Polygenic Overlap Between C-Reactive Protein, Plasma Lipids and Alzheimer's Disease

Abstract

Background

Epidemiological findings suggest a relationship between Alzheimer's disease (AD), inflammation and dyslipidemia, although the nature of this relationship is not well understood. We investigated whether this phenotypic association arises from a shared genetic basis.

Methods and Results

Using summary statistics (p-values and odds ratios) from genome-wide association studies of over 200,000 individuals, we investigated overlap in single nucleotide polymorphisms (SNPs) associated with clinically diagnosed AD and C-reactive protein (CRP), triglycerides (TG), high- (HDL) and low-density lipoprotein (LDL) levels. We found up to 50-fold enrichment of AD SNPs for different levels of association with CRP, LDL, HDL and TG SNPs using an FDR threshold < 0.05. By conditioning on polymorphisms associated with the four phenotypes, we identified 55 loci associated with increased AD risk. We then conducted a meta-analysis of these 55 variants across four independent AD cohorts (total n = 29,054 AD cases and 114,824 healthy controls) and discovered two genome-wide significant variants on chromosome 4 (rs13113697, closest gene HS3ST1, odds ratio (OR) = 1.07, 95% confidence interval (CI) = 1.05-1.11, p = 2.86 × 10⁻⁸) and chromosome 10 (rs7920721, closest gene ECHDC3, OR = 1.07, 95% CI = 1.04-1.11, p = 3.38 × 10⁻⁸). We also found that gene expression of HS3ST1 and ECHDC3 was altered in AD brains compared with control brains.

Conclusions

We demonstrate genetic overlap between AD, CRP, and plasma lipids. By conditioning on the genetic association with the cardiovascular phenotypes, we identify novel AD susceptibility loci including two genome-wide significant variants conferring increased risk for Alzheimer's disease.

INTRODUCTION

Late-onset Alzheimer's disease (AD) is the most common form of dementia with an estimated prevalence of 30 million people worldwide, a number that is expected to quadruple in the next 40 years.¹ Given the absence of disease-modifying therapies and increasing awareness that symptoms develop over many years, there is significant interest in identifying effective strategies for AD prevention. Delaying dementia onset by a modest 2 years could potentially lower the worldwide prevalence of AD by more than 22 million cases over the next 40 years, resulting in significant societal savings.¹

A growing body of evidence suggests an association between AD and potentially modifiable processes including dyslipidemia and inflammation. In observational studies, high serum cholesterol levels have been associated with increased risk of AD ^2,3 and molecular ⁴ and biomarker findings ⁵ suggest that phospholipids may play an integral role in modulating AD-associated pathogenesis. Complement factors and activated microglia are established histopathologic features in brains of AD patients ⁶ and epidemiological studies in older individuals indicate that high serum levels of inflammatory proteins are associated with cognitive decline ⁷ and may predict dementia risk. ⁸ Genome-wide association studies (GWAS) in late-onset AD have replicated the established association with apolipoprotein E (APOE) and identified single nucleotide polymorphisms (SNPs) implicated in lipid metabolism, such as CLU and ABCA7 and inflammatory processes, such as CR1 and HLA-DRB5. ^9,10 In addition, a rare sequence variant in TREM-2 with known anti-inflammatory function has recently been identified as conferring increased risk for AD. ^11,12 Taken together, these findings suggest that processes involved with lipid metabolism and inflammation may also impact Alzheimer's pathogenesis.

Combining GWAS from multiple disorders and phenotypes provides insights into genetic pleiotropy (defined as a single gene or variant being associated with more than one distinct phenotype) and could elucidate shared pathobiology. Using this approach, we have recently reported genetic overlap between a number of diseases and phenotypes and identified novel common variants associated with schizophrenia, ^13,14 bipolar disorder,¹³ prostate cancer,¹⁵ hypertension, ¹⁶ and primary sclerosing cholangitis. ¹⁷ Here, we applied this method to AD, taking advantage of several large GWASs, ^18-20 to identify SNPs associating with clinically diagnosed AD, C-reactive protein (CRP) levels, and plasma lipid levels (specifically triglycerides (TG), high- (HDL) and low-density lipoproteins (LDL)).

METHODS

Participant Samples

We evaluated complete GWAS results in the form of summary statistics (p-values and odds ratios) for clinically diagnosed AD, ¹⁸ CRP levels, ¹⁹ and plasma lipid levels (TG, HDL and LDL ²⁰ (see Table 1). The CRP GWAS summary statistic data consisted of 82,725 individuals drawn from 25 studies with genotyped or imputed data at 2,671,742 SNPs (for additional details see reference 19). The plasma lipids GWAS summary statistic data consisted of 188,577 individuals with genotyped or imputed data at 2,508,375 SNPs (for additional details see reference 20). We obtained publicly available AD GWAS summary statistic data from the International Genomics of Alzheimer's Disease Project (IGAP Stage 1 + 2, for additional details see Supplemental Information and reference 18). We used IGAP Stage 1 as our discovery cohort, which consisted of 17,008 AD cases (mean age = 74.7 ± 7.7 years; 59.4% female) and 37,154 controls (mean age = 76.3 ± 8.1 years; 58.6% female) drawn from four different consortia across North America and Europe with genotyped or imputed data at 7,055,881 SNPs (for a description of the AD cases and controls within the IGAP Stage 1 sub-studies, please see reference 18). To confirm our findings from IGAP Stage 1, we assessed the p-values of pleiotropic SNPs (conditional FDR < 0.05; see Statistical analysis below) from the discovery analyses in three independent AD cohorts, namely the IGAP Stage 2 sample, a cohort of AD cases and controls drawn from the population of Iceland (deCODE), and a cohort of AD cases and controls drawn from the population of Norway (DemGene). The IGAP Stage 2 sample consisted of 8,572 AD cases (mean age = 72.5 ± 8.1 years; 61% female) and 11,312 controls (mean age = 65.5 ± 8.0 years; 43.3% female) of European ancestry with genotyped data at 11,632 SNPs (for additional details see reference 18). Clinical diagnosis of probable AD within the IGAP Stage 2 cohort was established according to the DSM-III-R and NINCDS-ADRDA criteria. ²¹ The deCODE dataset was drawn from the Icelandic population and included 2,470 genotyped AD cases (age = 84.9 ± 7.2 years; 65.8 % female) and 65,347 genotyped controls (age = 68.8 ± 13.7 years; 57.8% females) (for additional details see reference 12). As previously described, ¹² patients from Iceland were diagnosed with definite, probable or possible Alzheimer's disease based on the NINCDS-ADRDA criteria ²¹ or according to guidelines for ICD-10 F00, and were compared to population controls. The Norwegian sample (DemGene) included 1,004 cases (age = 74.1 ± 9.6 years; 60.2 % female) and 1,011 controls (age = 74.6 ± 9.3 years; 57.7 % female) with genotyped data at 693,377 SNPs. Clinical diagnosis of AD and dementia within the DemGene sample was established using ICD-10 research criteria ²², the recommendations from the National Institute on Aging-Alzheimer's Association (NIA/AA) ²³ or the NINCDS-ADRDA criteria ²¹ (Supplemental Information). The relevant institutional review boards or ethics committees approved the research protocol of the individual GWAS used in the current analysis, and all human participants gave written informed consent.

Table 1.

Summary data from all GWAS used in the current study

Disease/Trait	N	# SNPs	Reference
Alzheimer's disease (AD) – IGAP Stage 1+2	74,046 (25,580 AD cases + 48,466 controls)	7,055,881 (Stage 1) 11,632 (Stage 2)	Lambert JC, Ibrahim-Verbaas CA, Harold D, et al. Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer's disease. Nat Genet. 2013;45:1452-8.
Alzheimer's disease (AD) – deCODE	67,817 (2,470 cases + 65,357 controls)	Whole-genome sequencing	Jonsson T, Stefansson H, Steinberg S, et al. Variant of TREM2 associated with the risk of Alzheimer's disease. N Eng J Med 2013;368:107-16.
Alzheimer's disease (AD) – DemGene	2,015 (1,004 cases + 1,011 controls)	693,377	N/A
Triglycerides (TG)	188,577	2,508,369	Teslovich TM, Musunuru K, Smith AV, et al. Biological, clinical and population relevance of 95 loci for blood lipids. Nature. 2010;466:707-13.
Low Density Lipoprotein (LDL)	188,577	2,508,375
High Density Lipoprotein (HDL)	188,577	2,508,370
C-Reactive Protein (CRP)	82,725	2,671,742	Dehgan A, Dupuis J, Barbalic M, et al. Meta-analysis of genome-wide association studies in >80 000 subjects identifies multiple loci for C-reactive protein levels. Circulation. 2011;123:731-8.

For gene expression analyses, we used publicly available total RNA expression data from 1647 autopsied brain tissues (from dorsolateral prefrontal cortex, visual cortex and cerebellum) in 549 brains of 376 AD patients and 173 non-demented healthy controls from the Gene Expression Omnibus (GEO) data set GSE44772. ²⁴ As described previously, ²⁴ all subjects were diagnosed at intake and each brain underwent extensive neuropathology examination. Tissues were profiled on a custom-made Agilent 44K array of 40,638 DNA probes.

Statistical analysis

Using recently developed statistical methods to evaluate pleiotropic effects, ^13-17 we evaluated SNPs associating with AD (discovery cohort - IGAP Stage 1) and CRP levels as well as AD and plasma lipid levels. For given associated phenotypes A and B, pleiotropic ‘enrichment’ of phenotype A with phenotype B exists if the proportion of SNPs or genes associated with phenotype A increases as a function of increased association with phenotype B. To assess for enrichment, we constructed fold-enrichment plots of nominal –log₁₀(p) values for all AD SNPs and for subsets of SNPs determined by the significance of their association with CRP and plasma lipids. We also utilized conditional quantile-quantile (Q-Q) plots, which are complementary to fold-enrichment plots and provide visualization of polygenic enrichment (for additional details see Supplemental Information). In fold-enrichment plots, the presence of enrichment is reflected as an upward deflection of the curve for phenotype A if the degree of deflection from the expected null line is dependent on the degree of association with phenotype B. To assess for polygenic effects below the standard GWAS significance threshold, we focused the fold-enrichment plots on SNPs with nominal –log₁₀(p) < 7.3 (corresponding to p > 5×10⁻⁸). The enrichment seen can be directly interpreted in terms of true discovery rate (TDR = 1 – False Discovery Rate (FDR)) (for additional details see Supplemental Information). To identify specific loci we computed conditional FDRs. ^13,14 The standard FDR framework derives from a model that assumes the distribution of test statistics in a GWAS can be formulated as a mixture of null and non-null effects, with true associations (non-null effects) having more extreme test statistics, on average, than false associations (null effects). The FDR can be interpreted, as the probability that a SNP is null given its p-value is as small or smaller than its observed p-value. The conditional FDR is an extension of the standard FDR, which incorporates information from GWAS summary statistics of a second phenotype to adjust its significance level. The conditional FDR is defined as the probability that a SNP is null in the first phenotype given that the p-values in the first and second phenotypes are as small as or smaller than the observed ones. It is important to note that ranking SNPs by standard FDR or by p-values both give the same ordering of SNPs. In contrast, if the primary and secondary phenotypes are related genetically, conditional FDR re-orders SNPs, and results in a different ranking than that based on p-values alone. We used an overall FDR threshold of < 0.05, which means 5 expected false discovery per hundred reported. Additionally, we constructed Manhattan plots based on the ranking of conditional FDR to illustrate the genomic location. In all analyses, we controlled for the effects of genomic inflation by using intergenic SNPs (see Supplemental Information). Detailed information on fold enrichment and conditional Q-Q plots, Manhattan plots, and conditional FDR can be found in the Supplemental Information and prior reports. ^13-17 For loci with conditional FDR < 0.05, we performed a fixed effects, inverse variance weighted meta-analysis²⁵ across all available AD cohorts (IGAP Stage 1 + 2, deCODE, and DemGene, total n = 29,054 AD cases and 114,824 healthy controls) using the R package meta (http://CRAN.R-project.org/package=meta). Briefly, the fixed effects, inverse variance weighted meta-analysis summarizes the combined the statistical support across independent studies under the assumption of homogeneity of effects. Individual study β estimates (log odds ratios) are averaged, weighted by the estimated standard error. ²⁶ The IGAP Stage 1+2 β estimates and standard errors were obtained from the publicly available summary statistics (for additional details, Online Methods and Supplementary Note within reference18). For the DeCODE and DemGene cohorts, β estimates and standard errors were estimated via logistic regression predicting AD case/control status from SNP risk alleles count and adjusting for appropriate covariates including principal components.

For the gene expression analyses, we focused on transcript expression (total RNA levels) of genes closest (within 500 kB) to the SNPs reaching genome-wide significance in our meta-analysis. Using logistic regression, we examined whether transcript expression of these genes significantly differed between AD cases and controls.

RESULTS

We observed SNP enrichment for AD (IGAP Stage 1 – discovery cohort) across different levels of significance with CRP, TG, HDL and LDL levels indicating a genetic association between AD and the four cardiovascular phenotypes (Figure 1). For progressively stringent p-value thresholds for AD SNPs (i.e. increasing values of nominal –log₁₀(p)), we found at least 50-fold enrichment using CRP, 30-fold enrichment using TG, 20-fold enrichment using HDL and 40-fold enrichment using LDL (Figure 1). Conditional Q-Q plots similarly demonstrated polygenic enrichment in AD as a function of CRP and plasma lipids (Supplemental Figure 1).

Figure 1.

Fold enrichment plots of enrichment versus nominal -log₁₀ p-values (corrected for inflation) in Alzheimer's disease (AD) below the standard GWAS threshold of p < 5×10⁻⁸ as a function of significance of association with C-reactive protein (CRP) (panel A), high-density lipoprotein (HDL) (panel B), low-density lipoprotein (LDL) (panel C), and triglycerides (TG) (panel D) at the level of -log₁₀(p) ≥ 0, -log₁₀(p) ≥ 1, -log₁₀(p) ≥ 2 corresponding to p ≤ 1, p ≤ 0.1, p ≤ 0.01, respectively. Blue line indicates all SNPs.

To identify AD-associated polymorphisms that are more likely to replicate, we ranked IGAP Stage 1 AD SNPs conditional on their genetic association with CRP and plasma lipids (conditional FDR). We restricted our analyses to SNPs found in both IGAP Stage 1 and 2 and focused on those AD variants that have not been previously described at a genome-wide significant level. At a conditional FDR < 0.05, we found 55 AD susceptibility loci from IGAP Stage 1 (Figure 2, Supplemental Table 1). For these 55 loci, we performed a meta-analysis across all available AD cohorts and found two novel genome-wide significant (p < 5 × 10⁻⁸) loci associated with increased risk for AD (Table 2). These two variants are: 1) rs13113697 (chromosome 4, closest gene HS3ST1, conditioning trait = TG, reference allele = T, OR = 1.07, 95% CI = 1.05-1.11, p = 2.86 × 10⁻⁸) (Figures 3a and 4a) and 2) rs7920721 (chromosome 10, closest gene ECHDC3, conditioning trait = TG, reference allele = G, OR = 1.07, 95% CI = 1.04-1.11, p = 3.38 × 10⁻⁸) (Figures 3b and 4b).

Figure 2.

‘Conditional Manhattan plot’ of conditional –log₁₀ (FDR) values for Alzheimer's disease (AD) alone (IGAP Stage 1 AD cohort) (black) and AD given C-reactive protein (CRP; AD|CRP, green), triglycerides (TG; AD|TG, aquamarine), high-density lipoprotein (HDL, AD|HDL orange), and low-density lipoprotein (LDL; AD|LDL, red). SNPs with conditional – log₁₀ FDR > 1.3 (i.e. FDR < 0.05) are shown with large points. A black line around the large points indicates the most significant SNP in each LD block and this SNP was annotated with the closest gene, which is listed above the symbols in each locus. For additional details, see Supplemental Information.

Table 2.

New loci reaching genome-wide significance at conditional FDR < 0.05. Odds ratios provided for the major allele.

SNP	Position	Chr	Nearest Gene	Reference Allele	Associated phenotype	Min Cond FDR	IGAP Stage 1+2 p- value	IGAP Stage 1+2 OR (95% CI)	deCODE p-value	deCODE OR (95% CI)	DemGene p-value	DemGene OR (95% CI)	Meta- analysis p- value	Meta- analysis OR (95% CI)
rs13113697	11711232	4	HS3ST1	T	TG	9.56E-03	5.03E-07	1.07 (1.04-1.10)	0.031	1.07 (1.01-1.14)	0.088	1.13 (0.98-1.31)	2.86E-08	1.07 (1.05-1.11)
rs7920721	11720308	10	ECHDC3	G	TG	4.49E-02	2.89E-07	1.07 (1.04-1.09)	0.12	1.05 (0.99-1.11)	0.08	1.12 (0.99-1.29)	3.38E-08	1.07 (1.04-1.11)

Figure 3.

Forest plots for (a) rs13113697 on chromosome 4 and (b) rs7920721 on chromosome 10.

Figure 4.

Regional association plots for (a) rs13113697 on chromosome 4, and (b) rs7920721 on chromosome 10. Linkage Disequilibrium measured in the 1000 genomes European Populations using plink v1.07.

The meta-analysis also revealed three suggestive AD susceptibility loci with p-values < 1 × 10⁻⁶ (Table 3, Supplemental Figure 2). These three loci are rs7396366 (on chromosome 11, closest gene AP2A2, conditioning trait = CRP, reference allele = C, OR = 0.94, 95% CI = 0.92-0.96, p = 6.8 × 10⁻⁷), rs3131609 (on chromosome 15, closest gene USP50, conditioning trait = CRP, reference allele = C, OR = 0.93, 95% CI = 0.91-0.96, p = 7.21 × 10⁻⁷) and rs2526378 (on chromosome 17, closest gene BZRAP1, conditioning trait = TG, reference allele = G, OR = 0.94, 95% CI = 0.92-0.96, p = 2.73 × 10⁻⁷).

Table 3.

SNPs showing suggestive association with AD at conditional FDR < 0.05. Odds ratios provided for the major allele.

SNP	Position	Chr	Nearest Gene	Reference Allele	Associated phenotype	Min Cond FDR	IGAP Stage 1+2 p- value	IGAP Stage 1+2 OR (95% CI)	deCODE p-value	deCODE OR (95% CI)	DemGene p-value	DemGene OR (95% CI)	Meta- analysis p-value	Meta- analysis OR (95% CI)
rs7396366	11711232	11	AP2A2	C	CRP	3.91E-02	2.89E-06	0.93 (0.91-0.96)	0.22	0.96 (0.91-1.02)	0.21	0.92 (0.91-0.96)	6.80E-07	0.94 (0.92-0.96)
rs3131609	11720308	15	USP50	C	CRP	4.49E-02	3.90E-07	0.93 (0.90-0.96)	0.94	1.0 (0.93-1.08)	0.95	0.99 (0.86-1.15)	7.21E-07	0.93 (0.91-0.96)
rs2526378	47336320	17	BZRAP1	G	TG	1.83E-03	8.34E-07	0.94 (0.91-0.96)	0.50	0.98 (0.93-1.03)	9.20E-04	0.80 (0.70-0.91)	2.73E-07	0.94 (0.92-0.96)

We additionally evaluated the directionality of allelic effects in SNPs associated with AD and the four cardiovascular phenotypes (SNPs with conditional FDR < 0.05). Across all 55 shared loci, we found the same direction of effect between SNPs associated with AD and 1) CRP in 72% (18 out of 25, p-value = 0.02) 2) HDL in 40% (4 out of 10, p-value = 0.62), 3) LDL in 20% (1 out of 5, p-value = 0.81), and 4) TG in 40% (6 out of 15, p-value = 0.69) (Supplemental Table 1). For HS3ST1 and ECHD3 variants, we found an opposite direction of allelic effect between increased AD risk and TG levels (Supplemental Table 1).

We assessed whether HS3ST1 and ECHD3 transcript levels are altered in AD brains compared with control brains (GEO dataset GSE 4472). We found significantly decreased HS3ST1 transcript expression (standardized β-coefficient = −0.09201, standard error (SE) = 0.01864, p = 9.99 × 10⁻⁷) and significantly increased ECHDC3 transcript expression (standardized β-coefficient = 0.12715, SE = 0.01829, p = 8.32 x×10⁻¹²) in AD brains compared with control brains.

DISCUSSION

In this study, we show that polymorphisms associated with CRP and plasma lipids (TG, HDL and LDL) are also associated with increased risk for AD (genetic pleiotropy). We found that genetic enrichment in AD based on SNP association with cardiovascular phenotypes results in improved statistical power for gene discovery. By conditioning on polymorphisms associated with CRP and plasma lipid levels, we identified 55 AD susceptibility loci. In meta-analyses across 4 independent cohorts, we found that two of these risk variants, namely rs13113697 (on chromosome 4, closest gene HS3ST1) and rs7920721 (on chromosome 10, closest gene ECHDC3), were genome-wide significant. We additionally observed that HS3ST1 and ECHDC3 transcript expression was different in AD brains compared with control brains.

Our findings provide novel insights into the relationship between AD pathogenesis, inflammation and dyslipidemia, beyond the known loci associated with AD. We found a consistent direction of allelic effect between SNPs associated with AD risk and CRP levels indicating overlapping pathobiology between AD and inflammation. These results are consistent with the hypothesis that inflammatory mechanisms influence Alzheimer's pathogenesis ^9,27-28 and may have implications for treatment and prevention strategies in AD. On the other hand, we did not find a consistent direction of allelic effect between SNPs associated with AD risk and plasma lipid levels (LDL, HDL and TG). Additionally, for HS3ST1 and ECHD3 variants, we found an opposite direction of allelic effect between increased AD risk and TG levels. One hypothesis for these findings is that the observed pleiotropy between AD and plasma lipids could be due to different haplotypes/gene alleles involving the same SNPs. Another equally plausible hypothesis is that the same haplotypes/gene alleles are involved for both AD and plasma lipids but the underlying biologic mechanisms are distinct. Based on these findings, it seems less likely that the pleiotropic SNPs detected in this study influence AD pathogenesis via cholesterol mediated pathways.

Unlike epidemiological studies, co-heritability analyses, ²⁹ or bivariate GWAS methods, ³⁰ one strength of our current approach is the ability to detect genetic pleiotropy even when there is no correlation of the signed effects (mixed directionality of effect); the conditional FDR method can detect SNPs that have a non-null effect in one trait and that also tend to have a non-null effect in another trait, independent of directionality. Another strength of this framework is leveraging genetic signal in one phenotype to identify variants in a second phenotype that would otherwise not be detected using a single phenotype approach. We note that the conditional FDR approach allows for re-ordering (and re-ranking) of SNPs based on p-value significance in the second phenotype (e.g. CRP or TG) thus enabling identification of novel SNPs in the primary phenotype (e.g. AD). In addition, as previously demonstrated, these genetic analysis methods result in improved sensitivity for a given specificity. ¹³ Using this ‘pleiotropic’ approach, we detected 55 novel variants indicating that genetic enrichment improves statistical power for gene discovery. In meta-analyses, we discovered two GWAS significant AD susceptibility loci. The closest genes associated with the two risk variants showed altered RNA levels in postmortem AD brains compared with control brains suggesting a functional role. The first variant (rs13113697) is closest to the HS3ST1 gene on chromosome 4 (Figure 4a), which encodes heparan sulfate glucosaminyl 3-O-sulfotransferase, an intraluminal Golgi protein enzyme with multiple biological activities. ³¹ The second variant (rs7920721) is closest to the ECHDC3 gene on chromosome 10 (Figure 4b), which encodes an enzyme called enoyl CoA hydratase domain containing 3. ³² We note that by conditioning on cardiovascular traits and evaluating additional AD cohorts (deCODE and DemGene), we were able to find genome-wide significant evidence for previously¹⁸ suggested signal close to HS3ST1 and ECHDC3. At a p-value < 1.0 × 10⁻⁶, we additionally found three suggestive variants on chromosome 11 (rs7396366, closest gene APA2A), chromosome 15 (rs3131609, closest gene USP50) and chromosome 17 (rs2526378, closest gene BZRAP1).

It is important to note that in this study the diagnosis of AD was established clinically. Postmortem evidence from community and population based cohorts indicates that vascular brain injury often presents concomitantly with Alzheimer's pathology and correlates with cognitive impairment above and beyond AD neuropathology. ³³ It is feasible that the clinically diagnosed AD individuals from the IGAP, deCODE and DemGene cohorts may have concomitant vascular brain disease, which may further contribute to their cognitive decline and dementia. As such, an alternative interpretation of our findings is that the susceptibility loci identified in this study may increase brain vulnerability to vascular and/or inflammatory insults, which in turn may exacerbate the clinical consequences of AD pathological changes.

In conclusion, we found polygenic overlap between AD, CRP and plasma lipids, and leveraged this association to identify two novel genome-wide significant variants associated with increased AD risk. Careful and considerable effort will be required to further characterize the novel candidate genes detected in this study and to detect the functional variants responsible for the association of these loci with Alzheimer's risk. Although no single common variant maybe informative clinically, a combination of variants involved with inflammation or lipid metabolism may help identify older individuals at increased risk for AD. Our findings may also have implications for Alzheimer's prevention trials involving anti-inflammatory agents.