Association with longevity of phosphatidylinositol 3-kinase regulatory subunit 1 gene (PIK3R1) variants stems from protection against mortality risk in men with cardiovascular disease

Abstract

Introduction:

Genetic variation in the phosphatidylinositol 3-kinase reregulatory subunit 1 gene (PIK3R1) is associated with longevity.

Objective:

To determine whether cardiovascular disease (CVD) affects this association.

Methods:

We performed a longitudinal study of longevity-associated PIK3R1 single nucleotide polymorphism rs7709243 genotype by CVD status in 3,584 elderly American men of Japanese ancestry.

Results:

At baseline (1991–1993), 2,254 subjects had CVD, and 1,314 did not. Follow-up until Dec 31, 2019 found that, overall, men with a CVD had higher mortality than men without a CVD (P=1.7×10⁻⁵). However, survival curves of CVD subjects differed according to PIK3R1 genotype. Those with longevity-associated PIK3R1 TT/CC had survival curves similar to those of subjects without a CVD (P=0.11 for TT/CC, and p=0.054 for TC), whereas survival curves for CVD subjects with the CT genotype were significantly attenuated compared with survival curves of subjects without a CVD (p=0.0000012 compared with TT/CC, and p=0.0000028 compared with TC). Men without CVD showed no association of longevity-associated genotype with lifespan (p=0.58). Compared to subjects without any CVD, hazard ratios for mortality risk were 1.26 (95% CI, 1.14–1.39; p=0.0000043) for CT subject with CVD, and 1.07 (95% CI 0.99–1.17; p=0.097) for CC/TT subjects with CVD. There was no genotypic effect on lifespan for 1,007 subjects with diabetes, and 486 with cancer.

Conclusion:

Our study provides novel insights into the basis for PIK3R1 as a longevity gene. We suggest that the PIK3R1 longevity genotype attenuates mortality risk in at-risk individuals by protection against cellular stress caused by CVD.

Introduction

Longevity can be accompanied by either healthy aging (i.e., an absence of life-threatening diseases, chronic conditions), or unhealthy aging, in which elderly individuals live long lives with chronic conditions. Genetic factors make a 15–40% contribution to lifespan [1], and may be as high as 48% in the oldest-old [2]. At very old age (> 90 years) specific longevity genes may dominate over environmental influences in determination of eventual lifespan, with evidence that there is a stronger effect in males [3]. There are a number of inherited “risk” genes that prevent living past the period of chronic disease of middle age and do not qualify as “longevity” genes.

Phosphatidylinositol 3-kinases (PI3Ks) are lipid kinases that regulate key intracellular signaling pathways and vesicle trafficking by generating phosphatidylinositol-3,4,5-tris phosphate (PIP₃) [4]. They comprise 8 isoforms spanning 3 classes. These are expressed broadly in tissues and have roles in cardiovascular physiology and cardiovascular disease (CVD), in particular platelet function and thrombosis [4]. Class 1a PI3K is an essential mediator of the metabolic actions of insulin and growth [5]. PI3K is composed of catalytic subunit (p110α, β, or δ) and regulatory subunit (p85α, p85β, or p55γ). The catalytic subunit (p110) can be referred to as a “tumor suppressor”, while the regulatory subunit (p85) as an “oncogene” [6]. Mutations in the PIK3 regulatory subunit α gene (PIK3R1) have been associated with SHORT syndrome (short stature, hyperextensibility, hernia, ocular depression, Rieger syndrome, and teething delay) [7], immunodeficiency [8], agammaglobulinemia-7 [9], insulin resistance [10], and cancer [11]. The p85α regulatory subunit of PI3KR1 modulates the cellular response to endoplasmic reticulum (ER) stress by promoting nuclear translocation of X-box binding protein 1 (XBP1) isoform 2 in an ER stress- and/or insulin-dependent manner during metabolic overloading in the liver and thereby improves glucose tolerance [12].

We showed previously an association with longevity for alleles of 3 adjacent SNPs within PIK3R1 (rs7713645-C, rs6881033-A, rs7709243-T; p=0.00408, p=0.00569, p=0.00083, respectively) in men in the Kuakini Honolulu Heart Program/Kuakini and Honolulu-Asia Aging Study (KHHP/KHAAS) cohort [13].

The present study addresses the hypothesis that the PIK3R1 longevity-associated alleles increase cell and organism resilience by protecting against cellular stress caused by CVD and possibly other chronic conditions of aging. To test this hypothesis, we compared the effect of longevity-associated alleles with lifespan in subjects with and without CVD (hypertension, or coronary heart disease (CHD), or cerebrovascular accident (CVA; stroke)), type 2 diabetes, and cancer. We examined genotypes of our sentinel SNP, rs7709243, in KHHP/KHAAS subjects in a longitudinal study with up to 29 years follow-up.

Methods

Study Cohort

Participants were American men of Japanese ancestry living on the island of Oahu, Hawaii. They were recruited in 1965–1968 from World War II Selective Service records for the Kuakini Honolulu Heart Program (KHHP) [14], which continued from 1991 onwards as the Kuakini and Honolulu-Asia Aging Study (HAAS) [14–17]. The analysis was conducted as part of the Kuakini Hawaii Lifespan Study and the Kuakini Hawaii Health-span Study, an embedded cohort study of healthy aging drawn from the original KHHP-KHAAS population. Subjects had parents who were both from a limited geographic area of Japan, mostly the western, central and southern regions [14,18]. They were recruited at the same time and place (Oahu), meaning there was no apparent reason why genetic background should be substantially different. The KHHP cohort is quite robust for phenotype-genotype associations, since the data collection was exceptionally accurate, and involved cross validation utilizing an expert Morbidity and Mortality Committee. The Hawaiian Japanese population is from Japan, with little outbreeding and, based on the authors’ unpublished data, exhibits a smaller degree of genetic diversity than the overall population of Japan.

All participants in the current study were interviewed at Examination 4 of the KHHP (1991–1993). Archived phenotypic data and blood samples from Examination 4 of the KHHP (1991–1993), which coincided with the commencement of the KHAAS, were used as the baseline examination for our study. The HAAS was begun as an expansion of the KHHP for the study of neurodegenerative diseases, cognitive function, and other aging phenotypes in elderly persons. During 1991–1993, all of the survivors of the KHHP cohort, ranging in age from 71–93 years (mean age: 77.9 ± 4.7 years), were invited to the 4th examination. Response rate was 80% of survivors (including clinic, home, and nursing home visits; n = 3,741).

The study involved 3,568 of 3,741 men aged 71 to 93 (mean age 77.7 ± 4.6 SD years) for whom we had banked DNA, so making them eligible for inclusion. Of the 3,568 men, 3,533 had died (mean age at death 88.5±6.1 SD years; range 72–107 years) and 35 were still alive (mean age 101.7 ± 1.9 SD years; range 100–108 years) at the end of the follow-up period, December 31, 2019.

The KHHP was a longitudinal observation study. Subjects from the KHHP/KHAAS population [15] had been followed with regular examinations and blood work until 2015, or death, up to the end of 2019.

Information on the prevalence of CHD, stroke, and cancer was identified by the KHHP surveillance system (review of hospital records by an expert panel or matching to Tumor Registry for Cancer). Hypertension was defined as systolic/diastolic blood pressure ≥160/95 mmHg or on anti-hypertensive medication at baseline. Diabetes was defined by fasting serum glucose ≥126 mg/dL or 2-hour post-load glucose ≥200 mg/dL or taking insulin and/or oral hypoglycemic medications at baseline.

Procedures performed were in accord with institutional guidelines and were approved by the Institutional Review Board of Kuakini Medical Center. Written informed consent was obtained from all study participants or from family representatives, if participants could not provide consent.

Genotyping

Leucocyte DNA obtained from participants was used for genotyping. Recruitment, demographic characteristics, and DNA extraction were as described previously [15,19]. Genotyping was performed by allelic discrimination assays using TaqMan^® (Applied Biosystems, Inc.) and a Life Technologies QuantStudio 12K Flex OpenArray system. Genotype data were managed through an integrated database sample management-data processing system of proven accuracy. All positive controls on each genotyping plate were evaluated for consistency. SNP call rates exceeded 98%. We implemented a series of quality control checks based on the Illumina metrics. For inclusion of data for a SNP its call rate had to exceed 0.95 and the Hardy-Weinberg equilibrium p value needed to be >0.01. Although 88% of participants were born in Hawaii, there is a theoretical possibility of confounding of case vs. control status for allele frequencies due to geographic origin. Therefore, for certain analyses, cases and controls were stratified by parental prefecture of origin using conditional logistic regression models.

Statistical Analyses

General linear models were used to compare age-adjusted indirect measurements between groups, and logistic models was used to compare the age-adjusted direct measurements. Cox proportional models were used to assess the association of PIK3R1 genotype on mortality stratified by disease status, such as by CVD, diabetes, and cancer, and by chronic conditions defined by presence of any of these. The Cox proportional hazard assumption was tested for each Cox model. The effect of interaction of disease with PIK3R1 genotype on mortality was tested in the Cox model. All statistical analyses were performed using the Statistical Analysis System version 9.4 [20]. Figures were generated using the STATA 12 Graphics [21].

Results

Characteristics of Subjects

Supplementary Table S1 shows age-adjusted baseline (1991–1993) characteristics of long-lived and average lifespan men in our prior case-control study of PIK3R1 genotypes [13]. By December 31, 2019, 3,533 out of 3,568 of the subjects had died during the 29 years of follow-up from exam 4. At exam 4 (baseline) among the 3,568 participants, 53% had hypertension, 21% had CHD, and 4.5% had cardiovascular accident (stroke) history (collectively, cardiovascular disease, CVD). Overall, 63% had at least one CVD. Proportion of the cohort with diabetes was 29%, and with cancer was 14%. Mean age at death was 88.5 ± 6.1 years for those with at least one CVD, and 89.3 ± 6.0 years for those who did not have a CVD (p = 0.00040). Previous analyses have shown that subjects in the KHHP/KHA cohort have an absence of population stratification and sub-structure [15].

Effect of Genotype on Mortality

Genotype frequencies of PIK3R1 SNPs in our cohort were similar to those in the population of Japan (Suppl. Table S2). The p-values for association with longevity in original case-control study by a major allele recessive model are shown in Suppl. Table S3. In the present longitudinal study, we examined SNP rs7709243, since it had the most statistically significant association with longevity in our case-control study [13]. Suppl. Table S4 shows baseline characteristics for each genotype of PIK3R1 SNP rs7709243 for our subjects. In Suppl. Table S5 baseline characteristics were divided into those who had a CVD and those who did not have a CVD. Table 1 shows age-adjusted baseline characteristics of men with CVD and men without a CVD for longevity-associated genotypes using a heterozygote disadvantage model; CC/TT and CT genotype, the latter genotype is not associated with longevity.

Table 1.

Age-adjusted baseline characteristics by CVD and PIK3R1 rs7709243 genotype

		With CVD			Without CVD
Variable	TC	TT/CC	P	TC	TT/CC	P
n	798	1456		477	837
Age (years)	77.6 ± 4.6	78 ± 4.6	0.05	77.4 ± 4.7	77.6 ± 4.6	0.38
BMI (kg/m2)	23.7 ± 3.1	23.7 ± 3	0.61	22.9 ± 3.2	23.1 ± 3	0.18
Fasting glucose (mg/dl)	115.8 ± 33.4	114.9 ± 29.3	0.50	108.8 ± 25.5	109.7 ± 27.3	0.56
Smoking (pack-years)	29.4 ± 36.8	26.5 ± 34	0.068	25.1 ± 33.3	23 ± 31.3	0.28
Alcohol intake (oz/mo)	19.2 ± 42	20.5 ± 44.4	0.52	17.9 ± 36	16.4 ± 34.3	0.47
Physical activity index	30.9 ± 4.4	30.8 ± 4.7	0.51	31 ± 4.4	31 ± 4.5	1.00
Depression	10.7	10.4	0.83	10.1	10.8	0.69
Diabetes (%)	33.9	32.5	0.47	17.9	23	0.031
Cancer (%)	13.4	12.8	0.70	12.4	15.2	0.16
CHD (%)	32.4	32.7	0.92	−	−	−
Stroke (%)	7.1	6.8	0.81	−	−	−
Hypertension (%)	84.2	84.8	0.71	−	−	−

Values shown are age-adjusted mean ± SD for indirect measures and proportion for direct measurements, except age.

Table 2 shows hazard ratios (HR) of PIK3R1 rs7709243 genotypes for total mortality by CVD, as well as diabetes and cancer for the 4 possible genetic models. The heterozygous disadvantage model and the major allele recessive model showed statistically significant associations by two Cox models with covariate adjustment. However, only the heterozygous disadvantage model remained significant after Bonferroni correction for multiple testing. Thus, the heterozygous disadvantage model was chosen for further analyses.

Table 2.

Hazard ratios (HR) of PIK3R1 rs7709243 genotypes for total mortality by diabetes, cancer and CVD for the 4 possible genetic models.

			With the disease		Without the diseases
Disorder (n with, n without)	Cox model *	Genetic model**	HR (95% CI)	p	HR (95% CI)	p
Diabetes	1	TC vs. TT/CC	1.05 (0.92–1.19)	0.49	1.11 (1.03–1.21)	0.0098
(1007, 2512)	2	TC vs. TT/CC	1.01 (0.87–1.16)	0.93	1.10 (1.01–1.20)	0.038
Cancer	1	TC vs. TT/CC	1.13 (0.93–1.36)	0.22	1.09 (1.01–1.17)	0.026
(486, 3082)	2	TC vs. TT/CC	1.20 (0.97–1.48)	0.095	1.06 (0.97–1.15)	0.19
CVD	1	TC vs. TT/CC	1.17 (1.07–1.28)	0.00038	0.98 (0.88–1.10)	0.78
(2254, 1314)	2	TC vs. TT/CC	1.17 (1.06–1.28)	0.0017	0.96 (0.84–1.08)	0.47
Diabetes	1	TC/CC vs. TT	1.04 (0.91–1.17)	0.59	1.09 (1.01–1.18)	0.035
(1007, 2512)	2	TC/CC vs. TT	1.01 (0.88–1.16)	0.88	1.07 (0.98–1.17)	0.13
Cancer	1	TC/CC vs. TT	1.10 (0.92–1.32)	0.31	1.07 (1.00–1.15)	0.07
(486, 3082)	2	TC/CC vs. TT	1.12 (0.91–1.38)	0.27	1.04 (0.96–1.13)	0.32
CVD	1	TC/CC vs. TT	1.14 (1.05–1.24)	0.0027	0.98 (0.88–1.10)	0.76
(2254, 1314)	2	TC/CC vs. TT	1.11 (1.01–1.22)	0.024	0.97 (0.86–1.10)	0.66
Diabetes	1	TT/TC vs. CC	1.04 (0.79–1.35)	0.80	1.08 (0.90–1.28)	0.41
(1007, 2512)	2	TT/TC vs. CC	0.98 (0.74–1.30)	0.90	1.11 (0.91–1.36)	0.28
Cancer	1	TT/TC vs. CC	1.06 (0.73–1.53)	0.76	1.06 (0.90–1.24)	0.50
(486, 3082)	2	TT/TC vs. CC	1.24 (0.80–1.91)	0.34	1.06 (0.89–1.26)	0.54
CVD	1	TT/TC vs. CC	1.08 (0.90–1.30)	0.39	1.01 (0.80–1.28)	0.94
(2254, 1314) Diabetes (1007, 2512) Cancer (486, 3082) CVD (2254, 1314)	2 1 2 1 2 1 2	TT/TC vs. CC Additive Additive Additive Additive Additive Additive	1.18 (0.96–1.45) 1.02 (0.92–1.13) 1.01 (0.90–1.13) 1.05 (0.91–1.21) 1.04 (0.88–1.22) 1.07 (1.00–1.15) 1.05 (0.97–1.13)	0.11 0.74 0.87 0.50 0.64 0.038 0.24	0.93 (0.71–1.20) 1.05 (0.98–1.12) 1.03 (0.96–1.11) 1.04 (0.98–1.10) 1.02 (0.96–1.09) 0.99 (0.90–1.08) 0.99 (0.90–1.10)	0.56 0.16 0.41 0.21 0.56 0.78 0.88

^*Cox models: Model 1 – adjusted for age. Model 2 – adjusted for age, BMI, glucose, smoking (pack-year), alcohol consumption (ounces per month), physical activity index (PAI), and depressive symptoms

^**Genetic models (in order): heterozygous disadvantage, minor allele dominant, major allele dominant, and minor allele additive.

Shown are results for 4 genetic models for 3 diseases (diabetes, cancer and cardiovascular disease (CVD, i.e., at least one of hypertension, CHD, and CVA (stroke)), thus total of 24 tests for covariate models.

P_B=0.0017*24 = 0.041 for heterozygote disadvantage model of CVD in covariate adjusted model. No other genetic models were statistically significant after Bonferroni’s correction.

Survival curves for men with a CVD and those without a CVD showed that the latter lived longer (Kaplan-Meier Log-rank χ² = 18.5; p = 1.7×10^–5). In Figure 1 we show survival curves according to genotype for these subjects according to a heterozygote disadvantage model: TC vs. TT/CC (where the frequency of the longevity-associated (T) allele was 10-times that of the minor (C) allele, so makes a greater genotypic contribution to any genotypic effect). These curves were determined using a Cox proportional hazard model adding an interaction term of PIK3R1 with CVD. Only in CVD subjects did the TT/CC genotype have an effect on longevity.

Figure 1.

Survival curves spanning the period from baseline (1991–1993) to Dec 31, 2019 for subjects with and without a CVD according to whether they had or did not have longevity-associated genotypes of SNP rs7709243. The survival probabilities were estimated from the Cox proportional hazard model (see Methods) h(t) = h(t0) * exp(β1*Age + β2*BMI + β3*glucose + β4*CVD + β5*PIK3R1_TC + β6*(CVD*PIK3R1_TC)) by fixing age at 75 years, BMI at the mean, 23.5 kg/m², and glucose at the mean, 113 mg/dL (where β6 is the effect of the interaction of CVD with PIK3R1_TC genotype (TC vs. TT/CC genotype) on mortality, giving P(β6) = 0.0050). The p-values for comparison of survival curves for PIK3R1_TT/CC and for PIK3R1_TC for those with a CVD versus those without any CVD, were P=0.24 and P=0.000069, respectively.

Effectively, lifespan of TT/CC subjects with CVD did not differ significantly from lifespan of subjects without CVD, whereas lifespan of TC subjects with a CVD was significantly attenuated. Furthermore, men without a CVD showed no significant difference in lifespan regardless of whether they had or did not have longevity-associated alleles of rs7709243. Figure 2 shows hazard ratios for mortality risk in subjects with a CVD according to whether they were carriers of the longevity-associated (TT/CC) genotype or the TC genotype, when compared with subjects who did not have a CVD. In men with a CVD and the PIK3R1 longevity genotype TT/CC, mortality risk did not differ from that of men without a CVD (HR 1.07; p= 0.097; Table 3).

Figure 2.

Mortality risk expressed as hazard ratio, adjusted for age, BMI, glucose, smoking (pack-years), alcohol consumption (oz/mo), physical activity index, diabetes and cancer for CVD subjects and subjects without a CVD according to PIK3R1 SNP rs7709243 genotype in (a heterozygous disadvantage model, TT/CC vs. CT, where T is the major allele. In men with CVD and the PIK3R1 longevity genotype, mortality risk was reduced to normal (i.e., did not differ significantly from that of men without CVD).

Table 3.

Hazard ratios (HR) of PIK3R1 rs7709243 homozygotes vs. heterozygotes with total mortality by CVD.

		With CVD (n=2254)		Without CVD (n=1314)
Cox model*	Genetic model**	HR (95% CI)	p	HR (95% CI)	p
1	TT vs. TC	0.86 (0.78–0.94)	0.00056	1.02 (0.91–1.14)	0.77
	CC vs. TC	0.84 (0.69–1.01)	0.065	1.00 (0.78–1.28)	0.99
2	TT vs. TC	0.87 (0.79–0.96)	0.0041	1.03 (0.91–1.17)	0.65
	CC vs. TC	0.75 (0.61–0.94)	0.010	1.11 (0.84–1.45)	0.46

^*Cox models: Model 1 – adjusted for age. Model 2 – adjusted for age, BMI, glucose, smoking (pack-year), alcohol consumption (ounces per month), physical activity index (PAI), and depressive symptoms

^**Genetic models (in order): heterozygous disadvantage, minor allele dominant, major allele dominant, and minor allele additive.

Shown are results for 4 genetic models for 3 diseases (diabetes, cancer and cardiovascular disease (CVD, i.e., at least one of hypertension, CHD, and CVA (stroke)), thus total of 24 tests for covariate models.

Functional Correlates

In an attempt to identify putative functional variants that might help to explain PIK3R1’s role in longevity and cell resilience we examined SNPs in near-perfect LD (Suppl. Figure S1) with our 3 original SNPs, rs7713645, rs6881033, and rs7709243, leading us to identify two additional SNPs, rs7707370 and rs7725370 (Suppl. Tables S6 and S7). Although rs7709243 does not modify transcription factor binding sites, neighboring SNPs rs7713645, rs7707370 and rs7725370 that are in LD with rs7709243 are predicted to significantly modify the binding efficiencies of 3 transcription factors (Suppl. Figure S2). Specifically, the major (i.e., common) allele of rs7713645 creates a DBX1 and ZFP187 binding site and abolishes DMRT2, DMRT3, DMRT4, DMRT5, DMRT7 and NCX2 sites. Variant rs7707370 creates CDX, FOXA, FOXJ2, FOXP1, and POU3F1 sites. Variant rs7725370 creates SMAD 2 and SMAD 3 and abolishes FOXA and FOXP1 sites. Suppl. Table S8 shows the effect that the common (major) allele has on transcription factor binding (abolish/create), biological pathways affected, and the major tissues in which the transcription factor is expressed. Major tissues for expression of these transcription factors include kidney, brain, gastrointestinal tract, and artery. All of the above SNPs were found to overlap regions of chromatin marks denoted by the H3K4me3 mark, according to the HaploReg database. H3K4me3 chromatin refers to tri-methylation of lysine 4 in histone H3 and is associated with active chromatin. The marks were reported in blood, adipose tissue, gastrointestinal tract, skin, brain, muscle, heart, and liver tissues (data not shown).

The Pik3r1 protein is expressed in most tissues, most notably in adipose, artery, brain, breast, gastrointestinal, lung, muscle, and skin, as shown in Suppl. Figure S3. Tissue expression of PIK3R1 transcript isoforms is shown in Suppl. Figure S4. Super-enhancers play a major role in determining cell lineage and differentiation. The locations of super-enhancers in PIK3R1 are shown in Suppl. Table S9 and Suppl. Figure S5. Common tissues for these super-enhancers include lymphocytes, brain, and adipose. However, since these data were obtained from permanent cell lines they may not reflect the locations in adult tissues.

Splice-site variants influence exon usage and ultimately the protein isoform. sQTLs (splice quantitative trait loci) are variants that are empirically shown to be associated with different transcripts in single tissues. Shown in Suppl. Table S10 are 167 sQTLs, many of which overlap in location with the SNPs listed in Suppl Table S7. These sQTLs were found to be prevalent in testis, cultured fibroblasts, and skeletal muscle. Collectively, these results support the contention that the resilience variants function by controlling the response to stress in adipose, gastrointestinal, kidney, skin, brain, muscle, heart, and liver tissues.

Discussion

Here we show for the first time that longevity-associated genetic variants of PIK3R1 are associated with a significant reduction in risk of mortality in elderly men who have CVD. However, there was no effect of PIK3R1 genotype on mortality for subjects with diabetes or cancer. In men who were free of CVD, mortality risk and lifespan were the same for each PIK3R1 genotype. We propose that longevity associated PIK3R1 variants confer cell resilience by mitigating the adverse effects associated with having CVD.

We originally chose to examine the role of PIK3R1 in aging because its expression is downregulated in liver in calorically restricted mice [22]. Although we noted an association of genetic variation in PIK3R1 with longevity in our cohort, a negative result was obtained in a GWAS of nonagenarians [23]. The PIK3R1 SNPs rs7713645 and rs7709243 have shown association with BMI [24]. Rs7713645 has been reported to be associated with body fat and serum leptin levels [24], and may modify the effect of body size and weight gain on breast cancer risk [25]. In our study, the 3 significant SNPs were associated with body weight, BMI and forced expiratory volume (FEV1). Downregulation of insulin signaling is associated with increased lifespan [26]. The major transcriptional effector downstream of PI3K is FoxO3 [27]. Inactivation of FoxO family members results in suppression of transcriptional programs that control cell proliferation and survival [28]. FoxO3 in turn upregulates the PIK3 catalytic subunit in response to loss of growth factor signaling [29].

There are many PIK3R1 mutations, both germline and somatic, that lead to various diseases such as SHORT Syndrome [7,30], immunodeficiency [8], and insulin resistance (with and without obesity/dyslipidemia) [10] and cancer [11]. Several variant proteins, such as p55α and p50α, are produced by differential splicing. The three different isoforms (including p85α) can relay the signal from IRS proteins to the p110 catalytic subunit with different efficiencies [31]. Loss of p85α results in compensation by upregulation of p55α and p50α, leading to hypoglycemia and increased insulin sensitivity in mice [32]. In insulin-resistant obese mice, expression of p85α in liver is decreased, while expression of p55α and p50α is greater than that in their lean littermates [33,34]. The alteration in expression of the regulatory subunits, in the heterozygous state, in the obese mice may be a compensatory reaction to the insulin-resistance. This suggests that different splice isoforms of p85α may be necessary for normal metabolism and development.

The biological basis for our finding of a detrimental effect on survival of CVD subjects who were heterozygous for PIK3R1 alleles (CT genotype) may relate in part to activation of the catalytic subunit, PIK3R2 [35]. Mutations found in three families with senescent T cells, lymphadenopathy and immunodeficiency (PASLI) disease were reported with a heterozygous splice site mutation in PIK3R1 [35]. The PIK3R1 splice site mutation causes skipping of exon 11. The mutant p85α protein was expressed at low levels in patient cells and activated PI3K signaling when overexpressed in T cells from healthy subjects due to qualitative and quantitative binding changes in the p85α-p110δ complex and failure of the C-terminal region to properly inhibit p110δ catalytic activity [35].

Our subjects were on average in their late 70s at examination 4. We considered whether some of the subjects who were homozygous for the minor allele died, so selectively exiting the study cohort, the premise being that this was due to the existence of a high-risk allele that was in linkage disequilibrium with our sentinel SNP. If that had happened, we would be seeing only the minor allele homozygotes that did not have this putative neighboring risk allele. To address this, we examined trajectories of the three genotypes over time, but found no significant difference: At examination 3 (1971–1975) genotype frequencies were 5.8% (CC), 35.0% (TC), and 59.2% (TT), compared with 5.6% (CC). 35.7% (TC) and 58.6% (TT) at examination 4 (1991–1993).

As a practical application of the findings, it would be interesting to examine the effect of common PIK3 inhibitors [36,37] on mortality risk in TC genotype men with CVD.

In conclusion, we have found, for the first time, that remaining lifespan of elderly men with longevity-associated alleles of PIK3R1 who have CVD is similar to that of elderly men who do not have CVD. Thus, favorable PIK3R1 genotype can abrogate the increased risk for mortality in men with prevalent CVD. Since PIK3R1 genotype had no effect on lifespan of men without CVD, the influence of PIK3R1 genotype on longevity of the population as a whole was because of an effect on at-risk elderly individuals only. While we do not know the exact mechanisms involved in the disease resilience proposed, we posit that the associated genetic variants are key to understanding the heritable genetic components to healthy aging.