APOE ε4-TOMM40 ‘523 haplotypes and the risk of Alzheimer’s disease in older Caucasian and African Americans

Patterns of linkage between the ε4 allele of Apolipoprotein E (APOE) and ‘523 poly-T alleles in the adjacent gene, TOMM40, differ between Caucasian and African Americans. The extent to which this difference affects the risk of Alzheimer’s disease (AD) is unclear. We compared the APOE ε4-TOMM40 ‘523 haplotypes between older Caucasian and African Americans, and examined their relationship with AD dementia. Data came from three community based cohort studies of diverse participants. APOE genotypes were determined by polymorphisms of rs429358 and rs7412. TOMM40 ‘523 genotypes were defined by the poly-T repeat length of rs10524523 (short [‘523-S]: poly-T ≤ 19, long [‘523-L]: 20 ≤ poly-T ≤ 29, and very long [‘523-VL]: poly-T ≥ 30). Cox proportional hazards models examined the effect of haplotype variation on the risk of incident AD dementia. A total of 1,848 Caucasian and 540 African American individuals were included in the study. In Caucasians, nearly none (0.8%) of the non-ε4 carriers and almost all (94.2%) of the ε4 carriers had ‘523-L. The classification was highly concordant. Each ε4 allele doubled the risk for AD dementia and the dose effect was evident. Almost identical effect size and effect pattern were observed for TOMM40 ‘523-L. In African Americans, nearly none (1.1%) of the non-ε4 carriers had ‘523-L, but only 47.8% of the ε4 carriers had ‘523-L. The concordance was weaker compared with Caucasians. The effect patterns on incident AD dementia differed distinctively between ε4 and ‘523-L carriers. Further, both genotypic and allelic data support that among African Americans the ε4-‘523-L haplotype had stronger effect on risk of AD dementia than other ε4-‘523 haplotypes.

Introduction Apolipoprotein E (APOE) ε4 is by far the most replicated risk allele for late onset Alzheimer's disease (AD). Compared with Caucasians, the allele frequency of ε4 is higher among African Americans but its effect on the disease susceptibility appears weaker [1][2][3]. This inconsistency may be attributable to race specific variation of haplotypes across APOE and neighboring genes. For instance, it has been reported that among Caribbean Hispanic ε4 carriers, subjects with the HpaI+ variant in the downstream APOC1 gene are more likely to have AD, but the same association is not observed among African Americans [4]. Recent data suggest that the haplotypes between APOE and a poly-T variant in another adjacent gene, Translocase of Outer Mitochondria Membrane (TOMM40) '523, together provide better precision in estimating the age of AD onset [5]. The '523 variant is located in the intron of TOMM40. Depending on the poly-T repeat length, individual '523 alleles can be grouped into 3 classes: Short ('523-S: Poly-T repeat 19), Long ('523-L: 20 poly-T repeat 29), and Very Long ('523-VL: poly-T repeat ! 30) [6]. Genetic studies show that varying '523 poly-T repeat length differentiates AD susceptibility among Caucasian ε3/ε3 carriers [7,8]. Notably, the APOE and TOMM40 '523 haplotypes differ between Caucasian and African Americans [9]. In Caucasians, ε4 is almost perfectly linked with the '523-L allele. By contrast, in African Americans, ε4 is commonly linked to '523-S in addition to the '523-L allele. The extent to which such difference affects AD risk in African Americans has not been reported.
In this study, using data from a large sample of prospective longitudinal studies of aging, we compared the distributions of TOMM40 '523 in linkage with APOE ε4 between older Caucasian and African Americans, and examined their relationship with incident AD dementia. Since APOE ε4 and TOMM40 '523-L are in strong linkage disequilibrium (LD) among Caucasians, it is expected that the two alleles will demonstrate a similar effect on AD dementia in this population. By contrast, the heterogeneity of the ε4-'523 haplotypes among African Americans allows us to investigate the role of the ε4-'523-L haplotype in relation to AD susceptibility.

Study participants
Participants came from three community-based longitudinal cohort studies of diverse participants, the Religious Orders Study (ROS) [10], the Rush Memory and Aging Project (MAP) [11], and the Minority Aging Research Study (MARS) [12]. The studies were approved by the Institutional Review Board of Rush University Medical Center. Participants were enrolled without known dementia and each participant signed an informed consent and agreed to annual clinical evaluations. The ROS and MAP participants were predominantly Caucasian Americans and the MARS participants were all African Americans. Importantly, all three studies are conducted by the same team of investigators and share a large common core of testing batteries and uniform structured clinical evaluations. This makes it possible for a combined analysis.
At the time of this study, 1,948 Caucasian participants free of AD dementia at baseline had completed at least 1 follow-up clinical evaluation. Genotyping data were available on 1,886 (96.8%) of the participants. Considering the conflicting effects between the ε2 and ε4 alleles on AD dementia, ε2/4 carriers (N = 38) were excluded. As the result, statistical analyses were conducted in the remaining 1,848 Caucasian participants. Separately, genotyping data were available on 568 (98.8%) of the 575 African American participants who were dementia free at baseline and had completed at least 1 follow-up evaluation. We excluded 28 ε2/4 carriers and focused the analyses on the remaining 540 African Americans. For the African American participants, the mean baseline age was 73.4 years (SD: 6.6, range: 58.9-97.6). 77.6% were females and the mean education was 14.8 years (SD: 3.4, range: 0-30). The average length of follow-up was 5.9 years (SD = 3.6, range: 1-16).
APOE ε4 and TOMM40 '523 genotyping DNA was extracted from peripheral blood mononuclear cells (PBMC) or post-mortem brain tissue. Genotyping was performed by Polymorphic DNA Technologies (Alameda, California), blinded to all clinical data. The APOE genotypes were determined by the two polymorphisms of rs429358 (codon 112) and rs7412 (codon 158) at exon 4 of the APOE gene. The TOMM40 '523 genotypes were determined by the polymorphism rs10524523 at intron 6 of the TOMM40 gene (chr19:44,899,792-44,899,826, human genome reference assembly GRCh38/hg38). Based on the the poly-T repeat length, each '523 allele was categorized into Short ('523-S), Long ('523-L) and Very Long ("523-VL), as previously defined [6]. In this study, we primarily focused on the effect of ε4 in linkage with TOMM40 '523.
Distinct from the exclusive linkage between ε4 and '523-L in Caucasian Americans, other '523 alleles, '523-S in particular, have been reported in linkage with ε4 among African Americans. We therefore obtained strand specific (phased) haplotype data in a subset of African American ε3/4 heterozygotes in an effort to further explore how the strand specific ε4-'523 haplotype variations differ in relation to AD dementia.

Clinical diagnosis of AD dementia
Participants underwent a detailed cognitive testing and a uniform structured clinical evaluation each year. Cognitive assessment results were reviewed by a neuropsychologist for signs of impairment in various domains. Participants were examined by a clinician. After reviewing all available data, an annual diagnosis was made by the clinician. Diagnosis of AD follows the recommendation of the joint working group of the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association [13] and the decision rules for implementing these criteria have been described previously [14]. Briefly, the AD diagnosis requires a history of cognitive decline and evidence of impairment in multiple cognitive domains including memory.

Statistical analysis
Student t and Chi-square tests described demographic and genetic differences between Caucasian and African Americans. Primary analyses were stratified by race. Cohen's κ coefficients [15] evaluated the concordance between APOE ε4 and TOMM40 '523-L genotypes. Cox proportional hazards models examined the associations of genetic variants with incident AD dementia. In these models, the outcome variable was time in years to incident AD dementia (event time). Time was right-censored at death or last clinical evaluation for participants who were never diagnosed with AD dementia. We first examined the ε4 effect by fitting a model including terms for ε3/4 heterozygous and ε4/4 homozygous. The dose effect was assessed by estimating the difference between the effect of ε4/4 and twice the effect of ε3/4. Next, we repeated the model for '523-L in lieu of ε4. Both models were adjusted for baseline age, sex, and education. Finally, by leveraging phased haplotype data from 83 African American ε3/4 heterozygotes, we applied Cox proportional hazards model to evaluate the associations of strand specific ε4-'523 haplotype variations with incident AD dementia.
All the analyses were conducted using SAS/STAT programs (SAS Institute, Cary, NC). Statistical significance was determined a priori at the nominal level of α = 0.05.

Characteristics of study participants
Characteristics of study participants were summarized in Table 1. Compared with African Americans, Caucasian Americans in this study were older and had more years of education (both ps <0.001). Percent female participants were higher among African Americans (p = 0.003). On average, Caucasian Americans were followed 2 years longer than African Americans (p<0.001). During the follow-ups, 27.8% of Caucasian Americans and 10.4% of African Americans were diagnosed with AD dementia.

APOE ε4 and TOMM40 '523-L with AD dementia in Caucasian Americans
Consistent with the linkage between ε4 and '523-L, nearly none (0.8%) of the Caucasian non-ε4 carriers had '523-L, whereas 94.2% of the Caucasian ε4 carriers had '523-L ( Table 2). The  Table 3). The difference between the estimated effect size of ε4/4 and twice the estimated effect size of ε3/4 was not significant (p = 0.632), supporting a dose effect of ε4 on AD susceptibility. Almost identical effect size and effect pattern were observed for TOMM40 '523-L in Caucasian Americans (Fig 1). The hazard ratios for the 523-L heterozygotes and homozygotes were 1.89 and 3.80 respectively (both ps<0.001). The dose effect of the '523-L allele was also evident, and the estimated effect of the 523-L homozygosity was equivalent to twice the effect of the 523-L heterozygosity (p = 0.859).
The Cox proportional hazards model shows that while the ε3/4 heterozygotes among African Americans had higher risk for AD dementia, the estimated hazard ratio was not significant (p = 0.195) ( Table 4). By contrast, the ε4/4 homozygotes increased the AD risk by over six fold (HR = 6.32, 95% CI = 2.76-14.45, p<0.001). The lack of dosage effect for APOE ε4 in African Americans may be due to small sample size, and the test that assessed the difference between the effect of ε4/4 and twice the effect of ε3/4 was inconclusive. Notably, The effect pattern of TOMM40 '523-L on AD dementia differed distinctively from that of APOE ε4 (Fig 2). The '523-L heterozygosity doubled the risk for incident AD dementia (HR = 2.21, 95% CI = 1.20-4.08, p = 0.011), but we did not detect a significant association for '523-L homozygosity (p = 0.650).
Since almost all of the non-ε4 African Americans did not carry '523-L allele and approximately half of the ε4 carriers had '523-L allele, we further split the ε4 carriers into those with and without the '523-L allele. 8.4% of the 358 African American non-ε4 carriers were diagnosed with AD dementia, and this percentage increased to 10.8% and 17.7% respectively for To assess the robustness of our findings considering differences between Caucasian and African Americans in this study, we performed secondary analyses using a subset of individuals that were 1-to-1 matched by age, sex, education, length of follow-up as well as vital status. A total of 464 Caucasian and 464 African Americans were included. Notably, while statistical difference was no longer observed in any of the matching variables (all ps>0.05), results were consistent with those using the full dataset. In Caucasians, there was a high concordance between ε4 and '523-L (Cohen's κ = 0.94). Nearly none (0.6%) of the non-ε4 carriers and almost all (92.4%) of the ε4 carriers had '523-L. The effect size and effect pattern on incident AD dementia were similar between ε4 and '523-L (Table A in S1 File). In African Americans, we observed a weak concordance between ε4 and '523-L (Cohen's κ = 0.55). Nearly none (1.3%) of the non-ε4 carriers had '523-L, but only 49% of the ε4 carriers had '523-L. The results in relation to incident AD dementia for African Americans were essentially the same as reported in the full dataset, such that the effect patterns on incident AD dementia differed between ε4 and '523-L carriers ( Table B in S1 File). The ε3/4 heterozygosity had higher risk for AD dementia, but was not significant. By contrast, the ε4/4 homozygosity increased the AD risk by over 7 fold, and was highly significant. The '523-L heterozygosity doubled the risk for incident AD dementia, but we did not detect a significant association for '523-L homozygosity. Further, ε4 coupled with '523-L had a stronger effect on the risk of AD dementia than ε4 in the absence of '523-L.

Discussion
In this study, we examined the haplotypes between APOE ε4 and the neighboring TOMM40 '523 variant in a large sample of community based older Caucasian and African Americans, and we interrogated their associations with incident AD dementia. Several observations were made. We confirmed the linkage between ε4 and '523-L in Caucasians Americans. This linkage is highly concordant, such that almost all the ε4 carriers had presence of the '523-L allele and cumulative hazards of incident AD dementia for representative female African Americans with mean age and education. For panel A, the blue solid curve represents cumulative hazard for non-ε4 carriers, the black dash curve for ε4 heterozygotes, and the red solid curve for ε4 homozygotes. For panel B, the blue solid curve represents cumulative hazards for non-'523-L carriers, the black solid curve for '523-L heterozygotes, and the red dash curve for '523-L homozygotes.
https://doi.org/10.1371/journal.pone.0180356.g002 almost all the non-ε4 carriers were absent of the '523-L allele. When modeled for the associations with incident AD dementia, ε4 and '523-L share very similar effect size and effect pattern as expected. In both cases, every single allele doubles the risk for AD dementia and the dose effect is evident. The two '523-L heterozygosities ('523 S/L and '523 L/VL) show the same strength in association with incident AD dementia. This strong linkage between ε4 and '523-L among Caucasian Americans probably accounts for the lack of association of the poly-T repeat length with age of AD onset in a previous report [16]. Some evidence suggests that the three TOMM40 '523 genotypes (S/S, S/VL and VL/VL) that are exclusive to the non-ε4 carriers may further differentiate age of onset for late onset AD [17], but the results have not been conclusive [7,18]. It is noted that the effect size of APOE ε4 on the risk of AD dementia in Caucasians can differ across studies due to a variety of reasons ranging from population or sample differences to phenotypic variation. A previous study reported an odds ratio of 2.7 for ε3/4 and 12.5 for ε4/4, relative to ε3/3 [2]. The study differs from ours in several important ways. First, the prior study is a meta-analysis that combined data from 40 participating centers with different study designs. As the result, the samples are more heterogeneous. Indeed, the authors reported that study specific odds ratios of AD for ε4 carriers versus non-carriers varied, and the associations were weaker for population/community based studies. Second, the prior study was based on cross-sectional data where prevalent AD cases are mixed with incident cases. By contrast, our data came from community based cohorts and we focused on incident AD. All individuals included in the analyses were free of dementia at baseline and followed longitudinally. The effect size of ε4 reported in our study is in general consistent with those from similar population or community based studies on incident AD [19][20][21]. Finally, our subjects are older and the effect of ε4 tends to attenuate in old age.
We observed that the frequencies of TOMM40 '523 genotypes differ distinctively between African and Caucasian Americans. There was an enrichment of '523-S allele in African Americans. In particular, the percent '523-S/S carriers in African Americans was more than doubled compared with Caucasian Americans. This result is consistent with an earlier report that the '523-S allele is more prevalent in African Americans than whites or Hispanics [22].
We also observed that the strong linkage between ε4 and '523-L haplotype was absent in African Americans. While nearly none of the non-ε4 carriers had the '523-L allele, only less than half of the ε4 carriers had '523-L. Consequently we could examine the effect patterns of haplotype variations on incident AD dementia. In fact, data at both genotypic and allelic level suggest that ε4-'523-L haplotype among African Americans has a stronger effect on the susceptibility of AD dementia than other ε4-'523 haplotypes. This finding helps further elucidate the relationship between APOE ε4 and AD dementia in African Americans.
The reports on ε4 and AD in African Americans have been inconsistent [1,[23][24][25][26]. Multiple studies show that the ε4 effect on AD tends to be weaker in African Americans than Caucasian Americans [2,27]. In addition, the strength of the ε4 allele is even weaker in African Yoruban than African Americans [28][29][30]. These findings raise the possibility that the ε4 effect in African Americans is likely affected by the degree of admixture with Caucasian or other populations. One hypothesis is that admixture leads to haplotype variations surrounding the APOE locus. The cis ε4 haplotype difference due to population stratification is not new. One earlier study reported an APOE ε4 and APOC1 HpaI+ haplotype that differs in frequency between ADs and controls, and such difference is only observed in Caribbean Hispanics, but not in African Americans [6]. Variation in TOMM40 '523-APOE haplotypes among African Americans may well be just another example. Indeed, while the ε4-TOMM40 '523-L haplotype predominates in Caucasians, two additional haplotypes have been reported in Yorubans. The linkage between ε4 and multiple '523 alleles in African Americans represents a genetic crossover between Caucasian and Western African populations [9]. Data from this study support the theory that, of the three ε4-TOMM40 '523 haplotypes, the two originated from the Western African population are less potent than the ε4-TOMM40 '523-L haplotype in affecting the susceptibility of AD. Since genetic risk factors in admixed populations including African Americans are highly dependent on ancestral background, future work is warranted to identify local ancestry in the APOE-TOMM40 region and to investigate its implication in the risk of AD.
Extensive evidence suggests that the APOE gene is involved in amyloid deposition, a neuropathologic hallmark of AD, via amyloid-β clearance [31,32]. The APOE ε2, ε3, and ε4 alleles encode three protein isoforms (apoE2, apoE3 and apoE4) which show differential response for amyloid-β kinetics. In mouse models, apoE4 increases amyloid-β deposition relative to apoE2 or apoE3 [33,34]. In humans, APOE ε4 carriers have more severe burden of amyloid plaques [35,36], and the effect of ε4 on cognition and AD dementia is largely mediated by amyloid and downstream neurofibrillary tangles [37,38]. The biochemical mechanism by which the TOMM40 '523 variant affects AD pathophysiology is an area of active research [39]. Structural DNA variations, especially those in intronic or intergenic regions like TOMM40 '523, most likely exert their effects by altering gene transcription efficiency, the timing of transcription, transcript stability, transcript splicing, or possibly by changing patterns of epigenomic modification [40][41][42][43][44][45][46]. There is support for the idea that short structural variants may be involved in human diseases [47][48][49]. It has been demonstrated that TOMM40 '523 affects the expression levels of APOE and TOMM40 mRNAs in the temporal and occipital cortexes of late onset AD patients and normal controls [50], and the effect of the TOMM40 poly-T variation on transcription regulation was recapitulated in a cell-based luciferase reporting system [51]. Gene expression studies suggest that the number of protein import channels per mitochondrion may be regulated by TOMM40 poly-T variants. Bekris et al. described a complex transcriptional regulatory region for TOMM40 and APOE expression that extends throughout both genes and is influenced by multiple polymorphisms including the TOMM40 poly-T locus [52]. It has been suggested that a relatively modest change in mRNA expression may produce pathology that accumulate over time and is expressed clinically in later age.
To our knowledge this is the first study that investigated the effect of ε4-TOMM40 '523 haplotype variations on the risk of incident AD dementia among African Americans. Our data came from a diverse group of community based older persons who were free of AD at enrollment and followed longitudinally for incident events. The prospective nature of the design avoids potential bias due to prevalent AD. Uniform and structured decision rules for AD diagnosis were applied across all three studies. Limitations of this study are also noted. These are voluntary cohorts and participants are older and have higher education than the general population. The findings await replications from other studies.
Supporting information S1 File. Table A. APOE ε4 and TOMM40 '523-L with AD dementia in Caucasian Americans (matched sample) Table B. APOE ε4 and TOMM40 '523-L with AD dementia in African Americans (matched sample). (DOCX)