine, California, USA; ¹⁰⁰Department of Epidemiology and Population Health, Stanford University, Stanford, California, USA; ¹⁰¹Department of Neurology and Neurological Sciences, Stanford University, Stanford, California, USA; ¹⁰²Vanderbilt Memory and Alzheimer's Center, Department of Neurology, Vanderbilt University Medical Center, Nashville, Tennessee, USA; ¹⁰³Vanderbilt Genetics Institute, Division of Genetic Medicine, Department of Medicine, Vanderbilt University Medical Center, Nashville, Tennessee, USA; ¹⁰⁴Department of Surgery, University of Texas Southwestern Medical Center, Dallas, Texas, USA; ¹⁰⁵Neurogenomics Division, Translational Genomics Research Institute, Phoenix, Arizona, USA; ¹⁰⁶Department of Pathology, Duke University, Durham, North Carolina, USA; ¹⁰⁷Department of Neurology, University of Texas Southwestern Medical Center, Dallas, Texas, USA; ¹⁰⁸Department of Neurological Surgery, University of Texas Southwestern Medical Center, Dallas, Texas, USA; ¹⁰⁹Department of Psychiatry, Washington University School of Medicine, St. Louis, Missouri, USA; ¹¹⁰Hope Center Program on Protein Aggregation and Neurodegeneration, Washington University School of Medicine, St. Louis, Missouri, USA; ¹¹¹Department of Genome Sciences, University of Washington, Seattle, Washington, USA; ¹¹²Department of Medicine (Medical Genetics), University of Washington, Seattle, Washington, USA; ¹¹³Department of Pathology and Laboratory Medicine, University of California Davis, Sacramento, California, USA; ¹¹⁴Department of Psychiatry, University of Pittsburgh, Pittsburgh, Pennsylvania, USA; ¹¹⁵Department of Human Genetics, University of Pittsburgh, Pittsburgh, Pennsylvania, USA; ¹¹⁶Alzheimer's Disease Research Center, University of Pittsburgh, Pittsburgh, Pennsylvania, USA; ¹¹⁷Department of Neurology, Albert Einstein College of Medicine, New York, New York, USA; ¹¹⁸Department of Neuroscience, Brigham Young University, Provo, Utah, USA; ¹¹⁹Department of Biology, Brigham Young University, Provo, Utah, USA; ¹²⁰Department of Neurology, Oregon Health and Science University, Portland, Oregon, USA; ¹²¹Department of Neurology, Portland Veterans Affairs Medical Center, Portland, Oregon, USA; ¹²²Department of Laboratory Medicine and Pathology, University of Washington, Seattle, Washington, USA; ¹²³Department of Pathology, Boston University, Boston, Massachusetts, USA; ¹²⁴Department of Neuropsychology, University of California San Francisco, San Francisco, California, USA; ¹²⁵Department of Epidemiology, University of Washington, Seattle, Washington, USA; ¹²⁶Department of Neurobiology and Behavior, University of California Irvine, Irvine, California, USA; ¹²⁷Department of Neurology, Emory University, Atlanta, Georgia, USA; ¹²⁸Kaiser Permanente Washington Health Research Institute, Seattle, Washington, USA; ¹²⁹Department of Pathology, University of Michigan, Ann Arbor, Michigan, USA; ¹³⁰National Center for PTSD at Boston VA Healthcare System, Boston, Massachusetts, USA; ¹³¹Department of Psychiatry, Boston University Chobanian & Avedisian School of Medicine, Boston, Massachusetts, USA; ¹³²Department of Psychiatry, Johns Hopkins University, Baltimore, Maryland, USA; ¹³³Department of Pathology, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA; ¹³⁴Cognitive Neurology and Alzheimer's Disease Center, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA; ¹³⁵Department of Medicine ‐ Pulmonary, New York University, New York, New York, USA; ¹³⁶Department of Neurology, Miller School of Medicine, University of Miami, Miami, Florida, USA; ¹³⁷Department of Pathology, University of California San Diego, La Jolla, California, USA; ¹³⁸School of Nursing Northwest Research Group on Aging, University of Washington, Seattle, Washington, USA; ¹³⁹Pfizer Worldwide Research and Development, New York, New York, USA; ¹⁴⁰Department of Pathology, Boston University Chobanian & Avedisian School of Medicine, Boston, Massachusetts, USA; ¹⁴¹Weill Institute for Neurosciences, Memory and Aging Center, University of California San Francisco, San Francisco, California, USA; ¹⁴²Department of Pathology, University of Southern California, Los Angeles, California, USA; ¹⁴³Department of Pathology, Stanford University School of Medicine, Stanford, California, USA; ¹⁴⁴Department of Pathology and Laboratory Medicine and Alzheimer's Disease Research Center, University of California Irvine, Irvine, California, USA; ¹⁴⁵Department of Neurology, Washington University, St. Louis, Missouri, USA; ¹⁴⁶Department of Psychiatry, Washington University School of Medicine, St. Louis Missouri, USA; ¹⁴⁷Department of Research Regulatory Compliance, College of Medicine, Howard University, Washington, DC, USA, tobisesan@Howard.edu; ¹⁴⁸Institute for Translational Research, University of North Texas Health Science Center, Fort Worth, Texas, USA; ¹⁴⁹Center for Mind and Brain and Department of Neurology, University of California Davis, Sacramento, California, USA; ¹⁵⁰Center for Population Health and Aging, Texas A&M University Health Science Center, Lubbock Texas, USA; ¹⁵¹Department of Family and Community Medicine, University of Texas Health Science Center San Antonio, San Antonio, Texas, USA; ¹⁵²Department of Laboratory Medicine and Pathology, Mayo Clinic, Rochester, Minnesota, USA; ¹⁵³Department of Psychiatry and Behavioral Sciences, University of Washington School of Medicine, Seattle, Washington, USA; ¹⁵⁴Pacific Health Research & Education Institute, VA Pacific Islands Healthcare System, Honolulu, Hawaii, USA; ¹⁵⁵Department of Neurology, University of Colorado School of Medicine, Aurora, Colorado, USA; ¹⁵⁶Department of Internal Medicine and Geriatrics, University of North Texas Health Science Center, Fort Worth, Texas, USA; ¹⁵⁷Department of Medical Education, TCU/UNTHSC School of Medicine, Fort Worth, Texas;, ¹⁵⁸Arizona Alzheimer's Consortium, Phoenix, Arizona, USA; ¹⁵⁹Banner Alzheimer's Institute, Phoenix, Arizona, USA; ¹⁶⁰Department of Psychiatry, University of Arizona, Phoenix, Arizona, USA; ¹⁶¹Alzheimer's Disease Center, New York University, New York, New York, USA; ¹⁶²Department of Neurology, University of California Los Angeles, Los Angeles, California, USA; ¹⁶³Tanz Centre for Research in Neurodegenerative Disease, University of Toronto, Toronto, Ontario, Canada; ¹⁶⁴Departments of Psychiatry, Medicine, Family and Community Medicine, and the Glenn Biggs Institute for Alzheimer's and Neurodegenerative Diseases, UT Health Science Center at San Antonio, San Antonio, Texas, USA; ¹⁶⁵Department of Neurology, Barrow Neurological Institute St. Joseph's Hospital and Medical Center, Phoenix, Arizona, USA; ¹⁶⁶Department of Medical Genetics, University of British Columbia, Vancouver, Canada; ¹⁶⁷Department of Radiology and Imaging Sciences, Indiana University, Indianapolis, Indiana, USA; ¹⁶⁸Department of Pathology (Neuropathology), Rush University Medical Center, Chicago, Illinois, USA; ¹⁶⁹Department of Psychiatry, University of Southern California, Los Angeles, California, USA; ¹⁷⁰Cambridge Institute for Medical Research, University of Cambridge, Cambridge, UK; ¹⁷¹Faculty of Medicine, Department of Medicine (Neurology), University of Toronto, Toronto, Ontario, Canada; ¹⁷²Maya Angelou Center for Health Equity, Wake Forest School of Medicine, Winston‐Salem, North Carolina, USA; ¹⁷³Center for Outreach in Alzheimer's, Aging and Community Health at North Carolina A&T State University, Greensboro, North Carolina, USA; ¹⁷⁴Center for Applied Health Research, Baylor Scott & White Health, Temple, Texas, USA; ¹⁷⁵Center for Population Health and Aging, Texas A&M University Health Science Center, Lubbock Texas, USA; ¹⁷⁶College of Medicine, Texas A&M University Health Science Center, College Station, Texas, USA; ¹⁷⁷Program in Cellular Neuroscience, Neurodegeneration and Repair, Yale University School of Medicine, New Haven, Connecticut, USA;, ¹⁷⁸Department of Psychiatry and Human Behavior, University of California Irvine, Irvine, California, USA; ¹⁷⁹Renaissance Computing Institute, University of North Carolina Chapel Hill, Chapel Hill, North Carolina, USA; ¹⁸⁰Department of Pathology, Johns Hopkins University, Baltimore, Maryland, USA; ¹⁸¹Department of Neurology, Aristotle University of Thessaloniki, Thessaloniki, Macedonia; ¹⁸²Sanders‐Brown Center on Aging, Department of Neuroscience, College of Medicine, University of Kentucky, Lexington, Kentucky, USA; ¹⁸³Department of Pathology and Laboratory Medicine, University of California Los Angeles, Los Angeles, California, USA; ¹⁸⁴Department of Psychiatry and Behavioral Sciences, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA; ¹⁸⁵Department of Psychiatry and Behavioral Sciences, Duke University, Durham, North Carolina, USA; ¹⁸⁶Department of Neurology, Section of Gerontology and Geriatric Medicine Research, Wake Forest School of Medicine, Winston‐Salem, North Carolina, USA; ¹⁸⁷Department of Psychiatry, New York University Grossman School of Medicine, New York, New York, USA; ¹⁸⁸Center for Cognitive Neurology and Departments of Neurology and Pathology, New York University Grossman School of Medicine, New York, USA; ¹⁸⁹Department of Pathology, Oregon Health and Science University, Portland, Oregon, USA; ¹⁹⁰Evelyn F. McKnight Brain Institute, Department of Neurology, Miller School of Medicine, University of Miami, Miami, Florida, USA; ¹⁹¹Department of Pathology, Case Western Reserve University, Cleveland, Ohio, USA; ¹⁹²Social Sciences & Health Policy, Wake Forest School of Medicine, Winston‐Salem, North Carolina, USA; ¹⁹³Department of Epidemiology, Boston University School of Public Health, Boston, Massachusetts, USA.

Blue EE, Broome J, Xue D, et al. Multi‐ancestry meta‐analysis identifies genetic modifiers of age‐at‐onset of Alzheimer's disease at known and novel loci. Alzheimer's Dement. 2025;21:e70489. 10.1002/alz.70489