Brain APOE expression quantitative trait loci-based association study identified one susceptibility locus for Alzheimer’s disease by interacting with APOE ε4

Some studies have demonstrated interactions of AD-risk single nucleotide polymorphisms (SNPs) in non-APOE regions with APOE genotype. Nevertheless, no study reported interactions of expression quantitative trait locus (eQTL) for APOE with APOE genotype. In present study, we included 9286 unrelated AD patients and 8479 normal controls from 12 cohorts of NIA Genetics of Alzheimer’s Disease Data Storage Site (NIAGADS) and Alzheimer’s Disease Neuroimaging Initiative (ADNI). 34 unrelated brain eQTLs for APOE were compiled from BRAINEAC and GTEx. We used multi-covariate logistic regression analysis to identify eQTLs interacted with APOE ε4. Adjusted for age and gender, substantia nigra eQTL rs438811 for APOE showed significantly strong interaction with APOE ε4 status (OR, 1.448; CI, 1.124–1.430; P-value = 7.94 × 10−6). APOE ε4-based sub-group analyses revealed that carrying one minor allele T of rs438811 can increase the opportunity of developing to AD by 26.75% in APOE ε4 carriers but not in non-carriers. We revealed substantia nigra eQTL rs438811 for APOE can interact with APOE ε4 and confers risk in APOE ε4 carriers only.

rs438811 confers risk in APOE ε4 carriers predominantly. We used multivariate logistic regression analysis to identify eQTLs which can confer AD risk in the 34 proxy eQTLs for APOE. After adjustment for age and gender, substantia nigra eQTL rs438811 (odds ratio [OR], 2.343; 95% confidence interval [CI], 2.205-2.490; raw P-value = 7.49 × 10 −167 ) was associated with AD (Table 2). After adjustment for age, gender and APOE ε4 status, rs438811 (OR, 1.049; 95% CI, 0.969-1.135; P-value = 0.237) was not associated with AD (Table 2). After introducing interaction item rs438811 genotype × APOE ε4 status into the model, rs438811 was found can confer AD risk by interacting with APOE ε4 status strongly (OR, 1.448; 95% CI, 1.231-1.704; P-value = 7.94 × 10 −6 ; Table 3). Sub-group analysis showed rs438811 confer risk predominantly in APOE ε4 carriers (OR, 1.267; 95% CI, 1.124-1.430; P-value = 1.12 × 10 −4 ; Table 3), which indicates carrying one minor allele T of rs438811 can increase the opportunity of developing to AD by 26.75% in APOE ε4 carriers. As shown in BRAINEAC, minor allele T of rs438811 was associated with increased APOE expression level in substantia nigra. The APOE eQTL P-values of rs438811 across the ten different brain regions were shown in Table S2. By querying Encyclopedia of DNA Elements (ENCODE), rs438811 was found to be target of transcription factors POLR2A and RPC155. rs483082, a brain eQTL for APOE which is in complete linkage with rs438811 and was also identified can interact with APOE ε4 status to confer AD risk in this study, was the target of multiple transcription factors: HNF4G, CEBPB, MXI1, HDAC2, SP1, RFX5, MAX, EP300, JUND, FOSL2, ZBTB7A and CEBPD. rs483082 was also found be to located in a DNase I hypersensitivity cluster.

Discussion
APOE has long been a widely-investigated gene since the identification of its association with AD. Many studies have reported the relations between APOE genotypes and AD-related traits, such as cerebral spinal fluid (CSF) biomarkers 20-22 , brain morphology changes [23][24][25] , and particular cognitive measures 23,26,27 . However, except for APOE ε4, no locus encompassing the APOE region, including brain eQTLs for APOE, were identified as a conferring risk for AD or AD-related traits. This study identified AD risk-associated brain eQTL for APOE by incorporating multiple GWAS cohorts.
The susceptibility locus rs438811 identified is in complete linkage with another brain eQTL for APOE-rs483082-which was associated with AD in summary statistics of International Genomics of Alzheimer's project (IGAP) and was reported to associate with AD in a Japanese population 15,28 . rs483082 was also associated with AD and confer AD risk in APOE ε4 carriers only in this study (data not shown). Furthermore, rs483082 was also reported to associate with lipid level 29 . Abnormal lipid metabolism has long been demonstrated to as being involved in AD pathology [30][31][32] . Our results revealed that the eQTL may influence the progression of AD APOE ε4 carriers by increasing the expression level of APOE. -491A/T, or rs449647, a polymorphism located in APOE transcriptional regulatory region, is the earliestly-reported APOE expression-associated variant by luciferase/β-galactosidase activity assay related to AD independent of APOE ε4 dosage 33 . In contrast to the earliest AD-associated APOE eQTL, the APOE eQTL identified in this study affect APOE expression in a brain region-specific manner. We also analyzed the association of -491A/T with AD (Table S3) and its interactive effect with APOE ε4 status on AD risk (Table S4). In consistent with the result from the study applied luciferase/β-galactosidase activity assay 33 , −491A/T confer AD risk independent of APOE ε4 status. rs438811 is a substantia nigra-specific APOE eQTL. As a brain substructure dysfunction of which contributes to extrapyramidal signs (EPS), pathological changes of substantia nigra are responsible for EPS and aggravated EPS in AD patients 34 . As to how the brain APOE eQTL influence the progression of AD, one most probable explanation is the neurotoxic effect of increased APOE4 expression level in substantia nigra. Nevertheless, pathological mechanism of the brain APOE eQTL still needs to be unveiled by molecular biological experiments.
In summary, this is a pilot study associating brain APOE eQTLs to AD risk. It identified a novel SNPs associated with AD by interacting with APOE ε4 status and potentially regulating expression level of APOE.

Methods
Compiling of brain APOE eQTLs. Brain APOE eQTLs studied in this multi-cohort gene-wide association study were collected by querying BRAINEAC (http://www.braineac.org/) and GTEx (https://www.gtexportal.org/home/) databases. BRAINEAC provides the gene expression across ten brain tissues (cerebellar cortex, frontal cortex, hippocampus, medulla, occipital cortex, putamen, substantia nigra, thalamus, temporal cortex and intralobular white matter) from 134 healthy control individuals. GTEx provides the gene expression across thirteen brain tissues (amygdala, anterior cingulate cortex, caudate, cerebellar hemisphere, cerebellum, cortex, frontal cortex, hippocampus, hypothalamus, nucleus accumbens, putamen, spinal cord and substantia nigra) with the sample sizes ranged from 80 to 154. The cis brain eQTLs with P-values less than 1 × 10 −3 and located within up-or down-stream 10 Mb boundary of the APOE gene were retrieved from the two eQTL databases. To reduce redundant computation for the AD association analysis, brain eQTLs within the same linkage disequilibrium (LD) block were pruned and one eQTL was chosen to serve as the proxy for the LD block. Proxy brain eQTLs were determined by LDproxy (https://analysistools.nci.nih.gov/LDlink/?tab=ldproxy) with a threshold of LD r 2 ≥ 0.8.

Subjects.
In this study, we included a total of 13 AD GWAS cohorts. 12 out of them were from NIAGADS.
The criteria for inclusion of cohorts from NIAGADS was carrying covariate information on age, gender and APOE genotype. For detailed information on the 12 cohorts from NIAGADS, please refer to Table S5. The last cohort was from ADNI. GWAS data from ADNI were generated as previously described and obtained from the ADNI database (http://www.loni.ucla.edu/ADNI/) 35 . Finally, a total of 10606 AD patients and 9981 healthy controls were included in this study.

Identification and exclusion of related individuals.
We used KING to identify and exclude duplicate samples and kin with a third degree (e.g. first cousin) or closer relationship within and across datasets 36   1.9 37 . For each dataset, SNPs with a call rate of less than 99%, minor allele frequency of less than 1% and violation of Hardy-Weinberg equilibrium in controls (P < 1 × 10 −4 ) were removed. Samples with a call rate of less than 90% were removed.

Imputation of chromosome 19.
Because all the brain eQTLs collected in this study are cis eQTLs which are located within up-or down-stream 10 Mb boundary of the APOE gene, only chromosome 19 was imputed to reduce computation burden. The SNPs of chromosome 19 were prephased using SHAPEIT2 for each dataset 38 . SNPs were then imputed to a reference panel of 1000 Genome Project Phase 3 by IMPUTE2 39 . An imputation info score cutoff of 0.9 was applied to exclude low-quality imputed SNPs. After imputation, collected eQTLs from the two databases for brain tissues were extracted.
Identification of population substructure. To adjust for confounding effect of population substructure in our data, we calculated eigenvectors of individuals through whole genome-wide principal component analysis (PCA) before chromosome 19 SNPs were extracted using Plink 1.9 37 . All Principal components (PCs, from PC1 to PC20) were used for confounding adjustment.
Combination of genotyped and imputed data. In this study, the genotyped and imputed brain eQTLs for APOE were inconsistent among different datasets. In order to combine all the GWAS datasets for AD association analysis, imputed high-quality brain eQTL with high confidence (genotype probability greater than 0.8) were converted to be simulated genotype data by fcGENE for each individual 40 . Simulated genotype data were then combined with originally genotyped data.
Statistical analysis. After adjusting for age, gender and all PCs, these proxy brain eQTLs were tested for associations with AD with or without adjustment for APOE ε4 status through multivariate logistic regression analysis. eQTLs were coded as 0, 1, or 2 according to their number of minor alleles. APOE ε4 status was coded as 0 or 1 according to absence (APOE ε2/2, ε2/3 and ε3/3 subjects) or presence (APOE ε2/4, ε3/4 and ε4/4 subjects) of APOE ε4. Interaction item SNP genotype × APOE ε4 status was then introduced into the model to investigate the interactive effect of SNP and APOE ε4 status on AD risk. To account for multiple testing, we used the Bonferroni correction and considered significant only those brain eQTLs for which P-value < 0.05/34 = 0.05/34 = 1.47 × 10 −3 . All statistical calculations were performed using R 41 .

Biological function annotation of identified APOE ε4 status-interactive brain eQTLs for APOE.
Biological function annotation of identified APOE ε4 status-interactive brain eQTLs for APOE was performed via querying Encyclopedia of DNA Elements (ENCODE, https://www.encodeproject.org/).