Independent associations of TOMM40 and APOE variants with body mass index

Abstract The TOMM40‐APOE variants are known for their strong, antagonistic associations with Alzheimer's disease and body weight. While a stronger role of the APOE than TOMM40 variants in Alzheimer's disease was suggested, comparative contribution of the TOMM40‐APOE variants in the regulation of body weight remains elusive. We examined additive effects of rs2075650 and rs157580 TOMM40 variants and rs429358 and rs7412 APOE variants coding the ε2/ε3/ε4 polymorphism on body mass index (BMI) in age‐aggregated and age‐stratified cohort‐specific and cohort‐pooled analysis of 27,863 Caucasians aged 20–100 years from seven longitudinal studies. Minor alleles of rs2075650, rs429358, and rs7412 were individually associated with BMI (β = −1.29, p = 3.97 × 10−9; β = −1.38, p = 2.78 × 10−10; and β = 0.58, p = 3.04 × 10−2, respectively). Conditional analysis with rs2075650 and rs429358 identified independent BMI‐lowering associations for minor alleles (β = −0.63, p = 3.99 × 10−2 and β = −0.94, p = 2.17 × 10−3, respectively). Polygenic mega‐analysis identified additive effects of the rs2075650 and rs429358 heterozygotes (β = −1.68, p = 3.00 × 10−9), and the strongest BMI‐lowering association for the rs2075650 heterozygous and rs429358 minor allele homozygous carriers (β = −4.11, p = 2.78 × 10−3). Conditional analysis with four polymorphisms identified independent BMI‐lowering (rs2075650, rs157580, and rs429358) and BMI‐increasing (rs7412) associations of heterozygous genotypes with BMI. Age‐stratified conditional analysis revealed well‐powered support for a differential and independent association of the rs429358 heterozygote with BMI in younger and older individuals, β = 0.58, 95% confidence interval (CI) = −1.18, 2.35, p = 5.18 × 10−1 for 3,068 individuals aged ≤30 years and β = −4.28, CI = −5.65, −2.92, p = 7.71 × 10−10 for 6,052 individuals aged >80 years. TOMM40 and APOE variants are independently and additively associated with BMI. The APOE ε4‐coding rs429358 polymorphism is associated with BMI in older individuals but not in younger individuals.


| INTRODUCTION
Studies report that deviation from the normal body weight is associated with cognitive decline and development of dementia and Alzheimer's disease (AD) (Emmerzaal, Kiliaan, & Gustafson, 2015).
Despite these results are challenged in Fitzpatrick et al. (2009), they suggest that there might be common underlying biological mechanisms involved in regulation of body weight and cognitive function and/or AD pathology (Hinney et al., 2014). The relationship between body weight and cognitive function is complex as it may vary with age from a positive relationship between mid-life obesity and AD to a negative relationship in late life (Emmerzaal et al., 2015).
High throughput genotyping of large human samples provides an opportunity for designing well-powered studies to examine whether the same genetic variants can be associated with body weight and AD. The 19q13.3 genomic region harboring the APOE (apolipoprotein E) and TOMM40 (translocase of outer mitochondrial membrane 40 homolog) genes represents an opportunity for such an analysis as it harbors pleiotropic variants conferring the strongest, well-documented risk of AD and robust associations with BMI (body mass index, kg/m 2 ) (Guo et al., 2013;Roses et al., 2010). Indeed, this locus harbors common APOE ε2/ε3/ε4 polymorphism, coded by rs429358 and rs7412 SNPs (single nucleotide polymorphisms). The APOE ε4 allele has been most consistently associated with AD in late life in various populations (Raichlen & Alexander, 2014). Genome-wide association studies (GWAS) also report an AD-increasing effect of the rs2075650 (TOMM40) minor allele and AD-decreasing effect of the rs157580 (TOMM40) minor allele (Bao, Wang, & Mao, 2016;Harold et al., 2009). The association of rs2075650 with AD is often attributed (Yu et al., 2007) to linkage disequilibrium (LD) of this SNP with rs429358, which encodes the APOE ε4 allele.
Large-scale meta-analysis of populations of European ancestry reported a genome-wide significant association of rs2075650 with BMI, with the AD-risk-increasing (minor) allele associated with smaller BMI (Guo et al., 2013). Whether the effect of this SNP should be attributed to TOMM40 or APOE and, particularly, to the ε4 allele, has not yet been addressed in large samples (Guo et al., 2013). Thus, we performed cohort-specific and cohort-pooled analysis of 27,863 participants from seven independent longitudinal studies to comprehensively examine the associations of rs2075650 and rs157580 TOMM40 variants and rs429358 and rs7412 variants coding the APOE ε2/ε3/ε4 polymorphism with BMI in a Caucasian ancestry population sample as well as in subpopulations of younger and older individuals.

| RESULTS
Data on 27,863 participants were obtained from seven longitudinal cohort studies (ARIC, CHS, CARDIA, MESA, HRS, LLFS, and FHS) (see Section 4). Baseline measurements of BMI, the number of longitudinal measurements of BMI used in the analysis, and basic demographic information for the genotyped participants in each cohort and the pooled sample are presented in Supporting Information Table S1). Frequency distributions of genotypes for rs2075650, rs157580, rs429358, rs7412, and APOE ε2/ε3/ε4 (coded by rs429358 and rs7412, Supporting Information Table S2) polymorphisms are presented in Supporting Information Table S3. All polymorphisms were in Hardy-Weinberg equilibrium.

| LD structure
The largest LD (r 2 = 49%) was observed between rs429358 and rs2075650 ( Figure 1) and was due to clustering of minor alleles of these SNPs rather than their minor and major alleles (Supporting Information Tables S4A and S4G). In contrast, LD between rs157580 and rs429358 was more modest (r 2 = 8%) and due to depleted clustering of minor alleles of these SNPs and preferable clustering of their minor and major alleles (Supporting Information Tables S4B and   S4H). For the other SNPs pairs, LD measured by r 2 was small.

| Univariate associations
In univariate analyses of single polymorphisms in the models, the rs2075650 minor allele was associated with decreased BMI in each cohort that resulted in a genome-wide significant association in the pooled sample of all cohorts (β = −1.29, p = 3.97 × 10 −9 ; Figure 2).
F I G U R E 1 LD pattern of four SNPs in the TOMM40-APOE locus. LD (%), r 2 (a) and D' (b), is shown in the pooled sample of all cohorts

APOE polymorphisms with BMI
Despite moderate LD between rs2075650 and rs429358 (Figure 1), bivariate mega-analysis with these SNPs identified their independent associations with BMI, although their effects became smaller ( Model 5) resembled the results of the bivariate analysis. Association of rs2075650 with BMI, independent of the ε4 allele, was confirmed F I G U R E 2 Genetic associations with BMI in each cohort and in the pooled sample of all cohorts (all). Associations of rs2075650 and rs157580 polymorphisms are from additive genetic model with minor allele as an effect allele. Statistical models for APOE were fitted considering the effect of the ε2 and ε4 alleles as compared with the ε3ε3 reference genotype. SE denotes standard error. Horizontal bars indicate 95% confidence intervals in the mega-analysis of carriers of the ε3ε3 genotype only (β = −1.01, p = 2.65 × 10 −2 ; Supporting Information Table S5).
Rs157580 did not attenuate the association of rs2075650, rs429358, rs7412, or ε2/ε3/ε4 polymorphisms with BMI (Supporting Information Table S6). The effect of rs157580 remained non-significant but its direction changed aligning with the effect of the ε4 allele in all models except that with rs7412 (Supporting Information Conditional mega-analysis, with rs2075650, rs157580, rs429358, and rs7412 in the model, showed that all heterozygotes were independently associated with BMI ( Table 2). Heterozygotes of rs2075650, rs429358, and rs157580 showed BMI-lowering associations, whereas the rs7412 heterozygote (coding the APOE ε2ε3 and ε2ε4 genotypes) showed BMI-increasing association. Conditional mega-analysis with rs2075650, rs157580, and APOE genotypes did not show significant effects for either the ε2ε3 (β = 0.53, p = 9.25 × 10 −2 ) or ε2ε4 (β = −0.11, p = 8.86 × 10 −1 ) genotypes alone (Supporting Information   Table S7) implying that examination of rs429358 and rs7412 SNPs rather than the APOE genotypes can help in separating the effects of the ε2 and ε4 alleles. The rs429358 minor allele homozygote (coding the APOE ε4ε4 genotype) was significantly associated with smaller BMI, whereas the BMI-lowering association for the minor allele homozygotes of the other SNPs did not attain significance (Table 2).

| Independent associations of rs2075650, rs157580, and APOE variants with BMI in younger and older individuals
Given uncertainty in defining "younger" and "older" people, we used varying cutoffs ranging from reproductive age (30 years) to the oldest-old age (90 years) in 5-year increments. We selected 30 and 90 years as global cutoffs because the analysis in groups of even younger and older ages was not sufficiently powered. In conditional analysis, with rs2075650, rs157580, rs429358, and rs7412 in the model, there were no significant differences between the effect estimates in younger and older carriers of the rs2075650 heterozygote at any given cutoff, as evidenced by well-overlapped confidence intervals (CIs) (Figure 4a). Nevertheless, there was a trend for a decreased magnitude of the effect size with age from β = −0.84, p = 1.39 × 10 −2 for individuals older than 30 years to β = −0.28, p = 8.55 × 10 −1 for individuals older than 90 years. The effects were virtually age independent for carriers of the rs157580 heterozygote (Supporting Information Table S10).
In contrast, we observed significant differences between the associations of the rs429358 heterozygote with BMI in younger and  Table S10). We did not observe significant differences between the effects in younger and older carriers of either the rs7412 genotypes (Supporting Information Table S10). The results were qualitatively the same in the models with the APOE genotypes and alleles.

| DISCUSSION
This study presents the results of the largest analysis so far of the associations of TOMM40 (rs2075650 and rs157580) and APOE (rs429358 and rs7412 coding the ε2/ε3/ε4 polymorphism) variants with BMI in seven cohorts of participants ranging in age from early reproductive age (about 20 years) to centenarians. These analyses leveraged longitudinal information on BMI assessed at different ages and addressed the impact of TOMM40 and APOE variants in the age-aggregated and age-stratified samples of older and younger individuals defined using varying cutoffs ranging from 30 years (i.e., people who were younger and older than 30 years at BMI assessment) to 90 years.
Given modest LD between rs2075650 and rs429358 (r 2 = 49% in our sample), the effect of rs2075650 is often considered as a proxy for that of the APOE ε4 allele. For example, studies of longevity indicate that strong effect of rs2075650 is likely due to the effect of the APOE isoforms (Murabito, Yuan, & Lunetta, 2012).
Studies of AD also indicate that TOMM40 variants are unlikely to have a major effect on AD (Yu et al., 2007), although the role of TOMM40-APOE haplotypes in AD is also acknowledged (Jazwinski et al., 2010;Lescai et al., 2011;Roses et al., 2010). The role of TOMM40 and APOE variants in body fat remains, however, unclear (Guo et al., 2013)  alleles of these SNPs, although it was smaller than that for compound genotypes. Accordingly, the analysis of compound variants was more useful in this study than that of polygenic score, constructed as sum of SNP-unspecified minor alleles, because it identified specific genetic profile of people with strong predisposition to lower BMI.
Previous studies reported that TOMM40-APOE variants other than rs2075650 may be involved in regulation of body fat (Lu et al., 2016 et al., 2006), the rs7412 heterozygote was associated with higher BMI in the current analysis.
TOMM40 encodes a protein localized in the outer membrane of the mitochondria, which is a part of a complex for mitochondrial protein import (Endo & Yamano, 2010). Mitochondria are a major player in ATP production, maintaining energy balance, and disposal of reactive oxygen species. Excessive energy can disrupt mitochondrial function that affects lipid and glucose metabolism (Bournat & Brown, 2010). Disruption of mitochondrial transport, followed by mitochondrial dysfunction, might be hypothesized as a plausible mechanism for TOMM40 in regulation of adipocyte function and overall body fatness (Gómez-Serrano et al., 2017). Studies suggest that APOE, a key protein in lipid metabolism, plays a role in adipocyte function. APOE has been implicated in the development of diet-induced obesity (Elosua et al., 2003;Feitosa et al., 2006;Kypreos et al., 2009 2008;Huebbe et al., 2015;Kuhel et al., 2013). The mechanism of influence of APOE on body fat traits might be through modulation on triglyceride content of adipocytes (Kypreos et al., 2009;Li & Liu, 2014).
Our age-stratified conditional analysis with rs2075650, rs157580, rs429358, and rs7412 in the model revealed well-powered support for a differential association of the APOE ε4-coding allele with decreased obesity in less-fitted children (Ellis et al., 2011) and non-significant trend in mid-aged adults (Volcik et al., 2006). Our previous study identified lower BMI in older ε4 allele carriers compared with the non-carriers in the FHS original cohort at ages >65 years . The differential role of the APOE ε4 F I G U R E 4 Independent associations of genetic variants with BMI in younger and older individuals. Cutoff shown on the x-axis defines younger (red dots) and older (blue diamonds) individuals. Effect sizes β and 95% confidence intervals for (a) rs2075650_Ag and (b) rs429358_Tc heterozygotes in conditional analysis with rs2075650, rs157580, rs429358, and rs7412 in the model. Upper-(lower-) case letter denotes major (minor) allele. Major allele homozygote was the reference. Numerical estimates are given in Supporting Information Table S10 allele in BMI in younger and older individuals is consistent with differential role of the ε4 allele in total cholesterol with age  and with change of frequency of this allele with birth years (Nygaard et al., 2014). This is also in line with age-specific associations of the ε4 allele with survival (Jacobsen et al., 2010;Tan et al., 2013), although this effect can be sex-specific (Joshi et al., 2016;Kulminski et al., 2014). No significant age-dependent associations of the TOMM40 SNPs and the APOE ε2 allele with BMI, that is difference in younger and older individuals, were identified.
Although, there was a non-significant trend on decreased magnitude of the effect for the rs2075650 older heterozygotes.
Our results on the differential role of the APOE ε4 allele in BMI in younger and older individuals contribute to better understanding of genetic architecture underlying differential role of elevated midand late-life BMI in risks of AD and dementia (Emmerzaal et al., 2015). They suggest that the ε4 allele may provide late-life-specific contribution to increased risk of AD through the mechanism of regulation of body fat, as discussed above, that is consistent with increasing risk of AD with age in general population (Akushevich, Kravchenko, Ukraintseva, Arbeev, & Yashin, 2012) and higher risk for underweight subjects to develop AD in old age (Joo et al., 2018).
These results warrant comprehensive longitudinal analysis of dynamic connections between body fat metabolism and progression to AD over the individuals' life course. Such analysis is necessary to examine genetic predisposition to complex interplay of changes in the body composition with age, aging process, and time trends in obesity and incidence of AD.
The results for the APOE ε2/ε3/ε4 polymorphism resembled those for rs429358 and rs7412 SNPs. Examination of rs7412 was, however, more beneficial in this study because of the ability to better handle the associations for the ε2 and ε4 alleles.

| Genotypes
Rs2075650 and rs157580 SNPs from TOMM40 gene were available

| Analysis
Measurements of BMI were natural-log-transformed to offset potential bias due to skewness of their frequency distributions and multiplied by 100 for better resolution. BMI was measured multiple times during follow-up of the same individuals in most cohorts (Supporting Information Table S1). We used all available measurements. Information on longitudinal measurements has multiple advantages including potential gain in statistical power in the analyses (Shi, Rice, Gu, & Rao, 2009). The analyses were performed in each cohort separately and the mega sample of all cohorts combined. To correct for repeated-measurements (all cohorts) and familial (FHS and LLFS) correlations, we used the linear mixed effects model (lme4 package in R). We evaluated the associations for SNPs given the measurements of BMI for individuals of a given age at each examination with available measurements. Because of longitudinal follow-up, the same individuals were used in age-stratified analyses in "old" and/or "young" group depending on his/her age at BMI assessment. The models were adjusted for sex and birth cohorts. Potential inter-study differences were handled by using mixed effects for studies. The models were not adjusted for principal components to control for population stratification in these Caucasian populations because of their trivial effects in these analyses.

ACKNOWLEDGMENTS
This work was supported by the National Institute on Aging (NIA) (grant numbers U01 AG023712, P01 AG043352, R01 AG047310), NIH, USA. The funding source did not participate in the writing of the manuscript or the decision to submit the manuscript for publication.
See also Supplemental Acknowledgment Text. The authors thank Arseniy P. Yashkin for help in preparation of phenotypes in HRS.

CONF LICT OF I NTEREST
The authors declare that they have no conflict of interest.