• Open Access

Association of common genetic variation in the insulin/IGF1 signaling pathway with human longevity

Authors


Elad Ziv, MD, 1701 Divisadero Street, Suite 554, UCSF Box 1732 San Francisco, CA 94115, USA. Tel.: +1 (415)353-7981; fax: +1 (415)353-7932; e-mail: elad.ziv@ucsf.edu

Summary

The insulin/IGF1 signaling pathways affect lifespan in several model organisms, including worms, flies and mice. To investigate whether common genetic variation in this pathway influences lifespan in humans, we genotyped 291 common variants in 30 genes encoding proteins in the insulin/IGF1 signaling pathway in a cohort of elderly Caucasian women selected from the Study of Osteoporotic Fractures (SOF). The cohort included 293 long-lived cases (lifespan ≥ 92 years (y), mean ± standard deviation (SD) = 95.3 ± 2.2y) and 603 average-lifespan controls (lifespan ≤ 79y, mean = 75.7 ± 2.6y). Variants were selected for genotyping using a haplotype-tagging approach. We found a modest excess of variants nominally associated with longevity. Nominally significant variants were then replicated in two additional Caucasian cohorts including both males and females: the Cardiovascular Health Study and Ashkenazi Jewish Centenarians. An intronic single nucleotide polymorphism in AKT1, rs3803304, was significantly associated with lifespan in a meta-analysis across the three cohorts (OR = 0.78 95%CI = 0.68–0.89, adjusted P = 0.043); two intronic single nucleotide polymorphisms in FOXO3A demonstrated a significant lifespan association among women only (rs1935949, OR = 1.35, 95%CI = 1.15–1.57, adjusted P = 0.0093). These results demonstrate that common variants in several genes in the insulin/IGF1 pathway are associated with human lifespan.

Introduction

Genes in the insulin/IGF1 signaling pathway affect lifespan in yeast, nematodes, fruit flies and mice (Kuningas et al., 2008). Mutation of genes in this signaling pathway confers greater resistance to oxidative stress and phenotypic characteristics consistent with delayed or slowed aging (Tatar et al., 2003).

The insulin/IGF1 pathway was first implicated in aging in Caenorhabditis elegans (Kenyon et al., 1993; Larsen et al., 1995; Morris et al., 1996). Mutation of daf-2, the nematode ortholog of the IGF1 and insulin receptors, results in extended lifespan and increased stress resistance (Kenyon et al., 1993; Kimura et al., 1997); these effects are mediated by the nuclear transcription factor, DAF-16, which is a homolog of the mammalian forkhead (FKHRL or FOXO) transcription factors (Ogg et al., 1997). Mutations in age-1, the homolog of mammalian PI3Kinase, which is involved in signal transduction from the insulin and IGF1 receptors (Dorman et al., 1995), also increase lifespan in C. elegans. Similarly, in Drosophila, mutations of the insulin receptor homolog, its substrate, chico (Clancy et al., 2001), and of the daf-16/foxo homolog (Hwangbo et al., 2004) reduce insulin signaling and extend lifespan.

In vertebrates, growth and glucose metabolism are regulated by two parallel pathways, with separate receptors for insulin and IGF1. In mice, mutations that result in growth hormone deficiency or mutations in growth hormone receptors cause reduced size, lower insulin levels, increased stress resistance and longer lifespan (Bartke, 2005). Heterozygous deletion of the IGF1R gene in mice causes a modest reduction in size, improves stress resistance and extends lifespan in females. Disruption of the INSR gene globally or in most tissues leads to insulin resistance and shortened lifespan (Joshi et al., 1996; Okamoto & Accili, 2003); however, mice homozygous for INSR deletion specific to fat cells have extended lifespan in both sexes (Bluher et al., 2002), indicating the presence of tissue-specific effects.

Results from model organisms suggest the hypothesis that genes in the insulin/IGF1 pathway may be involved in the control of human lifespan, and variants that affect their activity may be associated with human longevity. Although a few associations of insulin/IGF1 pathway genes with longevity have been reported (Bonafe et al., 2003; van Heemst et al., 2005; Kuningas et al., 2007; Suh et al., 2008; Willcox et al., 2008), most of these findings have not been replicated and the studies did not analyze genetic variation in the insulin/IGF1 pathway systematically.

We have comprehensively evaluated common variation in genes in the insulin/IGF1 signaling pathway for association with human lifespan. We performed a nested case–control association study of longevity in a cohort of elderly Caucasian women from the Study of Osteoporotic Fractures (SOF). We then replicated significant associations in two independent Caucasian cohorts of men and women.

Results

All subjects were Caucasian. The SOF cohort was all women, the Cardiovascular Health Study (CHS) cohort was 47% female, and the Ashkenazi Jewish Centenarians (AJC) study was 66% female (Table 1). More than half (55.6%) of the longevity cases were still alive in SOF, about a third (33.7%) were still alive in CHS and 24% of cases were still alive in AJC.

Table 1.   Characteristics of the three study cohorts
CharacteristicSOFCHSAJC
CasesControlsCasesControlsCasesControls
  1. Age refers to the age at death for deceased subjects, and age at last follow-up for subjects still alive.

  2. SOF, Study of Osteoporotic Fractures; CHS, Cardiovascular Health Study; AJC, Ashkenazi Jewish Centenarians; NA, not available.

N293603279797383363
Age (years) 95.3 ± 2.2 75.7 ± 2.6 94.5 ± 2.1 76.1 ± 3.1 99.8 (95–108) 85.5 (43–94)
Sex (% male) 0 0 47.0 52.6 25.2 42.7
% Alive 55.6 0 33.7 0 24 96
% Atherosclerosis death 17.7 30.2 16.8 26.1NANA
% Cancer death 4.1 40.8 12.9 37.5NANA
% Other death 22.5 29.0 36.6 39.1NANA

Association of genotype with longevity in SOF

We evaluated 291 single nucleotide polymorphisms (SNPs) in 30 genes in the insulin/IGF1 signaling pathway (Figure 1, Table S1). Twenty-one (7%) SNPs in 9 (30%) genes (AKT1, FOXO1A, FOXO3A, GHR, GHRHR, IGF1R, IGFBP3, IGFBP4, and PTEN) were associated with longevity at a nominal P < 0.05, a modest excess over the expected ∼15 (Table 2). Eight SNPs had a < 0.01 (an excess of ∼5 over the expected number), and three SNPs had P < 0.005 (∼1.5 more than expected). FOXO3A (7/16) and GHR (7/26) had the most SNPs associated with longevity.

Figure 1.

 Graphic of insulin/IGF1 signaling pathway showing proteins encoded by scanned genes.

Table 2.   Association of variants in insulin/IGF1 pathway genes with longevity
GeneSNPsP < 0.05P < 0.01P < 0.005
  1. Number of SNPs in each gene associated with longevity at a nominally significant level is shown. Results shown are for an additive model corrected for self-reported ethnicity. Total number of SNPs observed at three different significance thresholds is compared with the number expected. Genes with longevity-associated SNPs highlighted in bold.

AFX7000
AKT18100
AKT25000
AKT313000
FOXO1A7100
FOXO3A16741
GHR26731
GHRH1000
GHRHR13100
IGF16000
IGF1R34111
IGFBP111000
IGFBP28000
IGFBP35100
IGFBP47100
IGFBP511000
IGFBP64000
INS2000
INSR32000
INSRR4000
IRS16000
IRS213000
IRS41000
PDPK13000
PIK3CA5000
PIK3CB3000
PIK3R128000
PIK3R22000
PTEN9100
SHC11000
All genes, observed2912183
All genes, expected152.91.5

For all genes except GHR, haplotypes containing longevity-associated SNPs exhibited a similar effect size and significance level for association with longevity, suggesting that in each case, the putative causative variant was in linkage disequilibrium (LD) with a single haplotype-tagging SNP (data not shown). For GHR, we observed the most significant association for a three-SNP haplotype comprising rs9292854, rs13182117, and rs4866941 (P = 0.0005, compared with the best single-SNP association of P = 0.007 for rs4866941; analysis not corrected for race/ethnicity), suggesting that the putative causative allele resides on this haplotype.

Replication of association of AKT1 and FOXO3A SNPs with longevity

As the case–control associations we observed in SOF were not significant after correction for multiple hypothesis testing, we replicated the nominally associated SNPs in two independent Caucasian cohorts, the CHS and AJC. We selected all 21 SNPs individually associated with longevity in SOF (P < 0.05) as well as 4 SNPs significantly associated with longevity (P < 0.001) in 2-SNP interaction analyses for replication genotyping (Table 3). We observed the most consistent results across all three cohorts for 1 SNP in AKT1 (rs3803304), which was nominally associated with longevity in each of the three cohorts, with a consistent effect size (odds ratio, OR) of 0.77–0.79. In a meta-analysis of results from all three cohorts, this association was statistically significant (P = 0.0004; permutation-based correction for multiple testing P = 0.043).

Table 3.   Meta-analysis of association with longevity for SNPs genotyped in SOF, CHS, and AJC
All (males + females)SOF (n = 896)CHS (n = 1076)AJC (n = 746)Meta-analysis
SNPReasonGeneOR95%CIPOR95%CIPOR95%CIPOR95%CIP
  1. We show OR, 95%CI and P-values for an additive model for each study and for the meta-analysis. SOF analyses are adjusted for self-reported ethnicity, CHS and AJC analyses are adjusted for sex. Nominal P-values are shown for all analyses. For replication, only AKT1 rs3803304 meets the permutation-based threshold for statistical significance in the entire cohort; FOXO3A rs1935949 and rs4946935 meet the threshold for women only (data not shown); these 3 SNPs are highlighted in bold in the table.

  2. SOF, Study of Osteoporotic Fractures; CHS, Cardiovascular Health Study; AJC, Ashkenazi Jewish Centenarians; Reason, reason for replication; NA, not available.

rs3803304LongevityAKT10.790.62–0.990.0490.770.61–0.980.0330.770.59–0.980.0410.780.68–0.890.00040
rs1334241LongevityFOXO1A1.301.02–1.650.0320.840.65–1.080.170.900.67–1.200.481.010.87–1.170.87
rs1044891LongevityFOXO3A1.481.10–1.970.00830.910.66–1.220.53NANANA1.170.95–1.450.14
rs1935949LongevityFOXO3A1.251.00–1.560.0431.140.92–1.410.231.361.05–1.740.0161.241.09–1.410.0014
rs2153960LongevityFOXO3A1.311.05–1.630.0151.150.93–1.410.191.030.81–1.300.811.161.02–1.320.019
rs3778588LongevityFOXO3A1.511.13–1.990.00470.970.73–1.290.841.240.89–1.720.191.221.03–1.450.023
rs4946935LongevityFOXO3A1.261.00–1.570.0411.190.96–1.470.111.331.03–1.720.0291.251.10–1.430.00087
rs9486902LongevityFOXO3A1.471.12–1.920.00500.990.74–1.310.931.000.66–1.500.991.180.98–1.400.073
rs9486913LongevityFOXO3A1.491.12–1.960.00510.930.69–1.230.601.190.84–1.660.321.180.99–1.400.053
rs12153009LongevityGHR 0.760.60–0.950.0170.940.75–1.160.550.910.71–1.160.470.870.76–0.990.033
rs2940923LongevityGHR 1.311.06–1.610.00941.080.88–1.310.441.281.03–1.590.0231.211.09–1.370.0013
rs2940935InteractionGHR 1.070.82–1.380.611.371.07–1.730.0111.301.00–1.670.0461.241.08–1.440.0031
rs3764451LongevityGHR 0.680.50–0.900.00931.030.77–1.350.860.920.66–1.260.610.860.73–1.030.094
rs4292454LongevityGHR 1.251.00–1.540.0430.930.76–1.130.49NANANA1.070.92–1.230.39
rs4866941LongevityGHR 1.421.12–1.780.00270.880.70–1.100.27NANANA1.110.95–1.310.19
rs6883523LongevityGHR 1.301.00–1.670.0420.910.70–1.170.480.920.63–1.320.641.050.90–1.240.52
rs6887528LongevityGHR 1.721.12–2.620.0120.830.54–1.260.38NANANA1.190.88–1.610.25
rs2228078LongevityGHRHR2.681.00–7.140.0481.020.46–2.210.96NANANA1.480.81–2.720.21
rs2272037LongevityIGF1R1.391.10–1.740.00481.060.86–1.290.55NANANA1.191.03–1.390.021
rs6670LongevityIGFBP31.281.00–1.630.0470.890.70–1.130.360.940.74–1.190.631.020.89–1.180.75
rs1009728InteractionIGFBP41.090.87–1.360.431.030.82–1.280.781.100.89–1.350.351.080.95–1.220.24
rs7214466LongevityIGFBP40.750.59–0.930.0121.040.84–1.280.680.910.73–1.120.370.900.79–1.020.094
rs11575134InteractionIGFBP51.050.80–1.370.710.900.70–1.150.401.281.00–1.630.0431.070.93–1.240.36
rs3770472InteractionIGFBP51.010.78–1.310.910.870.68–1.090.231.291.00–1.630.0421.040.90–1.200.60
rs1022427LongevityPTEN1.651.04–2.610.0331.000.63–1.580.99NANANA1.280.93–1.780.13

Two SNPs in FOXO3A were nominally associated with longevity in SOF and in AJC, but not in CHS, although the magnitude and direction of the effect were consistent in CHS. The P-values in the meta-analysis were 0.0014 for rs1935949 and 0.0009 for rs4946935; these approached statistical significance in the meta-analysis after correction for multiple testing.

We observed a borderline effect of sex on the FOXO3A rs1935949 SNP association with longevity, suggesting a stronger effect of female sex (OR = 0.72, 95%CI = 0.52–0.99, P = 0.049), as also seen when comparing the odds ratio for association with longevity between the mixed-sex (OR = 1.24, 95%CI = 1.09–1.41) and female-only subsets (OR = 1.35, 95%CI = 1.15–1.57). When restricted to female subjects, the association of FOXO3A rs1935949 and rs4946935 with longevity in the meta-analysis achieved statistical significance (adjusted P-values were 0.0093 and 0.019 respectively).

Next, we evaluated the association with longevity for the replicated SNPs in a genotypic model (Table 4). For AKT1 rs3803304, the association with longevity appeared driven by the minor allele homozygous genotype (CC). Subjects homozygous for the minor allele were underrepresented among long-lived cases (OR = 0.41–0.5, P = 0.00016 in the meta-analysis). A single allele did not confer a significant effect (P = 0.12), suggesting a recessive effect, with homozygosity for the minor allele decreasing the probability of longevity. For the two FOXO3A SNPs, the effect appeared to be consistent with a multiplicative model; the probability of surviving to exceptional longevity increased with each additional copy of the minor allele. Results were similar when evaluated in women only (rs1935949: CT, OR = 1.33 (1.08–1.63), P = 0.067; TT, OR = 1.82 (1.25–2.65), P = 0.0017).

Table 4.   Meta-analysis of association with longevity for top SNPs genotyped in SOF, CHS, and AJC
SNPGenotypeSOF (n = 896)CHS (n = 1076)AJC (n = 746)Meta-analysis
NOR95%CIPNOR95%CIPNOR95%CIPOR95%CIP
  1. OR and 95%CI from a genotypic model are shown in men and women combined, adjusted for sex.

AKT1 rs3803304 G/CGG465ref607455 
GC3090.900.66–1.230.533450.840.62–1.130.252490.860.63–1.180.360.870.73–1.040.12
CC700.500.27–0.910.025670.500.25–0.970.040340.430.20–0.890.0230.480.32–0.700.00016
FOXO3A rs1935949 C/TCC420ref531429 
CT3661.290.95–1.750.103941.240.92–1.670.152741.300.95–1.770.101.271.07–1.520.0064
TT811.520.92–2.520.10871.160.70–1.940.56392.081.04–4.160.0391.461.07–2.010.019
FOXO3A rs4946935 G/AGG415ref521437 
GA3581.250.92–1.690.164031.341.00–1.800.0502111.340.96–1.870.0901.311.09–1.560.0035
AA781.610.97–2.670.066841.210.72–2.040.47391.780.90–3.520.101.481.07–2.030.018

Follow-up of significant findings in AKT1 and FOXO3A

Complete coverage (at r2 > 0.8) of common variation in both AKT1 and FOXO3A based on HapMap Caucasians (CEU) has been achieved (Fig. 2A, Fig. S1).

Figure 2.

Figure 2.

 Genomic context and linkage disequilibrium for (A, B) AKT1 and (C, D) FOXO3A. (A) Linkage disequilibrium plot for AKT1 of r2 in HapMap Caucasians. All HapMap SNPs at MAF > 5% are shown. SNPs genotyped in SOF are marked by black triangles. The longevity associated SNP rs3803304 is marked by a white triangle. (B) Close-up of 1kb area around rs3803304. UCSC genome browser tracks show mRNA isoforms, expressed RNAs, CpG islands, evolutionary and sequence pattern extraction through reduced representations regulatory potential (ESPRR) and conservation across 28 mammalian species. (C) Linkage disequilibrium plot for FOXO3A of r2 in HapMap Caucasians, showing all SNPs genotyped in SOF. The longevity- associated SNPs rs1935949 and rs4946935 are marked by numbers 141 and 142. A more detailed view showing all HapMap SNPs in the region is shown in Supplementary Figure 1. (D) Close-up of 8 kb area around rs1935949 and rs4946935. UCSC genome browser tracks show mRNA isoforms, expressed RNAs, regulatory potential (ESPRR) and conservation across 28 mammalian species.

Figure 2.

Figure 2.

 Genomic context and linkage disequilibrium for (A, B) AKT1 and (C, D) FOXO3A. (A) Linkage disequilibrium plot for AKT1 of r2 in HapMap Caucasians. All HapMap SNPs at MAF > 5% are shown. SNPs genotyped in SOF are marked by black triangles. The longevity associated SNP rs3803304 is marked by a white triangle. (B) Close-up of 1kb area around rs3803304. UCSC genome browser tracks show mRNA isoforms, expressed RNAs, CpG islands, evolutionary and sequence pattern extraction through reduced representations regulatory potential (ESPRR) and conservation across 28 mammalian species. (C) Linkage disequilibrium plot for FOXO3A of r2 in HapMap Caucasians, showing all SNPs genotyped in SOF. The longevity- associated SNPs rs1935949 and rs4946935 are marked by numbers 141 and 142. A more detailed view showing all HapMap SNPs in the region is shown in Supplementary Figure 1. (D) Close-up of 8 kb area around rs1935949 and rs4946935. UCSC genome browser tracks show mRNA isoforms, expressed RNAs, regulatory potential (ESPRR) and conservation across 28 mammalian species.

The longevity-associated AKT1 rs3803304 SNP is not in LD with any other HapMap SNPs (Fig. 2A). Therefore to follow-up, we selected other variants in the surrounding area from dbSNP. We genotyped eight additional AKT1 SNPs in SOF, including five coding SNPs, (one rare, four unvalidated), two synonymous SNPs, and one intronic SNP located 22 bp from an intron–exon junction. Three of these eight SNPs were polymorphic in SOF and none was associated with longevity (Table S1). As no coding SNPs in AKT1 were associated with longevity, we searched for potential regulatory sites in the region (Fig. 2B). AKT1 rs3803304 lies in an intron, 70 bp away from a conserved exon–intron boundary and in an area of high-predicted regulatory potential (0.3 score on the evolutionary and sequence pattern extraction through reduced representations (ESPERR) prediction metric) (King et al., 2005). Data from the Affymetrix Transcriptome Phase 3 gene tiling arrays (Kapranov et al., 2007) show a long (> 200 bp) RNA signal at this intron/exon border encompassing rs3803304 (Fig. 2B); a CpG island is also found here. It is possible that rs3803304 disrupts an RNA regulatory mechanism.

Initial SNP discovery in FOXO3A was by sequencing—no other common exonic SNPs were found. The two SNPs most strongly associated with longevity, FOXO3A rs1935949 and rs4946935, lie in intron 3, 1.7 kb from the intron–exon boundary of the 3′UTR, but not in an area of evolutionary conservation (Fig. 2D).

Interaction analyses

We included four SNPs in the replication genotyping that were not individually associated with longevity in SOF, but had a significant association with longevity in interaction analyses with another SNP. Two IGFBP5 SNPs (rs3770472 and rs11575134) had significant interactions with GHR SNP rs2940935 in SOF (both P < 0.00004); these interactions appeared to replicate in the AJC study (P = 0.012 for rs3770472 and 0.026 for rs11575134), but not in the CHS cohort. SNP rs1009728 was included in the replication genotyping because of a significant interaction with a SNP in PIK3R1, but this latter SNP could not be genotyped in the replication cohorts.

Association of lifespan-associated SNPs with cause-specific mortality

We evaluated associations of the longevity-associated SNPs with reported cause of death in SOF and CHS (Table 5). AKT1 rs3803304 was associated with deaths due to cardiovascular disease and deaths from noncardiovascular, noncancer causes. The FOXO3A SNP rs4946935 was associated with cardiovascular causes of death in SOF and with noncardiovascular, noncancer other causes of death in CHS. The FOXO3A SNP rs1935949 was associated with cardiovascular and cancer causes of death in SOF, and with other causes of death in CHS.

Table 5.   Association of genotype with cause of death in SOF and CHS
Cause of deathCohortSignificanceAKT1 rs3803304FOXO3A rs1935949FOXO3A rs4946935
  1. For this analysis, the case group comprises participants who died of specific causes and the control group comprises long-lived participants.

  2. Results for an additive model are shown. Analyses in SOF are adjusted for self-reported ethnicity; analyses in CHS are adjusted for sex. Nominally significant associations (P < 0.05) are highlighted in bold.

AtherosclerosisSOFP0.0690.0130.017
OR (95%CI)1.33 (0.98–1.81)0.68 (0.51–0.92)0.69 (0.51–0.94)
CHSP0.0400.490.73
OR (95%CI)1.38 (1.02–1.89)1.10 (0.84–1.45)1.05 (0.79–1.40)
Meta-analysisP0.00620.250.17
OR (95%CI)1.36 (1.09–1.69)0.89 (0.73–1.09)0.87 (0.70–1.07)
CancerSOFP0.0870.0450.12
OR (95%CI)1.28 (0.97–1.67)0.76 (0.58–0.99)0.81 (0.62–1.06)
CHSP0.370.380.16
OR (95%CI)1.14 (0.85–1.53)0.89 (0.70–1.15)0.83 (0.64–1.08)
Meta-analysisP0.0640.0440.037
OR (95%CI)1.21 (0.99–1.48)0.83 (0.69–0.99)0.82 (0.68–0.99)
OtherSOFP0.100.320.48
OR (95%CI)1.30 (0.95–1.78)0.86 (0.64–1.16)0.90 (0.66–1.21)
CHSP0.0360.0350.048
OR (95%CI)1.34 (1.02–1.75)0.76 (0.58–0.98)0.76 (0.58–1.00)
Meta-analysisP0.00790.0250.051
OR (95%CI)1.32 (1.08–1.62)0.80 (0.66–0.97)0.82 (0.67–1.00)

Discussion

We have comprehensively evaluated common variants in 30 genes in the insulin/IGF1 signaling pathway for their associations with human lifespan in a cohort of elderly Caucasian women. Replication of nominally significant findings in two independent Caucasian cohorts revealed a SNP in AKT1, which was significantly associated with lifespan, and two SNPs in FOXO3A, which were significantly associated with lifespan in women. Overall, the longevity-associated SNPs were associated with the three major categories (cardiovascular, cancer, other) of cause of death, suggesting that their association with lifespan is due to a general effect rather than to a specific disease process (Table 5).

Several previous studies have demonstrated associations between variants in other genes in the insulin/IGF1 signaling pathway and human aging phenotypes. Most of these studies investigated only a small number of SNPs or genes, and their results have not been replicated. Our study analyzed 291 SNPs in 30 genes providing the first comprehensive screen of the insulin/IGF1 signaling pathway in humans.

Association of genetic variation in AKT1 with human lifespan has not been previously reported; however several studies have reported associations with FOXO3A SNPs or haplotypes. A recent case–control association study in elderly Japanese men found an association with three SNPs in FOXO3A and longevity (Willcox et al., 2008); this study did not report replication of findings in independent cohorts. While the tag SNPs genotyped were different in our study, two of the three associated SNPs share strong LD (r2 > 0.8 in HapMap Caucasians) with the FOXO3A SNPs associated with longevity in our study. The effect size is larger in the Japanese cohort, but, as in our study, homozygosity for the minor allele is associated with longevity. The most strongly associated SNP from the Japanese cohort lies in the first intron and has lower LD in Caucasians (r2 = 0.48) than in Japanese (r2 = 0.71) with the top SNPs in our study, which are located in the last intron; this difference in LD patterns may account for the different localization of the strongest signal in our two studies.

The population-based Leiden 85-plus study (van Heemst et al., 2005) also found a FOXO3A haplotype, but no single SNPs, associated with an increase in mortality (Kuningas et al., 2007) in Caucasians. The FOXO3A haplotype identified contains some of the SNPs we found nominally associated with longevity in SOF, but resides in a different part of the gene from the two FOXO3A SNPs we found associated with longevity in our three-cohort meta-analysis. Taken together, these results suggest that 5 independent cohorts may be detecting an association between FOXO3A variants and longevity. A combined analysis in which all cohorts genotype the same SNPs will be useful to confirm whether these findings are true replications of the same effect, and to identify the region of the gene with the strongest signal to prioritize fine mapping efforts.

Other insulin/IGF1-pathway genes implicated in human longevity include FOXO1A and IGF1R. A genome-wide association scan in the Framingham Study showed association between age at death and two FOXO1A SNPs (Lunetta et al., 2007), which are in complete LD with FOXO1A rs1334241, the SNP we found to be nominally associated with longevity in SOF. However, this finding did not replicate in CHS or AJC. In the AJC study, rare nonsynonymous mutations of IGF1R were significantly more common in female centenarians (Suh et al., 2008). Female centenarians carrying those mutations had lower IGF1R levels and decreased IGF signaling compared with other female centenarians (Suh et al., 2008). In a study of long-lived Italians, those carrying a common variant allele at the IGF1R locus had lower plasma IGF1 levels than those without the allele; this allele was over-represented in long-lived Caucasians (Bonafe et al., 2003). We genotyped this variant but did not replicate the association with lifespan in the SOF cohort.

The association of lifespan with FOXO3A variants in several studies and the novel association we identified with a SNP in AKT1 strongly suggest that the same components of the insulin/IGF1 signaling pathway that affect lifespan in C. elegans, Drosophila and mice are also involved in determining lifespan in humans. FOXO3A is one of the three mammalian homologues of the C. elegans transcription factor DAF-16, which is required for the lifespan effects of the insulin/IGF1 signaling pathway. AKT1 is one of the tyrosine kinases in the signaling cascade that links IGF1 binding to its receptor to transcriptional activation and downstream effects on cell metabolism and oxidative stress response; AKT1 phosphorylates FOXO3A and prevents it from entering the nucleus thus inhibiting transcriptional activation (Brunet et al., 1999; Zheng et al., 2000). Thus, the two genes most strongly associated with longevity in our results are also functionally related (Figure 1).

How the variants we identified affect lifespan remains uncertain. All three longevity-associated SNPs that replicated across three independent cohorts are intronic SNPs of no known function, and they do not reside in motifs conserved through evolution. In the case of FOXO3A, the two longevity-associated SNPs are in LD with other variants across an area of approximately 67 kb (Fig. S1). As the LD is so strong in this area, it may not be possible to discern which is causative from association studies in Caucasians alone; association studies in other populations, locus re-sequencing and functional studies are needed.

The AKT1 SNP associated with lifespan is not in high LD with any other HapMap variants; evaluation of eight unlinked variants in the nearby area, including five amino-acid coding polymorphisms, did not reveal any further associations with lifespan. As AKT1 rs3803304 resides 70 bp away from an intron–exon boundary, and in an area of high-predicted regulatory potential (King et al., 2005), which potentially encodes a noncoding regulatory RNA (Kapranov et al., 2007), it may be a candidate SNP for influencing RNA splicing or other regulatory mechanisms. In vitro functional studies will be needed to explore such potential mechanisms.

Our study has several potential limitations. First, although we adjusted for self-reported ethnicity in the SOF study, population stratification may have affected our results. In addition, we did not have self-reported ethnicity information in CHS. Second, subtle effects on longevity might have been missed because of our relatively small sample size (896 participants in SOF); we had adequate (80%) power to detect an odds ratio of ∼1.7. As with other complex traits (Easton et al., 2007; Frayling, 2007; Harley et al., 2008), much larger sample sizes will be required to identify modest risk modifiers. Third, like most studies of longevity in humans, our study may be influenced by birth cohort effects, as controls were generally born later (e.g. in the 1910s–1920s) than the cases (in the 1890s–1900s). Finally, we only tested common SNPs and small insertion-deletions. We did not genotype copy number variants or large insertion-deletions and, with the exception of a few amino-acid coding variants, we only evaluated the effects of common variants (minor allele frequency > 0.05). The relative dearth of common SNPs associated with longevity in the genes we analyzed does not exclude the existence of other genetic effects in the pathway. To obtain a complete picture of the association of genetic variation in the insulin/IGF1 pathway with longevity phenotypes, discovery of rare variants by re-sequencing of the same set of candidate genes will be necessary.

Despite these limitations, replication among three independent Caucasian samples, including a founder population (AJC), strengthens our findings. It is unlikely that population stratification could lead to false-positive results that would be consistent across a European American/Caucasian American population and the Ashkenazi population, as the patterns of stratification would be different. Any birth cohort effects should be relatively the modest, as the differences in birth-year were small, and because both groups were born before there were substantial changes in early-life mortality rates due to advances in medical care.

In summary, we screened the insulin/IGF1 signaling pathway for association with human longevity and identified common variants in two genes, AKT1 and FOXO3A, which are consistently associated with longer lifespan in three independent Caucasian cohorts. Our results extend the remarkably consistent effect of this pathway on lifespan in model organisms to human populations. Although the effects we observed are subtle, they implicate the insulin/IGF1 signaling pathway in human aging, confirming that studies in model organisms can help us to gain insights into the biology of human aging.

Experimental procedures

Samples

Data from three studies were analyzed. Two are large cohort studies of older adults: SOF and CHS. The third study is a case–control study (AJC).

Study of Osteoporotic Fractures (SOF)

A total of 9704 ambulatory community-dwelling Caucasian women age 65 and older were recruited from September 1986 through October 1988 from four areas in the United States: Baltimore, MD: Monongahela Valley, PA; Minneapolis, MN; and Portland, OR (Cummings et al., 1990). The study protocol was approved by the appropriate institutional review committees and written informed consent was obtained from all participants. Self-reported ethnicity was recorded as Northern, Southern, or Central European, or mixed/other. The cohort has been followed for over 20 years with age and cause of death recorded for participants who have died. Eight hundred and ninety-six SOF participants were selected for this study. These included 293 long-lived cases (lifespan ≥ 92 years, mean ± SD = 95.3 ± 2.2 years) of whom approximately 56% were still living at the last study visit. We selected 603 controls who were the youngest to die in this cohort (lifespan ≤ 79 year, mean = 75.7 ± 2.6 year); however, the age of death in the controls is not substantially different compared to the general population, as all women had to be relatively healthy to enter the cohort (Cummings et al., 1990). Cause of death was adjudicated by an experienced internist based on review of death certificates and hospital records.

Cardiovascular Health Study (CHS)

The original CHS cohort includes 5201 Caucasian and African-American participants recruited from 1989 to 1990 in four communities in the United States (Fried et al., 1991; Tell et al., 1993) and 684 additional African-American participants recruited from 1992 to 1993. All participants were 65 years of age or older. At enrollment, participants were not institutionalized, wheelchair-bound at home, or receiving treatment for cancer. Informed consent was obtained from all participants. Deaths were ascertained through surveillance and at semiannual contacts. For all deaths, the CHS collected information from medical records, physician interviews, autopsy reports, coroner reports, and next of kin interviews. Data were reviewed by the CHS Events Committee and cause of death classified according to prespecified criteria (Ives et al., 1995). From the 4383 Caucasian CHS participants, we defined a case–control replication sample of 1076 for the longevity analysis using the same age cutoffs as in the SOF study. This subset included 279 long-lived cases (lifespan ≥ 92 year, mean = 94.5 ± 2.1 years) and 797 average-lifespan controls (lifespan ≤ 79 year, mean = 76.1 ± 3.1 year).

Ashkenazi Jewish Centenarians (AJC)

Subjects were of self-reported Ashkenazi Jewish ancestry (Barzilai et al., 2003; Atzmon et al., 2006; Suh et al., 2008). Informed written consent was obtained in accordance with the policy of the Committee on Clinical Investigations of the Albert Einstein College of Medicine. The 383 cases (74.8% female) were required to be living independently at 95 years of age as a reflection of good health (mean age 97.7 year, range 95–108 year). The 363 controls (mean age 79.5 years, age range 43–94 year, 57% female), all living at time of recruitment, were of self-reported Ashkenazi Jewish descent without a family history of unusual longevity (parents’ age < 85 year). Participants’ ages were defined by birth certificates or passports. Four percentage of controls and 76% of centenarians have died since recruitment.

Gene and variant selection

Variants were selected for genotyping in 30 genes encoding the key proteins (Figure 1) in the human insulin/IGF1 signaling pathway, including ligands (IGF1, insulin, GH1, GHRH), binding proteins (IGFBP1, IGFBP2, IGFBP3, IGFBP4, IGFBP5, IGFBP6), receptors (IGF1R, INSR, INSRR, GHR, GHRHR), signal transduction proteins (AKT1, AKT2, AKT3, IRS1, IRS2, IRS4, PDPK1, PIK3CA, PIK3CB, PIK3R1, PIK3R2, PTEN, SHC1), and transcription factors (AFX, FOXO1A, FOXO3A). GH1 was included initially, but all three SNPs available in dbSNP failed genotyping attempts. Initially, variants in FOXO1A and FOXO3A were identified by sequencing exons, intron–exon boundaries, and the promoter in 80 Caucasians and selecting haplotype-tagging variants. Subsequently, we utilized the International HapMap Consortium data from CEU Caucasians (The International HapMap Consortium 2005; Frazer et al., 2007) (http://www.hapmap.org) to select haplotype-tagging SNPs. Briefly, tag SNPs were picked for the gene ± approximately 10 kb genomic flanking sequence using the Tagger algorithm (de Bakker et al., 2005) implemented in the Haploview program (Barrett et al., 2005) (http://www.broadinstitute.org/mpg/haploview). The entire genomic sequence was tagged for all genes except IGF1R and INSR, where most of the large intron 1 was excluded. We set the r2 cutoff at 0.8 and minor allele frequency cutoff at 0.05 (except for amino-acid coding variants, which were all included). Pair-wise tag selection was used for all genes except IGFBP2, IGFBP5, IRS2, and PIK3R1, where two-marker haplotypes were used to reduce the number of tag SNPs. Amino-acid coding variants were prioritized and previously published candidate variants were included (Bonafe et al., 2003). Eight additional SNPs in AKT1 were selected for follow-up. We selected 339 SNPs overall; 293 SNPs were successfully genotyped.

To analyze the genomic regions surrounding FOXO3A and AKT1, we used the UCSC Genome Browser (http://www.genome.ucsc.edu) including data from several analysis tracks (ESPRR regulatory potential (King et al., 2005), Phastcons conservation, CpG islands, Affymetrix Transcriptome Phase 3 (Kapranov et al., 2007)) on the March 2006 assembly of the human genome.

Genotyping

Single nucleotide polymorphisms were genotyped in the SOF and CHS cohorts using two genotyping platforms: template directed primer extension with fluorescence polarization detection (FP-TDI, AcycloPrime II detection kit, Perkin Elmer, Boston, MA, USA) (Hsu & Kwok, 2003) and SNPstream 48plex (Beckman Coulter, Fullerton, CA, USA). Oligonucleotide primers were designed using http://www.autoprimer.com, primer3 (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi), and as previously described (Hsu & Kwok, 2003). Assay conditions are available upon request. Genotyping in AJC was performed at the Albert Einstein College of Medicine using the Sequenom platform and pyrosequencing.

Samples were genotyped by investigators blinded to phenotype. All data were scored by at least two investigators. DNA samples were genotyped on 384-well plates with negative and positive controls (duplicates) included in each plate. There were 19 duplicates among the SOF and 267 among the CHS samples. Plates with fewer than 90% successful genotypes or with duplicate discrepancies were repeated. Plates where duplicate discrepancies persisted after re-genotyping were excluded. Overall genotyping accuracy was > 99.5%.

Single nucleotide polymorphisms genotype distributions in SOF controls were tested for consistency with Hardy–Weinberg equilibrium using a γ2 test (Bonferroni-corrected significance threshold P = 1.7E−04). 2/293 SNPs, IRS4 rs1801162 and IGFBP2 rs9341156, were not in Hardy–Weinberg equilibrium and were removed from further analysis.

Statistical analyses

Association of SNP genotype with longevity

For all 291 SNPs analyzed, minor allele homozygotes were coded 2, heterozygotes were coded 1, and major allele homozygotes were coded 0. Logistic regression was used to evaluate the association between genotype and longevity case/control status. We assumed an additive model as our baseline model; significant associations were also tested with a genotypic model. In SOF, the logistic regression model was adjusted for self-reported ethnicity. In CHS and AJC, models were adjusted for sex; we also analyzed data separately for women in these two cohorts. Association of significant genotypes with the three main categories of cause of death (cardiovascular disease, cancer, or other) was evaluated in the two cohorts (SOF and CHS) that had this information. OR and 95%CI are reported. We also performed haplotype case–control association analyses using two- and three-SNP sliding window and haplotype analysis across entire LD-blocks using the Haploview program (Barrett et al., 2005).

Replication and meta-analysis

The logistic regression results from all three cohorts (or from two cohorts only in some cases) were combined using inverse variance weighting. To evaluate statistical significance in the replication, we corrected for all of the SNPs we replicated and used only the data from the two replication cohorts (AJC + CHS) to assess significance. We determined statistical significance by permutation. Case/control status was permuted randomly for all participants for the 25 SNPs genotyped in the replication stage. In each permutation, we recorded the maximum absolute value of the Z scores. We performed 2000 permutations. We then compared the Z scores from the observed data with the distribution of the maximum Z scores from the permutations. The percentile of the observed absolute z score among the permuted z scores was calculated as the adjusted P-value for determining statistical significance corrected for multiple testing. Statistical significance in the replication was defined as an adjusted P < 0.05 in the AJC + CHS replication cohorts.

Acknowledgments

We acknowledge D. Lind for coordination of pilot studies, J. Atilano, S.-W. Chan, E. Lovins, and T.J. Nguyen for technical assistance and the staff of SOF and CHS Coordinating Centers for study support. This work was supported by the Longevity Consortium, http://www.longevityconsortium.org (National Institute on Aging (NIA) 5U19AG023122 to S.R.C. and W.S.B.). The SOF is supported by the NIA (AG05407, AR35582, AG05394, AR35584, AR35583, R01 AG005407, R01 AG027576-22, 2 R01 AG005394-22A1, and 2 R01 AG027574-22A1). The CHS is supported by the National Heart, Lung, and Blood Institute (N01-HC-85079 through N01-HC-85086, N01-HC-35129, N01 HC-15103, N01 HC-55222, N01-HC-75150, N01-HC-45133, U01 HL080295) with additional contribution from the National Institute of Neurological Disorders and Stroke. A full list of principal CHS investigators and institutions can be found at http://www.chs-nhlbi.org/pi.htm.

Author contributions

Ludmila Pawlikowska: Study design, acquisition of data, interpretation of data, preparation of manuscript. Donglei Hu: Analysis and interpretation of data, preparation of manuscript. Scott Huntsman: Analysis and interpretation of data. Andrew Sung: Acquisition of data, preparation of manuscript. Catherine Chu: Acquisition of data, preparation of manuscript. Justin Chen: Acquisition of data, preparation of manuscript. Alexander H. Joyner: Interpretation of data, preparation of manuscript. Nicholas J. Schork: Interpretation of data, preparation of manuscript. Wen-Chi Hsueh: Interpretation of data, preparation of manuscript. Alexander P. Reiner: Interpretation of data, preparation of manuscript. Bruce M. Psaty: Interpretation of data, preparation of manuscript. Gil Atzmon: Interpretation of data, preparation of manuscript. Nir Barzilai: Interpretation of data, preparation of manuscript. Steven R. Cummings: Study design, interpretation of data, preparation of manuscript. Warren S. Browner: Study design, interpretation of data, preparation of manuscript. Pui-Yan Kwok: Interpretation of data, preparation of manuscript. Elad Ziv: Study design, analysis and interpretation of data, preparation of manuscript.

Ancillary