Key PointsQuestion听
Is circulating cholesterol level associated with early-onset Alzheimer disease, and if so, what is the underlying genetic mechanism?
Findings听
This case series of plasma samples from 2125 cases and controls found an association between early-onset Alzheimer disease and higher levels of low-density lipoprotein cholesterol independent of the effects of APOE, as well as demonstrated enrichment of rare coding variants of APOB, a gene known to influence plasma cholesterol levels, in samples from patients with early-onset Alzheimer disease.
Meaning听
Two parallel lines of evidence suggest that circulating low-density lipoprotein cholesterol levels may play a role in the pathogenesis of early-onset Alzheimer disease, and future work should elucidate whether APOB influences the risk for early-onset Alzheimer disease by acting through plasma low-density lipoprotein cholesterol levels or through other mechanisms.
Importance听
Early-onset Alzheimer disease (EOAD) is a rare form of Alzheimer disease (AD) with a large genetic basis that is only partially understood. In late-onset AD, elevated circulating cholesterol levels increase AD risk even after adjusting for the apolipoprotein E 蔚4 (APOE E4) allele, a major genetic factor for AD and elevated cholesterol levels; however, the role of circulating cholesterol levels in EOAD is unclear.
Objectives听
To investigate the association between circulating cholesterol levels and EOAD and to identify genetic variants underlying this possible association.
Design, Setting, and Participants听
In this case series, plasma cholesterol levels were directly measured in 267 samples from the AD research centers (ADRCs) of Emory University and University of California, San Francisco, collected from January 21, 2009, through August 21, 2014. The association between cholesterol and EOAD was examined using multiple linear regression. To determine the underlying genetic variants, APOB, APP, PSEN1, and PSEN2 were sequenced in samples from 2125 EOAD cases and controls recruited from 29 ADRCs from January 1, 1984, through December 31, 2015. Data were analyzed from November 23, 2016, through April 10, 2018.
Exposures听
Clinical diagnosis, age at clinical diagnosis, plasma cholesterol measures (total cholesterol, low-density lipoprotein cholesterol [LDL-C], high-density lipoprotein cholesterol, triglycerides, and apolipoprotein B), and genetic variants in APOE, APP, PSEN1, PSEN2, and APOB.
Main Outcomes and Measures听
The primary outcome was the association between EOAD and plasma cholesterol measures. The secondary outcome was the association between EOAD and the burden of genetic variants in APOB.
Results听
Of the 2125 samples that underwent genetic sequencing, 1276 were from women (60.0%) and 654 (30.8%) were from patients with EOAD (mean [SD] ages, 55.6 [4.3] years for cases and 72.0 [9.6] years for controls). APOE E4 explained 10.1% of the variance of EOAD. After controlling for APOE E4, EOAD cases had higher levels of total cholesterol (mean difference [SE], 21.9 [5.2] mg/dL; P鈥�=鈥�2.9鈥壝椻赌�10鈭�5), LDL-C (mean difference [SE], 22.0 [4.5] mg/dL; P鈥�=鈥�1.8鈥壝椻赌�10鈭�6), and ApoB (mean difference [SE], 12.0 [2.4] mg/dL; P鈥�=鈥�2.0鈥壝椻赌�10鈭�6) than controls in 267 frozen samples. Approximately 3% of EOAD cases carried known AD-causing mutations. Gene-based rare variant burden testing in 2066 samples showed that rare APOB coding variants were significantly more abundant in EOAD cases after adjusting for sex, APOE E4, genetic principal components, ADRC center, and batch (effect size, 0.20; P鈥�=鈥�4.20鈥壝椻赌�10鈭�4).
Conclusions and Relevance听
Elevated LDL-C levels were associated with higher probability of having EOAD, and EOAD cases were enriched for rare coding variants in APOB, which codes for the major protein of LDL-C. Collectively, these novel findings highlight the important role of LDL-C in EOAD pathogenesis and suggest a direct link of APOB variants to AD risk.
Early-onset Alzheimer disease (EOAD) is a rare form of Alzheimer disease (AD) that manifests before 65 years of age and has a large genetic basis with heritability of 91% to 100%.1 Studies of families with EOAD led to the discovery of AD-causing mutations in 3 genes鈥攁myloid precursor protein (APP [NCBI ]), presenilin 1 (PSEN1 [NCBI ]), and presenilin 2 (PSEN2 [NCBI ])鈥攖hat proved pivotal in establishing the amyloid hypothesis in AD.2 The precise contribution of APP, PSEN1, and PSEN2 to EOAD is unknown but is likely less than 10% of all incident EOAD cases,1-3 leaving approximately 90% of EOAD cases unexplained.1
To identify new causes of EOAD, we investigated the role of circulating cholesterol because elevated cholesterol levels have been linked to increased risk of late-onset AD (LOAD) in multiple lines of evidence. First, as with LOAD, EOAD is associated with the apolipoprotein E 蔚4 allele (APOE E4) [OMIM ]),4 and APOE E4 is known to raise circulating cholesterol levels, primarily by raising low-density lipoprotein cholesterol (LDL-C) levels.5 Second, epidemiologic studies of LOAD have found elevated midlife cholesterol levels increase the risk of AD and cognitive decline,6,7 even after adjusting for APOE E4.8-12 Third, several large studies reported that use of drugs to lower cholesterol levels is associated with a reduced AD risk,13-18 and the largest of these studies found that this effect was independent of APOE E4.18 These data suggest that cholesterol is a risk factor for AD independent of APOE; however, the role of cholesterol in EOAD and of other cholesterol-related genes in EOAD is not known.
Herein, we present a 3-part study to identify new causes of EOAD. First, we examined the contribution of known AD-causing mutations鈥�APP, PSEN1, PSEN2, and APOE E4鈥攊n 2125 patients with EOAD and controls recruited from across the United States. Second, we investigated whether different fractions of circulating plasma lipoproteins were associated with EOAD after adjusting for APOE E4 in 267 EOAD cases and controls from the Alzheimer鈥檚 Disease Research Centers (ADRCs) of Emory University, Atlanta, Georgia, and University of California, San Francisco (UCSF). Third, on finding that EOAD was associated with higher levels of total cholesterol, low-density lipoprotein cholesterol (LDL-C), and plasma apolipoprotein B (ApoB), we aimed to understand the mechanistic link between cholesterol and AD. To that end, we performed a large-scale targeted sequencing of the 2125 EOAD cases and controls for APOB because APOB is known to harbor rare variants that have strong effects on lowering or raising LDL-C levels.19-21
Additional details about the study are provided in the eMethods in the Supplement. All individuals were recruited from US ADRCs using established comprehensive research evaluations and diagnostic criteria.22-24 Patients with EOAD were defined as having probable or definite AD, using the National Institute of Neurological and Communicative Diseases and Stroke鈥揂lzheimer鈥檚 Disease and Related Disorders Association Work Group criteria,24 with symptoms beginning at 65 years or younger, and controls were cognitively normal individuals older than 60 years. The Emory University ADRC samples constituted the discovery data set, and the UCSF ADRC samples and the ADRC samples obtained from the National Cell Repository for Alzheimer鈥檚 Disease (NCRAD) constituted 2 separate replication data sets. Samples from the Emory University and UCSF ADRCs provided frozen plasma, and all samples had genetic material available. Samples were restricted to white research volunteers because only this group had sufficient sample size for meaningful testing. For the samples obtained from NCRAD, we obtained demographic data and APOE genotyping on available samples from the National Alzheimer鈥檚 Coordinating Center. All participants or their legal guardians gave written informed consent. Samples from each center were obtained using research protocols approved by the respective institutional review boards. All genetic analyses for this work were approved by the institutional review board of Emory University.
Plasma samples that had not previously been thawed were available for measurement of LDL-C levels in 267 of 667 of the Emory University and UCSF samples (40.0%). Samples were randomized with respect to sex, EOAD status, and age at draw for the Emory University and UCSF data sets, which were analyzed together. Lipid levels were analyzed by colorimetric method on a chemistry analyzer (AU480; Beckman Coulter) at the Emory Lipid Research Laboratory. Total cholesterol and triglyceride levels were assayed using reagents from Beckman Diagnostics, and LDL-C and high-density lipoprotein cholesterol (HDL-C) levels were determined by the homogeneous assay methods using reagents from Sekisui Diagnostics. Levels of ApoB were determined by immunoturbidometric methods using reagents from Sekisui Diagnostics.
Multiplex polymerase chain reaction primers were designed for the Access Array System by using the MPD software25 that targeted APOB, APP, PSEN1, PSEN2, 400 ancestrally informative markers, and 25 common X chromosome markers. The PEMapper and PECaller computational package26 was used for mapping and base calling, respectively, against the Genome Reference Consortium human genome build 38. Variants were annotated and filtered using Bystro,27 and quality control procedures are detailed in eMethods in the Supplement.
Data were analyzed from November 23, 2016, through April 10, 2018. Estimates of the liability to AD were generated using age at onset for patients with EOAD and age at enrollment for controls, sex, and AD presence or absence by linear interpolation, as described previously.1 Linear regression was used to test for association between EOAD and copies of APOE E4 alleles, adjusted for sex and sample source.
Linear regression was used to test for association between each plasma cholesterol measure (ie, total cholesterol level and directly measured ApoB, LDL-C, HDL-C, and triglyceride levels) and EOAD, adjusted for sex, batch, educational attainment, and copies of APOE E4 alleles. In a separate model, we included the aforementioned covariates plus age because age and diagnosis are correlated (ie, affected individuals are younger by definition). To further control for the APOE effect, a stratified analysis was performed on samples within each of the 2 most common genotypes, APOE E3/3 and APOE E3/4, adjusting for the same covariates.
Statistical testing of sequencing data was performed with a burden-based variable threshold test28 as implemented in RAREMETALWORKER (version 4.13.8) in the discovery data set and separately in each replication data set. Meta-analysis of all 3 data sets was performed using RAREMETAL (version 4.13.8).29 Based on the premise that sites of interest are likely to be rare (or would have been identified by genome-wide association study) and of at least moderate effect size (or we would not have power to detect), we focused our analysis on those in which the coding sequence is altered (ie, missense or nonsense mutations) with a Combined Annotation Dependent Depletion Phred scale score of greater than 20.30 We used a variable-threshold gene-based burden test with an adaptive frequency that selects a minor allele threshold to maximize the burden statistic. Analysis in each individual data set (ie, 1 discovery and 2 replication data sets) was adjusted for sex and the first 4 genetic principal components. The final meta-analysis was adjusted for sex, the first 4 genetic principal components, copies of the APOE E4 allele, study center, and sequencing batch. Our primary analysis was to test for enrichment of rare APOB coding variants in EOAD. A secondary analysis was to test for enrichment of rare coding variants in known AD-causing genes in EOAD.
A total of 2125 cases and controls from 29 ADRCs contributed to this study (1276 women [60.0%] and 849 men [40.0%]; mean [SD] ages, 55.6 [4.3] years for cases and 72.0 [9.6] years for controls) and were divided into the following 3 data sets: a discovery data set consisting of samples from the Emory University ADRC (n鈥�=鈥�381), a first replication data set consisting of samples from the UCSF ADRC (n鈥�=鈥�181), and a second replication data set consisting of samples from the 27 other ADRCs (excluding samples from the Emory University and UCSF ADRCs) (n鈥�=鈥�1563). Table 1 provides demographic details for each data set. Patients with EOAD were younger than controls (by design), had a higher burden of APOE E4 alleles (as expected), were more likely to be male in the Emory and other ADRCs, and tended to have fewer years of formal education for all three groups (Table 1). In the NCRAD data set, educational attainment data were present for all control subjects but were missing for 69% of patients with EOAD, making statistical adjustment for educational attainment unfeasible.
The Contribution of APP, PSEN1, and PSEN2 to EOAD
Deep sequencing of APP, PSEN1, and PSEN2 in all 2125 samples revealed 13 previously described AD-causing mutations in 23 individuals (3.4% of all samples), including 2 sites in APP (n鈥�=鈥�2), 10 sites in PSEN1 (n鈥�=鈥�18), and 2 sites in PSEN2 (n鈥�=鈥�3) (eTable 1 in the Supplement). A gene-based burden analysis of rare (minor allele frequency <1%) coding variants in APP, PSEN1, and PSEN2 (including known pathogenic sites) was performed. Only PSEN1 was significantly associated with EOAD, although well-described pathogenic mutations were found in all 3 genes (eTables 2-4 in the Supplement). For PSEN1, we found no evidence of enrichment in the Emory University (effect size, 鈭�0.11; P鈥�=鈥�.70) or UCSF (effect size, 0.42; P鈥�=鈥�.06) data set, but there was an association with the larger replication data set of 1563 samples (effect size, 0.41; P鈥�=鈥�1.6鈥壝椻赌�10鈭�5) that was even stronger in a meta-analysis of all 3 data sets (effect size, 0.37; P鈥�=鈥�1.1鈥壝椻赌�10鈭�5). To account for potential selection bias of known AD mutation carriers in the data sets, we performed a fully adjusted meta-analysis that accounted for number of APOE E4 alleles and study center. This analysis revealed a weaker association between PSEN1 and EOAD in 2088 samples (effect size, 0.21; P鈥�=鈥�2.60鈥壝椻赌�10鈭�3) (eTable 3 in the Supplement) and no association between APP or PSEN2 and EOAD. Moreover, to determine whether known pathogenic mutations in APP, PSEN1, or PSEN2 were responsible for the association between PSEN1 and EOAD, we excluded these known pathogenic mutations and repeated the gene-based rare variant burden analyses. They revealed no significant association between EOAD and APP, PSEN1, or PSEN2 (eTables 2-4 in the Supplement). These findings suggest that known AD-causing mutations account for the association between PSEN1 and EOAD in our data. Going forward, for our primary analysis described below, all known AD mutation carriers were excluded to reduce the genetic heterogeneity of EOAD and improve our power to detect new associations in the remaining samples.
Variance in Liability to EOAD
As expected, we found APOE E4 highly associated with EOAD in 2083 samples (APOE E4 allele frequency of 75.1% in cases vs 30.5% in controls; P鈥�&濒迟;鈥�2鈥壝椻赌�10鈭�16). We used sex-specific AD prevalence estimates and linear interpolation to find that APOE explains approximately 10.1% of the variance in liability to EOAD after adjusting for sex and study center.
Elevated Cholesterol Level Independent of APOE
Given the strong associations between APOE E4 and EOAD and between APOE E4 and elevated circulating LDL-C levels, we expected individuals with EOAD to have elevated LDL-C levels. In this analysis of 267 frozen samples from the Emory University and UCSF data sets, we found that even after accounting for the effect of APOE E4, EOAD cases had higher mean (SD) levels of total cholesterol (203.0 [42.7] vs 177.7 [37.2] mg/dL; P鈥�=鈥�2.9鈥壝椻赌�10鈭�5), LDL-C (131.6 [41.9] vs 104.4 [30.6] mg/dL; P鈥�=鈥�1.8鈥壝椻赌�10鈭�6), and plasma ApoB (82.2 [20.8] vs 69.3 [16.6] mg/dL; P鈥�=鈥�2.0鈥壝椻赌�10鈭�6) compared with controls (to convert cholesterol to millimoles per liter, multiply by 0.0259; ApoB to grams per liter, multiply by 0.01). Because total cholesterol largely consists of LDL-C and ApoB is the main lipoprotein of LDL-C, these findings are consistent with one another. On the other hand, EOAD showed no association with HDL-C (mean [SD], 62.7 [15.6] vs 65.1 [20.5] mg/dL; P鈥�=鈥�.39) and was only nominally associated with higher triglyceride levels (mean [SD], 141.3 [70.2] vs 127.3 [66.4] mg/dL; P鈥�=鈥�.05) (to convert triglycerides to millimoles per liter, multiply by 0.0113) (Table 2). All results were adjusted for study center, batch, sex, educational attainment, and copies of APOE E4. From these data, we estimated that LDL-C explains 7.6% of the variance in liability to EOAD independently of APOE E4.
To further demonstrate that the association between EOAD and LDL-C is independent of APOE, we performed 2 separate regression analyses stratified by APOE E3/3 or APOE E3/4 genotype. Both analyses confirmed that EOAD cases have significantly higher LDL-C levels compared with controls independently of APOE E3/3 in 153 samples (mean difference [SE], 25.6 [5.8] mg/dL; P鈥�=鈥�2.0鈥壝椻赌�10鈭�5) or APOE E3/4 in 82 samples (mean difference [SE], 19.8 [8.4] mg/dL; P鈥�=鈥�.02) (Figure, B and C).
Age is well known to influence LDL-C; however, adjusting for age is potentially problematic in our models because age and diagnosis were significantly correlated (Spearman 蟻鈥�=鈥夆垝0.34; P鈥�=鈥�1.3鈥壝椻赌�10鈭�9). Nevertheless, adjusting for age gives similar results in both the adjusted analysis (mean difference [SE] for LDL-C level, 21.3 [5.4] mg/dL; P鈥�=鈥�1.2鈥壝椻赌�10鈭�4) and stratified analyses for APOE E3/3 (mean difference [SE] for LDL-C level, 25.3 [7.0] mg/dL; P鈥�=鈥�4.3鈥壝椻赌�10鈭�4) and APOE E3/4 (mean difference [SE] for LDL-C level, 17.8 [10.1] mg/dL; P鈥�=鈥�.08). Age was not significantly associated with LDL-C in any of these models. Together, these results demonstrate that elevated levels of LDL-C (and ApoB) were significantly associated with increased EOAD risk, and this effect was only partially mediated by APOE E4 genotype.
Rare APOB Genetic Coding Variants
To identify genes underlying the association between LDL-C and EOAD independently of APOE, we performed deep resequencing of APOB because APOB is known to have variants that raise or lower LDL-C levels independently of APOE genotype,19-21 and ApoB is the main lipoprotein of LDL-C. Thus, we hypothesized that rare genetic variants in APOB partially underlie the association between elevated LDL-C levels and EOAD observed above. To test this hypothesis, we examined APOB variants associated with coding changes (ie, missense or nonsense variants) with a scaled Combined Annotation Dependent Depletion score of greater than 20. This cutoff was selected to focus on variants that likely alter structure, function, or expression.
In the 381 samples in the Emory University discovery data set, we found significant enrichment of APOB variants in EOAD after adjusting for sex and genetic principal components (effect size, 0.26; P鈥�=鈥�9.75鈥壝椻赌�10鈭�3) (Table 3). This association was replicated in the 1548 samples in the ADRC replication data set after adjusting for sex and genetic principal components (effect size, 0.21; P鈥�=鈥�.02) (Table 3) but not in 173 samples in the UCSF replication data set, possibly owing to small sample size (effect size, 鈭�0.07; P鈥�=鈥�.78) (Table 3). Results of the meta-analysis of 2102 samples in all 3 data sets showed a stronger association between APOB variants and EOAD (effect size, 0.23; P鈥�=鈥�6.40鈥壝椻赌�10鈭�4) (Table 3) than in individual data set analyses. A final meta-analysis that was additionally adjusted for study center and APOE E4, in addition to sex and genetic principal components, also showed a strong association between APOB and EOAD independent of APOE E4 in 2066 samples (effect size, 0.20; P鈥�=鈥�4.20鈥壝椻赌�10鈭�4) (Table 3).
Fifty-seven rare APOB coding variants were included in the final meta-analysis, with 33 variants in 31 EOAD cases, 24 variants in 24 controls, and no shared variants (eTable 5 in the Supplement). Five percent of patients with EOAD carried an allele, in comparison with 1.7% of controls (P鈥�=鈥�3.72鈥壝椻赌�10鈭�5 by Fisher exact test). Two affected individuals from the Emory University and combined ADRC data sets were compound heterozygotes, with the remainder being heterozygotes. Each compound heterozygote case was heterozygous for 2 different rare coding sites: p.150P>S and p.999T>I in one case and p.3074Y>H and p.3919S>T in the other case. Of these 4 variants, only p.150P>S has been previously described (minor allele frequency of 7.41鈥壝椻��10鈭�6) in the Genome Aggregation Database.31 Twenty-six variants were described in the Genome Aggregation Database with a mean allele frequency of 6.27鈥壝椻��10鈭�4 (range, 0.11鈥壝椻��10鈭�6 to 4.06鈥壝椻��10鈭�6). Because APOB mutations can cause hyperlipidemia, we queried the ClinVar database,32 a database of genomic variants and human health, for these APOB variants and found 11 annotated sites, of which 3 are labeled as 鈥渃onflicting interpretations of pathogenicity鈥� and 8 as 鈥渙f uncertain significance鈥� for 鈥渇amilial hypercholesterolemia.鈥� The sites with conflicting interpretations of pathogenicity were identified in 1 EOAD case (p.3294S>P []) and 2 controls (p.532R>W [] and p.3594W>R []). One of those, , was found to be associated with hypercholesterolemia in a Malaysian data set.33
We found no evidence that a particular ApoB region or domain has a clustering of variants found in EOAD cases or controls in any of the 5 known domains by using the Fisher exact test. To determine whether carriers of rare coding APOB variants accounted for the difference in LDL-C observed for EOAD in our data, we examined plasma lipid and sequencing data from samples obtained in 221 subjects with such variants. Among carriers of rare coding APOB variants, we found no statistically significant difference in LDL-C levels between cases and controls, likely owing to small sample size and thus low power (n鈥�=鈥�21) (mean difference [SE] for LDL-C level, 18.4 [16.5] mg/dL; P鈥�=鈥�.28). Among noncarriers of APOB variants, we continued to see higher LDL-C levels among patients with EOAD (n鈥�=鈥�200) (mean difference [SE], 18.1 [4.9] mg/dL; P鈥�=鈥�2.6鈥壝椻赌�10鈭�4) after adjusting for APOE variants and sex. This finding suggests that additional factors contribute to differences in LDL-C levels in patients with EOAD (eFigure in the Supplement).
Early-onset AD is an extreme phenotype of AD that accounts for approximately 10% of AD cases. In this study, we first examined the role of known AD-causing genes (ie, APP, PSEN1, and PSEN2) and the largest single genetic risk factor for EOAD (ie, APOE mutations) in a large data set including patients with EOAD (n鈥�=鈥�654) and controls (n鈥�=鈥�1471) from 29 US ADCs/ADRCs regardless of family history. In our data set, we found approximately 3.4% of all patients with EOAD carried a known AD-causing variant; however, only rare coding variants in PSEN1 were associated with EOAD, although pathogenic variants were found in all 3 genes. The association between EOAD and PSEN1 was accounted for by known pathogenic mutations.
As expected, we confirmed the strong association between APOE E4 and EOAD, with APOE E4 accounting for approximately 10.1% of the liability to EOAD. The association between APOE E4 and EOAD prompted us to examine the role of LDL-C in EOAD, given the known association between APOE E4 and elevated LDL-C levels. This association has not been previously examined, to our knowledge. Using frozen plasma samples from 267 patients with EOAD and controls from the Emory University and UCSF ADRCs, we found a robust association between elevated LDL-C levels and EOAD independent of APOE E4. This finding led us to hypothesize that genes with genetic variants that are known to be associated with higher LDL-C likely increase EOAD risk. To test this hypothesis, we performed large-scale targeted sequencing of APOB, a gene known to harbor rare variants that alter LDL-C, and found a novel genetic association between rare APOB coding variants and EOAD. Approximately 5.0% of patients with EOAD and 1.7% of controls were found to harbor a rare coding polymorphism in APOB that is likely to disrupt the structure, functions, or abundance of ApoB protein. Finally, we showed that the association between higher LDL-C levels and EOAD was not fully explained by APOE E4 or APOB, suggesting that additional genes and/or mechanisms contribute to those findings.
To place the genetic association between APOB and EOAD in context, we note that only 3.4% of all EOAD cases in our combined data set showed a known pathogenic mutation, and we found a stronger association between EOAD and rare coding variants in APOB compared with PSEN1 in our fully adjusted analysis. A likely explanation of this finding is ascertainment bias, that is, carriers with PSEN1 mutations are more likely to be recruited by an ADC because of their mutation status. In this case, we would expect to see attenuation in the association between known AD-causing mutations and EOAD after adjusting for ADC site, which is precisely what we observed.
Our finding of elevated LDL-C levels in EOAD is consistent with epidemiologic studies of LOAD.34 Likewise, our estimate that APOE variants account for 10.1% of the variance in EOAD is comparable with the reported estimate that approximately 13% of the genetic variance in LOAD is explained by APOE.35 Our finding of a significant association between rare coding variants in APOB and EOAD independently of APOE is novel, important, and consistent with multiple genome-wide association studies that revealed strong associations between LOAD and common intron markers of genes involved in brain cholesterol metabolism (eg, ABCA7, BIN1, CLU, SORL1).36-38 Also consistent with our findings is the report that rare loss-of-function variants in ABCA7 were enriched in LOAD.39 Furthermore, mice overexpressing ApoB show hyperlipidemia, neurodegeneration, increases in APP, accumulation of amyloid plaques, and cognitive impairment similar to mice overexpressing wild-type human APP.40,41 Collectively, these studies and our findings suggest an important role of cholesterol metabolism in AD pathogenesis.
Strengths and Limitations
A notable strength of the study is that samples were recruited by 29 US ADRCs that have standardized research and diagnosis protocols, which enables greater generalizability of our findings and mitigates the influence of the recruitment strategy of any given center. Our findings generated a number of questions, including whether APOB may harbor protective and deleterious variants and the role of APOB variants in LOAD. Our data showed that controls who carried specific APOB variants continued to have lower LDL-C levels. In particular, APOB p.154R>X was associated with an exceptionally low LDL-C level (39 mg/dL). Thus, APOB may harbor protective variants similar to the common APOE E2 allele. With regard to LOAD, the rarity of these APOB variants and the reduced heritability of LOAD (compared with EOAD) would make identifying risk and protective variants unlikely, given the power of current LOAD sequencing efforts.42
Genetic studies that are designed to probe for associations between a trait or disease (ie, EOAD) and a potential biological intermediate (ie, LDL-C) commonly infer causality vs genetic pleiotropy using mendelian randomization.43 In this study, the rarity of the alleles tested and the limited number of cases that included plasma LDL-C measurement made this process unfeasible. Thus, we were unable to conclude that the observed association is causal and not due to genetic pleiotropy; however, both APOB and APOE act on plasma circulating cholesterol, suggesting a connection, perhaps through effects on the brain vasculature. Moreover, APOE is a key gene in brain cholesterol homeostasis, whereas APOB is not expressed in the brain. Thus, variants in APOE that alter cholesterol metabolism may act both centrally and peripherally, whereas APOB variants may act predominantly if not exclusively through the periphery.
Another potential limitation of this study is that the LDL-C analysis may be confounded by unavailable data (eg, severity of AD, smoking, or use of drugs to lower cholesterol levels). However, the genetic association between APOB and EOAD is unlikely to be confounded by these factors; its statistical significance substantially tempers this concern.
Our APOB findings are not of exome-wide significance, but neither were the PSEN1 findings. The low prevalence of EOAD presents a challenge for obtaining enough samples to perform well-powered whole-exome or whole-genome sequencing studies. The existing LOAD sequencing efforts have excluded individuals with EOAD. Furthermore, our findings in plasma assays and in genetic sequencing were mutually consistent and complementary, strongly supporting an overall association between APOB and EOAD.
Our study provides evidence that circulating cholesterol is associated with EOAD independently of APOE E4. Furthermore, we have identified novel rare genetic coding changes in APOB that are associated with EOAD independently of APOE. The APOB variants we found do not fully explain the association between elevated LDL-C levels and EOAD, indicating that further studies are needed to identify additional genetic variants underlying the contribution of lipid metabolism to AD pathogenesis.
Accepted for Publication: January 18, 2019.
Corresponding Author: Thomas S. Wingo, MD, Department of Neurology, Emory University School of Medicine, 505K Whitehead Bldg, 615 Michael St NE, Atlanta, GA 30322 (thomas.wingo@emory.edu).
Published Online: May 28, 2019. doi:10.1001/jamaneurol.2019.0648
Author Contributions: Dr T. Wingo had full access to all the data in the study and takes responsibility for the integrity of the data and accuracy of the data analysis.
Concept and design: T. Wingo, Cutler, Miller, Lah, Levey.
Acquisition, analysis, or interpretation of data: T. Wingo, Cutler, A. Wingo, Le, Rabinovici, Lah, Levey.
Drafting of the manuscript: T. Wingo, Cutler, Lah.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: T. Wingo, Cutler, A. Wingo.
Obtained funding: T. Wingo, Lah, Levey.
Administrative, technical, or material support: Lah, Levey.
Supervision: T. Wingo, Miller, Lah, Levey.
Conflict of Interest Disclosures: Dr T. Wingo reports receiving grants from the National Institutes of Health (NIH), the National Institute of Aging (NIA), and the Veterans Administration (VA) during conduct of the study. Dr Cutler reported receiving grants from the NIH during the conduct of the study. Dr A. Wingo reported receiving grants from the VA during conduct of the study. Dr Rabinovici reported receiving grants from Avid Radiopharmaceuticals, Eli Lilly, Piramal Imaging, and GE Healthcare and personal fees from Eisai, Axon Neuroscience, Merck, Genentech, and Roche outside the submitted work. Dr Miller reported receiving grants from the NIH, NIA, Quest Diagnostics Dementia Pathway Collaboration, Cornell University Subcontract, and the Bluefield Project to cure frontotemporal dementia during the conduct of the study. Dr Lah reported receiving grants from the NIH, Avanir Pharmaceuticals, Genentech, AbbVie, Novartis, Biogen, Lilly, Cognito Therapeutics, and Roche during the conduct of the study. Dr Levey reported serving as a webinar speaker for Future Health LLC, and receiving grants from the NIH, the NIA, AbbVie, Genentech, Cognito Therapeutics, Biogen, Merck, Esai, Novartis, and Takeda outside the submitted work. No other disclosures were reported.
Funding/Support: This work was supported by grants IK2 BX001820 and IK2 CX000601 from the Veterans Health Administration; grants P50 AG025688, P50 AG023501, R01 AG056533, and R01 AG045611 from the NIH; the Emory Integrated Genomics Core, which is subsidized by the Emory University School of Medicine and is one of the Emory Integrated Core Facilities; the To Remember Foundation; the Douglas French Alzheimer鈥檚 Foundation; and contract 04-33516 from the State of California Department of Health Services, Alzheimer鈥檚 Disease Research Center of California.
Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the Veterans Health Administration or the NIH.
Additional Contributions: We thank contributors who collected samples used in this study, as well as patients and their families, whose help and participation made this work possible. Zoe White, BS, and Se Min Canon, BS, Wingo Laboratory, provided technical assistance, for which they were compensated.
Additional Information: Samples from the National Cell Repository for Alzheimer鈥檚 Disease (NCRAD), which receives government support under cooperative agreement grant U24 AG21886 from the NIA, were used in this study. The National Alzheimer鈥檚 Coordinating Center (NACC) database is funded by grant U01 AG016976 from the NIA/NIH. The NACC data are contributed by Alzheimer disease centers with the following grants from the NIA: P30 AG019610 (principal investigator [PI], Eric Reiman, MD), P30 AG013846 (PI, Neil Kowall, MD), P50 AG008702 (PI, Scott Small, MD), P50 AG025688 (PI, Allan Levey, MD, PhD), P50 AG047266 (PI, Todd Golde, MD, PhD), P30 AG010133 (PI, Andrew Saykin, PsyD), P50 AG005146 (PI, Marilyn Albert, PhD), P50 AG005134 (PI, Bradley Hyman, MD, PhD), P50 AG016574 (PI, Ronald Petersen, MD, PhD), P50 AG005138 (PI, Mary Sano, PhD), P30 AG008051 (PI, Thomas Wisniewski, MD), P30 AG013854 (PI, M. Marsel Mesulam, MD), P30 AG008017 (PI, Jeffrey Kaye, MD), P30 AG010161 (PI, David Bennett, MD), P50 AG047366 (PI, Victor Henderson, MD, MS), P30 AG010129 (PI, Charles DeCarli, MD), P50 AG016573 (PI, Frank LaFerla, PhD), P50 AG005131 (PI, James Brewer, MD, PhD), P50 AG023501 (PI, Bruce Miller, MD), P30 AG035982 (PI, Russell Swerdlow, MD), P30 AG028383 (PI, Linda Van Eldik, PhD), P30 AG053760 (PI, Henry Paulson, MD, PhD), P30 AG010124 (PI, John Trojanowski, MD, PhD), P50 AG005133 (PI, Oscar Lopez, MD), P50 AG005142 (PI, Helena Chui, MD), P30 AG012300 (PI, Roger Rosenberg, MD), P30 AG049638 (PI, Suzanne Craft, PhD), P50 AG005136 (PI, Thomas Grabowski, MD), P50 AG033514 (PI, Sanjay Asthana, MD, FRCP), P50 AG005681 (PI, John Morris, MD), and P50 AG047270 (PI, Stephen Strittmatter, MD, PhD).
1.Wingo
听TS, Lah
听JJ, Levey
听AI, Cutler
听DJ. 听Autosomal recessive causes likely in early-onset Alzheimer disease.听听Arch Neurol. 2012;69(1):59-64. doi:
2.Cacace
听R, Sleegers
听K, Van Broeckhoven
听C. 听Molecular genetics of early-onset Alzheimer鈥檚 disease revisited.听听Alzheimers Dement. 2016;12(6):733-748. doi:
3.Campion
听D, Dumanchin
听C, Hannequin
听D,
听et al. 听Early-onset autosomal dominant Alzheimer disease: prevalence, genetic heterogeneity, and mutation spectrum.听听Am J Hum Genet. 1999;65(3):664-670. doi:
4.van Duijn
听CM, de Knijff
听P, Cruts
听M,
听et al. 听Apolipoprotein E4 allele in a population-based study of early-onset Alzheimer鈥檚 disease.听听Nat Genet. 1994;7(1):74-78. doi:
5.Chasman
听DI, Par茅
听G, Mora
听S,
听et al. 听Forty-three loci associated with plasma lipoprotein size, concentration, and cholesterol content in genome-wide analysis.听听PLoS Genet. 2009;5(11):e1000730. doi:
6.Kivipelto
听M, Solomon
听A. 听Cholesterol as a risk factor for Alzheimer鈥檚 disease: epidemiological evidence.听听Acta Neurol Scand Suppl. 2006;185:50-57. doi:
7.Shepardson
听NE, Shankar
听GM, Selkoe
听DJ. 听Cholesterol level and statin use in Alzheimer disease, I: review of epidemiological and preclinical studies.听听Arch Neurol. 2011;68(10):1239-1244. doi:
8.Kivipelto
听M, Helkala
听EL, Laakso
听MP,
听et al. 听Apolipoprotein E 蔚4 allele, elevated midlife total cholesterol level, and high midlife systolic blood pressure are independent risk factors for late-life Alzheimer disease.听听Ann Intern Med. 2002;137(3):149-155. doi:
9.Notkola
听IL, Sulkava
听R, Pekkanen
听J,
听et al. 听Serum total cholesterol, apolipoprotein E 蔚4 allele, and Alzheimer鈥檚 disease.听听狈别耻谤辞别辫颈诲别尘颈辞濒辞驳测. 1998;17(1):14-20. doi:
10.Kalmijn
听S, Foley
听D, White
听L,
听et al. 听Metabolic cardiovascular syndrome and risk of dementia in Japanese-American elderly men: the Honolulu-Asia Aging Study.听听Arterioscler Thromb Vasc Biol. 2000;20(10):2255-2260. doi:
11.Whitmer
听RA, Sidney
听S, Selby
听J, Johnston
听SC, Yaffe
听K. 听Midlife cardiovascular risk factors and risk of dementia in late life.听听狈别耻谤辞濒辞驳测. 2005;64(2):277-281. doi:
12.Jarvik
听GP, Wijsman
听EM, Kukull
听WA, Schellenberg
听GD, Yu
听C, Larson
听EB. 听Interactions of apolipoprotein E genotype, total cholesterol level, age, and sex in prediction of Alzheimer鈥檚 disease: a case-control study.听听狈别耻谤辞濒辞驳测. 1995;45(6):1092-1096. doi:
13.Wolozin
听B, Kellman
听W, Ruosseau
听P, Celesia
听GG, Siegel
听G. 听Decreased prevalence of Alzheimer disease associated with 3-hydroxy-3-methyglutaryl coenzyme A reductase inhibitors.听听Arch Neurol. 2000;57(10):1439-1443. doi:
14.Li
听G, Larson
听EB, Sonnen
听JA,
听et al. 听Statin therapy is associated with reduced neuropathologic changes of Alzheimer disease.听听狈别耻谤辞濒辞驳测. 2007;69(9):878-885. doi:
15.Jick
听H, Zornberg
听GL, Jick
听SS, Seshadri
听S, Drachman
听DA. 听Statins and the risk of dementia.听听尝补苍肠别迟. 2000;356(9242):1627-1631. doi:
16.Sparks
听DL, Kryscio
听RJ, Sabbagh
听MN, Connor
听DJ, Sparks
听LM, Liebsack
听C. 听Reduced risk of incident AD with elective statin use in a clinical trial cohort.听听Curr Alzheimer Res. 2008;5(4):416-421. doi:
17.Cramer
听C, Haan
听MN, Galea
听S, Langa
听KM, Kalbfleisch
听JD. 听Use of statins and incidence of dementia and cognitive impairment without dementia in a cohort study.听听狈别耻谤辞濒辞驳测. 2008;71(5):344-350. doi:
18.Haag
听MD, Hofman
听A, Koudstaal
听PJ, Stricker
听BH, Breteler
听MM. 听Statins are associated with a reduced risk of Alzheimer disease regardless of lipophilicity: the Rotterdam Study.听听J Neurol Neurosurg Psychiatry. 2009;80(1):13-17. doi:
19.Peloso
听GM, Auer
听PL, Bis
听JC,
听et al; NHLBI GO Exome Sequencing Project. 听Association of low-frequency and rare coding-sequence variants with blood lipids and coronary heart disease in 56,000 whites and blacks.听听Am J Hum Genet. 2014;94(2):223-232. doi:
20.Lange
听LA, Willer
听CJ, Rich
听SS. 听Recent developments in genome and exome-wide analyses of plasma lipids.听听Curr Opin Lipidol. 2015;26(2):96-102. doi:
21.Lange
听LA, Hu
听Y, Zhang
听H,
听et al; NHLBI Grand Opportunity Exome Sequencing Project. 听Whole-exome sequencing identifies rare and low-frequency coding variants associated with LDL cholesterol.听听Am J Hum Genet. 2014;94(2):233-245. doi:
22.National Alzheimer鈥檚 Coordinating Center. Description of NCAA database. . Accessed April 10, 2018.
23.Beekly
听DL, Ramos
听EM, Lee
听WW,
听et al; NIA Alzheimer鈥檚 Disease Centers. 听The National Alzheimer鈥檚 Coordinating Center (NACC) database: the Uniform Data Set.听听Alzheimer Dis Assoc Disord. 2007;21(3):249-258. doi:
24.McKhann
听G, Drachman
听D, Folstein
听M, Katzman
听R, Price
听D, Stadlan
听EM. 听Clinical diagnosis of Alzheimer鈥檚 disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer鈥檚 Disease.听听狈别耻谤辞濒辞驳测. 1984;34(7):939-944. doi:
25.Wingo
听TS, Kotlar
听A, Cutler
听DJ. 听MPD: multiplex primer design for next-generation targeted sequencing.听听BMC Bioinformatics. 2017;18(1):14. doi:
26.Johnston
听HR, Chopra
听P, Wingo
听TS,
听et al; International Consortium on Brain and Behavior in 22q11.2 Deletion Syndrome. 听PEMapper and PECaller provide a simplified approach to whole-genome sequencing.听听Proc Natl Acad Sci U S A. 2017;114(10):E1923-E1932. doi:
27.Kotlar
听AV, Trevino
听CE, Zwick
听ME, Cutler
听DJ, Wingo
听TS. Bystro: rapid online variant annotation and natural-language filtering at whole-genome scale. Genome Biol. 2018;19(1):14.
28.Price
听AL, Kryukov
听GV, de Bakker
听PI,
听et al. 听Pooled association tests for rare variants in exon-resequencing studies.听听Am J Hum Genet. 2010;86(6):832-838. doi:
29.Feng
听S, Liu
听D, Zhan
听X, Wing
听MK, Abecasis
听GR. 听RAREMETAL: fast and powerful meta-analysis for rare variants.听听叠颈辞颈苍蹿辞谤尘补迟颈肠蝉. 2014;30(19):2828-2829. doi:
30.Kircher
听M, Witten
听DM, Jain
听P, O鈥橰oak
听BJ, Cooper
听GM, Shendure
听J. 听A general framework for estimating the relative pathogenicity of human genetic variants.听听Nat Genet. 2014;46(3):310-315. doi:
31.Lek
听M, Karczewski
听KJ, Minikel
听EV,
听et al; Exome Aggregation Consortium. 听Analysis of protein-coding genetic variation in 60,706 humans.听听狈补迟耻谤别. 2016;536(7616):285-291. doi:
32.Landrum
听MJ, Lee
听JM, Benson
听M,
听et al. 听ClinVar: public archive of interpretations of clinically relevant variants.听听Nucleic Acids Res. 2016;44(D1):D862-D868. doi:
33.Lye
听SH, Chahil
听JK, Bagali
听P,
听et al. 听Genetic polymorphisms in LDLR, APOB, PCSK9 and other lipid related genes associated with familial hypercholesterolemia in Malaysia.听听PLoS One. 2013;8(4):e60729. doi:
34.Anstey
听KJ, Ashby-Mitchell
听K, Peters
听R. 听Updating the evidence on the association between serum cholesterol and risk of late-life dementia: review and meta-analysis.听听J Alzheimers Dis. 2017;56(1):215-228. doi:
35.Ridge
听PG, Hoyt
听KB, Boehme
听K,
听et al; Alzheimer鈥檚 Disease Genetics Consortium (ADGC). 听Assessment of the genetic variance of late-onset Alzheimer鈥檚 disease.听听Neurobiol Aging. 2016;41:200.e13-200.e20. doi:
36.Hollingworth
听P, Harold
听D, Sims
听R,
听et al; Alzheimer鈥檚 Disease Neuroimaging Initiative; CHARGE consortium; EADI1 consortium. 听Common variants at ABCA7, MS4A6A/MS4A4E, EPHA1, CD33 and CD2AP are associated with Alzheimer鈥檚 disease.听听Nat Genet. 2011;43(5):429-435. doi:
37.Lambert
听JC, Ibrahim-Verbaas
听CA, Harold
听D,
听et al; European Alzheimer鈥檚 Disease Initiative (EADI); Genetic and Environmental Risk in Alzheimer鈥檚 Disease; Alzheimer鈥檚 Disease Genetic Consortium; Cohorts for Heart and Aging Research in Genomic Epidemiology. 听Meta-analysis of 74 046 individuals identifies 11 new susceptibility loci for Alzheimer鈥檚 disease.听听Nat Genet. 2013;45(12):1452-1458. doi:
38.Naj
听AC, Jun
听G, Beecham
听GW,
听et al. 听Common variants at MS4A4/MS4A6E, CD2AP, CD33 and EPHA1 are associated with late-onset Alzheimer鈥檚 disease.听听Nat Genet. 2011;43(5):436-441. doi:
39.Steinberg
听S, Stefansson
听H, Jonsson
听T,
听et al; DemGene. 听Loss-of-function variants in ABCA7 confer risk of Alzheimer鈥檚 disease.听听Nat Genet. 2015;47(5):445-447. doi:
40.Bereczki
听E, Bern谩t
听G, Csont
听T, Ferdinandy
听P, Scheich
听H, S谩ntha
听M. 听Overexpression of human apolipoprotein B-100 induces severe neurodegeneration in transgenic mice.听听J Proteome Res. 2008;7(6):2246-2252. doi:
41.L枚ffler
听T, Flunkert
听S, Havas
听D,
听et al. 听Impact of ApoB-100 expression on cognition and brain pathology in wild-type and hAPPsl mice.听听Neurobiol Aging. 2013;34(10):2379-2388. doi:
42.Naj
听AC, Schellenberg
听GD; Alzheimer鈥檚 Disease Genetics Consortium (ADGC). 听Genomic variants, genes, and pathways of Alzheimer鈥檚 disease: an overview.听听Am J Med Genet B Neuropsychiatr Genet. 2017;174(1):5-26. doi:
43.Smith
听GD, Ebrahim
听S. 鈥溙齅endelian randomization鈥�: can genetic epidemiology contribute to understanding environmental determinants of disease?听听Int J Epidemiol. 2003;32(1):1-22. doi: