Skip to main content


Research | Open | Published:

Genome-wide analysis of genetic predisposition to Alzheimer’s disease and related sex disparities



Alzheimer’s disease (AD) is the most common cause of dementia in the elderly and the sixth leading cause of death in the United States. AD is mainly considered a complex disorder with polygenic inheritance. Despite discovering many susceptibility loci, a major proportion of AD genetic variance remains to be explained.


We investigated the genetic architecture of AD in four publicly available independent datasets through genome-wide association, transcriptome-wide association, and gene-based and pathway-based analyses. To explore differences in the genetic basis of AD between males and females, analyses were performed on three samples in each dataset: males and females combined, only males, or only females.


Our genome-wide association analyses corroborated the associations of several previously detected AD loci and revealed novel significant associations of 35 single-nucleotide polymorphisms (SNPs) outside the chromosome 19q13 region at the suggestive significance level of p < 5E–06. These SNPs were mapped to 21 genes in 19 chromosomal regions. Of these, 17 genes were not associated with AD at genome-wide or suggestive levels of associations by previous genome-wide association studies. Also, the chromosomal regions corresponding to 8 genes did not contain any previously detected AD-associated SNPs with p < 5E–06. Our transcriptome-wide association and gene-based analyses revealed that 26 genes located in 20 chromosomal regions outside chromosome 19q13 had evidence of potential associations with AD at a false discovery rate of 0.05. Of these, 13 genes/regions did not contain any previously AD-associated SNPs at genome-wide or suggestive levels of associations. Most of the newly detected AD-associated SNPs and genes were sex specific, indicating sex disparities in the genetic basis of AD. Also, 7 of 26 pathways that showed evidence of associations with AD in our pathway-bases analyses were significant only in females.


Our findings, particularly the newly discovered sex-specific genetic contributors, provide novel insight into the genetic architecture of AD and can advance our understanding of its pathogenesis.


Alzheimer’s disease (AD) is a slowly progressive neurodegenerative disorder that usually manifests with insidious deterioration of cognitive functions such as memory, language, judgment, and reasoning. Visuospatial deficits and neuropsychiatric symptoms like anxiety, irritability, depression, delusion, and personality changes may occur in the course of the disease, and these are eventually followed by impairment of most daily activities [1, 2]. The median survival is 3.3–11.7 years after disease manifestation [3]. Except for some uncommon autosomal dominant forms, AD is mainly a complex disorder with a polygenic nature [2, 4] that predominantly affects elderly individuals, also known as late-onset AD. It is the most common cause of dementia in the elderly worldwide [5] and is the sixth leading cause of death in the United States [6]. Age is the main risk factor for AD. The annual incidence increases from 1% at age 65 years to 6–8% after 85 years [7], and its prevalence increases from 11% to 32% [5]. In addition, AD is more prevalent in females than males [7,8,9,10], with their lifetime risk of developing the disease being almost twice that of males [7]. This might be to some extent justified by different life expectancies of males and females. However, Genin et al. [11] suggested that the age-adjusted penetrance of Apolipoprotein E (APOE) was sex dependent as well. For instance, they found that the lifetime risks for homozygote APOE-ε4 carriers were 51% and 60% in males and females older than 85 years, respectively. The corresponding risks for heterozygote APOE-ε3ε4 carriers were 23% and 30%, respectively [11]. AD is also more severe in females than males [9]. Henderson and Buckwalter [12] reported that female AD patients had greater impairment of naming task, verbal fluency, and delayed recall compared to male patients. In another study, Barnes et al. [13] suggested that females were more likely to develop clinical AD compared to males in response to pathology changes (e.g., amyloid beta (Aβ) and neurofibrillary tangles) in the brain. They found that each additional unit of pathology in the brain would increase the odds of overt AD by 20-fold and 3-fold in females and males, respectively [13]. The underlying mechanisms of sex disparity in AD are not fully clear [9, 14]. This may raise the possibility that such sex disparities might be in part due to potential differences in the genetic bases of AD between males and females. Investigating such differences is important, particularly for tailoring more effective medical interventions [14, 15].

Give the considerable physical, emotional, and economic burdens imposed by AD on patients, their families, and societies, exploring the genetic and nongenetic mechanisms underlying its pathogenesis has become a public health priority. With increased life expectancy, the prevalence and global economic costs of AD are forecast to increase considerably by 2050 [5]. Many studies have investigated the genetic basis of AD. APOE was the first gene linked to late-onset AD [16], and, in particular, the dosage of its ε4 allele was implicated in increasing the risks of disease and earlier onset [17]. More susceptibility loci were detected with the advent of genome-wide association (GWA) methodology, although not all of them were consistently replicated in independent datasets. In addition to APOE, which was almost universally replicated, BIN1, CLU, CR1, CD2AP, CD33, MS4A4E, MS4A6A, EPHA1, and PICALM genes have been associated with the polygenic form of AD in different studies [18, 19]. The narrow-sense heritability (h2) of AD (i.e., the proportion of its phenotypic variance explained by additive genetic variance) has been estimated to be 58–79% by twin studies [20]. Furthermore, Ridge et al. [19], using a linear mixed models (LMMs) framework, found that 53% of phenotypic variance of AD can be explained by ~ 8 million single-nucleotide polymorphisms (SNPs). They also noticed that SNPs inside known AD-associated genes or within their 50 kb upstream/downstream regions can only explain ~ 31% of AD phenotypic variance (~ 59% of genetic variance) [19], leaving a sizable portion of its h2 to be explained.

In this study, we investigated the genetic architecture of polygenic AD through genome-wide association (GWA), transcriptome-wide association (TWA), gene-based, and pathway-based analyses in four independent datasets (two with family designs and two with population designs) using genetic information for approximately 2 million genotyped and imputed SNPs. Since exploring the genetic sex disparity of AD was of particular interest, in addition to analyzing the entire sample of males and females in each dataset, two alternative plans were also considered in which either only males or only females were included in analyses.


Study participants

Four independent datasets were used to fulfill the aims of this study: Late-Onset Alzheimer’s Disease Family Study from the National Institute on Aging (NIA-LOADFS) [21]; Framingham SNP Health Association Resource (SHARe) project from Framingham Heart Study (FHS) [22,23,24]; SNP Typing for Association with Multiple Phenotypes from Existing Epidemiologic Data (STAMPEED) project from Cardiovascular Health Study (CHS) [25]; and University of Michigan Health and Retirement Study (HRS) [26]. All four datasets were approved by the institutional review boards (IRBs) and had gathered data after obtaining written informed consent from participants or their legal guardians/proxies. Details about the designs of the NIA-LOADFS, FHS, CHS, and HRS studies can be found in the original publications. Briefly, the NIA-LOADFS is a family-based study primarily initiated to investigate late-onset AD risk factors. It recruited families with multiple affected members if the age at AD onset or diagnosis of proband was above 60 years. Controls were selected from unaffected individuals with a minimum age of 50 years who had no history of major neurological/psychiatric disorders or life-threatening conditions. Of 9468 participants with phenotype data, 5220 subjects (2319 affected with AD), predominantly Caucasians, were genotyped using Illumina’s Human 610-Quad array. The FHS is an ongoing longitudinal study with a family-based design that provides phenotype and genotype information on individuals from three-generational families with Caucasian ancestry. The main objective of the study was to investigate cardiovascular disorder risk factors. It was first initiated by recruiting 5209 participants (i.e., original cohort) between ages 30 and 62 years with no history of cardiac disease or stroke. Later, the cohort was expanded by adding the offspring of the original cohort and their spouses (5124 subjects as the offspring cohort) and their grandchildren (4095 subjects as the third generation). Of these, 9274 individuals (1529, 3852, and 3893 individuals from the three aforementioned generations, respectively) were genotyped using the Affymetrix Human Mapping 500 K array in the SHARe project. The CHS is a population-based longitudinal study with the main objective of investigating risk factors contributing to heart diseases. It was initiated by recruiting an original cohort of 5221 mainly Caucasian participants who were older than 65 years and had not been institutionalized. Later, a new cohort of 687 participants, predominantly African-Americans, was added to the study. Of these, 3989 and 803 individuals were genotyped by Illumina’s Human CNV370-Duo and Human Omni1-Quad arrays, respectively, in the STAMPEED project. The HRS is a population-based longitudinal study launched to provide age-related health and economic information on more than 20,000 individuals older than 50 years. The HRS makes use of administrative records such as Social Security and Medicare claims to gather information of interest about participants. The study was expanded in 2006 to include a biomarker and genetic component in which 12,595 individuals, predominantly Caucasian, were genotyped by Illumina’s Human Omni2.5-Quad array.

Our study focused on people of Caucasian ancestry from the four aforementioned studies to increase the sample size and power of the analyses. The LOADFS and FHS datasets directly identify cases with Alzheimer’s disease and unaffected controls. For the CHS and HRS datasets, the International Classification of Disease codes, ninth revision (ICD-9) were used to define cases and controls. Finally, to make the four datasets comparable in terms of participants age, we only included the original and offspring cohorts from the FHS dataset. Demographic information about the cohorts included in our study is presented in Table 1. Also, Additional file 1: Table S1 lists the numbers of cases and controls in these cohorts.

Table 1 Demographic information about the four cohorts under consideration

Imputation of genotype data

Since the four datasets of interest were genotyped using different platforms, imputation was conducted to generate a common set of 2,928,658 SNPs. Only autosomal SNPs were subject to imputation. Genome coordinates of SNPs in our data (NCBI build 38/UCSC hg38) were lifted over to NCBI build 37/UCSC hg19 using LiftOver software [27]. After removing duplicate SNPs, preimputation quality control (QC) was performed using PLINK software [28] to remove low-quality SNPs/subjects by setting the following QC criteria: minor allele frequency < 0.01, SNPs and subject call rates < 95%, and Hardy–Weinberg p < 1E–06. For the LOADFS and FHS cohorts that have family-based designs, a Mendel error rate of 2% was set to remove SNPs and subjects/families with high Mendelian errors. The SHAPEIT2 (i.e., Segmented Haplotype Estimation and Imputation Tool) package [29] was used to ensure that alleles were aligned to the same DNA strand in our and the reference data. Haplotype phasing was then conducted using SHAPEIT2 to estimate the haplotypes for subjects in each dataset. Finally, genotypes were imputed by Minimac3 software [30] over prephased haplotypes. SHAPEIT2 and Minimac3 were run using default values for input arguments and European population (EUR) haplotypes from 1000 Genomes Phase 3 data (release October 2014) as the reference panel.

Postimputation QC

Directly genotyped SNPs along with the imputed SNPs, for which the squared correlation (r2) between imputed and expected true genotypes was > 0.7, were selected for preanalysis QC. This step was performed based on the same criteria explained earlier for preimputation QC. Additional file 1: Table S2 contains information on the numbers of genotyped and imputed SNPs that remained in each of the four datasets of interest after QC.

Population structure

The top 20 principal components (PCs) of genotype data were obtained through principal component analysis (PCA) to be included in downstream genetic analyses to address potential population stratification. In each dataset, PCA was performed over a subset of unrelated individuals and a subset of SNPs that were not in high linkage disequilibrium (LD) measured by r2 [31]. KING (i.e., Kinship-based Inference for Genome-wide association studies) software [32] was used to obtain the subset of unrelated subjects by keeping one subject per family or relative cluster whose identity-by-descent (IBD) was > 0.0884 (i.e., closer than third-degree relatives). The genotyped autosomal SNPs on each chromosome were then pruned by PLINK software [28] in an unrelated set of subjects such that no SNP pairs with r2 > 0.2 were kept within any 100-SNP windows. PCA was then conducted over the selected low-LD SNPs with the GENESIS R package [33, 34]. Additional file 1: Table S3 contains genomic inflation factors (λ values) resulting from logistic regression models for the four datasets under consideration. The λ values were less than 1.1 in all cases, indicating a subtle impact of population structure on our analyses [35, 36].

Genetic analysis

GWA analysis

The associations between SNPs and AD were investigated by fitting logistic regression models. The genetic analyses of each dataset were performed under three alternative plans analyzing the entire sample, only males, and only females. The top five PCs and subject’s birth cohort (i.e., birth year) were included in the models as fixed-effects covariates. In addition, sex was considered a fixed-effect covariate under plan 1. Only additive genetic effects were modeled; dominance effects were ignored. The birth cohort is a proxy for the age and environmental exposures which are characteristic for a cohort. Thus, this adjustment controls for age and overtime trends in the incidence of AD. The logistic models were fitted using PLINK software (v1.07) [28]. It was previously suggested that for samples with a family-based design, ignoring family relationships would not generate considerable bias in effect sizes of SNPs but may increase type I error rates whose magnitude depends on pedigree complexity (e.g., nuclear family vs extended family) and trait heritability. For instance, the inflation of type I error rates has been suggested to be trivial in datasets with simple pedigrees. On the other hand, type I error rates may increase by a factor of 2–3 when family structure is ignored in a dataset with an extended family pedigree and trait heritability values of 0.6–0.9. Therefore, a two-step screening–validating approach could be used with such datasets to prevent inflation of type I error rates and decrease the computational burden of analysis [37]. For the LOADFS and FHS cohorts, we adopted a two-step approach in which the SNPs with p < 0.05 in the logistic models explained earlier were subjected to fitting generalized linear mixed models (GLMMs) by including all aforementioned fixed-effects covariates along with family IDs as a random-effects covariate. GLMMs were fitted using the lme4 R package [33, 38].

All GWA analyses were conducted in a discovery–replication manner. Each of the LOADFS, FHS, CHS, and HRS datasets was considered a discovery set to detect SNPs in significant associations with AD. Results from the discovery stage in a particular dataset were then subject to further replication in the remaining three datasets. At the discovery stage, a genome-wide significance level of p < 5E–08 was set to select statistically significant associations, and SNPs with p values between 5E–08 and 5E–06 were considered suggestive AD-associated markers. These significance levels are widely accepted by genome-wide association studies in order to decrease the type I error rate (i.e., false-positive findings) due to multiple testing issues arising from investigating associations of millions of SNPs [39, 40]. A Bonferroni-corrected significance threshold of 0.0167 (i.e., 0.05/3, where 3 is the number of replication datasets for validating any significant association signals from a discovery dataset) was considered at the replication stage.

Finally, a conventional fixed-effects meta-analysis, using the inverse variance method, was conducted over the results under each plan from the four investigated datasets to obtain combined statistics for the tested SNPs. To avoid missing heterogeneous associations of opposite directions of effects, we also performed a meta-analysis on absolute values of coefficients in addition to the conventional meta-test. The results from the meta-analysis on absolute values of coefficients were used just as an additional piece of information to determine how heterogeneous effects in different cohorts can affect the results of a conventional inverse-variance meta-analysis. The meta-analysis results were interpreted according to the significance level at the discovery phase. The meta-analysis was performed using GWAMA (i.e., Genome-Wide Association Meta-Analysis) software [41].

Also, for SNPs that had significant p values only in males or females (i.e., plans 2 or 3), a Wald χ2 statistic with 1 degree of freedom was calculated according to the following formula [42] to investigate whether their odds ratios were significantly different between the two sexes:

$$ {\chi}^2=\frac{{\left({b}_{\mathrm{m}}-{b}_{\mathrm{f}}\right)}^2}{se_{\mathrm{m}}^2+{se}_{\mathrm{f}}^2} $$

where bm and bf are the coefficients (i.e., the natural logarithm of odds ratios) for any SNP in males and females, respectively, and sem and sef are their corresponding standard errors.

The significant findings from GWA analyses were compared to previous studies using the GRASP (i.e., Genome-Wide Repository of Associations Between SNPs and Phenotypes) search tool (v2.0.0.0) [43]. Also, LD between significant SNPs and previously detected AD-associated loci in their 1-Mb flanking regions (r2 ≥ 0.4 or significant p value from χ2 test for LD) was investigated in the CEU population (i.e., Utah Residents with Northern and Western European Ancestry) through the HaploR R package [33, 44] and the LDlink web-tool [45]. The genes coordinate’s list provided by PLINK [28] was used to find the closest genes of the significant SNPs. The chromosomal regions (i.e., cytogenetic bands) were determined using the annotation database from UCSC Genome Browser [46].

Gene-based analysis

Under each of three aforementioned plans, gene-based analysis was performed over the meta-analysis results using the fastBAT (i.e., Fast set-Based Association Test) method [47] implemented in the GCTA (i.e., Genome-wide Complex Trait Analysis) package (v1.26.0) [48]. This method combines z-statistics for a set of SNPs corresponding to each gene into a quadratic form of a multivariate normal variable. SNPs located within a gene or its 50 kb upstream/downstream regions were considered as an SNP set for that gene. The HRS dataset was used as the reference panel for LD calculation (i.e., r2 metric) in order to remove one of each pair of SNPs with r2 > 0.9 from any given set. To deal with multiple-testing issues, the false discovery rate (FDR) method suggested by Benjamini and Hochberg [49] was used to rank and select significant findings. Genes with significant p values at the FDR level of 0.05 were considered novel AD-associated ones if there were no SNPs with p < 5E–08 in their 1-Mb upstream/downstream regions in the current or previous studies.

Pathway-based analysis

A pathway-based analysis was also performed using the fastBAT method using the pathways predefined by the REACTOME pathway knowledgebase [50] and PID (i.e., the Pathway Interaction Database) [51]. These were provided by the molecular signatures database (MSigDB) at the Broad Institute gene set enrichment analysis (GSEA) website [52, 53]. Here, a SNP set corresponding to a particular pathway was defined as the SNPs within 50 kb of the genes in that pathway. As with the gene-based analysis, the HRS cohort was used to prune the SNP sets based on the pairwise LD measures of SNPs. The significant results were interpreted at the FDR levels of 0.05 (plans 1 and 2) and 0.025 (plan 3) to ensure that the number of possible false-positives was < 1 under each analysis plan.

TWA analysis

Results from conducted meta-analyses along with summary data from a publicly available expression quantitative trait loci (eQTLs) study on peripheral blood [54] were used to perform a transcriptome-wide association analysis using SMR (i.e., Summary-data-based Mendelian Randomization) software (v0.68) [55]. The eQTLs summary data were downloaded from the SMR software website. Both cis-eQTLs and trans-eQTLs were of interest. Trans-eQTLs were defined as eQTLs located at least 5 Mb away from a probe on the same chromosome or located on other chromosomes. Probes for which at least one eQTL with p < 5E–08 had been detected by Lloyd-Jones et al. [54] were included in our analyses provided that the corresponding eQTLs were among the genotyped or imputed SNPs in our study. This resulted in the inclusion of sets of up to 8257 probes with cis-eQTLs and 2763 probes with trans-eQTLs.

The significance of p values resulting from SMR testing (i.e., PSMR) was interpreted at an FDR level of 0.025–0.05. The appropriate FDR level for each of three analysis plans was chosen so we can ensure that the number of possible false-positive findings among significant probes was < 1. To identify the pleiotropic effects of SNPs on gene expression levels and AD development, probes with significant PSMR values were then subject to heterogeneity testing (i.e., the HEIDI test) which can differentiate pleiotropy from linkage [55, 56]. Genes corresponding to probes that passed both the SMR and HEIDI tests (i.e., significant PSMR and PHEIDI ≥ 0.05) were deemed significant as their expression profiles might be associated with AD because of the pleiotropic effect of a single variant that affects both probe expression and AD susceptibility. Selected genes were considered potentially novel AD genes if there were no SNPs with p < 5E–08 within their 1-Mb upstream/downstream regions in the current or previous studies.

Finally, we also performed TWA analyses using summary results from a publically available tissue-specific eQTLs study [57] which contains eQTLs data for several regions of the brain, including the amygdala, anterior cingulate cortex (BA24), basal ganglia (e.g., caudate, nucleus accumbens, and putamen), cerebellar hemisphere, cerebellum, cortex, frontal cortex (BA9), hippocampus, hypothalamus, and substantia nigra. Once again, probes that had significant eQTLs with p < 5E–08 were included in our analyses. This resulted in the inclusion of sets of 597–3566 probes with cis-eQTLs (based on the brain region). The results of brain-specific TWA analyses were interpreted at a FDR level of 0.05.


dbGaP:; GCTA:; GENESIS R Package:; GRASP:; GSEA:; GWAMA:; HaploR R package:; KING:; LDlink:; LiftOver:; Lme4 R Package:; Minimac3:; PLINK:; SHAPEIT:; SMR:; 1000 Genomes:;


GWA analysis

GWA analyses were performed in four independent datasets (i.e., LOADFS, FHS, CHS, and HRS). Each of these datasets served as a discovery set to detect SNPs with significant association signals (at either a genome-wide significance level of p < 5E–08 or a suggestive level between 5E–08 and 5E–06), which were then subject to further replication (at the significance level of 0.0167) in the other three datasets. These analyses provided replicated and nonreplicated sets of SNPs. Finally, results from the individual datasets were combined through meta-analysis and interpreted according to the significance level at the discovery phase. Additional file 1: Tables S4–S12 provide an overview of replicated, nonreplicated, and meta-analysis sets of SNPs that were significantly associated with AD in males and females combined (plan 1) or males and females separately (plans 2 and 3). As seen in these tables, most of the newly detected AD-associated SNPs, particularly those in nonreplicated and meta-analysis sets, had significant p values only in one of the three study plans. For instance, among 44 and 72 newly detected SNPs in males and females, 36 and 51 SNPs had sex-specific significant p values, respectively. Additional file 1: Figures S1–S6 show the Manhattan and QQ plots of the GWA results in the four investigated datasets, as well as in the conducted meta-analyses under these three plans. In general, SNPs with p values smaller than the genome-wide significance threshold were mostly located on chromosome 19.

Replicated sets of SNPs

The replicated sets of SNPs under plans 1–3 contained 31, 20, and 23 SNPs, respectively (Additional file 1: Tables S4–S6). These SNPs had significant p values at the genome-wide level or a suggestive level of associations at the discovery stage and were then replicated in another dataset. Additional files 23, and 4 contain detailed information (e.g., allele frequencies, odds ratios (ORs), p values, etc.) about the replicated SNPs in the four tested datasets under the three analysis plans. Notably, 12, 8, and 8 replicated SNPs, respectively, had not been previously associated with AD. The other SNPs had some evidence of direct association signals [43]. Among previously detected SNPs, rs9882471 (plan 2) was nominally associated with AD in previous studies (5E–06 ≤ p < 5E–02) [58].

Most of the newly detected SNPs were located inside a previously well-known susceptibility region for AD on chromosome 19q13 (i.e., APOE cluster gene region) and were mostly significant under different analysis plans. This subset of newly detected SNPs mostly had p < 5E–08, the same directions of effects in discovery and replication datasets, and significant p values (at genome-wide or suggestive levels of significance) in the meta-analysis.

Table 2 summarizes information about the four newly detected SNPs located outside the chromosome 19q13 region. Among these SNPs, rs62402815 was significant under plan 1 (i.e., males and females) and plan 3 (i.e., only females); and rs9918162 and rs726411 were significant only in males (i.e., plan 2). Their association signals were significant only at the suggestive level of associations (except rs62402815, which had a genome-wide level significant p = 1.2E–08 in females) in the discovery stage. The two SNPs that were significant in males did not have p < 5E–06 in conventional fixed-effects meta-analyses, which might be partially due to the heterogeneity of their effects across different datasets. These heterogeneous effects were reflected by high i2 inconsistency metrics and significant Q-statistics in Cochran’s heterogeneity test (Pq < 0.05). A meta-analysis based on the absolute values of the coefficients confirmed a substantial role of heterogeneity by providing smaller p values for most of these SNPs.

Table 2 Newly detected replicated and meta-analysis sets of significant SNPs located outside chromosome 19q13 region

Also, rs62402815 and rs726411 had the same direction of effects in the discovery and replication datasets. The directions of effects of rs9918162 were opposite in the discovery and replication sets. While genetic variants that have the same direction of effects in multiple independent cohorts are generally of more interest, those with opposite effects can be important as well because they may be indicative of the genetic heterogeneity of the studied trait in different cohorts arising, for example, from the epistasis or differences in LD patterns [59,60,61].

Although no evidence of direct association with AD was found in previous studies for the newly detected subsets of replicated SNPs, their 1-Mb upstream/downstream regions harbor AD-associated SNPs. We therefore investigated their LD with AD-associated loci in their 1-Mb flanking regions in the CEU population [45]. Newly detected SNPs were considered informative AD markers if their p values were smaller than those of the top AD-associated SNPs in their neighborhood or they were not in LD with previously AD-associated loci whose p values were smaller than those detected in this study. Additional file 1: Table S13 contains LD information about those newly detected SNPs for which proxy AD-associated loci have been reported. As seen in Additional file 1: Table S13, all newly detected SNPs on chromosome 19q13 had larger p values than the top AD-associated loci in their neighborhood and were in LD with them. Therefore, they were likely to relay the same information as their neighboring AD-associated SNPs.

On the other hand, the p values of SNPs located outside the chromosome 19q13 region were mostly smaller than the previously detected association signals in their flanking regions and were not in LD with such loci. As seen in Table 2, among the closest genes to these SNPs, only ADCY8 (corresponding to rs726411 located in the 8q24.22 region) was associated with AD in previous GWAS at a suggestive level of associations (rs263238 with p = 2.40E–06 [62]). In addition, none of the chromosomal regions (i.e., cytogenetic bands) in which other SNPs are located contained any previously AD-associated SNPs with p < 5E–06 [43]. Detailed information about the genes and chromosomal regions corresponding to the newly detected SNPs that contain previously AD-associated SNPs can be found in Additional files 23, and 4.

Nonreplicated sets of SNPs

Additional file 1: Tables S7–S9 (corresponding to plans 1–3) show that 54, 40, and 46 SNPs had significant p values at genome-wide or suggestive levels of associations in only one of the four datasets of interest. Most of them were newly detected (41, 33, and 40 SNPs, respectively), as there was no evidence of their direct association with AD in previous studies [43]. Also, they were mostly plan specific and demonstrated evidence of sex disparity. Most were located in chromosomal regions other than 19q13 and were significant at a suggestive level of associations. Detailed information about nonreplicated sets of SNPs (e.g., allele frequencies, ORs, p values, etc.) can be found in Additional files 23, and 4. Of those SNPs previously associated with AD, rs11038106, rs9597722, rs723804, rs17697225 [63], rs2065706 [64] (plan 1), rs4679840 [58] (plan 2), and rs1359176 [65] (plan 3) were only nominally significant (5E–06 ≤ p < 5E–02) in previous studies. Once again, SNPs located outside the chromosome 19q13 region either had smaller p values than previously detected AD-associated loci in their proximity or were not in LD with them, except for rs34779859 on chromosome 2 (plan 3) which was significant in females. LD information about those newly detected SNPs for which proxy AD-associated loci have been previously identified can be found in Additional file 1: Table S13.

Among the closest genes to newly detected SNPs outside the chromosome 19q13 region, BIN1, FRMD4A, and CDH4 that were significant under plan 3 were previously associated with AD with p < 5E–06 (rs744373 with p = 2.60E–14 [66], rs7921545 with p = 5.40E–07 [67], and rs4925189 with p = 6.30E–07 [68], respectively). Also, several other genes were located in AD-associated chromosomal regions. Information about these genes/regions is summarized in Additional files 23, and 4.

Meta-analysis sets of SNPs

Additional file 1: Tables S10–S12 show that 17, 4, and 24 SNPs that were not among replicated or nonreplicated sets of significant SNPs under analysis plans 1–3 passed the significance threshold in the meta-analysis. Additional files 23, and 4 summarize the GWA results for these SNPs. The meta-analysis p values of these SNPs were mostly significant at the level of suggestive associations, except for rs76366838, rs115881343 (plan 1), rs73048293, rs57537848, and rs76366838 (plan 3) on chromosome 19q13 which had p < 5E–08. Also, they were mostly located outside chromosome 19q13 and were plan specific (i.e., they were not among replicated, nonreplicated, or meta-analysis sets of significant SNPs under other plans). For example, significant SNPs in males were not significant in females and vice versa. In addition, most SNPs (14, 3, and 24 SNPS under plans 1–3, respectively) were not associated with AD in previous studies [43]. Summary information about the newly detected subset of meta-analysis sets of SNPs that were outside chromosome 19q13 is presented in Table 2. As with the replicated and nonreplicated sets of SNPs, most of the newly detected SNPs not on chromosome 19q13 had smaller p values than the ones reported for their nearby AD-associated loci or were not in LD with them. These SNPs, therefore, were considered novel and informative AD markers. On the other hand, proxy AD-associated SNPs were found for all newly detected SNPs that were located on chromosome 19q13 (Additional file 1: Table S13).

As seen in Table 2, among the closest genes to newly detected SNPs outside the chromosome 19q13 region, AP2A2 (corresponding to rs10794342), MYO16 (corresponding to rs9555561 and rs912322 in the 13q33.3 region) and STK32B (corresponding to rs17675640, rs6838792, and rs895681 in the 4p16.2 region) were previously associated with AD with p < 5E–06 (rs17393344 with p = 1.70E–08; and rs78647349 with p = 5.20E–07, respectively [69]). In addition, several chromosomal regions including 3p14.1 (KBTBD8), 6p21.33 (TNXB), 7q22.1 (TRIM56), 12q24.33 (SFSWAP), 18q12.1 (MIR302F), 21q21.3 (MIR155HG, LINC00515, and MRPL39), and 23q21.31 (KLHL4) were associated with AD at a suggestive level of associations by previous GWAS. However, no AD-associated SNPs with p < 5E–06 have been previously detected in chromosomal regions corresponding to C9orf92, PAX5, LHX1, and LINC00158 genes (i.e., 9p22.3, 9p13.2, 17q12, and 21q21.2, respectively) that were significant under plan 1; and ANTXR1 and SYK genes (i.e., 2p13.3 and 9q22.2, respectively) [43]. Detailed information about these AD-associated genes and chromosomal regions is provided in Additional files 23, and 4.

Nominally significant sets of SNPs

Under each of the three analysis plans, there were several SNPs associated with AD at a nominal level of significance (5E–06 ≤ p < 5E–02) in all datasets they were present in. They were mostly present in three datasets as they were excluded from one dataset by the QC procedure. These SNPs (30, 28, and 28 SNPs under plans 1–3, respectively) are listed in Additional files 23, and 4. Although they did not have highly significant p values, they are reported here due to the consistency in their association signals that was observed in multiple tested datasets. With the exception of rs575088, which had nominally significant p values in all datasets under plans 1 and 3, the significance pattern of the other SNPs was observed under only one plan. Also, rs2282079 (detected in females) was among the meta-analysis set of SNPs under plan 1 as well. None of these SNPs had p < 5E–06 in the conducted meta-analyses. The lack of meta-analysis power could be due to the small sample size, weak association signals, absence of some SNPs in one dataset, or heterogeneous effects of some SNPs across the different datasets as evidenced by their high i2 values, significant Q tests, and smaller p values in meta-analysis on absolute values of coefficients. The SNPs whose associated signals were reported here for the first time were not in LD with previously detected AD-associated loci (p < 5E–06) in their 1-Mb flanking regions (Additional file 1: Table S13). Interestingly, 22 out of 28 SNPs detected in males had the opposite pattern of significance in females (i.e., p > 0.05 in all datasets). Also, 26 out of 28 SNPs detected in females had the opposite pattern of significance in males (Additional file 5). Not all SNPs with opposite patterns of significance in females-only vs males-only analyses had the same pattern in the meta-analysis. Closest genes to some of these SNPs were located in chromosomal regions that were previously associated with AD with p < 5E–06. Information about these genes/regions can be found in Additional files 23, and 4.

Adjustment by APOE SNPs

For the AD-associated SNPs that were located on chromosome 19, we further investigated whether their association signals may change after adjustment for APOE genotypes in the models. For each subject, the APOE genotype was determined based on its genotypes at rs429358 and rs7412 loci using the coding schema provided in Additional file 1: Table S14. We found that none of the tested SNPs had p < 5E–06 once APOE was added as a covariate to the models.

Additional file 1: Table S15 summarizes the information regarding the LD between SNPs detected in our study and APOE SNPs. Among newly AD-associated SNPs, only six SNPs were in LD with one or both of the APOE SNPs. Others were not in LD with the two APOE SNPs (i.e., r2 = 0.001–0.072) [45]. Therefore, it should be noted that despite a major impact of the APOE genotypes on the associations of other SNPs inside the chromosome 19q13 region with AD, this result would not automatically imply that the APOE SNPs (i.e., rs429358 and rs7412) are the only contributors to AD pathogenesis because APOE-adjusted models highlighted the statistical correlations rather than biological (i.e., genetic) linkage. Further analyses such as those examining the role of haplotypes and epistatic interactions would be helpful to more comprehensively dissect the genetic heterogeneity of this region, and to elucidate the biological relevance of the APOE-adjusted models [70].

Sex-specific effects

We also investigated the sex-specific effects of SNPs that were significantly associated with AD only in males or females by performing a Wald χ2 test to determine whether their odds ratios were significantly different between males and females. Additional file 1: Tables S16 and S17 summarize the results from this test for replicated, nonreplicated, and meta-analysis sets of AD-associated SNPs. We found that the differences between odds ratios of the SNPs in males and females were significant (p < 0.05) in most cases, except rs62405605, rs1062851, rs62510850, rs7000333, rs6572843 (among nonreplicated set of SNPs in females), and rs12386284 (among meta-analysis set of SNPs in females). Detailed information about the results from the Wald χ2 test can be found in Additional file 6.

In addition, the SNPs that had significant p values only in males or females were searched against the GRASP catalog [43] to find out whether they were among the known sex-linked autosomal SNPs or were associated with any other diseases/traits at suggestive level of associations. We noticed that there was no evidence of such associations in previous studies.

Gene-based analysis

The significant findings from gene-based analyses corresponding to plans 1–3 are summarized in Table 3. Under all plans, most genes with significant p values at the FDR of 0.05 were located in the chromosome 19q13 region. Since the chromosome 19q13 region harbors several SNPs with p < 5E–08 in both current and previous studies, significant genes in this region are not discussed here as they do not meet the criteria set for detecting novel AD genes. The only significant genes outside the APOE cluster region were LINC00158 under plan 1 and LINC00158, MIR155HG, MIR155, LINC00515, MRPL39, and JAM2 under plan 3 that were located in the chromosome 21q21.3 region. None of the SNPs inside or within 1-Mb flanking regions of these genes had significant p values at the genome-wide level in our study, although several had suggestive-level p values in conducted meta-analyses under plans 1 and 3. Also, SNPs in 1-Mb nearby regions of these genes were only nominally associated with AD (8.0E–04 < p < 5E–02) in previous studies [58, 65, 71,72,73]. However, the chromosome 21q21.3 region was associated with AD by previous GWAS at a suggestive level of associations (rs239713 with p = 5.00E–07 [68]). This SNP is located ~ 1.6 Mb away from significant genes reported in our study.

Table 3 Significantly AD-associated genes from gene-based analyses

Pathway-based analysis

We found that 19, 10, and 19 pathways were significantly associated with AD under plans 1–3, respectively (Table 4). The proper FDR levels at which the numbers of possible false-positives were less than 1 were 0.05 under plans 1 and 2, and 0.025 under plan 3. We found that 12 pathways were significant under two or three analysis plans (i.e., they were not plan specific). There were also seven pathways that were significant only under plan 1 (males and females), and seven others were significant only in females (i.e., plan 3). No pathways were specifically significant in males (i.e., plan 2).

Table 4 Significantly AD-associated pathways from pathway-based analyses

TWA analysis

Analyzing probes with cis-eQTLs

Using eQTLs data from peripheral blood, we found that four, eight, and four probes/genes passed both the SMR (PSMR < 6.03E–05) and HEIDI (PHEIDI ≥ 0.05) tests under plans 1–3, respectively. The significant FDR level for interpreting the results from the SMR test was set to 0.05 under plan 1 and 0.025 under plans 2 and 3 to ensure that the expected number of false-positive findings was < 1. Table 5 presents information about these 16 probes/genes, their top eQTLs, and the respective p values. The top eQTLs corresponding to these probes/genes were all nominally significant in our GWA analyses (2.01E–04 ≤ PGWAS ≤ 2.47E–02). Moreover, we did not identify any SNPs with significant p values at the genome-wide significance level within 1 Mb of these genes. However, several SNPs within 1 Mb of MS4A6A [64, 66, 74, 75] and UQCC [76] were associated with AD with p < 5E–08 in previous studies. Among 14 other genes, SNPs in regions around TRA2A [64], IRAK3 [77], and ESPN [78] were previously associated with AD at the suggestive level of associations. In addition, ATG10 [77] and LPXN [74] were located in chromosomal regions (i.e., 5q14.1 and 11q12.1) that contained AD-associated SNPs with p < 5E–06.

Table 5 Significantly AD-associated probes/genes from transcriptome-wide analyses

Our TWA analyses on brain-specific eQTLs data revealed associations of two probes/genes with AD in males (i.e., CRIPAK and PRDM10), and two others in females (i.e., AHSA2 and ATG10) at the FDR level of 0.05 (Table 6). No probe/gene passed the SMR and HEIDI tests under analysis plan 1. The probe corresponding to the AHSA2 gene was significantly associated with AD in several brain regions (i.e., caudate basal ganglia, cerebellum, cortex, hypothalamus, nucleus accumbens, putamen basal ganglia, and substantia nigra). Also, ATG10 was significantly associated with AD in the nucleus accumbens and putamen basal ganglia. The corresponding top eQTLs were nominally significant in our GWA analyses in males and females (4.30E–05 ≤ PGWAS ≤ 9.33E–02). There were no SNPs with significant p values at the genome-wide significance level within 1 Mb of these genes in our study; however, SNPs with significant p values at the suggestive level of significance were found in flanking regions of ATG10 in the nonreplicated set of SNPs in females (see Additional file 4). In addition, the SNPs within 1 Mb of these four genes were only nominally associated with AD in previous studies [43]. In terms of chromosomal regions, in addition to ATG10 as explained earlier, SNPs in the chromosome 11q24.3 region (PRDM10 gene) were also previously associated with AD at a genome-wide significance level [69].

Table 6 Significantly AD-associated probe/genes from transcriptome-wide analyses on brain tissue data

Analyzing probes with trans-eQTLs

Using eQTLs data from peripheral blood, one probe mapping to the SFN gene on chromosome 1p36 had significant PSMR at the FDR level of 0.05, and passed the HEIDI test under plan 2 (Table 5). The corresponding top eQTL was located on chromosome 4p16 in the intronic region of the MAEA gene and was nominally associated with AD in our study (PGWAS = 4.10E–04). There were no significant association signals at the genome-wide significance level in the SFN gene or its 1-Mb flanking regions in current or previous studies [43].


The genetic architecture of AD has been widely studied in recent years, and so far more than 60,000 SNPs have been associated with AD with p < 0.05. Of these, 281 SNPs (mapped to 49 genes) and 593 SNPs (mapped to 165 genes) had significant p values at the genome-wide and suggestive levels of associations, respectively [43]. Despite these efforts, a major proportion of h2 of AD has remained unexplained. Exploring the genetic risk factors contributing to AD is highly important from a precision medicine perspective where the goal is to personalize diagnostic and therapeutic interventions. The current study provides further insight into the genetic architecture of AD through GWA, TWA, gene-based, and pathway-based analyses of four independent datasets. These datasets, particularly the LOADFS cohort, were partially used in previous genetic studies of AD [21, 72, 75, 79,80,81,82,83,84,85,86,87].

Our GWA analyses corroborated the associations of a number of previously detected AD loci and revealed some significant novel association signals. Among previously detected AD-associated SNPs, we found several SNPs with p values that were smaller than those reported before. Also, the significant association signals for three SNPs inside the chromosome 19q13 region (i.e., nonreplicated rs2965169 SNP under plan 1, rs10426423 from the meta-analysis set of SNPs under plan 1, and rs769450 from the replicated set of SNPs under plan 1 and the nonreplicated sets of SNPs under plans 2 and 3) were previously reported only in African-Americans (p = 2.6E–8, p = 9.9E–7, and p = 5.3E–27, respectively [88]). Most newly detected AD-associated SNPs, particularly those outside the chromosome 19q13 region, can be considered informative AD markers because their p values in our study were smaller than those for other AD-associated loci in their 1-Mb upstream/downstream regions and they were not in LD with such loci. For instance, as seen in Table 2 that summarizes the replicated and meta-analysis sets of SNPs, 11, 4, and 21 novel AD-associated SNPs were detected under plans 1–3, respectively. These SNPs were mapped to 21 genes in 19 chromosomal regions (i.e., cytogenetic regions). Of these, four genes had been associated with AD in previous GWAS with p < 5E–06. Also, nine genes were located in eight chromosomal regions that contained previously AD-associated SNPs that were > 1 Mb away from the SNPs detected in our study. The other eight genes/regions had not been associated with AD in previous studies at genome-wide or suggestive levels of associations [43].

Our GWA analyses also revealed associations of a number of SNPs (41, 33, and 40 SNPs under plans 1–3, respectively) with AD that were present only in one of the four investigated cohorts. While successful replication of a discovered association in an independent cohort has become the gold standard in genome-wide association studies for substantiating the real genetic effects, failure to replicate SNP–disease associations does not necessarily indicate that they are false-positive findings. Instead, they might be real genetic contributors that confer population-specific risks due to the genetic heterogeneity of the disease [2, 60, 89, 90]. Other reasons for nonreproducibility can be the lack of statistical power due to insufficient sample sizes, the presence of environmental or gene–gene interactions, and a lack of genotyping information for particular loci in different studies. For instance, small between-population allele frequency differences at an interacting locus may result in a lack of power to detect the main effect of a genuine association signal in independent cohorts [60]. These reasons can also justify why not all previously discovered AD-associated SNPs were replicated in our study.

Of particular interest was to investigate the sex disparity in the genetic basis of AD. Addressing sex differences in biomedical research has been emphasized by the National Institutes of Health as an approach that can eventually bolster the personalized medicine paradigm [14, 15]. Our results revealed a number of new sex-specific genetic contributors to AD at the SNP, gene, and transcriptome levels. For instance, most of the newly detected SNPs, particularly SNPs outside chromosome 19q13, were sex specific as they had significant p values either in males or females and, in addition, their odds ratios were significantly different between the two genders. Interestingly, there were two additional subsets of SNPs that were nominally associated with AD in all datasets in one sex while they were nonsignificant in all datasets in the other. Such consistent sex-specific association signals, although weak, might be important in exploring the differences in genetic risk factors of AD between males and females and may demonstrate genome-wide significance in larger samples. Another level of sex disparity was observed in the gene-based and TWA analyses where several genes were significantly associated with AD in either males or females. Also, there were several pathways that were specifically significant in females. These will be further discussed in the following paragraphs.

In the gene-based analysis, LINC00158, MIR155HG, MIR155, LINC00515, MRPL39, and JAM2 were significantly associated with AD when the entire sample of individuals and/or only females were analyzed. These genes are located near each other on chromosome 21q21.3 in a ~ 332-kb region. The APP gene implicated in early onset familial AD or Down syndrome-related AD [4] is also located 163–449 kb from these genes. There were no AD-associated SNPs with p < 5E–08 within their 1 Mb in current or previous studies [43]. However, there were several SNPs with significant p values at the suggestive level of associations in that chromosomal region among meta-analysis sets of SNPs under plan 1 (i.e., rs76252969 and rs2298369) and plan 3 (i.e., rs12386284, rs1783012, rs1783013, rs926963, rs1893650, rs2226326, rs2829803, rs2298369, rs2829823, and rs2829832). The SNPs in the 1-Mb upstream/downstream regions of these genes were previously associated with some potential AD risk factors such as type 2 diabetes, hypertension, coronary artery disease, and lipid profile changes at the genome-wide significance level. They have also been associated with traits such as alcohol and nicotine codependence, age at onset of Parkinson’s disease, and pattern recognition memory at the suggestive significance level of association [43]. Furthermore, functional studies have provided insight into the potential roles of some of these genes in AD pathogenesis. For instance, MIR155HG and MIR155 encode two microRNAs. MIR155 overexpression was previously implicated in downregulation of complement factor H (CFH) expression in AD and other neurodegenerative diseases which in turn may prevent spontaneous immune system activation [91]. MRPL39 encodes a mitochondrial ribosomal protein involved in the oxidative–phosphorylation pathway. Impaired mitochondrial function has been reported in neurons of patients with AD [92, 93]. Lunnon et al. [92] reported that the expression levels of MRPL39 and another nearby gene (i.e., ATP5J involved in the oxidative–phosphorylation pathway) were slightly reduced in AD patients compared to controls. JAM2 encodes a membrane protein found at the tight junctions of epithelial and endothelial cells that acts as an adhesive ligand for immune cells. It belongs to the immunoglobulin superfamily of adhesive molecules that has been implicated in AD pathogenesis [94]. Also, duplication of an ~ 600-kb region on chromosome 21 containing the JAM2, ATP5J, and APP genes has been reported in autosomal dominant AD [95].

In TWA analyses using brain-specific eQTLs data, four probes/genes were associated with AD (two in males and two in females). Also, using eQTLs data from peripheral blood, the expression level of 17 probes/genes passed both the SMR and the HEIDI tests, indicating that variants influencing the expression of these genes may also have pleiotropic effects on developing AD [55, 56]. It should be noted that due to the tissue-specific expression of genes, using data from eQTLs studies on blood is not ideal for capturing associations between the transcriptome levels and AD. However, it increases the power of SMR analysis since such studies take advantage of more samples compared to brain-specific eQTLs studies [55]. Significant SNPs with p < 5E–08 were detected within 1 Mb of MS4A6A and UQCC genes (significant in TWA analyses of blood eQTLs data) in our GWAS or previous reports [43]. SNPs with p < 5E–06 were present only in 1-Mb upstream/downstream regions of ATG10 (significant in brain-specific TWA analyses) in our GWA analyses of females, although several AD-associated SNPs with p < 5E–06 were reported in regions around TRA2A [64], IRAK3 [77], and ESPN [78]. This is likely indicative of the lack of power of conducted GWAS due to insufficient sample sizes [55].

Taken together, all AD-associated genes in our TWA analyses except MS4A6A and UQCC can be considered novel potential AD-associated genes. Further functional analyses are needed to explore their potential roles in AD pathogenesis as detected associations do not imply causation. Instead, they provide a list of prioritized candidates for follow-up studies. SNPs in 1-Mb upstream/downstream regions around these genes have been previously associated with some other traits (e.g., autoimmune diseases or serum cholesterol levels) with p < 5E–06. Examples include associations of SNPs corresponding to ABCB9 with college completion and years of education, ATG10 with vascular dementia, C9orf72 with amyotrophic lateral sclerosis, frontotemporal lobar degeneration, and response of rheumatoid arthritis patients to anti-TNF treatment, GNAI3 with total and low-density lipoprotein cholesterol (LDL-C) and major depression, LPXN with inflammatory bowel disease, MED30 with rheumatoid arthritis and fasting blood glucose, PRDM10 with type 2 diabetes, and SFN with high-density lipoprotein cholesterol (HDL-C) [43].

Notably, none of the novel AD-associated genes detected in males were among the significant genes in females and vice versa. Among the significant genes detected in females, a pathologic hexa-nucleotide repeat expansion in the C9orf72 gene has been linked to frontotemporal dementia and may contribute to AD pathogenesis [96,97,98,99]. Also, the GNAI3 gene was reported to be overexpressed in AD intact mice compared to AD impaired ones [100]. CRIPAK, which was among significant genes detected in brain-specific TWA analyses in males, is an inhibitor of the PAK1 gene [101]. The PAK gene family was found to play roles in learning and memory, and the dysregulations were implicated in AD, Huntington disease, and mental retardation [102]. Also, rs1923775 located ~ 700 kb away from CRIPAK has shown relatively strong association (p = 5.60E–6) with AD in African Americans [88].

Of 26 pathways that were significantly associated with AD in our pathway-based analyses, 12 were not plan specific, seven were specifically significant only under plan 1 (males and females), and seven were specifically significant only in females (i.e., plan 3). Pathways that were significant in more than one plan were mostly involved in processes such as mitochondrial function, lipid metabolism, cell junctions, and immune and inflammatory responses that were implicated in AD [93, 103,104,105,106]. There are several lines of evidence in previous empirical studies substantiating the potential roles of some of the detected plan-specific pathways in AD pathogenesis. For instance, it was suggested that deactivation of the epidermal growth factor receptor (EGFR) signaling pathway may attenuate the Aβ-induced memory loss in Drosophila and mice models [107]. Also, the fragmentation and dysfunction of Golgi apparatus, an organelle involved in the posttranslational modifications and trafficking of proteins, has been implicated in AD pathogenesis [108, 109]. The upregulation of the Fas signaling pathway, involved in the apoptosis and modulating immune responses, was reported to contribute to the Aβ-induced cell death and neurodegeneration in AD [110, 111]. Also, dysregulation of the platelet-derived growth factor (PDGF) signaling pathway was suggested to increase Aβ production and contribute to the neurodegeneration in AD [112, 113].

Among the female-specific pathways, G-protein activation is a signal transduction pathway that can modulate the production and action of different intracellular effector proteins. The G protein-coupled receptors play important roles in the initiation and regulation of inflammatory responses such as phagocyte chemotaxis and cytokine production [50, 114]. The pathologically increased inflammatory responses were reported in the brain of patients with AD [93]. Gβγ signaling through the PI3Kγ pathway is involved in the regulation of immune system responses and platelet activation [115]. Also, the ADP signaling, signal amplification, and prostacyclin signaling pathways are involved in the regulation of platelets activation in response to injury or in healthy blood vessels [50]. Platelets, as the major sources of amyloid precursor protein (APP) and Aβ in blood, were reported to be overactivated in AD patients possibly due to their stimulation by injured cerebral endothelial cells or by their cell membrane abnormalities [116, 117]. The glucagon-type ligand receptors are found in the gastrointestinal epithelium and brain neurons. Glucagon-like peptide-1 (GLP-1) has been suggested as a potential treatment to reverse the neurodegeneration in AD and Parkinson’s disease [118, 119].


In summary, our study revealed significant associations of several SNPs at genome-wide or suggestive levels of significance which were not reported before. Most of the SNPs that were located outside the APOE cluster gene region were not in LD with previously discovered AD-associated polymorphisms that had p < 5E–06 (Table 2). These SNPs were mapped to 21 genes in 19 chromosomal regions. Of these, 8 genes/regions had not been associated with AD in previous GWAS with p < 5E–06. Also, 26 genes located outside the chromosome 19q13 region, and 26 pathways, showed evidence of associations with AD at the FDR level of 0.05 in our TWA, gene-based, and pathway-based analyses. Thirteen of these 26 genes were located in chromosomal regions with no AD-associated SNPs at the genome-wide or suggestive level of significance. Most of the significantly detected SNPs and genes as well as several AD-associated pathways were sex specific, indicating sex disparities in the genetic basis of AD. By detecting a number of novel potential AD-associated SNPs and discovering suggestive associations of several genes and transcripts, our study provides new insight into the genetic architecture of AD. Particularly, identifying sex-specific genetic contributors can advance our understanding of AD pathogenesis.

Despite the rigor of this study, there are some limitations. The case/control status in the four cohorts used in this study was mainly determined clinically. The routine clinical diagnosis of AD based on the symptoms and neurologic examinations may not provide the optimal case/control classification. Instead, the National Institute on Aging and the Alzheimer’s Association suggested that integrating additional paraclinical tests (e.g., histopathologic findings in brain biopsy, measuring AD-related cerebrospinal fluid (CSF) biomarkers, or detecting neurodegeneration by the imaging study) into the diagnostic protocols can aid researchers to more accurately identify AD patients and healthy controls [120, 121]. Beach et al. [122] investigated the accuracy of clinical diagnosis of AD by comparing such diagnoses to the histopathology findings from brain autopsies in a sample of 1198 subjects. They found that the sensitivity and specificity of clinical diagnostic classification were 70.9–87.3% and 44.3–70.8%, respectively, indicating a relatively high possibility of clinically false-negative and false-positive classification of subjects as controls and cases, respectively [122]. Finally, since the power of GWA analyses is affected by the sample sizes, and in particular the number of cases, the current study with 2741 cases and 14,739 controls may not have the optimal power. Further studies, possibly with larger sample sizes, are needed to clarify the genotype–phenotype relationships in AD.



Alzheimer’s disease


Apolipoprotein E


Amyloid precursor protein


Amyloid beta


Brodmann Area 24


Brodmann Area 9


Utah Residents with Northern and Western European Ancestry


Cardiovascular Health Study


Database of Genotypes and Phenotypes


Expression quantitative trait loci


European population haplotypes


Fast Set-Based Association Test


False discovery rate


Framingham Heart Study


Genome-wide Complex Trait Analysis


Generalized linear mixed model


Glucagon-like peptide-1


Genome-wide Repository of Associations between SNPs and Phenotypes


Gene set enrichment analysis


Genome-wide association analysis


Genome-wide association meta-analysis


Genome-wide association study


High-density lipoprotein cholesterol


Heterogeneity in Dependent Instruments


Human Genome Build 19


Human Genome Build 38


Health and Retirement Study




International Classification of Disease Codes, ninth revision




Institutional review board


Kinship-Based Inference for GAWS


Linkage disequilibrium


Low-density lipoprotein cholesterol


Molecular Signatures Database


National Center for Biotechnology Information


Late-Onset Alzheimer’s Disease Family Study from the National Institute on Aging


Odds ratio


Principal component


Principal component analysis


Pathway Interaction Database


Quality control


Segmented Haplotype Estimation and Imputation Tool


Framingham SNP Health Association Resource


Summary-data-based Mendelian Randomization


Single-nucleotide polymorphism


SNP Typing for Association with Multiple Phenotypes from Existing Epidemiologic Data


Transcriptome-wide association analysis


University of California, Santa Cruz


  1. 1.

    Kasper DL, Fauci AS, Hauser SL, Longo DL, Jameson JL, Loscalzo J. Harrison’s principles of internal medicine. 19th ed. New York: McGraw-Hill Education/Medical; 2015.

  2. 2.

    Yashin AI, Fang F, Kovtun M, Wu D, Duan M, Arbeev K, et al. Hidden heterogeneity in Alzheimer’s disease: insights from genetic association studies and other analyses. Exp Gerontol. 2018;107:148–60.

  3. 3.

    Todd S, Barr S, Roberts M, Passmore AP. Survival in dementia and predictors of mortality: a review. Int J Geriatr Psychiatry. 2013;28:1109–24.

  4. 4.

    Bird TD. Genetic aspects of Alzheimer disease. Genet Med. 2008;10:231–9.

  5. 5.

    Alzheimer’s Association. 2016 Alzheimer’s disease facts and figures. Alzheimers Dement. 2016;12:459–509.

  6. 6.

    National Center for Health Statistics. Health, United States, 2016: with chartbook on long-term trends in health. Hyattsville: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention; 2017. p. 488. Report No.: 2017–1232.

  7. 7.

    Mayeux R. Epidemiology of neurodegeneration. Annu Rev Neurosci. 2003;26:81–104.

  8. 8.

    Andersen K, Launer LJ, Dewey ME, Letenneur L, Ott A, Copeland JR, et al. Gender differences in the incidence of AD and vascular dementia: the EURODEM Studies. EURODEM Incidence Research Group. Neurology. 1999;53:1992–7.

  9. 9.

    Carter CL, Resnick EM, Mallampalli M, Kalbarczyk A. Sex and gender differences in Alzheimer’s disease: recommendations for future research. J Women's Health (Larchmt). 2012;21:1018–23.

  10. 10.

    Mielke MM, Vemuri P, Rocca WA. Clinical epidemiology of Alzheimer’s disease: assessing sex and gender differences. Clin Epidemiol. 2014;6:37–48.

  11. 11.

    Genin E, Hannequin D, Wallon D, Sleegers K, Hiltunen M, Combarros O, et al. APOE and Alzheimer disease: a major gene with semi-dominant inheritance. Mol Psychiatry. 2011;16:903–7.

  12. 12.

    Henderson VW, Buckwalter JG. Cognitive deficits of men and women with Alzheimer’s disease. Neurology. 1994;44:90–6.

  13. 13.

    Barnes LL, Wilson RS, Bienias JL, Schneider JA, Evans DA, Bennett DA. Sex differences in the clinical manifestations of Alzheimer disease pathology. Arch Gen Psychiatry. 2005;62:685–91.

  14. 14.

    Ronquillo JG, Baer MR, Lester WT. Sex-specific patterns and differences in dementia and Alzheimer’s disease using informatics approaches. J Women Aging. 2016;28:403–11.

  15. 15.

    Clayton JA, Collins FS. Policy: NIH to balance sex in cell and animal studies. Nature News. 2014;509:282-3.

  16. 16.

    Pericak-Vance MA, Bebout JL, Gaskell PC, Yamaoka LH, Hung WY, Alberts MJ, et al. Linkage studies in familial Alzheimer disease: evidence for chromosome 19 linkage. Am J Hum Genet. 1991;48:1034–50.

  17. 17.

    Corder EH, Saunders AM, Strittmatter WJ, Schmechel DE, Gaskell PC, Small GW, et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science. 1993;261:921–3.

  18. 18.

    Raghavan N, Tosto G. Genetics of Alzheimer’s disease: the importance of polygenic and epistatic components. Curr Neurol Neurosci Rep. 2017;17:78.

  19. 19.

    Ridge PG, Hoyt KB, Boehme K, Mukherjee S, Crane PK, Haines JL, et al. Assessment of the genetic variance of late-onset Alzheimer’s disease. Neurobiol Aging. 2016;41:200.e13-200.e20.

  20. 20.

    Gatz M, Reynolds CA, Fratiglioni L, et al. Role of genes and environments for explaining Alzheimer disease. Arch Gen Psychiatry. 2006;63:168–74.

  21. 21.

    Lee JH, Cheng R, Graff-Radford N, Foroud T, Mayeux R. Analyses of the national institute on aging late-onset Alzheimer’s disease family study: implication of additional loci. Arch Neurol. 2008;65:1518–26.

  22. 22.

    Dawber TR, Meadors GF, Moore FE. Epidemiological approaches to heart disease: the Framingham study. Am J Public Health Nations Health. 1951;41:279–86.

  23. 23.

    Feinleib M, Kannel WB, Garrison RJ, McNamara PM, Castelli WP. The Framingham offspring study: design and preliminary data. Prev Med. 1975;4:518–25.

  24. 24.

    Splansky GL, Corey D, Yang Q, Atwood LD, Cupples LA, Benjamin EJ, et al. The Third Generation Cohort of the National Heart, Lung, and Blood Institute’s Framingham Heart Study: design, recruitment, and initial examination. Am J Epidemiol. 2007;165:1328–35.

  25. 25.

    Fried LP, Borhani NO, Enright P, Furberg CD, Gardin JM, Kronmal RA, et al. The cardiovascular health study: design and rationale. Ann Epidemiol. 1991;1:263–76.

  26. 26.

    Sonnega A, Faul JD, Ofstedal MB, Langa KM, Phillips JW, Weir DR. Cohort profile: the health and retirement study (HRS). Int J Epidemiol. 2014;43:576–85.

  27. 27.

    Hinrichs AS, Karolchik D, Baertsch R, Barber GP, Bejerano G, Clawson H, et al. The UCSC genome browser database: update 2006. Nucleic Acids Res. 2006;34:D590–8.

  28. 28.

    Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet. 2007;81:559–75.

  29. 29.

    Delaneau O, Marchini J, Zagury J-F. A linear complexity phasing method for thousands of genomes. Nat Methods. 2012;9:179-81.

  30. 30.

    Das S, Forer L, Schönherr S, Sidore C, Locke AE, Kwong A, et al. Next-generation genotype imputation service and methods. Nat Genet. 2016;48:1284-87.

  31. 31.

    Verma SS, de Andrade M, Tromp G, Kuivaniemi H, Pugh E, Namjou-Khales B, et al. Imputation and quality control steps for combining multiple genome-wide datasets. Front Genet. 2014;5:370.

  32. 32.

    Manichaikul A, Mychaleckyj JC, Rich SS, Daly K, Sale M, Chen W-M. Robust relationship inference in genome-wide association studies. Bioinformatics. 2010;26:2867–73.

  33. 33.

    R Core Team. R: a language and environment for statistical computing. Vienna: R Foundation for Statistical Computing; 2013.

  34. 34.

    Conomos MP, Miller MB, Thornton TA. Robust inference of population structure for ancestry prediction and correction of stratification in the presence of relatedness. Genet Epidemiol. 2015;39:276–93.

  35. 35.

    Wellcome Trust Case Control Consortium. Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature. 2007;447:661–78.

  36. 36.

    Winkler TW, Day FR, Croteau-Chonka DC, Wood AR, Locke AE, Mägi R, et al. Quality control and conduct of genome-wide association meta-analyses. Nat Protoc. 2014;9:1192–212.

  37. 37.

    McArdle PF, O’Connell JR, Pollin TI, Baumgarten M, Shuldiner AR, Peyser PA, et al. Accounting for relatedness in family based genetic association studies. Hum Hered. 2007;64:234–42.

  38. 38.

    Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. J Stat Softw. 2015;67:1-48.

  39. 39.

    Ziegler A, König IR, Thompson JR. Biostatistical aspects of genome-wide association studies. Biom J. 2008;50:8–28.

  40. 40.

    Reed E, Nunez S, Kulp D, Qian J, Reilly MP, Foulkes AS. A guide to genome-wide association analysis and post-analytic interrogation. Stat Med. 2015;34:3769–92.

  41. 41.

    Mägi R, Morris AP. GWAMA: software for genome-wide association meta-analysis. BMC Bioinformatics. 2010;11:288.

  42. 42.

    Allison PD. Comparing logit and probit coefficients across groups. Sociol Methods Res. 1999;28:186–208.

  43. 43.

    Leslie R, O’Donnell CJ, Johnson AD. GRASP: analysis of genotype-phenotype results from 1390 genome-wide association studies and corresponding open access database. Bioinformatics. 2014;30:i185–94.

  44. 44.

    Zhbannikov IY, Arbeev K, Ukraintseva S, Yashin AI. haploR: an R package for querying web-based annotation tools. F1000Res. 2017;6:97.

  45. 45.

    Machiela MJ, Chanock SJ. LDlink: a web-based application for exploring population-specific haplotype structure and linking correlated alleles of possible functional variants. Bioinformatics. 2015;31:3555–7.

  46. 46.

    Casper J, Zweig AS, Villarreal C, Tyner C, Speir ML, Rosenbloom KR, et al. The UCSC Genome Browser database: 2018 update. Nucleic Acids Res. 2018;46:D762–9.

  47. 47.

    Bakshi A, Zhu Z, Vinkhuyzen AAE, Hill WD, McRae AF, Visscher PM, et al. Fast set-based association analysis using summary data from GWAS identifies novel gene loci for human complex traits. Sci Rep. 2016;6:32894.

  48. 48.

    Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome-wide complex trait analysis. Am J Hum Genet. 2011;88:76–82.

  49. 49.

    Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B Methodol. 1995;57:289–300.

  50. 50.

    Fabregat A, Jupe S, Matthews L, Sidiropoulos K, Gillespie M, Garapati P, et al. The reactome pathway knowledgebase. Nucleic Acids Res. 2018;46:D649–55.

  51. 51.

    Schaefer CF, Anthony K, Krupa S, Buchoff J, Day M, Hannay T, et al. PID: the pathway interaction database. Nucleic Acids Res. 2009;37:D674–9.

  52. 52.

    Mootha VK, Lindgren CM, Eriksson K-F, Subramanian A, Sihag S, Lehar J, et al. PGC-1α-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet. 2003;34:267–73.

  53. 53.

    Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. PNAS. 2005;102:15545–50.

  54. 54.

    Lloyd-Jones LR, Holloway A, McRae A, Yang J, Small K, Zhao J, et al. The genetic architecture of gene expression in peripheral blood. Am J Hum Genet. 2017;100:228–37.

  55. 55.

    Zhu Z, Zhang F, Hu H, Bakshi A, Robinson MR, Powell JE, et al. Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets. Nat Genet. 2016;48:481–7.

  56. 56.

    Pavlides JMW, Zhu Z, Gratten J, McRae AF, Wray NR, Yang J. Predicting gene targets from integrative analyses of summary data from GWAS and eQTL studies for 28 human complex traits. Genome Med. 2016;8:84.

  57. 57.

    GTEx Consortium. Genetic effects on gene expression across human tissues. Nature. 2017;550:204–13.

  58. 58.

    Hu X, Pickering E, Liu YC, Hall S, Fournier H, Katz E, et al. Meta-analysis for genome-wide association study identifies multiple variants at the BIN1 locus associated with late-onset Alzheimer’s disease. PLoS One. 2011;6:e16616.

  59. 59.

    Lin P-I, Vance JM, Pericak-Vance MA, Martin ER. No gene is an island: the flip-flop phenomenon. Am J Hum Genet. 2007;80:531–8.

  60. 60.

    Greene CS, Penrod NM, Williams SM, Moore JH. Failure to replicate a genetic association may provide important clues about genetic architecture. PLoS One. 2009;4:e5639.

  61. 61.

    Kulminski AM, Kernogitski Y, Culminskaya I, Loika Y, Arbeev KG, Bagley O, et al. Uncoupling associations of risk alleles with endophenotypes and phenotypes: insights from the ApoB locus and heart-related traits. Aging Cell. 2017;16:61–72.

  62. 62.

    Furney SJ, Simmons A, Breen G, Pedroso I, Lunnon K, Proitsi P, et al. Genome-wide association with MRI atrophy measures as a quantitative trait locus for Alzheimer’s disease. Mol Psychiatry. 2011;16:1130–8.

  63. 63.

    Wijsman EM, Pankratz ND, Choi Y, Rothstein JH, Faber KM, Cheng R, et al. Genome-wide association of familial late-onset Alzheimer’s disease replicates BIN1 and CLU and nominates CUGBP2 in interaction with APOE. PLoS Genet. 2011;7:e1001308.

  64. 64.

    Li H, Wetten S, Li L, St Jean PL, Upmanyu R, Surh L, et al. Candidate single-nucleotide polymorphisms from a genomewide association study of Alzheimer disease. Arch Neurol. 2008;65:45–53.

  65. 65.

    Heinzen EL, Need AC, Hayden KM, Chiba-Falek O, Roses AD, Strittmatter WJ, et al. Genome-wide scan of copy number variation in late-onset Alzheimer’s disease. J Alzheimers Dis. 2010;19:69–77.

  66. 66.

    Hollingworth P, Harold D, Sims R, Gerrish A, Lambert J-C, Carrasquillo MM, et al. Common variants at ABCA7, MS4A6A/MS4A4E, EPHA1, CD33 and CD2AP are associated with Alzheimer’s disease. Nat Genet. 2011;43:429–35.

  67. 67.

    Lambert J-C, Grenier-Boley B, Harold D, Zelenika D, Chouraki V, Kamatani Y, et al. Genome-wide haplotype association study identifies the FRMD4A gene as a risk locus for Alzheimer’s disease. Mol Psychiatry. 2013;18:461–70.

  68. 68.

    Han M-R, Schellenberg GD, Wang L-S. Alzheimer’s Disease Neuroimaging Initiative. Genome-wide association reveals genetic effects on human Aβ42 and τ protein levels in cerebrospinal fluids: a case control study. BMC Neurol. 2010;10:90.

  69. 69.

    Sherva R, Tripodis Y, Bennett DA, Chibnik LB, Crane PK, de Jager PL, et al. Genome-wide association study of the rate of cognitive decline in Alzheimer’s disease. Alzheimers Dement. 2014;10:45–52.

  70. 70.

    Kulminski AM, Huang J, Wang J, He L, Loika Y, Culminskaya I. Apolipoprotein E region molecular signatures of Alzheimer’s disease. Aging Cell. 2018;23:e12779.

  71. 71.

    Gerrish A, Russo G, Richards A, Moskvina V, Ivanov D, Harold D, et al. The role of variation at AβPP, PSEN1, PSEN2, and MAPT in late onset Alzheimer’s disease. J Alzheimers Dis. 2012;28:377–87.

  72. 72.

    Hollingworth P, Sweet R, Sims R, Harold D, Russo G, Abraham R, et al. Genome-wide association study of Alzheimer’s disease with psychotic symptoms. Mol Psychiatry. 2012;17:1316–27.

  73. 73.

    Jun G, Moncaster JA, Koutras C, Seshadri S, Buros J, McKee AC, et al. δ-Catenin is genetically and biologically associated with cortical cataract and future Alzheimer-related structural and functional brain changes. PLoS One. 2012;7:e43728.

  74. 74.

    Antúnez C, Boada M, González-Pérez A, Gayán J, Ramírez-Lorca R, Marín J, et al. The membrane-spanning 4-domains, subfamily A (MS4A) gene cluster contains a common variant associated with Alzheimer’s disease. Genome Med. 2011;3:33.

  75. 75.

    Naj AC, Jun G, Beecham GW, Wang L-S, Vardarajan BN, Buros J, et al. Common variants at MS4A4/MS4A6E, CD2AP, CD33 and EPHA1 are associated with late-onset Alzheimer’s disease. Nat Genet. 2011;43:436–41.

  76. 76.

    Poduslo SE, Huang R, Huang J, Smith S. Genome screen of late-onset Alzheimer’s extended pedigrees identifies TRPC4AP by haplotype analysis. Am J Med Genet B Neuropsychiatr Genet. 2009;150B:50–5.

  77. 77.

    Potkin SG, Guffanti G, Lakatos A, Turner JA, Kruggel F, Fallon JH, et al. Hippocampal atrophy as a quantitative trait in a genome-wide association study identifying novel susceptibility genes for Alzheimer’s disease. PLoS One. 2009;4:e6501.

  78. 78.

    Carrasquillo MM, Zou F, Pankratz VS, Wilcox SL, Ma L, Walker LP, et al. Genetic variation in PCDH11X is associated with susceptibility to late-onset Alzheimer’s disease. Nat Genet. 2009;41:192–8.

  79. 79.

    Seshadri S, DeStefano AL, Au R, Massaro JM, Beiser AS, Kelly-Hayes M, et al. Genetic correlates of brain aging on MRI and cognitive test measures: a genome-wide association and linkage analysis in the Framingham study. BMC Med Genet. 2007;8:S15.

  80. 80.

    Seshadri S, Fitzpatrick AL, Ikram MA, DeStefano AL, Gudnason V, Boada M, et al. Genome-wide analysis of genetic loci associated with Alzheimer disease. JAMA. 2010;303:1832–40.

  81. 81.

    Irie F, Fitzpatrick AL, Lopez OL, Kuller LH, Peila R, Newman AB, et al. Enhanced risk for Alzheimer disease in persons with type 2 diabetes and APOE epsilon4: the Cardiovascular Health Study Cognition Study. Arch Neurol. 2008;65:89–93.

  82. 82.

    Jun G, Naj AC, Beecham GW, Wang L-S, Buros J, Gallins PJ, et al. Meta-analysis confirms CR1, CLU, and PICALM as alzheimer disease risk loci and reveals interactions with APOE genotypes. Arch Neurol. 2010;67:1473–84.

  83. 83.

    Sweet RA, Seltman H, Emanuel JE, Lopez OL, Becker JT, Bis JC, et al. Effect of Alzheimer disease risk genes on trajectories of cognitive function in the Cardiovascular Health Study. Am J Psychiatry. 2012;169:954–62.

  84. 84.

    Miyashita A, Koike A, Jun G, Wang L-S, Takahashi S, Matsubara E, et al. SORL1 is genetically associated with late-onset Alzheimer’s disease in Japanese, Koreans and Caucasians. PLoS ONE. 2013;8:e58618.

  85. 85.

    Reitz C, Jun G, Naj A, Rajbhandary R, Vardarajan BN, Wang L-S, et al. Variants in the ATP-binding cassette transporter (ABCA7), apolipoprotein E ϵ4,and the risk of late-onset Alzheimer disease in African Americans. JAMA. 2013;309:1483–92.

  86. 86.

    Lambert JC, Ibrahim-Verbaas CA, Harold D, Naj AC, Sims R, Bellenguez C, et al. Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer’s disease. Nat Genet. 2013;45:1452–8.

  87. 87.

    Mez J, Marden JR, Mukherjee S, Walter S, Gibbons LE, Gross AL, et al. Alzheimer’s disease genetic risk variants beyond APOE ε4 predict mortality. Alzheimers Dement (Amst). 2017;8:188–95.

  88. 88.

    Logue MW, Schu M, Vardarajan BN, Buros J, Green RC, Go RCP, et al. A comprehensive genetic association study of Alzheimer disease in African Americans. Arch Neurol. 2011;68:1569–79.

  89. 89.

    Shriner D, Vaughan LK, Padilla MA, Tiwari HK. Problems with genome-wide association studies. Science. 2007;316:1840–2.

  90. 90.

    Kulminski AM, Loika Y, Culminskaya I, Arbeev KG, Ukraintseva SV, Stallard E, et al. Explicating heterogeneity of complex traits has strong potential for improving GWAS efficiency. Sci Rep. 2016;6:35390.

  91. 91.

    Lukiw WJ, Alexandrov PN. Regulation of complement factor H (CFH) by multiple miRNAs in Alzheimer’s disease (AD) brain. Mol Neurobiol. 2012;46:11–9.

  92. 92.

    Lunnon K, Keohane A, Pidsley R, Newhouse S, Riddoch-Contreras J, Thubron EB, et al. Mitochondrial genes are altered in blood early in Alzheimer’s disease. Neurobiol Aging. 2017;53:36–47.

  93. 93.

    Querfurth HW, LaFerla FM. Alzheimer’s Dis. N Engl J Med. 2010;362:329–44.

  94. 94.

    Leshchyns’ka I, Sytnyk V. Synaptic cell adhesion molecules in Alzheimer’s disease. Neural Plast. 2016;2016:6427537.

  95. 95.

    Antonell A, Gelpi E, Sánchez-Valle R, Martínez R, Molinuevo JL, Lladó A. Breakpoint sequence analysis of an AβPP locus duplication associated with autosomal dominant Alzheimer’s disease and severe cerebral amyloid angiopathy. J Alzheimers Dis. 2012;28:303–8.

  96. 96.

    Bieniek KF, Murray ME, Rutherford NJ, Castanedes-Casey M, DeJesus-Hernandez M, Liesinger AM, et al. Tau pathology in frontotemporal lobar degeneration with C9ORF72 hexanucleotide repeat expansion. Acta Neuropathol. 2013;125:289–302.

  97. 97.

    Cacace R, Van Cauwenberghe C, Bettens K, Gijselinck I, van der Zee J, Engelborghs S, et al. C9orf72 G4C2 repeat expansions in Alzheimer’s disease and mild cognitive impairment. Neurobiol Aging. 2013;34:1712.e1–7.

  98. 98.

    Harms M, Benitez BA, Cairns N, Cooper B, Cooper P, Mayo K, et al. C9orf72 hexanucleotide repeat expansions in clinical Alzheimer disease. JAMA Neurol. 2013;70:736–41.

  99. 99.

    Khan BK, Yokoyama JS, Takada LT, Sha SJ, Rutherford NJ, Fong JC, et al. Atypical, slowly progressive behavioural variant frontotemporal dementia associated with C9ORF72 hexanucleotide expansion. J Neurol Neurosurg Psychiatry. 2012;83:358–64.

  100. 100.

    Neuner SM, Wilmott LA, Hoffmann BR, Mozhui K, Kaczorowski CC. Hippocampal proteomics defines pathways associated with memory decline and resilience in normal aging and Alzheimer’s disease mouse models. Behav Brain Res. 2017;322:288–98.

  101. 101.

    Talukder AH, Meng Q, Kumar R. CRIPak, a novel endogenous Pak1 inhibitor. Oncogene. 2006;25:1311–9.

  102. 102.

    Ma Q-L, Yang F, Frautschy SA, Cole GM. PAK in Alzheimer disease, Huntington disease and X-linked mental retardation. Cell Logist. 2012;2:117–25.

  103. 103.

    Stamatovic SM, Keep RF, Andjelkovic AV. Brain endothelial cell-cell junctions: how to “open” the blood brain barrier. Curr Neuropharmacol. 2008;6:179–92.

  104. 104.

    El-Amraoui A, Petit C. Cadherins as targets for genetic diseases. Cold Spring Harb Perspect Biol. 2010;2:a003095.

  105. 105.

    Rikitake Y, Mandai K, Takai Y. The role of nectins in different types of cell–cell adhesion. J Cell Sci. 2012;125:3713–22.

  106. 106.

    Liu Q, Zhang J. Lipid metabolism in Alzheimer’s disease. Neurosci Bull. 2014;30:331–45.

  107. 107.

    Wang L, Chiang H-C, Wu W, Liang B, Xie Z, Yao X, et al. Epidermal growth factor receptor is a preferred target for treating Amyloid-β-induced memory loss. Proc Natl Acad Sci U S A. 2012;109:16743–8.

  108. 108.

    Baloyannis SJ. Golgi apparatus and protein trafficking in Alzheimer’s disease. J Alzheimers Dis. 2014;42(Suppl 3):S153–62.

  109. 109.

    Joshi G, Bekier ME, Wang Y. Golgi fragmentation in Alzheimer’s disease. Front Neurosci. 2015;9:340.

  110. 110.

    Su JH, Anderson AJ, Cribbs DH, Tu C, Tong L, Kesslack P, et al. Fas and Fas ligand are associated with neuritic degeneration in the AD brain and participate in β-amyloid-induced neuronal death. Neurobiol Dis. 2003;12:182–93.

  111. 111.

    Reich A, Spering C, Schulz JB. Death receptor Fas (CD95) signaling in the central nervous system: tuning neuroplasticity? Trends Neurosci. 2008;31:478–86.

  112. 112.

    Masliah E, Mallory M, Alford M, Deteresa R, Saitoh T. PDGF is associated with neuronal and glial alterations of Alzheimer’s disease. Neurobiol Aging. 1995;16:549–56.

  113. 113.

    Gianni D, Zambrano N, Bimonte M, Minopoli G, Mercken L, Talamo F, et al. Platelet-derived growth factor induces the beta-gamma-secretase-mediated cleavage of Alzheimer’s amyloid precursor protein through a Src-Rac-dependent pathway. J Biol Chem. 2003;278:9290–7.

  114. 114.

    Sun L, Ye RD. Role of G protein-coupled receptors in inflammation. Acta Pharmacol Sin. 2012;33:342–50.

  115. 115.

    Hawkins PT, Stephens LR. PI3Kgamma is a key regulator of inflammatory responses and cardiovascular homeostasis. Science. 2007;318:64–6.

  116. 116.

    Sevush S, Jy W, Horstman LL, Mao WW, Kolodny L, Ahn YS. Platelet activation in Alzheimer disease. Arch Neurol. 1998;55:530–6.

  117. 117.

    Catricala S, Torti M, Ricevuti G. Alzheimer disease and platelets: how’s that relevant. Immun Ageing. 2012;9:20.

  118. 118.

    Perry T, Lahiri DK, Chen D, Zhou J, Shaw KTY, Egan JM, et al. A novel neurotrophic property of glucagon-like peptide 1: a promoter of nerve growth factor-mediated differentiation in PC12 cells. J Pharmacol Exp Ther. 2002;300:958–66.

  119. 119.

    Mayo KE, Miller LJ, Bataille D, Dalle S, Göke B, Thorens B, et al. International Union of Pharmacology. XXXV. The glucagon receptor family. Pharmacol Rev. 2003;55:167–94.

  120. 120.

    McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Kawas CH, et al. The diagnosis of dementia due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7:263–9.

  121. 121.

    Jack CR, Bennett DA, Blennow K, Carrillo MC, Dunn B, Haeberlein SB, et al. NIA-AA research framework: toward a biological definition of Alzheimer’s disease. Alzheimers Dement. 2018;14:535–62.

  122. 122.

    Beach TG, Monsell SE, Phillips LE, Kukull W. Accuracy of the clinical diagnosis of Alzheimer disease at National Institute on Aging Alzheimer Disease Centers, 2005-2010. J Neuropathol Exp Neurol. 2012;71:266–73.

Download references


This manuscript was prepared using limited access datasets obtained though dbGaP (accession numbers: phs000168.v2.p2 (LOADFS), phs000007.v28.p10 (FHS), phs000287.v5.p1 (CHS), and phs000428.v2.p2 (HRS)) and the University of Michigan. Phenotypic HRS data are available publicly and through restricted access (from The authors thank Arseniy P. Yashkin for help preparing the HRS phenotypes. See also Additional file 1: Supporting Acknowledgment.


This research was supported by Grants from the National Institute on Aging (P01AG043352 and R01AG047310). The funders had no role in the study design, data collection and analysis, decision to publish, or manuscript preparation. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Availability of data and materials

The LOADFS, FHS, CHS, and HRS datasets are available through the dbGaP repository for qualified researchers (

Author information

The authors’ responsibilities were as follows: AN and AMK designed the study; AN analyzed data; AMK and AIY provided critical feedback; AN, AMK, and AIY wrote the manuscript; and all authors read and approved the final manuscript.

Correspondence to Alireza Nazarian or Alexander M. Kulminski.

Ethics declarations

Ethics approval and consent to participate

The four studies from which data were used (i.e., LOADFS, FHS, CHS, and HRS) were approved by the institutional review boards (IRBs) and were conducted after obtaining written informed consent from the participants or their legal guardians/proxies.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Additional files

Additional file 1:

Table S1. Cases and controls included. Table S2. QC-passed SNPs analyzed in datasets. Table S3. Genomic inflation factors (λ values) from logistic regression models. Table S4–S6. Replicated set of SNPs detected under analysis plan 1 (males and females), plan 2 (only males), and plan 3 (only females). Table S7–S9. Nonreplicated set of SNPs detected under analysis plan 1 (males and females), plan 2 (only males), and plan 3 (only females). Table S10–S12. Meta-analysis set of SNPs detected under analysis plan 1 (males and females), plan 2 (only males), and plan 3 (only females). Table 13. LD information about newly detected SNPs under plans 1–3 for which proxy AD-associated loci exist in 1-Mb flanking regions [8, 9]. Table S14. Coding schema used to determine APOE genotypes. Table 15. Information about LD between APOE SNPs and AD-associated SNPs located on chromosome 19 [8]. Table S16–S17. Wald χ2 test to compare ORs of SNPs between males and females for SNPs that were specifically significant in males and in females. Figure S1–S6. Manhattan plot and QQ plot of genome-wide association results under analysis plan 1 (males and females), plan 2 (only males), and plan 3 (only females). Supporting Acknowledgment. Furthre information about the four cohorts under consideration. (DOCX 284 kb)

Additional file 2:

Detailed information about the AD-associated SNPs under analysis plan 1 (males and females). (XLSX 72 kb)

Additional file 3:

Detailed information about the AD-associated SNPs under analysis plan 2 (only males). (XLSX 57 kb)

Additional file 4:

Detailed information about the AD-associated SNPs under analysis plan 3 (only females). (XLSX 72 kb)

Additional file 5:

Detailed information about the nominally AD-associated SNPs with the opposite pattern of significance in males and females. (XLSX 40 kb)

Additional file 6:

Detailed information about the results from the Wald χ2 test to compare ORs of SNPs between males and females. (XLSX 107 kb)

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (, which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark


  • Alzheimer’s disease
  • Sex disparities
  • Genome-wide association study
  • Meta-analysis
  • Gene-based analysis