Novel Loci for Alzheimer Disease Identified by Genome Wide Association Study in Ashkenazi Jews

Abstract

INTRODUCTION:

Most Alzheimer disease (AD) loci were discovered in European ancestry individuals.

METHODS:

We applied principal component analysis using Gaussian mixture models and an Ashkenazi Jewish (AJ) reference genome-wide association study (GWAS) dataset to identify AJs ascertained in GWAS (n=42,682), whole genome sequencing (WGS, n=16,815) and whole exome sequencing (WES, n=20,504) datasets. Association of AD was tested genome-wide (GW) in the GWAS and WGS datasets and exome-wide (EW) in all three datasets (EW). Gene-based analyses were performed using aggregated rare variants.

RESULTS:

In addition to APOE, GW analyses (1,355 cases and 1,661 controls) revealed associations with TREM2 R47H (P=9.66x10⁻⁹), rs541586606 near RAB3B (P=5.01x10⁻⁸) and rs760573036 between SPOCK3 and ANXA10 (P=6.32x10⁻⁸). In EW analyses (1,504 cases and 2,047 controls), study-wide significant association was observed with rs1003710 near SMAP2 (P=1.91x10⁻⁷). A significant gene-based association was identified with GIPR (P=7.34x10⁻⁷).

DISCUSSION:

Our results highlight the efficacy of founder populations for AD genetic studies.

1. Introduction

Alzheimer disease (AD), the most common neurodegenerative disorder in the world, affects individuals of all races and ethnicities; however, most genetic research for AD has been performed in individuals of European ancestry (EAs) [1, 2] with a limited number of large-scale genetic studies in other populations [1, 3-5]. Trans-ethnic studies have shown that population differences in genetic background can be leveraged to make novel discoveries that might require a sample size several orders of magnitude larger to achieve similar success studying a single population [5, 6]. Similarly, studies of small samples from founder populations (i.e., ethnic or religious groups whose origins can be traced to a limited number of ancestors and thus have a more homogeneous genetic background) have successfully detected robust and subsequently validated associations of AD with several genes [6-8].

For many centuries, Ashkenazi Jews (AJs) lived in communities in Eastern Europe and were genetically isolated from their non-Jewish neighbors. Because some rare autosomal recessive disorders manifesting in childhood (including Tay-Sachs disease, Gaucher disease, familial dysautonomia, Canavan disease, Bloom syndrome and spinal muscular atrophy), as well as particular gene mutations conferring high risk of common disorders such as early-onset breast cancer [9] and multiple gastrointestinal cancers [10-13], are found predominantly or at a much higher frequency in AJs compared to other populations, we hypothesized that some AD susceptibility variants are much more frequent, and thus more likely to show statistically significant associations, in an AJ sample compared to much larger and more genetically heterogeneous EA cohorts. We conducted a genome-wide association study (GWAS) of AD in a group of AJs who were discerned from large EA cohorts of AD cases and controls.

2. Methods

An overview of the study design and analysis workflow is shown in Figure 1.

Figure 1.

Overview of study design and analysis workflow. Number of subjects and variants are shown in each step. GWAS = genome-wide association study, WGS = whole genome sequencing, WES = whole exome sequencing, AJs = Ashkenazi Jews.

2.1. Study participants

Genome-wide variant and phenotypic data were obtained from the Azheimer’s Disease Genetic Consortium (ADGC) [14], Genomic Research at Fundació Ace (GR@ACE) [15] and Azheimer’s Disease Sequencing Project (ADSP) [16, 17]. The study sample includes GWAS data from 35 ADGC cohorts (35,273 non-Hispanic EAs [1, 18] and 7,409 Spanish participants of the GR@ACE GWAS study (https://ega-archive.org/studies/EGAS00001003424)). Genotyping in these datasets was performed as previously described [14, 15] and genotypes were imputed using the TOPMed reference panel [19] and retained if the imputation quality score (r²) was >0.3. Genotypes for subjects in the ADSP whole genome sequencing (WGS, n=16,815) and whole exome sequencing (WES, n=20,504) datasets (NG00067.v5) were joint called using its single nucleotide polymorphism (SNP) / Indel Variant Calling Pipeline and data management tool [20]. Ascertainment, diagnostic procedures, and previous studies using these data are described elsewhere [16, 21-23]. Characteristics of the GWAS, WGS and WES datasets are shown in Table S1. Studies of individual cohorts were approved by the appropriate institutional review boards, and all participants provided written informed permission or, for those with substantial cognitive impairment, consent was provided by a caregiver, legal guardian, or other proxy. A GWAS dataset including 3,096 AJs selected from the Hebrew University Genetic Resource (http://hugr.huji.ac.il) for a GWAS of schizophrenia [24] was obtained from dbGaP (phs000448.v1.p1) and used as a reference panel for clustering purposes. These subjects reported that all four grandparents were Ashkenazi.

2.2. Pre-processing AJ reference data

The AJ reference sample was genotyped with Illumina HumanOmni1-Quad arrays with A/B coding. We used GenGen [25] software to convert the coding to ATGC format and liftOver python scripts (https://github.com/knmkr/lift-over-vcf) to convert the genome positions to NCBI build 38. Genotypes were imputed using the TOPMed reference panel [19] and those with r² >0.3 were retained. Reference sample outliers who likely have a genetic background with substantial admixture with non-AJs were identified with Aberrant [26] and excluded from subsequent analyses, after which 2,452 subjects remained for clustering analysis.

2.3. Clustering analysis

Population clustering was conducted sequentially for the AJ GWAS reference dataset combined separately with the GWAS, WGS and WES datasets by principal components (PC) analysis [27] using Gaussian Mixture Models after excluding related individuals in each dataset (GWAS:1,158; WGS:1,114; WES:1,514) identified by IBD ($\widehat{\pi }<0.2$) and variants with missingness >10%, divergence from Hardy-Weinberg Equilibrium (P<10⁻⁶), and minor allele frequency (MAF) <5%. The genetic profile of each individual was summarized as a PC coordinate and assigned to k Gaussian distributions, each representing a subpopulation (cluster) with a certain mean vector and variance matrix, that were determined based on the individuals assigned to the cluster using the Expectation-Maximization algorithm implemented in the Python sklearn.mixture package [28]. The optimal k was determined based on the minimum number of Gaussian distributions that maximized the number of reference AJ subjects in a single cluster. Individuals assigned to the subpopulation containing the AJ reference subjects were included in subsequent analyses, noting that none of the related individuals in the parent GWAS, WGS and WES datasets who were excluded from the clustering procedure were related to the sample identified as AJ. We also applied the ancestry estimation method implemented in the ADMIXTURE software [29] to assess the effect of particular clustering approaches on the group of individuals identified as AJ.

2.4. Association Analysis

To maximize power, genome-wide (GW) and exome-wide (EW) datasets were created by combining the AD datasets according to the type of variants they contained. The GW dataset included AJs in the GWAS and WGS datasets, and the EW dataset included AJs in all three datasets and variants called from exome capture. Duplicate individuals across datasets were identified by IBD analysis ($\widehat{\pi }<0.95$) and the genetic data retained for analysis were selected according to the following priority scheme: GW analyses: WGS > GWAS; EW analyses: WGS > WES > GWAS. Separate quality control (QC) criteria were applied to the GW and EW datasets to account for the distinct types of genotype data. PLINK [30] was used to filter out individuals with >10% missing genotypes and variants with >10% missing or not in Hardy-Weinberg equilibrium (HWE, p<1.0x10⁻⁶ in the GW dataset, and variants that were monomorphic or not in HWE in the EW dataset. After QC, the GW dataset included 13,135,971 variants and 3,016 individuals, and the EW dataset included 562,040 variants and 3,551 individuals. Among these individuals, 2,044 were common to both datasets.

The association of AD with each bi-allelic common variant (MAF ≥ 0.01) was assessed in PLINK using logistic regression for an additive model including covariates for sex, age, and the first four PCs that were recalculated in the AJ dataset. A genome-wide significant (GWS) threshold was set at P=5.0x10⁻⁸ for the GW analyses and a study-wide significance (SWS) threshold was calculated as P=2.74 x10⁻⁷ for the EW analyses based on a Bonferroni correction for the number of tested variants. Gene-based rare variant association was tested for genes with ≥ 2 variants after excluding variants with MAF ≥ 0.01 using the same models as described above and the SKAT-O program in RVTESTS [31]. The Bonferroni-corrected significance threshold was set at P=2.13x10⁻⁶ for GW analyses and P=1.77x10⁻⁶ for EW analyses based on the number of tested genes. Regional visualization of genome-wide and exome-wide association scan results were visualized using LocusZoom [32].

2.5. Differential gene expression

Differential expression of genes at top-ranked loci was evaluated in dorsolateral prefrontal cortex area tissue from 627 participants (380 autopsy-confirmed AD cases and 247 controls) of the Religious Orders Study and Rush Memory and Aging Project, temporal cortex area tissue from 162 participants (82 autopsy-confirmed AD cases and 80 controls) of the Mayo Clinic Study of Aging, and frontal cortex tissue from 208 participants (64 autopsy-confirmed AD cases and 129 controls) of the Framingham Heart Study and Boston University Alzheimer’s Disease Research Center. Details regarding RNA sequencing, quality control procedures, quantification of gene expression, and differential gene expression analysis are described elsewhere [33]. Results across datasets were combined by a sample size-weighted meta-analysis with log2 of fold change (logFC) as direction using the software METAL [34].

3. Results

3.1. Identification of Ashkenazi Jews in Multi-ethnic Cohorts

PC analysis conducted for each AD dataset combined with the AJ reference GWAS dataset revealed a subset of 5,169 individuals (GWAS −3,150; WGS - 866; WES - 1,153) who most closely aligned with the AJ reference group contained in one of five clusters (Figure 2), many of whom were identified in more than one dataset (Figure 3). AJs accounted for approximately 7.3% of individuals in the total GWAS dataset comprised of EAs only, but a slightly smaller proportion in the WGS (5.2%) and WES (5.6%) datasets which included cohorts of African Americans and Caribbean Hispanics who have substantial European and African ancestry. After excluding 574 and 1,170 duplicate individuals from the GW and EW datasets, respectively, and 426 and 448 individuals from the GW and EW datasets, respectively, who didn’t pass genotype and phenotype QC filtering, the final sample for the genome-wide analysis contained 1,355 AD cases and 1,661 controls, and for the exome-wide analysis included 1,504 AD cases and 2,047 controls (Table 1). Application of the ADMIXTURE method for identifying AJs resulted in a highly overlapping sample suggesting that the results from the clustering analysis using GMM are robust (Figure S1).

Figure 2.

3-dimensional principal component (PC) plots showing clustering of Ashkenazi Jews (AJs) in the GWAS, whole genome sequencing (WGS) and whole exome sequencing (WES) datasets. Distinct population clusters are color coded. Red dots indicate the AJ reference subjects. Yellow dots indicate the clustered AJ samples from each dataset.

Figure 3.

Venn diagram showing the overlap of Ashkenazi Jews among the GWAS, whole genome sequencing (WGS) and whole exome sequencing (WES) datasets. AJs = Ashkenazi Jews.

Table 1.

Distribution of Alzheimer disease (AD) cases and controls of detected Ashkenazi Jewish individuals in each dataset and analysis.

Dataset	Genome-wide Analysis			Exome-wide Analysis
	Total	AD	Control	Total	AD	Control
GWAS	2,625	1,143	1,482	1,783	748	1,035
WGS	391	212	179	732	368	364
WES	NA	NA	NA	1,036	388	648
Total	3,016	1,355	1,661	3,551	1,504	2,047

GWAS: genome-wide association study; WGS: whole genome sequencing; WES: whole exome sequencing.

3.2. Genetic Associations with previously known and novel AD Loci

A genome-wide scan revealed several GWS and suggestive (P<1x10⁻⁶) associations with little evidence of genomic inflation (λ =1.044) (Figure S2). Among previously established AD loci, GWS associations were observed with many SNPs in the APOE region including the ε4 variant rs49358 (P=3.95x10⁻⁵⁴) and the TREM2 R47H variant (rs75932628, P=9.66x10⁻⁹) (Table 2). Results were available for 57 of the top SNPs at 76 independent GWS loci in the GWAS conducted by Bellenguez et al. [2] (Table S2). Associations of AD with 11 of these SNPs (19.3%) were at least nominally significant (p<0.05) and seven were significant at the p<0.01 level including TREM2 R47H and BIN1 SNP rs6733839 (P=4.97x10⁻⁵). As one might expect, associations that were significant in the AJ sample had effect sizes larger than in the EA GWAS. Further comparisons between the two GWAS revealed that associations for 44 of the 57 loci (77.2%) with results in the AJ dataset, including all nominally significant ones, had the same effect direction and similar odds ratios. Multiple significant associations were also observed with novel loci. A borderline GWS association was observed with a SNP near RAB3B (rs541586606, P=5.01x10⁻⁸), as well as suggestive associations with 10 other novel loci (Table 2). This variant is not in LD with other SNPs in the region that may explain the modest corroborating evidence for this association (Figure S3).

Table 2.

Top-ranked associations in the genome-wide analysis.

Gene	ID	Chr : position	Function	EA	RA	Minor Allele Frequency			OR	P-value *
						AD	Control	EUR
APOE	rs429358	19:44908684	exonic	C	T	0.30	0.12	0.16	3.07	3.95x10−54
TREM2	rs75932628	6:41161514	exonic	T	C	0.035	0.011	0.0023	3.17	9.66x10−9
RAB3B	rs541586606	1:51996478	intergenic	C	G	0.011	0.035	0.0010	0.32	5.01x10−8
SPOCK3/ANXA10	rs760573036	4:167607429	intergenic	CT	C	0.0034	0.023	0.0014	0.15	6.32x10−8
LRRC3B	rs6767457	3:26645626	intronic	A	G	0.17	0.12	0.20	1.49	8.09x10−8
HFM1	rs560945840	1:91344655	intronic	G	A	0.0037	0.023	0.0024	0.17	1.09x10−7
JAKMIP2-AS1	rs180825664	5:147562155	intronic	G	A	0.022	0.036	0.0008	0.35	1.33x10−7
PPM1H/AVPR1A	rs559501118	12:63117195	intergenic	T	A	0.0055	0.028	0.0022	0.23	1.77x10−7
CALCB/INSC	rs117925493	11:15092251	intergenic	T	C	0.0085	0.029	0.0067	0.29	2.39x10−7
NCOA1	rs572781170	2:24672133	intronic	G	A	0.010	0.033	0.0011	0.33	3.36x10−7
NFX1/AQP7	rs60257421	9:33378000	intergenic	G	A	0.12	0.079	0.088	1.56	4.46x10−7
BZW2	rs818581	7:16647544	intronic	A	G	0.44	0.50	0.62	0.76	4.55x10−7
CNTN5/LOC100128386	rs80135755	11:100624743	intergenic	T	A	0.059	0.033	0.095	1.92	4.90x10−7
PIEZO2	rs72865387	18:10955083	intronic	G	A	0.18	0.23	0.24	0.72	4.98x10−7

^*significance threshold P=5.0x10⁻⁸

Chr: chromosome; EA: effect allele; RA: reference allele; EUR: European reference population; OR: odds ratio.

Fewer associations were found in the analysis of the exome-wide scan (Figure S4) and these included the same APOE and TREM2 variants (P=4.52x10⁻⁵² and P=2.64x10⁻¹⁰, respectively) identified in the genome-wide scan (Table 3). In addition, the association with SMAP2 SNP rs1003710 (P=1.91x10⁻⁷) was SWS and the association with ZNF890P SNP rs200698976 (P=3.49x10⁻⁷) was nearly SWS (Table 3, Figures S5 and S6). These associations were evident but less significant in the smaller GW dataset. There were also suggestive associations (P<1x10⁻⁵) with SNPs in five other loci. Further inspection of the top-ranked results showed that the minor alleles of most SNPs were greater than four times more frequent in AJ controls than in the general European ancestry population. The differences were at least 9-fold for the variants at the RAB3B, HFM1, NCO1, TRAPPC8, and OR7C2 loci, and may explain why associations with these variants, especially those showing a protective effect, were not detected in much larger GWAS samples.

Table 3.

Top-ranked associations in the exome-wide analysis

Gene	ID	Chr : position	Function	EA	RA	Minor Allele Frequency			OR	P-value *
						AD	Control	EUR
APOE	rs429358	19:44908684	exonic	C	T	0.28	0.12	0.16	2.79	4.52x10−52
TREM2	rs75932628	6:41161514	exonic	T	C	0.036	0.011	0.0023	3.16	2.64x10−10
SMAP2	rs1003710	1:40374681	5’ UTR	A	G	0.40	0.46	0.44	0.77	1.91x10−7
ZNF890P	rs200698976	7:5132414	intronic	TG	T	0.28	0.34	0.31	0.76	3.49x10−7
TRAPPC8	rs149840013	18:31839471	intronic	A	G	0.0042	0.028	0.0022	0.16	1.53x10−6
DMWD	rs138963867	19:45786302	exonic	A	G	0.021	0.0077	0.0013	2.98	1.78x10−6
OR7C2	rs113508813	19:14942161	exonic	A	G	0.033	0.054	0.000071	0.55	1.81x10−6
SLC43A3	rs117212150	11:57414602	intronic	T	C	0.099	0.068	0.015	1.52	2.19x10−6
DOCK9	rs2296984	13:98805177	exonic	G	T	0.21	0.17	0.20	1.34	3.26x10−6

^*significance threshold P=2.74x10⁻⁷

Chr: chromosome; EA: effect allele; RA: reference allele; EUR: European reference population; OR: odds ratio

Aggregated rare variant gene-based tests revealed a GWS association with GIPR (P=7.34x10⁻⁷, Table S3) and a nearly exome-wide significant association with MAT2B (P=7.26x10⁻⁶, Table S4).

3.3. Differential gene expression at AD-associated loci

Differential expression in brain between AD cases and controls was evaluated for 25 genes containing or closest to the 21 showing loci GWS or highly suggestive evidence for association with AD (Table 2, Table 3). Expression data were unavailable for JAKMIP2-AS1, LOC100128386, and OR7C2. Significant differences (P_adj<0.0022) were found for genes at more than half of the loci (10/19). Higher expression in AD cases was observed for NCOA1 (P=1.50x10⁻⁵), TREM2 (P=2.00x10⁻⁴), SLC43A3 (P=1.20x10⁻¹¹) and PIEZO2 (P=3.70x10⁻⁶), whereas expression of for SMAP2 (P=1.80x10⁻⁶), RAB3B (P=3.10x10⁻⁶), BZW2 (P=8.30x10⁻⁸), CALCB (P=2.70x10⁻⁵), CNTN5 (P=1.00x10⁻⁴), and PPM1H (P=6.70x10⁻⁵) was significantly reduced (Table 4).

Table 4.

Differential expression in brain of genes at top-ranked loci between AD cases and controls.

Gene	Chr: pos	Z-score	P-value *
SMAP2	1:40344850	−4.77	1.80x10−6
RAB3B	1:51907956	−4.66	3.10x10−6
HFM1	1:91260766	−2.04	0.042
NCOA1	2:24491914	4.34	1.50x10−5
LRRC3B	3:26622806	−0.11	0.91
SPOCK3	4:166733384	1.45	0.15
ANXA10	4:168092515	1.62	0.11
TREM2	6:41158506	3.71	2.00x10−4
ZNF890P	7:5121239	−2.32	0.020
BZW2	7:16646131	−5.36	8.30x10−8
NFX1	9:33290511	−0.76	0.45
AQP7	9:33383179	0.24	0.81
CALCB	11:14904997	−4.20	2.70x10−5
INSC	11:15112424	0.14	0.89
SLC43A3	11:57406954	6.79	1.20x10−11
CNTN5	11:99020953	−3.88	1.00x10−4
PPM1H	12:62643982	−3.99	6.70x10−5
AVPR1A	12:63142759	1.88	0.060
DOCK9	13:98793429	1.94	0.052
PIEZO2	18:10666483	4.63	3.70x10−6
TRAPPC8	18:31829173	−0.66	0.51
APOE	19:44908684	2.06	0.039
DMWD	19:44905754	−0.26	0.80

Chr: chromosome; Pos: start position.

^*significance threshold P=0.0022

4.0. Discussion

4.1. Novel AD loci identified in Ashkenazi Jews

We identified approximately 3,500 individuals whose ancestry is almost exclusively Ashkenazi Jewish among much larger European ancestry samples using a robust clustering approach that compared genetic signatures with members of an AJ reference GWAS dataset. Genome-wide and exome-wide scans for AD conducted in this AJ sample identified GWS associations with established AD loci including APOE and TREM2. Notably, the association with the TREM2 R47H variant has a similar p-value (P= 2.64 x10⁻¹⁰) and shows a much stronger effect in the AJ sample (OR=3.16, [2.14-4.70]) compared to a very large AD GWAS sample containing undifferentiated European ancestry individuals (n=82,771, OR=2.39, [2.09-2.73]) [18]. The strength of this finding in the AJ sample is likely due to the MAF of this variant in the AJ controls which is about five times higher than in other European ancestry populations. We also detected significant associations with several novel loci including RAB3B, SMAP2 and ZNF890P.

Confidence in these results is bolstered by evidence for association of AD with 10 other previously established AD loci and similar pattern of association for SNPs at 82% (47/57) of the GWS loci identified by GWAS in a sample of approximately 789,000 EAs [2]. Not surprisingly, few of the known loci were highly significant in the AJ sample that had little power for detecting associations with even the most common variants having odds ratios <1.2 (i.e., accounting for nearly all of the common variant locus associations). Our findings are also supported by evidence for significantly expression differences in brain between AD cases and controls for genes at more than one-half of the top-ranked loci including two novel ones that were GWS (RAB3B and SMAP2).

Among the novel genome-wide or study-wide significant findings, SMAP2 encodes a GTPase-activating protein (GAP) for Arf1 with a putative clathrin-binding domain that binds the clathrin adaptor protein complex 1 (AP-1) [35]. Clathrin adaptor AP-1-mediated Golgi export of amyloid precursor protein is crucial for the production of neurotoxic amyloid fragments [36]. Previous AD GWAS identified other genes in this pathway including PICALM and BIN1 [37]. RAB3B encodes one of four closely related small GTP-binding proteins (Rab3A, Rab3B, Rab3C and Rab3D) that are believed to involved in presynaptic vesicle trafficking and priming steps for regulating synaptic transmission and plasticity, an idea supported by patch-clamp electrophysiology experiments conducted in neuronal cultures from Rab3-deficient mouse hippocampus [38]. RAB3B overexpression improves dopamine handling and storage and confers protection to dopaminergic neurons [39]. The function of ZNF890P is unknown, however a GWS association was reported in a Japanese cohort between a SNP located approximately 11kb from ZNF890P and elevated high-sensitivity cardiac troponin level [40]. Troponin is a well-known plasma marker of myocardial injury and has been associated with cognitive decline in the elderly and incident AD or dementia [41]. We also obtained a GWS association of AD with aggregated rare variants in GIPR which encodes a G-protein coupled receptor for gastric inhibitory polypeptide. GIPR stimulates insulin resistance in the presence of elevated glucose. Activation of GIPR has been incorporated in the development of compounds to treat type 2 diabetes [42], a major risk factor for AD [43]. Multiple studies showed that GIPR knockout mice sustained impairment of synaptic plasticity and cognitive deficits [44, 45].

Several loci showing suggestive evidence for association also are relevant to AD. SPOCK3 encodes a member of a family of calcium-binding proteoglycan proteins that is strongly expressed in cerebral cortex and hippocampus and may have roles in regulation of action potential in neurons, neurotransmitter uptake, memory, and neuroimmunity [46]. Variants in SPOCK3 have been associated with verbal declarative memory in a GWAS [46] and variability in cytokine secretion in response to smallpox vaccine [47]. The rs117925493 SNP near INSC that was associated with AD in this study was recently reported as a GWS association with several cognitive traits and a significant expression quantitative trait locus for INSC [48]. NCOA1 has been associated with the Aβ42/Aβ40 ratio measured in plasma via a gene-based test including five rare variants in 370 middle aged African American participants of the Atherosclerosis Risk in Communities-Neurocognitive Study [49]. Contactin 5 (CNTN5) is a neuronal membrane protein that contributes to axonal targeting, synaptic formation and plasticity. Biffi et.al identified an FDR-corrected significant association of a CNTN5 SNP with white matter lesion volume in a sample of 746 ADNI participants [50].

4.2. Advantages of founder populations for AD gene discovery

Our study illustrates the greatly increased power for detection of GWS associations in founder communities in which disease-associated variants may be much more frequent compared to samples ascertained from large admixed populations (e.g., 4.6-771 fold difference for 12 of the highly significant findings shown in Tables 2 and 3). Moreover, some genetic association signals for complex diseases like AD are likely to be stronger in founder populations that are relatively genetically homogeneous. There are numerous examples of successful genome-wide scans for complex diseases and traits in religious and geographic isolates including the Amish [51-54], Hutterites [8, 55], and Icelanders [56-58]. Previously, we identified a GWS association (P=3.4x10⁻⁹) of AD with a two-SNP haplotype in ACE in 258 AD cases and controls from an Israeli-Arab community of >81,500 people who trace their ancestors to 14 founder families [7], many years before this locus emerged as GWS in a much larger dataset [37]. The advantage of founder populations for AD gene discovery was recently demonstrated in a study that identified GWS association with MGMT independently in a small portion of a Hutterite kindred and in the large ADGC GWAS sample [8]. The most pronounced finding in an ADSP study focused on rare functional variants was a missense mutation in NOTCH3 that was observed in 10 AD cases but no controls [59]. Previously, more than 125 NOTCH3 mutations had been linked to a rare dementia syndrome called Cerebral Autosomal Dominant Arteriopathy with Sub-cortical Infarcts and Leukoencephalopathy (CADASIL), but not to AD [60]. Subsequent analysis of the NOTCH3 mutation carriers showed that they all had a single haplotype and a high frequency of two mitochondrial DNA haplotypes found primarily in AJs [59]. This mutation is exceedingly rare in all populations surveyed except AJs.

4.3. Study limitations

This study has several caveats. Most notably, the sample size is small compared to other AD GWAS conducted in genetically heterogeneous EA samples containing several ancestral populations, but relatively few AJs. We applied relatively stringent limits on the PC boundaries of the AJ cluster that was derived from large EA datasets thereby excluding some individuals with substantial AJ ancestry and further reducing the sample. However, this strategy likely increased the genetic homogeneity of the sample which increased the sensitivity for detecting associations enhanced by a founder effect. In addition, because the ADGC GWAS dataset contained multiple independently ascertained and genotyped cohorts that necessitated genotype imputation to be performed for each cohort separately, relatively infrequent variants may not have been imputed well, particularly in the smaller cohorts and thus excluded by the variant QC procedure. The sample size was further reduced for genome-wide analyses because a portion of the data was derived from whole exome sequencing. Despite these limitations, we identified several GWS and SWS associations with common and infrequent variants in previously established and novel loci. Future studies of larger AJ samples will be required to detect associations with additional loci and variants that are rare in AJs.

4.4. Conclusions

Our results highlight the efficacy of conducting GWAS for AD in founder populations which may have a significantly higher frequencies for some variants that are rare in genetically heterogeneous populations, and can lead to effective discovery of genetic associations for specific ancestry populations.

RESEARCH IN CONTEXT.

1. Systematic review: The authors reviewed the literature using traditional (eg., PubMed) as well as preprinted (e.g., medRxiv) sources on genetic association in Alzheimer disease (AD), genetics of AD and other diseases in founder populations, as well as the role of novel loci in AD.

2. Interpretation: We show that APOE, the TREM2 R47H mutation, and variants in several novel genes including RAB3B, SMAP2, ZNF890P and GIPR are associated with risk of AD in Ashkenazi Jews. These findings highlight the efficacy of conducting GWAS for AD in founder populations.

3. Future directions: Follow-up studies in independent cohorts of Ashkenazi Jews to confirm these findings and discovery additional loci are warranted. Future research should also focus on mechanisms that explain the association of AD with the novel genes and provide insight into potential novel therapeutic approaches.

Abbreviations

AD

Alzheimer disease

ADGC

Alzheimer’s Disease Genetics Consortium

ADSP

Alzheimer’s Disease Sequencing Project

AJ

Ashkenazi Jewish

EA

European ancestry

EW

exome-wide

GW

genome-wide

GWAS

genome-wide association study

GWS

genome-wide significant

HWE

Hardy-Weinberg equilibrium

MAF

minor allele frequency

QC

quality control

SNP

single nucleotide polymorphism

SWS

study-wide significant

WES

whole exome sequencing

WGS

whole genome sequencing

Data Availability

ADGC GWAS data, ADSP WGS and WES data, and summarized results are available from the National Institutue on Aging Genetics of Alzheimer Disease Storage site (NIAGADS; https://www.niagads.org).