|Home | About | Journals | Submit | Contact Us | Français|
The ε4 allele of the apolipoprotein E gene (APOE) is associated with increased risk and earlier age at onset in late onset Alzheimer’s disease (AD). Other factors, such as expression level of apolipoprotein E protein (apoE), have been postulated to modify the APOE related risk of developing AD. Multiple loci in and outside of APOE are associated with a high risk of AD. The aim of this exploratory hypothesis generating investigation was to determine if some of these loci predict cerebrospinal fluid (CSF) apoE levels in healthy non-demented subjects. CSF apoE levels were measured from healthy non-demented subjects 21–87 years of age (n = 134). Backward regression models were used to evaluate the influence of 21 SNPs, within and surrounding APOE, on CSF apoE levels while taking into account age, gender, APOE ε4 and correlation between SNPs (linkage disequilibrium). APOE ε4 genotype does not predict CSF apoE levels. Three SNPs within the TOMM40 gene, one APOE promoter SNP and two SNPs within distal APOE enhancer elements (ME1 and BCR) predict CSF apoE levels. Further investigation of the genetic influence of these loci on apoE expression levels in the central nervous system is likely to provide new insight into apoE regulation as well as AD pathogenesis.
Apolipoprotein E (apoE) is involved in lipid transport and binds to cell surface receptors to mediate lipoprotein uptake. It is the major apolipoprotein synthesized in the brain . The apoE protein exists in three extensively studied isoforms, E2, E3, and E4 that are the result of two non-synonymous SNPs, rs429358 and rs7412, located in exon 4 of the APOE gene. Structural consequences of the exon 4 APOE haplotype appear to be that the apoE E4 protein binds preferentially to plasma very low density lipids (VLDLs) whereas apoE E2 and E3 bind preferentially to plasma high density lipoproteins (HDLs) . In addition, apoE isoforms appear to influence plasma cholesterol levels [5,11], neuronal growth [15,28,41] and amyloid deposition [9, 27,31].
The APOE ε4 allele, defined by the rs429358 SNP, and age are currently the only risk factors strongly associated with late onset Alzheimer’s disease (AD) . However, a family history of dementia is associated with an increased risk of AD in the elderly only among APOE ε4 carriers and a large proportion of APOE ε4 carriers who survive into old age remain cognitively normal suggesting other genetic factors besides APOE ε4 play a role in AD [12,14].
There is no clear consensus on whether cerebrospinal fluid (CSF) apoE levels are associated with AD. Studies thus far indicate CSF apoE is either lower in AD [4, 17,21] or has no association with AD [20,26]. CSF apoE levels in healthy populations appear to be associated with age, but not with gender [16,20,45] although one study did not find an association with age . In addition, APOE ε4 appears not to be associated with CSF apoE levels in healthy populations or AD [4,17, 20,21,26,45], but is associated with lower apoE plasma levels in healthy subjects [21,37]. Interestingly, Fukumoto et al. report that apoE plasma levels are lower in AD compared to the mildly cognitively impaired  and APOE e3/e4 heterozygotes have a higher proportion of the apoE E3 isoform than the E4 in plasma . In contrast, the opposite proportion is present in CSF, suggesting a differential metabolism or regulation of apoE isoforms depending on the biological compartment measured; plasma or CSF . Because plasma apoE is produced by the liver and CSF apoE is secreted by brain glial cells [22,33], it appears that apoE and possibly apoE isoforms may be metabolized or regulated differently in the two compartments.
AD risk has been reported to be associated with APOE promoter polymorphisms directly upstream of the transcription start site. Characterized APOE promoter polymorphisms include −491, −427, −219 (Th1/E47cs), and +113. Establishing AD risk associated with APOE promoter polymorphisms independent of APOE ε4 has been difficult, with contrasting reports likely due to strong linkage disequilibrium (LD) in the region [3,6,30,34,42].
Regulatory sites distal to APOE have been characterized. Multienhancer 1 and 2 (ME1 and ME2) are homologous regulatory regions are located 3.3 and 15.9 kb distal to APOE and influence APOE regulation in macrophages and adipocytes [24,38]. Two homologous hepatic control regions (HCR1 and HCR2) play a role in APOE regulation in the liver and are located approximately 15 kb and 27 kb distal to APOE [1,39]. Recently, a brain control region (BCR) has been described and is located 41.9 kb distal to APOE between APOC2 and CLPTM1 .
The influence of APOE promoter SNPs on apoE levels in plasma and brain has been characterized  but, to our knowledge, the influence of APOE promoter SNPs and APOE distal regulatory regions on CSF apoE levels has not been investigated.
Several SNPs within the TOMM40 gene, in a region 15 kb proximal to the APOE promoter, are associated with AD risk, but are also in strong LD with APOE ε4 [25,46]. When these SNPs are entered into a logistic regression model that includes APOE ε4, only the effect of APOE ε4 remains significantly associated with AD . These observations diminish the possibility that loci in the TOMM40 gene, proximal to APOE, have a major effect on AD risk independent of APOE ε4. LD with APOE ε4 exists distal to the APOE gene in a region that contains the regulatory element ME1, but is weak in the region of the HCR2 and does not exist in the BCR, suggesting that the ME1 regulatory element is dependent on the presence or absence of APOE ε4, but HCR2 and BCR may be independent of APOE ε4 .
Because SNPs upstream of APOE (proximal), as well as SNPs downstream of APOE (distal) within ME1, HCR2 and BCR, are either within a LD block surrounding the APOE gene or are associated with APOE regulation, we hypothesized that genetic elements in the region proximate to APOE along with the APOE promoter and distal regulatory elements (ME1 and BCR), predict apoE expression in CSF, in a manner that can be dependent or independent of APOE ε4. We further hypothesized that HCR2, a hepatic control region of APOE, would not have an effect on the central nervous system and, thus, would not predict CSF apoE levels.
Thus, the aim of this exploratory investigation was to evaluate the potential influence of multiple genetic loci, within and surrounding the APOE gene, on CSF apoE levels in a cognitively normal population. The first aim was to characterize CSF apoE levels according to age, gender and APOE ε4 status in a sample of cognitively normal subjects. The second aim was to explore the influence of the APOE proximal, promoter and distal region SNPs on CSF apoE levels.
All procedures were approved by the institutional review boards of the participating institutions. Subjects were 148 healthy carefully evaluated non-demented adults age 21–87 years. All subjects provided written consent. Mini-Mental State Examination (MMSE) scores were between 26–30 and Clinical Dementia Rating Scale (CDR) scores were 0 as previously described . All 148 subjects were genotyped for 22 SNPs within a 70 kb region within and surrounding the APOE gene. Fourteen subjects lacked complete data for all 22 SNPs leaving 134 subjects to be utilized in the analysis. The mean age, gender ratio, APOE ε4 status, and mean CSF apoE levels are described in Table 1.
CSF was collected in the morning after an overnight fast using Sprotte 24-g atraumatic spinal needles as previously described . Samples were frozen immediately on dry ice and stored at −80°C until assay at which time apoE concentrations (mg/dl) were measured using a nephelometer (Dade Behring). Nephelometry is an antibody-based automated method that uses the BN II Nephelometer and a kit from Dade Behring for apoE measurement. Briefly, 10 μls of the CSF sample is used undiluted in a reaction between antigen (apoE) and antibody (human anti-apoE). Antibody positive and negative controls are provided with the kit. Nepholometry is a highly standardized method with an average CSF apoE inter-assay variability of less than 1%. No values with an inter-assay variability of greater than 5% are accepted.
Genomic DNAs, 5ng each, were dispensed into individual wells of a 384 plate and air dried overnight. Plates were then covered with adhesive lid tape and stored at 4°C. For TaqMan allelic discrimination detection on the 384 well plates, using a final volume of 3 μl per well. For each reaction; 0.075 μl of 20× SNP TaqMan Assay (Applied Biosystems, CA), 1.5 μl of TaqMan Universal PCR Master Mix (Applied Biosystems, CA) and 1.425 μl of dH2O were pipetted into each well. PCR reactions were carried out using a 9700 Gene Amp PCR System (Applied Biosystems, CA). The PCR profile was 95°C for 10 min and then 50 cycles at 92°C for 15 sec and 60°C for 90 sec. Plates were then subjected to end-point read in a 7900 Real-Time PCR System (Applied Biosystems, CA). Twenty one SNPs were genotyped including APOE proximal SNPs, APOE promoter SNPs, SNPs within the APOE gene and APOE distal SNPs (Table 3). The results were first evaluated by cluster variations, the allele calls were then assigned automatically before transferred and integrated into the genotype database, except for the ME1 and HCR2 SNPs.
The ME1 and HCR2 SNPs are located within a duplicated region, which has extremely high homology with ME2 and HCR1 elements, and thus were first PCR amplified (95°C for 15 min, 30 cycles of 92°C for 20 sec, 55°C for 30 sec, 72°C for 15 min) using primer sets specific to the one region surrounding the SNPs (not the duplication) and then subsequently DNA sequenced to determine the nucleotide content at each SNP site. BigDye terminator cycle sequencing kit (Applied Biosystems, CA) in a final volume of 10 μl, and the sequence information were collected from an 377 DNA Sequencer (AB). Primer sequences for region-specific PCR and sequencing is listed in Supplementary Table 1.
SNPs were selected in the region surrounding the APOE gene. SNPs included SNPs proximate to the APOE gene that have been previously described , SNPs distal to APOE, and SNPs within ME1, HCR2 and BCR. All SNPs are described in Table 2.
We examined the relationship between CSF apoE levels and APOE region SNPs, gender, and age using backward stepwise linear regression models. One SNP (rs157582) was omitted from all analyses because it was perfectly correlated with another SNP (rs157581) leaving 21 SNPs for the analyses. Except for rs429358, SNP genotypes were coded as 1, 2, or 3 and treated as continuous variables. The contribution of the rs429358 SNP was modeled using APOE ε4 status (coded as 1 or 2, where 2 denotes no ε4 alleles are present).
Backward elimination was performed using a variable selection procedure in which all variables are entered into the equation and then sequentially removed until the stopping criterion is reached . The variable with the smallest partial correlation with the dependent variable was considered first for removal. Variables were removed listwise from the model depending on the significance of the partial F-value (p-value > 0.10). After the first variable was removed, the variable remaining in the equation with the smallest partial correlation was considered next. The procedure stopped when there were no variables in the equation that satisfied the removal criteria (p-value > 0.10). R2 values were determined to assess the proportion of variability in CSF apoE levels that is explained by the predictor variables. The backward regression p-values do not take into account multiple comparisons. Bonferroni multiple comparison corrections were performed (Table 3, Table 5). Backwards regression based on the Akaike Information Criterion was also performed with the same results as the backward regression shown in Table 3 (data not shown) .
To determine if CSF apoE levels in our study sample (n = 134) are consistent with previous studies in which CSF apoE levels were not found to be associated with APOE ε4 [4,17,20,21,26]or gender but were associated with age [16,20,45], CSF apoE levels were stratified by APOE genotype, the presence or absence APOE ε4, gender and age (Fig. 1). No significant difference in CSF apoE levels between APOE genotypes (p = 0.59 in ε2/ε3/ε4 genotype and p = 0.23 in ε4 allele, Fig. 1, panel A and panel B) and gender (p = 0.18, Fig. 1, panel C) was found. In contrast, a significant increase in CSF apoE levels with increasing age was observed (p = 0.002, Fig. 1, panel D).
SNPs (Table 2) were selected either on the basis of our previous study that found most of these SNPs to be in LD with APOE ε4 as well as associated with AD risk , or on the basis of their position within ME1, HCR2 or BCR. These SNPs (n = 21) span approximately 70 kb and range from a proximal SNP (rs6857) located 5′ of the TOMM40 gene, to a distal SNP (rs7247551) within the APOE BCR.
All SNPs (n = 21) are in Hardy Weinberg equilibrium (data not shown). In addition, we determined the LD structure, (utilizing D′ and R2), within this study sample and found it to be similar to our previous findings (Table 4) . These results suggest the presence of LD across the region, which is consistent with our previous observation .
We applied backward regression models to assess the effects of the 21 SNPs on CSF apoE levels (Table 3) while taking into account APOE ε4 status, gender and age. The models show that six SNPs may predict CSF apoE levels, including three SNPs in the proximal region (rs11556506, n17664883, rs157584), one in the promoter (rs449647; −491), and two in the distal region (n17684509; ME1 and rs7247551; BCR) of APOE. The proportion of variance explained by the model, as calculated by R2, is 19%. Backward regression models based on the Akaike Information Criterion gave similar results where the same six SNPs predicted CSF apoE levels (data not shown).
In models in which only age and each of the six SNPs are entered alone, without other SNPs, into a linear regression model, either with or without APOE ε4 present in the model, only rs449647, independent of APOE ε4, and n17664883, in the presence of APOE ε4, predict CSF apoE levels (data not shown).
To determine if carriers of the more rare alleles (minor-allele-homozygoteplus heterozygote) have low CSF apoE levels, each SNP was collapsed into “non-variant” group (with major-allele-homozygote only) and “variant” group (with minor-allele-homozygote plus heterozygote) and linear regression was performed while taking into account age but not APOE ε4. In this model n17664883 (p-value, 0.002) and n17684509 (p-value, 0.021) predict CSF apoE levels while rs449647 marginally predicts CSF apoE levels (p-value, 0.088) (Fig. 2). However, after Bonferroni correction for multiple comparisons, only n17664883 remains significant (p-value, 0.030). Mean CSF apoE levels for SNP variant carriers compared to non-variant carriers while taking into account age are shown in Fig. 2.
To further address the question of whether a combination of SNPs may contribute to CSF apoE levels the six positive SNPs identified in the backward regression models were analyzed for their ability to predict CSF apoE levels. All possible combinations of these six SNPs were analyzed. Five haplotypes were found to predict CSF apoE levels while taking into account age and while taking into account all the haplotype combinations within the haplotype tested (Table 5: see haplotype p-values). After correcting for multiple comparisons only two of these haplotypes predict CSF apoE levels (see Bonferroni p-values for haplotype: n17684509, rs7247551 and haplotype: n17664883, rs449647, n17684509, rs7247551) (Table 5).
Because APOE ε4 and age are important AD risk factors, the characterization of regulatory elements surrounding the APOE gene that may predict the level of apoE E4 protein, is vital in understanding AD pathogenesis. The influence of APOE promoter SNPs on apoE levels in plasma and brain has been studied [18, 19] but, to our knowledge, the influence of APOE promoter SNPs or APOE distal regulatory regions on CSF apoE levels has not been characterized. In a previous investigation, SNPs proximal to APOE were found to be associated with AD risk  leading us to wonder if SNPs in the region proximal to APOE lay within an un-characterized APOE regulatory element. Thus, in this investigation we evaluated the differences in healthy subjects CSF apoE levels while taking into account age, gender and APOE ε4 status, as well as associations with 21 SNPs in a large 70 kb region surrounding the APOE gene. This study was an exploratory investigation to test the hypothesis that multiple genetic loci surrounding APOE can predict CSF apoE levels. The results indicate that in addition to the APOE promoter, SNPs in the TOMM40 gene, the ME1 and the BCR regions may also predict CSF apoE levels.
Consistent with the previous reports, CSF apoE levels increased with increasing age (Fig. 1, Panel D); and the levels were not associated with APOE ε4 genotype (Fig. 1, Panel A) or allele (Fig. 1, Panel B) or gender (Fig. 1, Panel C) [4,16,17,20,21,26,45].
Backward linear regression models indicate that out of the 21 SNPs entered into the models, six SNPs may predict CSF apoE levels (three SNPs within TOMM40; rs11556505, n17664883, rs157584, one APOE promoter SNP, rs449647, one SNP within ME1; n17684509, and one SNP within the BCR; rs7247551; Table 3). Of these six SNPs only two SNPs predict CSF apoE levels either with APOE ε4 (n17664883) or without APOE ε4 (rs449647) present in the model (data not shown).
Interestingly, previous reports indicate that the APOE −491 promoter polymorphism (rs449647) is associated with AD risk, although it is unclear whether LD is obscuring these results [6,30,42]. In our investigation the non-variant (AA genotype) carriers of −491 polymorphism appeared to have higher CSF apoE levels than the variant (AT and TT genotypes) carriers both alone and as part of a haplotype (Fig. 2; Table 5) which is consistent with previous reports by Laws et al. where higher brain apoE levels  and plasma apoE levels  are associated with the −491 AA genotype. To our knowledge this is the first investigation of the association between CSF apoE levels and APOE promoter SNPs. These results reported here, along with previous reports, implicate the −491 promoter polymorphism as an important regulator of apoE levels.
The association with the novel proximal SNP, n17664883, suggests that an APOE regulatory element exists in the region distantly proximal to APOE that may predict CSF apoE levels. This region within intron 4 of the TOMM40 gene, may contain a regulatory element that contributes to high CSF apoE levels when the major homozygote genotype (non-variant) is present (Fig. 2, Table 5). Whereas the minor homozygote or the heterozygotes (variant) may contribute to lower CSF apoE levels (Fig. 2, Table 5). Alternatively, CSF apoE levels may be consequences of TOMM40 gene action. The TOMM40 gene encodes the TOM40 protein which is the pore subunit of the mitochondrial outer membrane protein translocator . Currently, there is no evidence of protein-protein interaction between TOM40 and apoE. However, recent reports suggest that AβPP may be targeted to the mitochondria and translocated across the mitochondrial membrane via the TOM40 protein ; additionally, the translocation arrest of AβPP in the TOM40 pore may lead to mitochondria dysfunction and neuronal loss in AD . Therefore, it is possible that variants of TOM40 may be more susceptible to the translocation arrest of AβPP and may lead to more profound decline of mitochondrial function and neuronal damage. The biological feedback mechanism would then be activated and produce more apoE for the neuronal repair or regeneration, a well-known function of apoE.
We also assessed whether the APOE distal regulatory elements (HCR, ME and BCR) can predict CSF apoE levels. Because evidence suggests that the HCR is primarily active in the liver [1,39], we hypothesized that only the SNPs within the ME1 (rs483082, n17684509, rs584007) and BCR (rs7247551) but not HCR2 (rs35136575), would predict apoE expression in CSF. Indeed, in our analyses the HCR2 SNP did not have an effect on CSF apoE levels, but SNPs from both ME1 and BCR did (Table 3, Fig. 2, Table 5). Interestingly, only one of the three SNPs within ME1, a small region which spans 620 bp, is predicted to be associated with CSF apoE levels (Table 3). This result may be attributed to high correlation of other SNPs in the model with SNPs in the ME1 region, thus, leading to the elimination of other ME1 SNPs from the regression model. This result does not necessarily reflect a lack of a biological influence on CSF apoE levels by the other ME1 SNPs.
The hypothesis generated by these results is that there may be additional regulators of APOE in the proximal region as far upstream as 15 kb. The proximal region may act in combination with regulatory elements in the distal region, such as the ME1 and the BCR to increase the activity of the APOE promoter. An example of distant proximal enhancers can be demonstrated by the APOB gene, which has a cis-element enhancer located 54 to 62 kilobases 5′ to the structural gene . Further support of our hypothesis is demonstrated by our finding whereby multiple loci together have effects on CSF apoE levels (Table 3, Fig. 2, Table 5) implicating an influence on CSF apoE levels by a large haplotype. Such a concept is in line with studies suggesting that promoter haplotypes of APOE can influence plasma apoE levels [40,44]. But, our study goes beyond previous studies by investigating contributions by distal regulatory regions, such as ME1, HCR2 and BCR, on CSF apoE levels.
There were limitations of this exploratory investigation. First, our study sample size (n = 134) may be too small to detect small effects contributed by a few of the SNPs tested that have low minor allele frequencies. Even though we required all minor allele frequencies to be equal to or greater than 2%, and collapsed genotypes into variant and non-variant groups, some of the haplotype numbers were low (Table 5). Second, the statistical results should be approached with caution because stepwise linear regression models do not take into account multiple comparisons so that p-values do not represent true significance until corrected for multiple comparisons (Table 3). Thus, it is important to note that this is an exploratory investigation intended to generate further hypotheses.
In summary, linear regression models were used to search for APOE regulatory SNPs that predict CSF apoE levels. These SNPs are located within a large region surrounding the APOE gene. Six SNPs were found to predict CSF apoE levels; three TOMM40 SNPs (rs11556505, n17664883, rs157584), one APOE promoter SNP (rs449647; −491) and two APOE distal SNPs (ME1; n17684509, BCR; rs7247551). For two of these SNPs, the novel SNP within the TOMM40 gene (n17664883) and the APOE promoter SNP (rs449647; −491), there is a significant difference in CSF apoE levels between genotypes and these two SNPs also appear to contribute to haplotype CSF apoE levels. These results support the hypothesis that modestly penetrant SNPs within APOE regulatory elements may explain part of the variation in CSF apoE protein levels. Our data indicate that a multigenetic approach may be more powerful in explaining the variation in apoE levels than a monogenetic approach. However, the total contribution of several SNPs together was modest, with a large proportion of the variation remaining unaccounted for (R2 value, 0.19; Table 3), which may suggest that future evaluation of molecular haplotypes in the APOE gene region in a larger study population is required to explain more specifically the variation in apoE levels. In addition, given that APOE proximal SNPs within the TOMM40 gene predict CSF apoE levels, a possible independent influence on CSF apoE levels by the TOM40 protein may exist.
The mechanism whereby APOE ε4 increases the risk of AD is uncertain. APOE ε4 carriers show substantial variance for age at onset of AD. Some individuals who are ε4 homozygotes may be spared from AD even if they live into their 9th or 10th decade [12,14]. Factors that alter apoE protein expression, such as an APOE ε4 haplotype that includes proximal and distal regulatory elements, may help to explain this variance. In addition to aging, we have now shown that certain SNPs appear to be associated with levels of apoE protein in CSF in cognitively normal individuals. If these SNPs are also associated with younger age of onset in AD, this would implicate levels of apoE as a factor in AD pathogenesis. An extension of this study would be an investigation of familial AD cases to evaluate the influence of family history of AD on AD age-at-onset and associations between SNPs within and around APOE, APOE ε4 status, CSF apoE levels as well as other AD biomarkers such as CSF Aβ40, Aβ42, and tau.
In conclusion, this exploratory investigation has generated further hypotheses regarding the possible influence on CSF apoE levels by multiple genetic loci within and surrounding the APOE gene suggesting that the actual effect is likely to be determined by these loci’s haplotype structure.
Supported by grants from the National Alzheimer’s Association (IIRG-03-4750), NIA (R21 AG24486-01), Veterans Affairs Biomedical Laboratory Research Development Merit Review (ID1127558), Ruth L Kirschstein National Research Service Award (NR-SA) (5T32AG000258-07), NINDS (R01 NS48595), VISN20 MIRECC, NIA (AG05136), NIH P01HL30086, Alzheimer’s Center Grant (AG08017), National Alzheimer’s Coordinating Center, Friends of Alzheimer’s Research, Alzheimer’s Association of Western and Central Washington, an anonymous foundation, and the Department of Veterans Affairs. Dr. Kaye is on the advisory board of Myriad Pharmaceuticals and receives grant support from the Intel Corporation. Dr. Farlow is a consultant for Abbott Labs, Accera Inc., Adamas Pharm, Cephalon Inc., Comentis, Eisai, GlaxoSmithKline, Medivation, Inc., Memory Pharm, Merck and Co., Neurochem, Novartis, OctaPharma, Sanofi-Aventis Groupe, Shering Plough, Takada Pharmaceuticals, Talecris Biotherapeutics, and Worldwide Clinical Trials. Additionally, Dr. Farlow receives grant support from Ono Pharmanet, Eli Lilly and Co., Wyeth, and Elan Pharmaceuticals. Dr. Quinn is a member of the speaker’s bureau of Pfizer and Forest.
Communicated by Thomas Montine