Introduction

In the past 50 yr, the mean life expectancy in most developed countries has increased from around 50 yr to 80 yr. This change suggests that environmental changes may influence on extending the human lifespan. At the same time, variation in maximum lifespan among species and genetic variations in lifespan across human populations clearly points toward a genetic basis for lifespan (Christensen et al., 2006; Bergman et al., 2007).

According to a birth cohort study from north-east America, having a centenarian sibling increased the chance of survival beyond ninety years old by four times, indicating a strong familial aggregation to longevity (Perls et al., 1998). The heritability of lifespan was around 33% and did not vary by gender in Danish twins (McGue et al., 1993). Linkage study is often considered inappropriate for complex traits; however, a region of chromosome 4 was shown to be linked with exceptional longevity (Puca et al., 2001). Gene-environment interactions (e.g. ACE, angiotensin I converting enzyme (peptidyl-dipeptidase A) 1 and exercise; IL6, interleukin 6 (interferon, beta 2) and zinc) also contribute to the development of aging-related phenotypes (e.g. physical decline and inflammation, respectively) (Capri et al., 2006).

The frequency of centenarians is approximately 1/100,000 people worldwide. According to the Korea National Statistical Office (KNSO) Report in 2005 (http://www.kosis.kr.), the frequencies of nonagenarians and centenarians were 131 and 2/100,000 Koreans, respectively (KNSO, 2005). Centenarians may not be free of disease, but manage to survive with or without treatment owing to their resistance to disease (Franceschi et al., 2000). The incidences of major common diseases slow down or even decline around age 85-90 yr (Bergman. et al., 2007). Gender also accounts for important differences in longevity as well as in the occurrence of a variety of age-related disease. The male/female ratios were about 1:4 in nonagenarians and 1:8.2 in centenarians in Korea (KNSO, 2005). Such data raise the possibility of gender-specific genetic determinants of human longevity.

Approximately 30 candidate genes for longevity have been reported in previous studies since Herman (1956) proposed the free radical theory of aging, where oxidative stress causes aging (Salvioli et al., 2006). Genes involved in inflammation and the immune response (e.g. IL6; LMP2, low molecular mass protein 2; TNFA, tumor necrosis factor (TNF superfamily, member 2); TGFB1, transforming growth factor, beta 1; and PPARγ, peroxisome proliferator-activated receptor gamma), genome maintenance and repair (e.g. P53), lipid metabolism (e.g. APOE, apolipoprotein E; CETP, cholesteryl ester transfer protein, plasma; MTTP, microsomal triglyceride transfer protein, and PON1, paraoxonase 1), glucose metabolism (e.g. IGF1, insulin-like growth factor 1 (somatomedin C); and HFE, hemochromatosis), oxidative stress (e.g. SOD1, superoxide dismutase 1, soluble; PARP, poly (ADP-ribose) polymerase 1; and GSTT1, glutathione S-transferase theta 1), mitochondrial mutation, premature aging syndromes such as Werner syndrome (i.e. WRN), and telomere length have been considered with respect to both age-related diseases and longevity (Atzmon et al., 2006; Capri et al., 2006; Christensen et al., 2006; Salvioli et al., 2006).

Genes involved in increased lifespan may play protective roles in the development of age-related diseases. The presence of deleterious genotypes of age-related diseases may be a surrogate indicator of aging. Genes associated with aging-related diseases often have pleiotropic effects, while epistatic interactions affect human longevity (Christensen et al., 2006). For instance, significant decreases of paraoxonase (i.e. PON1) activity on the surface of HDL have been observed in both diabetes mellitus (DM) and cardiovascular disease (CVD) (Mackness et al., 2004). Although association between DM and cancer remains controversial, an increased risk of lung cancer has been observed among diabetic Korean women (hazard ratio 1.39, 95% confidence interval 1.10-1.76) (Jee et al., 2005) and such experimental evidence has suggested that both insulin and insulin-like growth factors (IGFs) can stimulate tumor cell proliferation (Rousseau et al., 2006).

Because the difficulty in finding centenarians or nonagenarians limits the replication of genome-wide association (GWA) studies using an extremely large number of randomly spaced markers, we adopted a high-throughput candidate gene approach to identify susceptibility variants controlling lifespan in Koreans.

Results

Among 565 polymorphic SNPs analyzed, 102 SNPs (18%) with MAF < 2%, HWE P < 10-4, or CR < 95% in the young age group did not passed the threshold of our quality test for genotypes. Thus, a total of 463 informative markers located in 176 genes were included in the subsequent analyses. For the comparison of the individuals above age 90 versus the young healthy controls, our study had a sample size adequate to reach a power of 80% to significantly detect a genotypic OR over 1.7/2.4 (heterozygote/ homozygote) under the assumptions shown above (see the Methods). The result obtained from centenarian women was shown in Supplemental Data Tables.

Allelic- and genotypic association

A total of 43 SNPs located in 33 genes (18.8% of 176 candidate genes) had a nominally significant association with lifespan (P < 0.05) in the allelic χ2 tests (data not shown). Among these, 34 SNPs of 28 genes showed a significant allelic OR and 7 genes had a P value less than 0.01 in at least one age/gender group (i.e. IL12RB2, interleukin 12 receptor, beta 2; PPP1R14C, protein phosphatase 1, regulatory (inhibitor) subunit 14C; EGFR, epidermal growth factor receptor; PAX4, paired box 4; MMP1, matrix metallopeptidase 1 (interstitial collagenase); PPP1R1A, protein phosphatase 1, regulatory (inhibitor) subunit 1A; and ALDH2, aldehyde dehydrogenase 2 family (mitochondrial)). A total of 128 SNPs located in 72 genes (41% of 176 genes) were statistically associated with one or more age/gender groups (P < 0.05) using regression models for five modes of inheritance (data not shown). Among these, the genotypic ORs of 95 SNPs located in 68 genes were nominally significant in at least one of age/gender groups and 13 genes had a P value less than 0.01 (Table 1). We present the P values corrected by genomic control where the λ is greater than 1 (i.e. female centenarians, log-additive model, λ = 1.05). As shown in Table 1, the genotypic effect on longevity was stronger than the corresponding allelic effect in either direction (i.e. either increased or decreased lifespan).

Table 1 Results from the allele-and genotype-based analyses for individual SNPs showing evidence for association with longevity (P < 0.01). aNumber of informative markers for the gene; bUTR, un-translated region; cAncestral alleles were shown in bold; dBest fitting mode of inheritance: Only the SNPs showing a significant genotypic OR from the most significant regression model (P < 0.01) for the gene were shown
Full size table

A total of 27 SNPs located in 25 genes showed evidence for association with lifespan in both gender groups (data not shown). Although the allelic effects of the matrix Gla protein (MGP) and the tumor necrosis factor (ligand) superfamily, member 11 (TNFSF11) were insignificant, these genes showed notable genotypic effects in the overdominant model. As shown in Table 1, TG heterozygotes of the rs1054016 were observed at lower frequencies among the old age group compared to the young age group (OR 0.45 and 0.32 and 0.30; P = 4 × 10-4 and 0.001 for each group of nonagenarians adjusted for gender and males ≥ 90, respectively). All 27 genes consistently showed evidence for association with lifespan in both allele- and genotype-based analyses. In particular, six genes showing significant association with a P value less than 0.01 (ADCY2, adenylate cyclase 2 (brain), PPP1R1A; LYN, V-yes-1 Yamaguchi sarcoma viral related oncogene homolog; PAX4, ALDH2, IL12RB2) attracted considerable attention as candidate genes affecting lifespan in Koreans. The results revealed significant gender differences in genetic associations with human longevity.

Nonagenarian men

A total of 43 SNPs located in 31 genes including four missense mutations located in each of eukaryotic translation initiation factor 4 gamma, 1 (EIF4G1), ATP-binding cassette, sub-family C (CFTR/MRP), member 8 (ABCC8), ALDH2, and telomerase-associated protein 1 (TEP1) showed evidence for genotypic association and 9 genes yielded P values less than 0.01 (Table 1B). By way of example, a male with an A allele of the rs671 (Lys504Glue) located in the ALDH2 increased the chance of living longer than 90 yr by 2.11 times compared to men having a G allele (P = 0.009) and this SNP fitted best under a dominant model (OR 2.63, P = 0.004). No allelic effect of the rs757110 (Ser1369Ala) in the ABCC8 gene was identified; however, the effect of the TG heterozygous genotype was greater than that of the rare GG homozygous type (OR 3.12, P = 6×10-4).

Nonagenarian women

A total of 37 SNPs located in 31 genes showed evidence for genotypic associations in either age group of women (i.e. women aged over 90 yr and centenarian women). Genes such as PPP1R14C (P = 0.005) and the EGFR (P = 0.007) had lower allelic P values in the centenarian group with a small sample size (data not shown), while IL12RB2 was significant only in the younger group with more subjects (P = 0.008). However, the rare T allele of rs2030071 was not sufficiently common to determine the genetic model (Table 1C).

Haplotypic association

Among the 69 genes that showed significant genotypic association, we found significant association with longevity in 12 genes by using tagSNPs for 32 genes with at least 3 genotyped SNPs. Among nine genes that consistently yielded significant evidence of allelic, genotypic, and haplotypic associations with longevity (i.e. HK2, hexokinase 2; ERBB4, v-erb-a erythroblastic leukemia viral oncogene homolog 4 (avian); PCSK1, proprotein convertase subtilisin/kexin type 1; ITK, IL2-inducible T-cell kinase; EGFR; PAX4; LPL, lipoprotein lipase; LYN; and CDK10, cyclin-dependent kinase 10), only the PCSK1 (P = 0.008), EGFR (P = 0.003), PAX4 (P = 0.008), and LYN (P = 0.002) remained significant after the Bonferroni correction for multiple testing of haplotypes in each gene (Table 2). Figure 1 summarizes the -log10 (P value) obtained from three different analyses of association for these four genes.

Table 2 Results from the haplotype-based analyses for candidate genes showing evidence for association with longevity (P < 0.05/Number of haplotypes). aNumber of tagSNPs among all SNPs genotyped for a candidate gene, bThe P value for each haplotype using a general linear model (GLM)
Full size table
Figure 1
figure 1

Summary results with gene and SNP information for four genes that consistently showed evidence for association with lifespan in allelic, genotypic, and haplotypic analyses.

Full size image

The presence of the G allele of rs155979 in the CGA haplotype of PCSK1 decreased lifespan among the group of nonagenarian adjusted by sex (OR 0.54). The G allele of rs712700 increased lifespan in the GG haplotype of PAX4 (OR 5.14) among nonagenarian men. Although only the rs1980042 (C) allele in haplotypes of LYN seemed to confer longevity, the effect size of 2.5 for the CC genotype increased to 6.2 for the CTCCGCA haplotype among nonagenarian men. Among female centenarians, the OR of 3.12 for the rs2293347 (GG) genotype increased to 4.11 for the CGCA haplotype of the EGFR gene when compared with the CACA reference haplotype.

Gene-gene interactions

Supplement Data Table S2 presents significant P values for the epistatic interaction LRT under a best fitting model among five genetic models tested for each age/gender group. Supplement Data Table S3 shows the ORs and their 95% CIs for the genotype combinations composed of each candidate SNP pair for two genes along with the numbers of cases and controls. For each age/gender group, 6, 1, 3, and 2 pairs of loci for each group yielded statistical significance in both the regression model and ORs for interaction. In particular, the interaction of platelet derived growth factor C (PDGFC) and glycogen synthase 2 (liver) (GYS2) for males older than 90 yr as well as the interaction of TNFSF11 and CDK10 in the group adjusted by gender showed a dose-response relationship with increasing copies of the target allele for each SNP.

Discussion

It is well known that genetic variation exists across ethnic groups. Indeed, significant association of genes for human leukocyte antigen (HLA), ACE, APOE, MTTP, and CETP with human longevity in Caucasian populations have not been replicated in studies using Korean centenarians (Choi et al., 2003). The Korean longevity study carried out since 1998 reported significant differences in the occurrence of centenarians by gender, location, lifestyle, and diet (Lee et al., 2005). In 2005, the ratios of the Jeola province to the Kyongsang province were 1:0.97 for nonagenarians and 1:1.27 for centenarians (KNSO, 2005). Interestingly, a reverse relationship between the incidence of cancer and degree of longevity has been observed across provinces, indicating that prevention of cancers may be an important strategy to enhance lifespan (Kwon and Park, 2005). This notion could be applied to other age-related diseases such as DM and CVD.

According to the theory of evolution, differential survival reflects the effect of deleterious genotypes of age-related diseases. However, some longevity genes (e.g. klotho that is associated with low HDL) buffer the deleterious effect of age-related diseases genes (e.g. lipoprotein, Lp(a), a susceptibility gene for vascular disease), which may explain an observed paradoxical increase of deleterious genotypes in centenarians (Bergman et al., 2007). Lunetta et al. (2007) conducted a genome-wide association study of age-related phenotypes using Affymetrix 100 K SNP Chips in the context of a cross-sectional study design. However, none of the associations with 5 aging traits achieved genome-wide significance using 1,345 Framingham Study participants.

In the current study, a total of 60 genes out of 179 candidates were nominally associated (P < 0.05) with longevity in at least one of age/gender groups. Among eleven genes that consistently yielded statistical evidence of allelic, genotypic and haplotypic association with lifespan, only four genes remained significant after the Bonferroni correction for haplotypes. The protein encoded by the PCSK1 gene, located in 5q15-q21, is a proinsulin-processing enzyme that plays a key role in regulating insulin biosynthesis (Ohagi et al., 1996). Mutations in this gene are associated with obesity, tumorigenesis, and metastasis (Tzimas et al., 2005). The PAX4 gene, located in 7q32, is a member of the paired box family of transcription factor that mediates differentiation of insulin-producing beta-cells in the pancreas and plays critical roles during cancer growth (Li et al., 2006). EGFR is a transmembrane glycoprotein that functions as a tyrosine protein kinase (Kondo and Shimizu, 1983). Mutations in the EGFR gene, located in the 7p12.3-p12.1 region, have oncogenic potential in the development of non-small cell lung cancer (Tai et al., 2006). The LYN, located on 8q13, encodes a tyrosine kinase that is essential in immunoglobulin-mediated signaling, particularly in establishing B cell tolerance (Parravicini et al., 2002). Even aging mice with a Lyn up/up phenotype do not display hematologic malignancies, while Lyn-/- mice are more susceptible to tumorigenesis (Harder et al., 2001). To the best of our knowledge, this is the first reported evidence associating these genes with human longevity.

Several limitations of this study deserve mention. First, the influence of environmental factors could not be controlled in this study due to lack of information in the control group. Second, we could not evaluate the function of any SNP that was not a missense mutation but did have statistical evidence for association. Third, given the small number of centenarians, the statistical power may not be sufficient to significantly detect genes with weak effect, particularly, for the gene-gene interaction. There are further issues concerning the interpretation of the findings using the over-dominant model and the control group consisted of 12-14 yr old middle school students. To eliminate the impact of different substructure between the long-lived group and the young control group, the P values were corrected by genomic control for each age-gender group. In the nonagenarian group, the Bonferroni threshold for P < 0.05 with 463 SNPs was 1.08 ×10-4 and the significance level was relaxed by 1000 permutations to 1.69 × 10-4, 1.56 × 10-4, 2.16 × 10-4, 1.50 × 10-4, and 2.42 ×10-4 under the co-dominant, dominant, recessive, over-dominant, and log-additive models, respectively. None of the SNPs achieved genome-wide significance. However, the underlying LD structure among the markers reduced the number of effectively independent tests, and so the genome-wide threshold of significance may be too stringent particularly in the context of candidate gene approach (Rao and Gu, 2001).

For genetic variants with strong effects, relatively small sample size is sufficient to achieve statistical power to detect association. The results in this study warrant further replication studies. However, the expectation for replication of results may need to be relaxed in studies of extremely old subjects, who are rarely observed. Under such circumstances, additional information gathered from laboratory techniques, bioinformatics, and a priori insights of biological pathways could be used to provide plausibility for interpreting genetic association findings (Chanock et al., 2007). To increase statistical power of a test for gene by gene or gene by environment interaction, international collaboration is essential due to the rarity of centenarians. Continuing efforts may be necessary to find a functional variant in the associated region.

Nevertheless, the results from various statistical analyses consistently supported that the presence of the target allele of a potent SNP conferred a greater likelihood of longevity in Koreans. The effects of gender mediated genetic association with the probability of achieving longevity, as shown in previous studies (Franceschi et al., 2000) and the observed gender difference in our findings supports, in part, the gender-specific probability of achieving longevity. Interestingly, a few candidate gene variants with no or insignificant allelic effect revealed strong evidence for the allele with a large heterozygote effect on longevity (e.g. MGP, TNFSF11, and ABCC8).

As far as we know, this is the first association study of longevity susceptibility genes adopting a high-throughput candidate gene approach in nonagenarians and centenarians who rarely appear in Korea. An understanding of the mechanisms underlying aging and longevity by discovering novel susceptibility genes may lay the foundations for prediction, prevention, and treatment of age-related diseases. Thus, the findings of the current study may provide a starting point to unravel genetic factors controlling longevity for future studies.

Methods

Subjects

We identified 137 long-lived individuals over 90 yr old including 35 centenarians from the Korean Centenarian Study, which began in 1999 (Choi et al., 2003). We also selected 213 healthy middle school students from individuals who visited the Samsung Seoul Hospital for psychiatric tests in 2003 to be used for comparison. Eight control samples from a three-generation family were included to test genotype quality. After informed consent was obtained from all subjects, a clinician interviewed study subjects and drew a 10 ml fasting venous blood sample in an EDTA tube. The institutional review board (IRB) of the Samsung medical center reviewed the research protocols and approved this study. The gender and age composition of study subjects were shown in Supplemental Data Table S1.

Candidate genes and SNP selection

Candidate genes, mostly for DM, CVD, and cancers, were selected based on searches for their functions in public databases (OMIM, Gene, etc.) and previous literature reviews. SNP markers within or near candidate genes were identified from the literature, the dbSNP database (http://www.ncbi.nlm.nih.gov/projects/SNP/), the HapMap database (http://www.hapmap.org/), and Vector NTI® software (Invitrogen, Inc., Carlsbad, CA). We included 565 informative single nucleotide polymorphism (SNP) markers with high design scores provided by Illumina, Inc. (San Diego, CA) and heterozygosity above 0.05 using 300 independent healthy Korean samples. Information for these 565 SNPs located in 194 genes is presented in Supplemental Data Table S2.

DNA preparation and SNP genotyping

Genomic DNA samples were prepared from peripheral blood using the QIAamp DNA Blood Maxi kit following the manufacturer's instructions (Qiagen, Inc., Valencia, CA). For each sample, an average concentration of DNA from three repeated measures was determined using a SpectraMax Plus 384 spectrophotometer (Molecular Devices, Inc., Sunnyvale, CA). A 250 ng aliquot of DNA was genotyped for SNP markers using Golden-Gate chemistry on Sentrix® Array Matrices (Illumina, Inc.) at the Samsung Biomedical Research Institute. Genotypes showing significant statistical evidence for association with longevity were confirmed using an ABI PRISM 3100 sequencer (Applied Biosystems, Inc., Foster City, CA).

Statistical analysis

At each SNP, the genotyping call rate (CR > 95%), minor allele frequency (MAF > 2%), and deviation from Hardy-Weinberg equilibrium (HWE P < 0.0001) were separately computed in the long-lived group and the young healthy group. Ancestral alleles for each SNP were determined using the dbSNP database. The SNPs-based whole genome association studies (SNPassoc) program implemented in the free downloadable statistical software environment R v. 2.6.2 (http://www.r-project.org) was used to perform the quality tests for genotypes (González et al., 2007). We performed preliminary analyses of linkage disequilibrium (LD) patterns using both D' and r2 and then selected a set of tagging SNPs within a gene for the haplotype-based analysis using Haploview v. 4.0 (http://www.broad.mit.edu/mpg/haploview/index.php/).

We performed subsequent statistical analyses for each of four groups stratified by age and gender (nonagenarians, male nonagenarians, female nonagenarians, and female centenarians). We assessed allelic association with the χ2-tests for individual SNPs and calculated the allelic odds ratio (OR) and 95% confidence interval (CI) in both adjusted and stratified analyses using publicly available subroutines 'genassoc' and 'gamenu' in the STATA/SE v. 9 software package (http://www.gene.cimr.cam.ac.uk/clayton/software/stata/).

We performed genotypic association tests and logistic regression analyses implemented in the SNPassoc package in five different genetic models: codominant, dominant, recessive, over-dominant, and log-additive. To account for the impact of substructure or genetic relatedness among the subjects, we estimated the 'inflation factor (λ)' for each age-gender group and corrected P values by genomic control (Devlin and Roeder, 1999).

We used general linear models for the regression of longevity on ambiguous haplotypes. Rare haplotypes with an allele frequency of less than 1% were combined in a group to be analyzed. We applied logistic regression models (LRMs) to perform two-way interaction analyses between two SNPs for each gene (Briollais et al., 2007). For each of four age/gender groups, we screened the most promising pairs of SNPs that yielded P < 0.01 for the interaction likelihood ratio test (i.e. LRTij = -2(logLfull(i,j)-logLadditive(i,j)) under the five genetic models. Subsequently, we obtained the ORs and 95% CIs under the best fitting model for both marginal effects and interactions of the pair of SNPs.

We computed the statistical power of our sample sizes for each genotype using the web browser 'Genetic Power Calculator' (http://pngu.mgh.harvard.edu/~purcell/gpc/cc2.html) under an additive model with the assumption of disease allele frequencies of 0.3, D' of 1, and a longevity prevalence of 0.13% and 0.002% for the Korean nonagenarians and centenarians, respectively (Purcell et al., 2003).