Varying Effects of APOE Alleles on Extreme Longevity in European Ethnicities

Abstract

APOE is a well-studied gene with multiple effects on aging and longevity. The gene has three alleles: e2, e3, and e4, whose frequencies vary by ethnicity. While the e2 is associated with healthy cognitive aging, the e4 allele is associated with Alzheimer’s disease and early mortality and therefore its prevalence among people with extreme longevity (EL) is low. Using the PopCluster algorithm, we identified several ethnically different clusters in which the effect of the e2 and e4 alleles on EL changed substantially. For example, PopCluster discovered a large group of 1,309 subjects enriched of Southern Italian genetic ancestry with weaker protective effect of e2 (odds ratio [OR] = 1.27, p = .14) and weaker damaging effect of e4 (OR = 0.82, p = .31) on the phenotype of EL compared to other European ethnicities. Further analysis of this cluster suggests that the odds for EL in carriers of the e4 allele with Southern Italian genetic ancestry differ depending on whether they live in the United States (OR = 0.29, p = .009) or Italy (OR = 1.21, p = .38). PopCluster also found clusters enriched of subjects with Danish ancestry with varying effect of e2 on EL. The country of residence (Denmark or United States) appears to change the odds for EL in the e2 carriers.

Apolipoprotein E (APOE) is a class of proteins involved in lipid metabolism with functions determined by alleles of the gene APOE. The gene has three alleles e2, e3, and e4 defined by combinations of genotypes of the single-nucleotide polymorphisms (SNPs) rs7412 and rs429358 (1,2). APOE is a well-studied gene with multiple effects on aging and longevity. The e4 allele is a well-established risk factor for late onset of Alzheimer’s disease (3–5). We and others have demonstrated that having an APOE e4 allele has a deleterious effect on longevity that decreases the odds to reach extreme human life span (1,6) independently of e2. The e3 allele is the “neutral allele” in many ethnicities, while e2 is the allele that promotes longevity and healthy aging, independently of e4 (1,6–8).

The frequency of the APOE alleles varies among human populations (9). For example, the most common e3 allele frequency varies from 54% in African Pygmies to 90% in Southern Italians and Sardinians. The frequency of the e4 allele varies from 5% in Sardinians to 41% in African Pygmies. It has been reported that the frequency of the e4 allele increases with latitude due to the natural selection to protect against low-cholesterol levels (10). Additionally, the e4 allele is associated with better resistance to adverse nonindustrialized environments, specifically to parasites and infections in children (11,12).

Studies have suggested that APOE e4 has different effects on the risk of Alzheimer’s disease in Europeans, African Americans, and Hispanics (5,13,14), while the role of ethnicity on the effect of APOE e2 on longevity and neuroprotection is unknown. To investigate the ethnic-specific effect of APOE e2 and e4 on extreme longevity (EL), we used our new algorithm PopCluster (15) to search for ethnically different clusters of Europeans in which the effect of APOE e2 and e4 on EL change.

Methods

Study Populations

We used genome-wide genotype data from a consortium of four studies of EL and healthy aging: the Southern Italian Centenarian Study (SICS) (16), the Longevity Gene Project (LGP) (17), the Long Life Family Study (LLFS) (18), and the New England Centenarian Study (NECS) (19) (Table 1). The SICS is a study of longevity that focused enrollment of long-lived individuals in the South of Italy (16). The LGP is a study of longevity that enrolled long-lived individuals who were of Ashkenazy Jewish descent, survived to at least age 95 years old, and were dementia free at the time of enrollment (20). Some siblings, offspring and spouses of offspring were also enrolled and additional unrelated population controls were selected based on lack of familial longevity. The LLFS is a family-based study of healthy aging and longevity that recruited approximately 550 families and 5,000 family members selected for familial longevity (18,21). Participants were enrolled at three American field centers (Boston, Pittsburgh and New York), and a European field center in Denmark. The NECS is a study of centenarians, some of the long-lived siblings, offspring, offspring spouses, and additional unrelated controls selected because their parents died before reaching the median age survival of their birth year cohort (19). The study recruits centenarians in worldwide.

Table 1.

Study Characteristics

Study	Cases (median age, range)	Controls
SICS	174 (100, 96–109)	540
LLFS	574 (100, 96–111)	2,561
LGP	308 (102, 96–113)	621
NECS	1,087 (103, 96–119)	3,103
Total	2,143	6,825

Note: SICS = Southern Italian Centenarian Study; LLFS = Long Life Family Study; LGP = Longevity Genes Project; NECS = New England Centenarian Study. Cases were defined as individuals who lived past the 1 percentile survival age from the 1900 birth year cohort based on U.S. Social Security Administration cohort life tables (26). Controls were defined either as individuals who died before reaching the threshold age, or as random subjects from the general population.

Genotype Data

Genome-wide genotype data for all studies were generated using Illumina SNP arrays (22) and imputed to the 1,000 genomes haplotypes Phase 1 using IMPUTE2 and standard protocol l (23). Imputation was preceded by prephasing with ShapeIT (24). More details on the datasets can be found in in (25). APOE alleles were inferred from SNPs rs7412 and rs429358 that were either genotyped using real time PCR (LLFS samples) or imputed using IMPUTE2 (SICS, LGP, and NECS samples; Supplementary Table S1). The imputation quality scores of the two SNPs when imputed were > 0.9 (6). Cases were defined as individuals who lived past the 1 percentile survival age from the 1900 birth year cohort based on U.S. Social Security Administration cohort life tables (26), that is, age 96 years and greater for males, and 100 years and greater for females. Controls were defined either as individuals who died before reaching the threshold age, or as random subjects from the general population. The combined datasets contain several European ethnicities and information about place of birth and mother/father tongue.

Statistical Analysis

We used PopCluster (15) to find ethnic-specific clusters of subjects with varying effect of APOE e2 and e4 on EL. PopCluster discovers subsets of individuals characterized by significant effects of a genetic variant. The algorithm uses hierarchical clustering of the first four principal components calculated from the genome-wide genotype data to generate the smallest clusters with at least 100 subjects and uses logistic regression to estimate adjusted OR for EL in each cluster. Then, a recursive procedure is used to test whether the ORs in connected clusters are significantly different using Z-tests. By design, the algorithm finds nonoverlapping clusters; thus, no cluster contains subjects from another cluster. We used logistic regression adjusted by sex and cluster-specific principal components calculated from the genome-wide genotype data to estimate cluster-specific associations between APOE alleles and EL. To adjust p values for multiple testing, we used p < .05/(number of clusters returned by the algorithm). Our evaluation showed that this correction maintains a family-wide error rate < 5% (15).

To evaluate the effect of APOE e2 and e4 on EL independently of each other, we conducted two analyses. In one analysis, we removed all carriers of e4, and used an additive genetic model with e3e3 coded as 0 (5,901 subjects), e2e3 as 1 (1,362 subjects), and e2e2 as 2 (56 subjects). Similarly, to evaluate the effect of e4 on EL, we removed all carriers of e2, and used an additive genetic model with e3e3 coded as 0 (5,901 subjects), e3e4 as 1 (1,497 subjects), and e4e4 as 2 (126 subjects). Total genotype counts are presented in Supplementary Table S2. All power analyses were done using the G*Power software (27).

Annotation of Clusters by Ethnicity

Ancestry information was available for 68% of the subjects, specifically: 3,717 NECS subjects had information on either their birth places or native languages of their grandparents, or both (28), all SICS subjects were assumed to be South Italians, all LGP subjects were assumed to be Ashkenazi Jews, and 735 subjects from the LLFS were recruited in Denmark and assumed to be of Danish descent. Additionally, when the data set is limited to Caucasians (as is the case here), subjects’ ethnicities can be inferred from their location on the scatter plots of first four genome-wide principal components (22,28–30). We labeled each cluster with ethnicity based on the following rules followed in order: (i) If more than 50% of subjects in a cluster are enriched of a certain ethnicity, the cluster was labeled by that ethnicity. (ii) If there is no known ethnicity with more than 50% subjects represented in a cluster, but the scatter plots of genome-wide principal components (Figures 1 and 2 and Supplementary Figures S1–S3) of a cluster are localized, the cluster was labeled based on the location on the scatter plots (31). (iii) If neither of two previous conditions hold, a cluster was labeled as mixed.

Figure 1.

Scatter plots of principal components PC1-PC2 and PC3-PC4 from genome-wide genotype data of all subjects in the study of extreme longevity. Grey (red in the online version) points are subjects in clusters with a significant effect of APOE e2: (A) cluster with 977 subjects enriched of Ashkenazi Jews; (B) cluster with 437 subjects enriched of Danish subjects.

Figure 2.

Scatter plots of principal components PC1-PC2 and PC3-PC4 from genome-wide genotype data of all subjects in the study of extreme longevity. Grey (red in the online version) points are subjects in clusters with a significant effect of APOE e4: (A) cluster with 1416 subjects enriched of British ancestry; (B) cluster with 559 subjects enriched of Danish ancestry; (C) cluster with 974 subjects with mixed ancestry; (D) cluster with 977 subjects enriched of Ashkenazi Jews; (E) cluster with 635 subjects enriched of Danish subjects.

Results

We analyzed the ethnic-specific association of EL with APOE alleles in 2,143 cases of EL, and 6,825 controls summarized in Table 1.

Ethnic-Specific Effect of APOE e2

PopCluster identified two clusters with a significant, positive effect of APOE e2 on EL after correction for multiple comparisons (p < .05/13 = .0038; Table 2), and an additional cluster with nominally significant effect of APOE e2 on EL (Supplementary Table S3). In all other clusters, the genetic association between APOE e2 and EL did not reach statistical significance although several clusters had sample size > 180 that is the minimum size required to have 80% power to detect an odds ratio (OR) of 2, assuming a level of significance of 0.0038. Two pairs of these 13 clusters (Ashkenazi Jews_1 and British_2, Danish_1 and British_2 from Supplementary Table S3) had significantly different effects of APOE e2 on EL using a false discovery rate (FDR) < 7% as level of significance to correct for multiple testing. Figure 1 displays the scatter plots of the first four genome-wide principal components for the two clusters in Table 2. In both clusters, carriers of e2 have increased odds for EL compared to carriers of e3e3: OR = 2.24, 95% confidence interval (CI): 1.35, 3.73 in the cluster enriched of Danish ancestry, and OR = 2.12, 95% CI: 1.44, 3.13 in the cluster enriched of Ashkenazi Jewish ancestry. The effects of e2 on EL in the two clusters were not significantly different (p = .87), although sample and effect sizes provided only 6% power to detect whether the difference between these two clusters was significant.

Table 2.

Significant Positive Associations Between APOE e2 and EL in Ethnic-Specific Clusters

Ethnicity Label	No. Subjects	(Enriched ethnicity) No. Enriched Ethnicity/No. Subjects with Labels	OR (e2 vs e3)	95% CI	p
Ashkenazi Jews_1 (*)	977	(Ashkenazi Jews) 411/646 ≈ 0.64	2.12	1.44, 3.13	.0002
Danish_1 (*)	437	(Danish) 319/344 ≈ 0.93	2.24	1.35, 3.73	.002

Note: Ethnicity label: This was inferred by either the enriched ethnicity of subjects (>50% in the cluster) (*) or based on the PCA plots for this cluster (**) (underscore and number mean that there are more than 1 distinct cluster that are labeled the same). No. subjects: total number of subjects in the cluster; (Enriched ethnicity) No. enriched ethnicity / No. subjects with labels: proportion of subjects with enriched ethnicity from the subjects with known information on their ancestry; OR (e2 vs e3): odds ratio for EL comparing carriers of one copy of APOE e2 to e3e3 carriers; 95% CI: 95% confidence interval. p: p value (association is significant if p < .05/13 = .004). Given sample and effect sizes, there was not enough power (6%) to detect whether the difference between these two clusters was significant. CI = Confidence interval; OR = Odds ratio.

Ethnic-Specific Effect of APOE e4

Similarly, PopCluster identified four ethnic-specific clusters with a nominally significant effect of APOE e4 on EL (Supplementary Table S4), and five clusters in which APOE e4 was significantly and negatively associated with EL after correcting for multiple comparisons (p < .05/12 = .004; Table 3). In the remaining three clusters (South Italians, Polish, British_4), the genetic association did not reach statistical significance and it is noticeable that in the cluster enriched of Southern Italians with N = 1,309, the effect of e4 on EL was substantially smaller than in other ethnic groups (OR = 0.82, p = .31; Supplementary Table S4). Two pairs of these 12 clusters (British_3 and South Italians, British_3 and British_4 from Supplementary Table S4) had significantly different effects of APOE e4 on EL using an FDR < 7% as level of significance to correct for multiple testing. Figure 2 displays the scatter plots of the first four genome-wide principal components for the five clusters in which the association between APOE e4 and EL is statistically significant, after correction for multiple testing. In all five clusters, the effect of APOE e4 is deleterious on longevity with worst effect in subjects of British ancestry (OR = 0.3, 95% CI: 0.21, 0.44) and slightly less severe effect in subjects with North Eastern Europeans ethnicities. None of the pairwise difference of genetic effects between the five clusters was statistically significant. However, when the effect of the cluster enriched of British ancestry was compared to the effect of the other four clusters combined, the difference was borderline significant (p = .06). Given sample and effect sizes, there was not enough power (5%–43%) to detect whether the difference between each pair of the five significant clusters was significant (27).

Table 3.

Significant Negative Associations Between APOE e4 and EL in Ethnic-Specific Clusters

Ethnicity Label	No. Subjects	(Enriched ethnicity) No. Enriched Ethnicity/No. Subjects with Labels	OR (e4 vs e3)	95% CI	p
British_3 (**)	1,416	(British) 101/995 ≈ 0.10	0.3	0.21, 0.44	4.42E−10
Danish_4 (**)	559	(Danish) 57/175 ≈ 0.33	0.44	0.26, 0.72	.001
Mixed_1 (**)	974	(Danish) 134/286 ≈ 0.47	0.49	0.32, 0.78	.002
Ashkenazi Jews_1 (*)	977	(Ashkenazi Jews) 411/646 ≈ 0.64	0.48	0.30, 0.77	.003
Danish_5 (*)	635	(Danish) 444/484 ≈ 0.92	0.47	0.28, 0.78	.004

Note: Ethnicity label: This was inferred by either the enriched ethnicity of subjects (>50% in the cluster) (*) or based on the PCA plots for this cluster (**) (underscore and number mean that there are more than 1 distinct cluster that are labeled the same). No. subjects: total number of subjects in the cluster; (Enriched ethnicity) No. enriched ethnicity / No. subjects with labels: proportion of subjects with enriched ethnicity from the subjects with known information on their ancestry; OR (e4 vs e3): odds ratio for EL comparing carriers of one copy of APOE e4 to e3e3 carriers. 95% CI: 95% confidence interval. p: p value (association is significant if p < .05/12 = .004). Given sample and effect sizes, there was not enough power (5%–43%) to detect whether the difference between each pair of these five clusters was significant. CI = Confidence interval; OR = Odds ratio.

Effect of APOE in Italians

The cluster enriched of Southern Italians with N = 1,309 includes 77% of subjects of South Italian ancestry (Supplementary Tables S3 and S4 and Supplementary Figure S2A) with 805 subjects from the Southern Italian Centenarian Study (SICS; Table 1) who live in South Italy, and 504 subjects who live in the United States (Figure 3). In this cluster, neither the effect of APOE e2 nor e4 were statistically significant when the data were analyzed without adjustment to the country of residence (carriers of e2: OR for EL = 1.27, 95% CI: 0.92, 1.76; carriers of e4: OR for EL = 0.82, 95% CI: 0.56, 1.20). To investigate the interaction between APOE alleles and country of residence, we analyzed the data in this cluster using a logistic regression model that included the APOE effect, sex, country of residence indicator variable coded as 0 for subjects living in United States, and 1 for subjects living in Italy, and the interaction term between the indicator variable and the genetic effect. We did not detect a statistical significant interaction between residence and the effect of APOE e2. However, the model with the APOE e4 allele had a significant interaction between residence and the e4 effect (Supplementary Table S5). EL was 71% less likely in Italians with one copy of e4 versus e3e3 who live in the United States (OR = 0.29, 95% CI: 0.11, 0.73). However, there was no significant association between e4 and EL in Italians who live in Italy (OR = 1.21, 95% CI: 0.79, 1.86), although a sample size of 805 provides more than 75% power to detect an OR > 1.2. To reduce genetic heterogeneity between the two groups of Italian subjects shown by significant differences in the principal components (Supplementary Table S6), we also reanalyzed the data after removing 157 subjects who mostly live in the United States, but whose ethnicity is more consistent with Northern and Central Europeans based on the PC1-PC2 plot (Supplementary Figure S2A, Figure 3, and Supplementary Table S6). Removing these subjects reduced the variability and differences between PC1-PC4 of the two groups of subjects, and the difference between the statistical estimates slightly increased. Specifically, EL was 81% less likely in Italians with one copy of e4 versus e3e3 who live in the United States (OR = 0.19, 95% CI: 0.04, 0.80), and there was no significant association between e4 and EL in Italians who live in Italy (OR = 1.21, 95% CI: 0.79, 1.85).

Figure 3.

Scatter plots of principal components PC1-PC2 and PC3-PC4 from genome-wide genotype data of all subjects in the extreme longevity (EL) study. Subjects are colored in lighter grey (blue in the online version) if they are in the cluster of 1,309 subjects of Southern Italian descent and live in South Italy. Subjects in darker grey (red in the online version) denote individuals in this cluster who live in the United States. On the scatter plot in the left panel, subjects with PC1 ≥ −0.005 are more consistent with Northern Italian ethnicity, rather than Southern Italian ethnicity.

Effect of APOE in Danish

There were five distinct clusters enriched of subjects of Danish ancestry found to have ethnic-specific effects of APOE alleles on EL (Supplementary Tables S3 and S4), that is, clusters with 437 (Danish_1), 198 (Danish_2), and 766 (Danish_3) subjects for APOE e2, and clusters with 559 (Danish_4) and 635 (Danish_5) subjects for APOE e4. LLFS enrolled participants in both the United States and Denmark, and therefore, we could examine how country of residence modifies the effects of APOE on EL. Table 4 summarizes the distribution of Danish subjects in the five clusters by residence (Denmark or United States). Interestingly, while e2 was significantly positively associated with EL in the Danish_1 cluster that includes 72% of Danish living in Denmark (OR = 2.24, 95% CI: 1.35, 3.73), this association in the Danish_3 cluster that includes only 7% of Danish living in Denmark was not significant (OR = 1.13, 95% CI: 0.75, 1.69), and the ORs were significantly different (p = .04;Supplementary Table S7). The OR for EL in carriers of e4 in the Danish_4 cluster that includes 90% of Danish living in the United States (OR = 0.44, 95% CI: 0.26, 0.72) was slightly smaller than the OR for EL in carriers of e4 in the Danish_5 cluster that includes only 29% of Danish living in the United States (OR = 0.47, 95% CI: 0.28, 0.78), and the ORs were not significantly different (p = .86).

Table 4.

Distribution of Countries Where Subjects Live for Clusters Enriched of Danish Ethnicity

Cluster, No. Subjects	Live in the United States	Live in Denmark
Danish_1, 437	28%	72%
Danish_2, 198	40%	60%
Danish_3, 766	93%	7%
Danish_4, 559	90%	10%
Danish_5, 635	29%	71%

Note: Detailed information for each cluster can be found in Supplementary Tables S3 and S4. We refer to the clusters in which large majority (>65%) of subjects live in the United States or Denmark as clusters with Danish living in the United States (clusters Danish_3 and Danish_4) versus Danish living in Denmark (clusters Danish_1 and Danish_5), respectively.

Discussion

APOE e2 and e4 alleles are known to have an effect on EL (1,6,7) but the analyses in this manuscript suggest that the magnitude of these associations is ethnic-specific among Europeans. We used a novel algorithm to search for clusters of individuals characterized by specific genetic ancestry and varying genetic effects. Our analysis discovered multiple ethnicities with no significant effect of APOE e2 and e4 alleles on EL, one group of North, Eastern European ancestry with a strong protective effect of APOE e2 on EL, and two groups of North European ancestry with different, deleterious effects of APOE e4 on EL. While with larger sample sizes the genetic association between APOE e2 and EL could become statistically significant in more European ethnicities, our analysis suggests that the protective effect of APOE e2 on EL in most European ethnicities is smaller than the effect in Ashkenazi Jewish/Northern European and Danish subjects living in Denmark.

Conomos et al. (32) have shown that the PCA of genetic data might capture family relatedness instead of population structure when applied to the datasets with relatedness. Even though our combined dataset contains ~14% related individuals (Table 1), the evaluation of PopCluster on this very dataset with and without the related subjects performed similarly well when false-positive rate was evaluated. More details of this and other evaluations of PopCluster can be found in ref. (15).

We also provided evidence that the genetic effect of APOE alleles changes based on country of residence in addition to genetic ancestry, suggesting the presence of environmental risk factors of a place of residence that modify the genetic effects of APOE after controlling for genetic ancestry. For example, our analysis showed that there was no deleterious effect of APOE e4 in subjects with Southern Italian ancestry living in the South of Italy. These results suggest that factors related to living in the South of Italy may mitigate the deleterious effect of APOE e4. Our hypothesis, which needs to be further investigated, is that the Mediterranean diet that is followed in Italy contributes to the difference. The results are consistent with previous findings that the Mediterranean diet reduces the risk of Alzheimer’s (33), and APOE e4 carriers versus noncarriers might have an exaggerated or different response to nutrition and other factors in relation to Alzheimer’s and cognitive function (34–36).Similarly, our analyses showed that the protective effect of the e2 allele in subjects with Danish ancestry is stronger in those individuals who live in Denmark and becomes much weaker in individuals of Danish ancestry who live in the United States. The overall diet composition (energy/protein/fat/carbohydrate amounts) in Denmark and the United States is comparable (37,38). A major difference between two countries is that Denmark is one of the world’s happiest countries to live in—Denmark was ranked second in the United Nations’ 2019 World Happiness Report as compared with the United States being ranked 19th (39). These differences are suggestive of complex gene–environment interaction of APOE and nutrition on EL that could lead to the development of natural interventions for healthy aging.

The APOE protein is essential for healthy cholesterol metabolism and central nervous system cholesterol transport. Total APOE levels in plasma in very old individuals were found to be associated with lower total cholesterol and LDL cholesterol levels, which in turn were associated with the APOE e2 allele (40). The APOE e4 allele has been associated with abnormal lipid metabolism in cerebrospinal fluid, and reduced capacity to deliver neuronal cholesterol (5). Detrimental effects of APOE e4 may be alleviated through diet interventions (41), specifically Mediterranean diet (increased omega-3 fatty acids) (36,42). Additionally, APOE e4 carriers may be more sensitive to cholesterol and saturated fatty acids (43). APOE e2 carriers with metabolic syndrome might benefit from diet interventions as well (44).

Our results are consistent with the hypothesis of an interaction between APOE and nutrition that differs by European ethnicity. Future investigations into diet and APOE genotype interaction might point at viable nutrition interventions to reduce the deleterious effect of APOE e4 allele. Additionally, accounting for ethnic-specific differences in the drug development process would contribute to higher drug efficacy for more populations (45).

Funding

This paper was published as part of a supplement sponsored and funded by AARP. The statements and opinions expressed herein by the authors are for information, debate, and discussion, and do not necessarily represent official policies of AARP

Conflict of Interest

None reported.