Genetic correlates of longevity and selected age-related phenotypes: a genome-wide association study in the Framingham Study

Genetic factors associated with human longevity and healthy aging remain largely unknown. Heritability estimates of longevity derived from twin registries and large population-based samples suggest a significant but modest genetic contribution to the human lifespan (heritability ~15 to 30%) [1‐4]. However, genetic influences on lifespan may be greater once an individual achieves age 60 years [5]. Moreover, the reported magnitude of the genetic contribution to other important aspects of aging such as healthy physical aging (wellness)[6], physical performance [7, 8], cognitive function [9], and bone aging [10] are much larger. Both exceptional longevity and a healthy aging phenotype have been linked to the same region on chromosome 4 [11, 12], suggesting that although longevity per se and healthy aging are different phenotypes, they may share some common genetic pathways.

A number of potential candidate genes in a variety of biological pathways have been associated with longevity in model organisms. Genes involved in the regulation of DNA repair and genes in the evolutionarily conserved insulin/insulin-like growth factor signaling pathway [13, 14] are emerging as holding great promise in the future elucidation of the underlying physiology controlling lifespan. Many of these genes have human homologs and thus have potential to provide insights into human longevity [15‐20]. Although numerous candidate genes have been proposed, studies in humans are limited and initial findings often fail replication [21, 22]. More recently genome-wide association studies (GWAS) have become feasible and offer a more comprehensive and untargeted approach to detect genes with modest phenotypic effects that underlie common complex conditions [23].

We had the opportunity to use the Framingham Heart Study (FHS) Affymetrix 100K SNP genotyping resource for a GWAS of longevity and aging-related phenotypes. The FHS offers the unique advantage of a longitudinal family-based community sample with participants who have been well-characterized throughout adulthood with respect to prospectively ascertained risk factors and diseases and continuously followed until death. We report several strategies for 100K SNP associations: 1) a simple low p-value SNP ranking strategy; 2) SNP selection due to associations with more than one related phenotype; and 3) SNP associations within candidate genes and regions previously reported to be associated with longevity in model organisms or humans.

The longevity and aging traits available in the FHS 100K SNP resource are listed in Table 1. In this report, we consider only five of the traits listed in Table 1: multivariable-adjusted age at death, morbidity-free survival at age 65 years, age at natural menopause, walking speed, and biologic age by OSS. These traits include a pooled sample of Original Cohort and Offspring participants, with the exception of walking speed, which is reported in Offspring participants only. Details of the sample size and covariate adjustment for each trait are provided in Table 1.

For each of the five phenotypes, Table 2a and 2b provides the top five SNPs ranked in order by lowest p-value for the GEE and FBAT models (all associations can be viewed on the web http://​www.​ncbi.​nlm.​nih.​gov/​projects/​gap/​cgi-bin/​study.​cgi?​id=​phs000007). If multiple SNPs in linkage disequilibrium (LD r² > 0.80) were included in the top 5, additional SNPs were included until a set of 5 independent associations were listed. Eight SNPs on chromosome 1 were associated with age at death in the FBAT analysis; all with p-value < 10^-4 and two with p-value < 10^-5. The 8 SNPs consisted of two sets of SNPs (rs10493513, rs10493514, rs6689491, rs6657082, rs1405051) and (rs10493515, rs10493518, rs10493517), clustered in two regions approximately 500 kb apart. There was exceptionally high LD across this 500 kb region: the minimum r² between pairs of the eight SNPs was 0.58. The nearest genes in this region existing in public databases were >500 kb from any of these SNPs [31, 32].

There were several additional associations not listed in Table 2a and 2b that were of interest. For age at death in the GEE analysis, SNP associations ranked numbers 9 and 13 were rs10507486 (p-value 0.000128) and rs4943794 (p-value 0.000277), both are intronic FOXO1A SNPs. For age at natural menopause, top ranked SNP associations in the GEE model included number 11, rs6910534 (p = 0.00003) near FOXO3a and number 18, rs3751591 (p = 0.00006) in CYP19A1.

Table 2c presents the LOD scores ≥2.0 and the corresponding 1.5-LOD support interval from genome-wide linkage for the three quantitative aging traits. None of the regions overlapped with SNPs associated with these aging traits in the FBAT and GEE analyses. Of note for biologic age by OSS the linkage peak on chromosome 21 confirmed a prior Framingham Study report using a genome-wide scan with 401 microsatellite markers [10].

Table 3 provides all SNP associations with a GEE or FBAT p < 0.01 for both traits within the two pairs of related traits. For age at death and morbidity-free survival at age 65 years, FBAT models identified 7 SNPs and GEE models identified 9 SNPs associated with both traits including rs2374983 near PON1 (Tables 3a and 3b). For biologic age by OSS and walking speed, 13 SNPs in FBAT models and 6 SNPs in GEE models were associated with both traits (Tables 3c and 3d).

We identified from the literature 79 potential candidate genes and regions associated with longevity (see Additional file 1 for listing). Of these, 12 genes had no SNPs and 67 genes had 1 to 45 SNPs within 60 kb of the gene on the 100K Affymetrix GeneChip. There were 2036 SNPs in the LGV1 region on chromosome 4 previously linked to exceptional longevity [11]. Table 4 shows the candidate genes with SNPs associated with an FBAT or GEE p-value < 0.01 for age at death including: FOXO1a, GAPDH, KL, LEPR, PON1, PSEN1, SOD2, and WRN and for morbidity-free survival at 65 years including:GHR, LEPR, MORF4L1, PON1, PTH, and WRN. Biologic age by OSS shared 2 SNPs in common with age at death: rs4943794 intronic to FOXO1a and rs911847 near SOD2.

To our knowledge, this is the first dense GWAS of longevity and aging traits in a community-based sample of adults from two generations of the same families. Over 1300 men and women have detailed longevity and aging-related phenotypes and 100K SNP genotyping results available on the web. This resource has the potential to detect novel susceptibility genes for human longevity and aging and to examine the relevance of promising candidate gene associations reported in animal models to human aging. We describe several strategies to prioritize SNP associations in this unique resource to enhance the discovery of various genes and pathways that contribute to the control of human longevity. Furthermore, FHS investigators are part of the NIA sponsored Longevity Consortium http://​www.​longevityconsort​ium.​org which offers the opportunity of collaboration with other investigators to replicate important findings in additional cohorts.

In our untargeted approach of ranking SNP associations by the strength of the p-value, 2 intronic FOXO1a SNPs were associated with age at death. One of these SNPs (rs4943794) also was associated with biologic age by OSS in our a priori evaluation of select candidate genes. FOXO (forkhead box group O) transcription factors are targets of insulin-like signaling and are involved in a diverse set of physiological functions including DNA repair and resistance to oxidative stress [33, 34]. Further, FOXO plays a role in lifespan extension in C. elegans and Drosophila [35]. Studies of this gene in humans are limited; two case-control studies have not identified an association between FOXO1a and longevity [36, 37]. However, the prospective population-based Leiden 85-plus Study found that FOXO1a was associated with increased mortality attributable to diabetes related deaths in participants aged 85 years and older [38]. The Leiden 85-plus Study also reported that genetic variation causing a reduction in insulin/IGF-1 signaling resulted in improved old age survival among women [20]. However, that report examined other genes in the insulin/insulin-like signaling pathway and did not specifically examine FOXO1a. Finally, the untargeted approach to SNP selection also identified a SNP near FOXO3a associated with age at natural menopause. This gene has been implicated in oocyte death, depletion of functioning ovarian follicles, and infertility in mice [39, 40] and thus represents a plausible candidate gene for menopause. Most positive common gene variant-disease association studies have failed replication [41] including reports on exceptional longevity. Haplotype-based fine mapping of the region on chromosome 4 linked to human longevity initially suggested the MTP gene, a gene important in lipoprotein synthesis, was associated with longevity [21]. However, this association failed replication in a French cohort of long-lived individuals and subsequent case-control studies of nonagenarians [22, 42]. Beekman, et al [43] found neither linkage to chromosome 4 nor association with the MTP gene and longevity among nonagenarians in the Leiden Longevity Study. Meta-analyses implicated admixture of the control sample in the original report as an explanation for the presumed false-positive association. Thus, our findings are hypothesis generating and their importance can not be determined without evidence of consistent replication in other populations.

We examined pleiotropic effects by identifying SNP associations across two pairs of related traits. One SNP near PON1 emerged as associated with both age at death and morbidity-free survival. Surprisingly, there were relatively few SNPs associated with both traits; prior work had suggested that longevity per se and healthy aging may share common genetic pathways [11, 12]. However, morbidity-free survival was measured at age 65 years, it is possible that as our participants age morbidity-free survival defined at age 75 or 85 years will share additional SNP associations with our longevity trait, age at death. A SNP near SOX5, a gene potentially related to musculoskeletal function was associated with both biologic age by OSS and walking speed.

Our strategy of selecting SNPs in candidate genes and regions previously reported to be associated with longevity yielded interesting findings. For age at death, we identified SNPs in or near several genes including KL, LEPR, PON1, SOD2, and WRN. Defects in the WRN gene are the cause of Werner Syndrome, an autosomal recessive disorder characterized by premature aging. A longitudinal study of ageing Danish twins recently reported a possible association between a successful aging trait and 3 SNPs in the WRN gene [44]. We were unable to determine if our SNP (rs2725369) was in LD with the SNPs in the prior report because the SNPs were not included in HapMap. Mutations in the KL (Klotho) gene in the mouse lead to a syndrome resembling human aging [45‐47]. There has been one report linking a functional variant of the KL gene to human longevity [15]. Thus, results from this GWAS may direct resources to the most relevant candidate genes and pathways for further investigation in humans.

Several important limitations merit comment. First, we acknowledge that there may be a survival bias as participants in this sample had to survive to provide DNA (first systematic DNA collection began 1995) and hence are likely healthier than the full FHS sample. To ameliorate this issue, we adjusted for covariates using the full Framingham sample, and used the residual traits for the subset of individuals genotyped using the 100K Affymetrix GeneChip to test for association with the SNPs using linear regression models. Residual traits from Cox and logistic models typically are not ideally distributed for linear regression models, but our adjustment method using the full sample precludes the testing of SNP associations with age at death and morbidity-free survival using Cox and logistic models. Second, the 100K Affymetrix GeneChip provides limited coverage of the genome; many of our a priori candidate genes did not have any SNP coverage on the chip. For example, several genes that have been studied in model organisms or even in humans such as ACE, Lamin A, SIRT2 and SIRT3, had no SNPs within 60 kb of the gene on the 100K Affymetrix GeneChip. However, genotyping is near-complete for the NHLBI funded 550 K genome-wide scan on all FHS participants. This will enable deeper exploration of our initial 100K SNP associations in a larger sample with denser coverage of the genome. Third, in this analysis we did not examine epistasis or gene-environment interactions which may modify the associations in this study. Importantly, this study is hypothesis generating. Our findings need to be replicated in other samples.

In summary, the untargeted genome-wide approach to detect genetic associations with longevity and aging traits provides an opportunity to identify novel biologic pathways related to lifespan control. GWAS also have the potential to direct investigators of human aging to the most promising candidate gene associations and biologic pathways reported to regulate lifespan in animal models. Enhancing our understanding of the mechanisms responsible for aging may in turn identify directions for health promotion and disease prevention efforts in middle-aged and older adults so that older persons can enjoy more time in good health. These data generate hypotheses regarding novel biologic pathways contributing to longevity and healthy aging and serve as a resource for replication of findings from other population-based samples.