Author's Department/Program

Biology and Biomedical Sciences: Human and Statistical Genetics


English (en)

Date of Award

Summer 9-1-2014

Degree Type


Degree Name

Doctor of Philosophy (PhD)

Chair and Committee

Alison Goate


Alzheimer's disease (AD) is the most common form of dementia affecting the health of more than 5 million Americans in 2013. Understanding how genetic variants contribute to AD is important to develop effective therapeutics for delaying and eventually curing the disease. Recent sequencing studies have identified rare variants p.V232M in PLD3 and p.R47H in TREM2 significantly associated with AD risk. Additionally, genome-wide association studies (GWAS) uncovered common variants in ABCA7, BIN1, CD33, CD2AP, CLU, CR1, EPHA1, MS4A4A, and PICALM that contribute to AD; however, the most-significant variants in these loci are located within non-coding regions and have no direct functional impact on AD pathogenesis. We hypothesized that if PLD3 and TREM2 are truly AD risk genes, they will carry additional functional variants that can substantially affect AD risk. Moreover, we hypothesized that GWAS genes carry additional risk alleles across the frequency spectrum so that common, low frequency or rare variants within these genes may contribute to AD risk. We undertook pooled sequencing of exonic and flanking intronic sequence for the aforementioned genes in 3,730 European Americans (EA) (2,082 AD cases and 1,648 controls) and 336 African Americans (AA) (204 AD cases and 132 controls). We found rare variants in PLD3 and TREM2 are more frequently seen in cases than in controls of EA descent. Single-variant analyses showed that p.M6R and p.A442A in PLD3 and p.R62H in TREM2 are significantly associated with AD risk besides p.V232M in PLD3 and p.R47H in TREM2. We found rare variants in PLD3 (PSKAT-O=1.44×10-11) and TREM2 (PSKAT-O=5.37×10-7) are genome-wide significantly associated with AD. Additionally, we found a significant association for PLD3 coding variants with AD risk in AA (PSKAT-O=1.40×10-3). However, we did not find evidence of association for TREM2 variants with AD risk in AA. We validated 90 coding variants in GWAS-identified genes and bioinformatic analyses implicated that a large proportion of these variants are functional.

The International Genomics of Alzheimer's Project recently performed meta- and gene-wide analyses and identified 23 loci associated with AD risk, of which 13 were novel. However, these loci's role in affecting the molecular pathways of AD remains unknown. To determine whether these loci are also associated with cerebrospinal fluid (CSF) amyloid-beta 1-42 (Aβ42) and phosphorylated tau181 (ptau181) levels, we combined CSF biomarker datasets from several studies and performed single-variant, set-based, and conditional analyses for each locus. In the APOE locus, rs769449 is genome-wide significantly associated with CSFAβ42 and ptau181 levels independently of APOE and the association for CSF ptau181 levels was not driven by Aβ metabolism. We found rs7937331, within the CELF1 fine-mapping region, tags the same signal as the IGAP top SNP (rs62003531) and is significantly associated with CSF Aβ42 levels and AD risk. Additionally, rs62003531, located in the intronic region of FERMT2, tags the same association as the IGAP top SNP (rs17125944) and is associated with CSF Aβ42 levels. None of the SNPs within the IGAP-identified AD risk loci except the APOE locus are significantly associated with CSF ptau181 levels after multiple test correction. In investigating the potential regulatory functions associated with IGAP top SNPs and CSF top SNPs, most of GWAS top SNPs have no significant regulatory potential and are unlikely to be functional variants for AD risk. However, RegulomeDB predicts that several proxy SNPs in linkage disequilibrium (LD) with rs7937331 may be cis-acting expression quantitative trait loci (eQTLs) for nearby genes. The IGAP study also identified an intergenic polymorphism near TREML2 suggestively associated with AD risk; however, due to the study design, it was not possible to uncover the underlying functional variant or to determine whether this observed association was driven by the known AD risk allele, TREM2 p.R47H, or represented a novel locus. We performed analyses using whole-exome sequencing data, CSF biomarker analyses and meta-analyses to demonstrate that the AD risk association is likely driven by a TREML2 variant p.S144G (rs3747742) independently of TREM2 p.R47H risk for AD.

Finally, we sought to functionally characterize the effects of novel TREM2 variants on TREM2 cell surface transport. We transduced a T cell hybridoma cell line with virus containing TREM2 wild type (WT) and risk variants and measured TREM2 cell surface expression with a TREM2-specific monoclonal antibody. We found cells expressing p.T66M and p.R136W have a robust effect on TREM2 cell surface expression but cells expressing p.R47H and p.R62H are similar to hTREM2 WT. Additionally, since polymorphisms in the CELF1 fine-mapping region were implicated to be eQTLs for nearby genes, we performed cis-eQTL analysis for mRNA expression levels in several brain regions using four publicly available datasets to identify genetic determinants of gene expression in human brains. We found several C1QTNF4-expression-associated SNPs which tag the same signal are in LD with rs7937331, the top CSF Aβ42 SNP in the CELF1 fine-mapping region. Additionally, we found evidence of differential expression in the C1QTNF4 transcript between AD cases and controls in human brains. These findings provide additional evidence that genes involved in the inflammatory response play an important role in AD pathogenesis.


This work is not available online per the author’s request. For access information, please contact or visit

Permanent URL: