Background

Alzheimer's disease (AD) is the most common cause of dementia and the most prevalent neurodegenerative disorder. An estimated 10 percent of Americans over the age of 65 and half of those over age 85 have AD. More than 4.5 million Americans currently suffer from the disease. In autosomal dominant early-onset Alzheimer's disease (EOAD, age of onset < 60 years), three susceptible genes (APP, PSEN1, and PSEN2) have been identified [1, 2]. Late-onset AD (LOAD) has ~80% heritability, and is strongly associated with apolipoprotein E (APOE) [3]. APOE has three major alleles (ε2/ε3/ε4) that have different effects on the risk of LOAD, with ε4 having between 10 and 30 times of risk of developing AD by 75 years of age [4].

In addition, several genetic studies have identified putative susceptible loci and genetic variants, including sortilin-related receptor (SORL1) [5, 6], death-associated protein kinase 1 (DAPK1) [7], ubiquilin 1 (UBQLN1) [8], adenosine triphosphate-binding cassette transporter 1, subfamily A (ABCA1) [9], and low-density lipoprotein receptor-related protein 6 (LRP6) [10]. Besides these findings, a large meta-analysis from the AlzGene database [11] reported 598 potential AD-susceptibility genes. For the past few years, genome-wide genotyping association studies brought considerable success by reporting new susceptible loci for AD such as Golgi membrane protein 1 (GOLM1) [12–15]. Recently two groups published the two largest LOAD GWAS [16, 17]; Harold et al. [16] reported the association of SNPs in clusterin (CLU) and phosphatidylinositol binding clathrin assembly protein (PICALM), and Lambert et al. [17] reported association of clusterin (CLU) with LOAD and additionally reported a novel association with complement component (3b/4b) receptor 1 (CR1). These new findings have provided valuable insights in the genetics, neuropathologic mechanisms and pathways associated with AD.

The cerebrospinal fluid (CSF) components β-amyloid peptide (Aβ1-42), total tau protein (T-tau) and phosphorylated tau (P-tau181P) are biomarkers for AD and can be used to aid in diagnosis and to predict progression from mild cognitive impairment (MCI) to AD [18, 19]. These biomarkers can potentially be used in future applications to predict the development of MCI in cognitively normal subjects, progression to AD in MCI patients, and to monitor AD progression [20–23]. These biomarkers may also be used to reveal genes that are important in AD pathogenesis. In the present study, we assessed the several putative AD genes associated with CSF biomarkers that were identified from major public GWAS dataset for Alzheimer's disease, the Alzheimer's disease Neuroimaging Initiative (ADNI). This initiative is the most comprehensive effort to identify neuroimaging measures and biomarkers associated with cognitive and functional changes in healthy elderly people and in people who have MCI and AD [24]. The ADNI data is useful for researchers who are searching for genes that contribute to the development of Alzheimer's disease, which currently affects more than 4.5 million people in the United States alone. We also investigated possible associations of CSF biomarkers (Aβ1-42, T-tau, and P-tau181P) with the number of APOE ε4 allele, age and APOE genotype in order to improve the characterization of the CSF biomarkers and genome-wide SNP genotyping data from the ADNI cohort. This is the first genome-wide association study to use these AD-related biomarkers to identify genes critical the pathogenesis of AD.

Methods

Sample

Data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database http://www.loni.ucla.edu/ADNI. ADNI was launched in 2003 by the National Institute on Aging (NIA), the National Institute of Biomedical Imaging and Bioengineering (NIBIB), the Food and Drug Administration (FDA), private pharmaceutical companies and non-profit organizations, as a $60 million, 5-year public-private partnership. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer's disease (AD). Determination of sensitive and specific markers of very early AD progression is intended to aid researchers and clinicians to develop new treatments and monitor their effectiveness, as well as lessen the time and cost of clinical trials.

The Principle Investigator of this initiative is Michael W. Weiner, M.D., VA Medical Center and University of California - San Francisco. ADNI is the result of efforts of many co-investigators from a broad range of academic institutions and private corporations, and subjects have been recruited from over 50 sites across the U.S. and Canada. The initial goal of ADNI was to recruit 800 adults, ages 55 to 90, to participate in the research -- approximately 200 cognitively normal older individuals to be followed for 3 years, 400 people with MCI to be followed for 3 years, and 200 people with early AD to be followed for 2 years. For up-to-date information see http://www.adni-info.org.

The GWAS dataset (downloaded from ADNI website in June 2009) consisted of 243 normal, 235 MCI and 340 AD subjects genotyped using the Illumina Human Genome 610 Quad BeadChips. 410 of these subjects (119 Normal, 115 MCI and 176 AD; 247 males and 163 females) have CSF Aβ1-42, T-tau, and P-tau181P levels. Detailed protocols for subject recruitment and biomarker accrual are available at the ADNI website http://www.adni-info.org/. Demographics, CSF biomarkers, and APOE genotype data of the ADNI dataset are summarized in Table 1, Table 2 and Additional file 1.

Table 1 Demographic, clinical and biomarker data for each subject group after removing 20 outliers (n = 390: Normal, MCI and AD)
Full size table
Table 2 APOE genotype data for each subject group (Normal, MCI and AD)
Full size table

Genotyping and sample quality control

Quality control for the genotyping data was performed using PLINK http://pngu.mgh.harvard.edu/~purcell/plink/[25] as follows. 498,205 SNPs were retained after excluding SNPs with Minor Allele Frequency (MAF) < 5%, call rate < 98%, or significant in Hardy-Weinberg Equilibrium test (p ≤ 10-6). All samples had genotyping call rate > 95% and were retained. We then examined population stratification by visual inspection using the first two dimensions from principal components analysis, using SmartPCA from EIGENSTRAT http://genepath.med.harvard.edu/~reich/Software.htm[26, 27]. Self-reported ethnicity and racial identities for ADNI subjects were used to highlight samples in the PCA plot and are summarized in Additional file 2. 390 samples were retained after SmartPCA excluded 20 samples as outliers. We computed the top five principal component coordinates using SmartPCA to correct for stratification in association analysis. SmartPCA removed all but one of the Asian samples and retained Black/African Americans (Additional file 3); Visual inspection suggests that the first principal component (PC0), which explains the most variance in the data, separates the Caucasians and non-Caucasians reasonably well (setting threshold PC0 < = 0.01 can exclude all non-Caucasians). Finally, we excluded 52 non-Caucasian samples as outliers; the genomic control variance inflation factor λ was 1.00983, suggesting minimal population admixture in the final sample used for association analysis. We performed association analysis using age and APOE ε4 genotype as covariate and did not incorporate principal components. Quantile-Quantile plots for each of the three test groups with log10-transformed level of three CSF biomarkers (Additional file 4 and Additional file 5) suggested that population stratification having negligible bias on the genetic associations (Additional file 6). Finally, for the whole analysis we performed in the following method section, the study sample of 390 individuals with three CSF biomarkers was used after removing 20 outliers. The level of three CSF biomarkers was log10-transformed.

Effect of APOEε4 copy number on CSF biomarker levels

We performed Kruskal-Wallis test of log10-transformed CSF biomarkers (Aβ1-42, T-tau, and P-tau181P) stratified by the number of APOE ε4 alleles and the APOE genotypes across all 390 samples or within each of the three diagnostic groups.

Association testing of SNPs and CSF analysis

We tested SNP association with the three CSF biomarkers (Aβ1-42, T-tau, and P-tau181P) by PLINK using samples and SNP markers passing QC. CSF biomarker levels were log10-transformed so they become normally distributed. The association analysis used a full linear model comprising three genetic effects: additive effects of allele dosage (ADD), dominance deviation from additivity (DOMDEV) (negative means the allele is recessive), and 2-df joint test of both additive and dominance (GENO_2DF). In addition, we used age (of being recruited by the study) and APOE ε4 genotype (number of APOE ε4 allele; 0, 1, 2) as covariates after removing outliers. To ensure the significance is not due to population stratification, we also incorporated the top five PCA principal components in the linear regression to further control for the population structure, but found the addition has very little effect on the statistical significance.

Gene ontology and E-SNP analysis

We carried out gene ontology analysis of SNP association results using ALIGATOR (Association LIst Go AnnoTatOR) [28], to find gene-ontology terms enriched with significant SNPs. We used p-value cutoff < 10-3 for SNPs, 5000 replicate gene lists and 1000 permutations as parameters to run ALIGATOR. We examined the top associated SNPs and examined nearby SNPs in linkage disequilibrium (LD) that are associated with gene expression from published eQTL studies [29–33].

Results and Discussion

Association of SNPs with CSF biomarkers

We summarize the top SNPs (p-value < 10-6) with and without using covariates (APOE genotype and age) in Table 3. Without using covariates, we found some genes near the top SNPs, including CYP19A1 (rs2899472, p = 1.86 × 10-7) and TOMM40 (rs2075650, p = 3.03 × 10-7) from Aβ1-42 in normal subject. Several genetic studies have identified those genes as putative susceptible loci and genetic variants associated with Alzheimer's disease [34, 35]. However, close examination of nearby SNPs showed rs2899472 and rs2075650 were not supported by nearby SNPs in LD (nearby SNPs in LD are all non-significant). Because APOE genotypes are strongly associated with AD and TOMM40 is physically close to APOE, we focused on SNPs from adjusted results that consider APOE genotype and age. Here, we found 10 SNPs significantly associated with CSF biomarker level of Aβ1-42, 3 SNPs from T-tau, and 2 SNPs from P-tau181P at 10-6 significance level in normal subjects (Figure 1). In addition, we found 1 SNP significantly associated with CSF biomarker level of Aβ1-42, 3 SNPs from T-tau, and 2 SNPs from P-tau181P at 10-6 significance level in MCI subjects (Figure 2). No SNPs were found at 10-6 significance level in AD subjects (Figure 3). For normal subjects, we found genes near the top SNPs, included CYP19A1 (rs2899472, p = 1.90 × 10-7), NCAM2 (rs1022442, p = 2.75 × 10-7) for Aβ1-42 association and UPP2 (rs2074955, p = 2.07 × 10-7) for P-tau181p association. Again, close examination of SNPs in LD with rs2899472 (CYP19A1) did not support rs2899472. The SNP rs1022442 was in close to genome-wide significance, supported by nearby SNPs (Figure 4), and NCAM2 (neural cell adhesion molecule 2) gene was reported as a genetic factor related to the progression of AD in the Japanese population [36]. The Aβ1-42 level grouped by the SNP rs1022442 genotype over all three cohorts (normal, MCI and AD) supports our finding (Figure 5). Boxplots of Aβ1-42 levels in normal subjects stratified by rs1022442 genotype showed significant differences between AA, AB and BB. Previous study indicated an increased risk associated with rs2899472 in AD patients, which was amplified in APOE ε4 carriers in their study [35]. For MCI subjects, we found several genes near the top SNPs, included FLJ21511 (rs2768975; p = 1.96 × 10-7, rs6850199; p = 3.18 × 10-7) by T-tau association and CHN2 (rs121724, p = 1.45 × 10-7), MTUS1 (rs7842088, p = 2.12 × 10-7) by P-tau181p association.

Table 3 Linear regression result for SNPs with CSF biomarkers in Normal, MCI and AD subjects (Top SNPs with p-value < 10-6)
Full size table
Figure 1
figure 1

Manhattan plots of the quantitative trait (CSF biomarkers: T-tau, P-tau 181P and Aβ 1-42 ) genome wide association analysis in normal subjects. Colors on x-axis indicate an autosomal chromosome (from chromosome 1 to chromosome 22). The y-axis indicates p-values (-log10(observed p-values)). Red arrows indicate rs2899472 on chromosome 15 and blue arrow indicates rs1022442 on chromosome 21.

Full size image
Figure 2
figure 2

Manhattan plots of the quantitative trait (CSF biomarkers: T-tau, P-tau 181P and Aβ 1-42 ) genome wide association analysis in MCI subjects. Colors on x-axis indicate an autosomal chromosome (from chromosome 1 to chromosome 22). The y-axis indicates p-values (-log10(observed p-values)).

Full size image
Figure 3
figure 3

Manhattan plots of the quantitative trait (CSF biomarkers: T-tau, P-tau 181P and Aβ 1-42 ) genome wide association analysis in AD subjects. Colors on x-axis indicate an autosomal chromosome (from chromosome 1 to chromosome 22). The y-axis indicates p-values (-log10(observed p-values)).

Full size image
Figure 4
figure 4

Upper image shows genes in linkage disequilibrium with the SNP rs1022442 and lower plot shows the distribution of SNP rs1022442 and nearby SNPs. In lower plot, the x-axis indicates loci on chromosome 21 and y-axis indicates p-values (-log10(observed GENO_2DF p-values)). Green arrow indicates rs1022442 (p = 2.75 × 10-7) on loci 21277717 and pink arrow indicates rs2826629 (p = 8.40 × 10-5) on loci 21265887.

Full size image
Figure 5
figure 5

Boxplots of Aβ 1-42 levels in normal, MCI and AD subjects stratified by rs1022442 genotype. The x-axis indicates AA, AB and BB respectively.

Full size image

Association of APOEwith CSF biomarkers

Previous studies suggesting that CSF Aβ1-42 and T-tau levels are correlated with the number of APOE ε4 alleles [37]. We analyzed the distribution of levels stratified by diagnosis and the number of APOE ε4 in the ADNI cohort after QC (Figure 6) and reached the same conclusion. In the AD group, Aβ1-42 level was inversely correlated with APOE ε4 allele dose. The APOE e4 was not associated with T-tau or P-tau181P levels. Analysis of APOE genotypes showed that ε4/ε4 is associated with CSF biomarker level of Aβ1-42 , T-tau and P-tau181P (Figure 7).

Figure 6
figure 6

Boxplots of the APOE ε4 copy number with CSF biomarkers in normal, MCI and AD subjects. The x-axis indicates number of APOE ε4 alleles and y-axis indicates CSF biomarkers. P-values are produced by the Kruskal-Wallis test.

Full size image
Figure 7
figure 7

Boxplots of the APOE genotype with CSF biomarkers in normal, MCI and AD subjects. The x-axis indicates five different APOE genotypes and y-axis indicates CSF biomarkers. P-values are produced by the Kruskal-Wallis test.

Full size image

Pathway analysis of CSF biomarkers

We ran ALIGATOR to identify top gene ontology terms associated with genes containing SNPs with higher statistical significance, and summarized the results in Additional file 7. We found cerebral cortex development, methionine metabolic process, actinin binding, and pallium development to be among the most significant gene ontology terms associated with CSF biomarker level of Aβ1-42 in normal subjects. Elevation in Aβ in the cerebral cortex has been implicated in the pathophysiology of AD but its mechanism of action is unknown [38]. It has been known that mammals have a fully developed cortex, but the structure it evolved from pallium which is present in all vertebrates as well as the most primitive ones [39]. The medial pallium forms the precursor of the hippocampus. Since hippocampal disruption is one of the earliest signs for AD, pallium development might be involved in the pathophysiology AD.

E-SNP analysis of CSF biomarkers

We collected eSNPs (SNPs known to be associated with the expression level of some genes) in several published expression quantitative linkage (eQTL) studies, and examined regions around significant SNPs (p < 10-4) (Additional file 8). We found significant SNPs from Dixon et al. [29], Stranger et al. [31], and Gibbs et al. [33], but we could not find any top SNPs that are associated with gene expression from other papers [30, 32]. Dixon et al. [29] used lymphoblastoid cell lines (LCLs) derived from children both with and without asthma.

NCAM2 (Neural Cell Adhesion Molecule 2)

We obtained strong evidence that NCAM2 genotypes are associated with Aβ1-42 levels. NCAM2 (Ensemble: ENSG00000154654, OMIM: 602040;) is a 541Kb gene at 21q21.1 with no known alternative spicing forms. Novartis SymAtlas human tissue survey shows the gene (Affymetrix probeset ID 205669_at) is ubiquitously expressed and is highly expressed in cardiac myocytes, blood cells, and appendix. Among neuronal tissues NCAM2 has higher expression levels in prefrontal cortex, superior cervical ganglion, and hypothalamus. The transcript (Ensembl: ENST00000400546) consists of 18 exons and encodes a 93 k-Da, 835-residue plasma membrane protein (NP_004531.2). The NCAM2 protein architecture includes 5 IgC2 (Immunoglobulin C-2 type) domains followed by two FN3 (fibronectin type 3) domains and a transmembrane domain. The gene is conserved in chimpanzee, dog, cow, mouse, rat, chicken, zebrafish, and fruit fly; the eight-domain protein architecture is also conserved in all these organisms except for cow which has only four IgC2 domains. Little is known about NCAM2 except that the protein interacts with prion protein [40] and estrogen receptor 1 (ESR1) [41], is involved in neuron adhesion and fasciculation of neurons, and may be involved in AD [36], prion disease, and Down syndrome [42, 43].

TOMM40 (translocase of outer mitochondrial membrane 40 homolog (yeast)) and CYP19A1 (Cytochrome P450, family 19, subfamily A, polypeptide 1)

It is interesting to note that we found no significant association in APOE (rs769451, chr19: 50102751, p = 0.6682 for association with Aβ1-42 level in normal subjects), but a SNP with strong association in the nearby TOMM40 (intronic SNP rs2075650, chr19:50087459, p = 3.03 × 10-7 for association with Aβ1-42 level in normal subjects), when age and the number of APOE e4 alleles were not included in the regression. The TOMM40 gene is related to how easily molecules can get into and out of the surface of the mitochondria, the energy center of cells. This gene is a transporter of proteins across the mitochondrial membrane, and Sortillin-related receptor, which functions to partition amyloid precursor protein away from β-secretase and -secretase [44]. This is consistent with observations that levels are reduced in the brains of patients with Alzheimer's disease and MCI [44–46]. The TOMM40 gene has been reported in numerous studies in the study of AD genetics; for example, Yu et al. [47] reported possibility that loci in the TOMM40 gene may have a less effect on the risk for LOAD in Caucasians [47], and recently Roses et al. [48] found evidence supporting a poly-T polymorphism (rs10524523, chr19:50094889) in TOMM40 affecting the AD age of onset in two independent clinical cohorts. The potential association of TOMM40 and Aβ1-42 may be how the gene affects the risk and onset age of AD and should be further investigated.

The CYP19A1 gene is localized on chromosome 15q21.2 and spans 123 kb. This gene encodes a member of the cytochrome P450 superfamily of enzymes. Cytochrome P450 aromatase is an enzyme that catalyses the conversion of androgens, such as testosterone, to oestrogens, which act as sex steroid hormones but also function during growth and differentiation. There are high levels of expression in both the gonads and the brain [35]. Huang et al. [35] indicated an increased risk associated with SNP rs2899472 in the total number of AD patients, which was amplified in APOE ε4 carriers.

Conclusions

Our analysis of the ADNI genome-wide association study identified several putative loci that are in genetic association with Aβ1-42, T-tau and P-tau181P levels in cerebrospinal fluids. In particular an intronic SNP rs1022442 of gene NCAM2 is close to genome-wide significance in association with Aβ1-42 in normal subjects. Although the gene is poorly characterized in the literature, prior studies have implicated roles of NCAM2 in prion disease, Down syndrome, and AD. Our findings suggest NCAM2 could be part of the pathway on the pathogenesis of senile plaques in human brains with AD.

With only 119 normal subjects and 410 overall, the GWAS dataset is clearly underpowered. The most significant associations were identified using normal subjects since the variances of the CSF biomarker levels are much smaller in MCI and AD subjects due to dementia. Nonetheless, increasing the number of CSF biomarker measurements is challenging especially for normal subjects. An alternative that will substantially increase the sample size is to examine protein levels in blood instead of CSF, given levels of these proteins in blood are informative about AD pathology or prognosis [49].

Our analysis clearly demonstrates that quantitative trait linkage of biomarkers via genome-wide screening can reveal additional insights into the mechanism that connects known genetic factors to the disease [50]. Moving along this path requires further efforts by the research community towards larger sample size and accrual of additional biomarkers in AD and other neurodegenerative disorders.