Genetic Variation in the Raptor Gene Is Associated With Overweight But Not Hypertension in American Men of Japanese Ancestry

Abstract

BACKGROUND

The mechanistic target of rapamycin (mTOR) pathway is pivotal for cell growth. Regulatory associated protein of mTOR complex I (Raptor) is a unique component of this pro-growth complex. The present study tested whether variation across the raptor gene (RPTOR) is associated with overweight and hypertension.

METHODS

We tested 61 common (allele frequency ≥ 0.1) tagging single nucleotide polymorphisms (SNPs) that captured most of the genetic variation across RPTOR in 374 subjects of normal lifespan and 439 subjects with a lifespan exceeding 95 years for association with overweight/obesity, essential hypertension, and isolated systolic hypertension. Subjects were drawn from the Honolulu Heart Program, a homogeneous population of American men of Japanese ancestry, well characterized for phenotypes relevant to conditions of aging. Hypertension status was ascertained when subjects were 45–68 years old. Statistical evaluation involved contingency table analysis, logistic regression, and the powerful method of recursive partitioning.

RESULTS

After analysis of RPTOR genotypes by each statistical approach, we found no significant association between genetic variation in RPTOR and either essential hypertension or isolated systolic hypertension. Models generated by recursive partitioning analysis showed that RPTOR SNPs significantly enhanced the ability of the model to accurately assign individuals to either the overweight/obese or the non-overweight/obese groups (P = 0.008 by 1-tailed Z test).

CONCLUSION

Common genetic variation in RPTOR is associated with overweight/obesity but does not discernibly contribute to either essential hypertension or isolated systolic hypertension in the population studied.

Essential hypertension (EHT) increases in prevalence from middle age onward. In the elderly, isolated systolic hypertension (ISH) is a common condition. A characteristic feature of hypertension is hypertrophy of vascular and cardiac smooth muscle. This feature ties in with the hyperfunction theory of aging, which postulates that in later life the once beneficial processes that contributed to the growth of the organism early in life become deleterious because of their contribution to pathologies characterized by hypertrophy and hyperplasia.¹ Based on data from model organisms, the major mechanisms involved in hypertrophy and hyperplasia include key growth stimulating pathways, particularly those involving insulin/IGF-1 signaling and the mechanistic (formerly “mammalian”) target of rapamycin (mTOR) complex 1, which are pivotal to exacerbation of age-dependent chronic conditions.^1–4 To date, the role of these intracellular pathways in overweight and hypertension has received little attention.

mTOR is an evolutionarily conserved serine/threonine phosphoinositide 3-kinase-related kinase with crucial roles in cell growth and metabolism in response to nutrients, growth factors, cellular energy, and stress,^1,5–7 being positioned at a central hub of nutrient sensing signaling mechanisms.^5,8 The inhibition of mTOR kinase activity^9,10 by administration of rapamycin to middle-aged mice is thought to improve health by reducing the risk of cardiovascular disease.^11,12

mTOR is a component of 2 structurally and functionally distinct multiprotein complexes—TOR complex 1 (TORC1) and TOR complex 2 (TORC2). In mammals, 2 accessory proteins—regulatory-associated protein of mTOR (raptor/RPTOR) and rapamycin-insensitive companion of mTOR (rictor/RICTOR)—distinguish TORC1 from TORC2, respectively.^13,14 In response to nutrients, TORC1 stimulates organ and body growth by regulating translation, transcription, ribosome biogenesis, nutrient transport, and autophagy, whereas most of the effects of TORC2 are the opposite.⁷ As well as having a key role in aging, the mTOR signaling pathway has been implicated in cardiovascular disease, obesity, diabetes, and cancer.^15–20 The enhanced proliferative capacity of vascular smooth muscle with aging contributes to hypertension.²¹

In a recent study investigating tagging single nucleotide polymorphisms (SNPs) spanning several key mTOR pathway genes, we noted genotypic associations between 23 of 61 RPTOR tagging (tag) SNPs and body weight, blood pressure (BP), and hypertension, but 20 other phenotypes showed no such association.²² These data led us to hypothesize that genetic variation in RPTOR could be responsible in part for the genetic basis of overweight/obesity, and thence EHT, especially given its critical role in an important growth pathway in the cell in accord with the hypertrophic hyperfunction theory of aging.¹ RPTOR is located at chromosome 17q25.3, which contains a gene for EHT.²³ Allele frequencies of RPTOR variants exhibit particularly strong correlations with latitude.²⁴ They represent strong signatures of adaptation to different climates via metabolism and, in today’s world, may predispose to common metabolic disorders.²⁴ The allele most strongly associated with temperature variables conferred lower RPTOR expression.²⁵ Interestingly, adipose-specific Rptor knockout mice are resistant to diet-induced obesity, this effect involving an increase in mitochondrial uncoupling in white adipose tissue.^17,26

Here, we report the results of genetic association studies of RPTOR tagSNPs with overweight/obesity, EHT, and ISH.

METHODS

Study population

Subjects were drawn from the Honolulu Heart Program cohort. Recruitment, design, and procedures involved in this long-running longitudinal study have been outlined in detail elsewhere.^27,28 Briefly, American men of Japanese ancestry residing on the Hawaiian island of Oahu were recruited in 1965–1968 (examination 1) when they were aged 45–68 years (mean age 54 years). Analyses showed no evidence of population stratification in the data set.² Clinical parameters such as BP and body mass index (BMI) were obtained at study entry. A more extensive set of clinical parameters that included blood work was obtained at examination 4 (1991–1993) when the men were aged 71–93 years (mean age 78 years), as described previously.² Blood samples were collected by venipuncture at the antecubital fossa during examination 4 and stored at −70 °C. These samples were used for genotyping as described below.

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.

Subjects used for the association studies

Association studies of RPTOR SNPs and overweight/obesity, EHT, and ISH were performed using 2 different cohorts. One consisted of 374 male subjects who had lived a normal lifespan of <81 years at time of death, a limit chosen on the basis of the 3.5 years longer life expectancy of Japanese American men in Hawaii²⁹ compared with the 1910 US birth cohort-specific survival for middle aged American men in general of 77 years. The mean age at death for our normal lifespan cohort was 78.1±1.8 SD years (range 73–81). The other cohort consisted of 439 long-lived subjects whose lifespan exceeded 95 years (296 having died (age of death = 97.0±2.2 SD years; range 95–106) and 144 being still alive (mean age = 96.3±1.6 SD; range 95–106 years)). For the study of overweight/obesity, subjects were allocated to an overweight/obesity group based on BMI ≥ 23kg/m² and a lean/normal weight group in whom BMI was <23kg/m², where a BMI of 23kg/m² is the cutoff commonly set for being overweight in Asian subjects.³⁰ For the EHT study, subjects were considered to have EHT based on a systolic/diastolic BP of ≥140/≥90mm Hg, or who were receiving antihypertensive medication subsequent to an initial diagnosis of EHT on this basis. EHT subjects were compared with normotensive (NT) subjects (systolic/diastolic BP of <140/<90mm Hg and not taking antihypertensive medication). For the ISH study, ISH was defined as a systolic BP of ≥140mm Hg and a diastolic BP of <90mm Hg. These were compared with NT subjects.

Genotyping

We isolated total leukocyte DNA using the PureGene system (Gentra Systems, Minneapolis, MN). DNA was quantified using PicoGreen staining (Molecular Probes, Eugene, OR). The SNPs genotyped in RPTOR were located within the gene and 5kb of flanking DNA. These were tagSNPs selected using Haploview, a program that defines a haplotype based on high correlation between the first and last markers and all intermediate markers. Figure 1, plotted using Haploview,³¹ shows the linkage disequilibrium (LD) blocks captured by the tagSNPs used, where red squares denote blocks that have a Hedrick’s multiallelic D′ = 1, while pink or white denote blocks that have a D′ value < 1.³² Our selection of SNPs was restricted to those having a minor allele frequency of ≥0.1 in the Japanese population (as indicated by the HapMap database: http://hapmap.ncbi.nlm.nih.gov/) (HapMap release 27/phase II + III, February 2009 on NCBI B36 assembly, dbSNP b126—http://www.ncbi.nlm.nih.gov/projects/SNP/). The minimal coefficient of determination (r ²) value at which all alleles were to be captured was set to a threshold of 0.8 for the identification of all tagSNPs. SNPs were genotyped at the University of Hawaii Cancer Center on the Illumina GoldenGate platform (high-throughput SNP genotyping on universal bead arrays³³). We implemented a series of quality control checks based on Illumina metrics. In order for data for a SNP to be included, its call rate had to exceed 0.95 and the Hardy–Weinberg equilibrium P needed to be >0.01. Of 61 tagSNPs genotyped, all met these criteria and were included. In addition, >96% concordance was observed for all SNPs for HapMap samples that were assayed for quality assurance. Genotype data were managed through an integrated database system sample management-data processing system that we have found to be accurate in other successful studies.² All positive controls on each genotyping plate were evaluated for consistency. Call rates for markers exceeded 98%.

Figure 1.

LD plot showing the 61 tagging SNPs in RPTOR on chromosome 17p25.3. Shown is the region spanning nucleotides 76122143–76565844 (from HapMap data release 27, phase II + III, February 2009, on NCBI B36 assembly, dbSNP b126). It is estimated that 406 alleles with a mean r ² of 0.92 are captured using the tagSNPs in the present study. The figure was plotted using Haploview. In the figure, red squares denote blocks that have a Hedrick’s multiallelic D′ = 1, whereas pink or white denote blocks that have a D′ value < 1.^29–32 Abbreviations: LD, linkage disequilibrium; SNP, single nucleotide polymorphism.

Statistical analyses

Genotype frequencies of each SNP were evaluated for deviation from Hardy–Weinberg equilibrium. For each SNP, allele frequencies in case and control groups were compared by Pearson chi-square analysis using the SAS statistical package.³⁴ The odds ratios (ORs) of SNPs by hypertension status were estimated by conditional logistic models (treating longevity and average lived subjects as different strata). The general linear model and analysis of covariance were further used to compare proportion of healthy study participants by RPTOR SNPs.

Logistic regression is commonly used when the response variable of interest can be expressed as present/absent or “yes”/“no” (coded as 1 or 0, respectively). Increasing the number of variables and their potential interactions in any parametric model increases the risk of over-fitting the model and thereby reducing its predictive accuracy when applied to new data. To reduce that risk, we conducted an alternative method of analysis, namely recursive partitioning (RP).³⁵ RP is an “assignment” or “classification” procedure that provided an alternative way to test whether genetic variation in RPTOR enhanced the prediction of an individual having an overweight/obesity, EHT, or ISH phenotype. RP generates “random forests”³⁶ (a collection of up to 100 decision trees in our case) where each tree in the forest and each branch in the tree³⁷ is a randomly selected subset of the available predictor variables (a mix of binary and continuous in our case) that are then used to maximally separate the 2 subpopulations of the binary response variable (e.g., “1” if EHT and “0” if NT). The size of the tree varies by endpoint because we used the “early stopping” option in the JMP statistical software (JMP 10 Pro 2014, SAS Institute, Cary, NC). This option causes the forest to stop growing new trees when they no longer improve the model fit as defined by the validation statistics (e.g., entropy R-squared, generalized R-squared, mean Log P, root mean square error approximation, mean absolute deviation, and misclassification rate). The value of the stopping rule is that it reduces the chance of over fitting the model. In our analyses, there were a minimum of 10 splits per tree. Although RP analyses are exceedingly complex, their final results can be easily summarized by a 2×2 contingency table referred to as a “confusion matrix.” The rows of the matrix are observed membership (e.g., “1” if EHT or “0” if NT) and the columns are predicted membership (EHT or NT). The “0,0” cell and the “1,1” cell reveal the proportion of correct assignments or “accuracy.” Two RP analyses were performed; 1 analysis involved a backward elimination of anthropometric variables only (“reduced model”) and the other analysis (“full model”) augmented the reduced model by including all of the RPTOR SNP variables. During the backward elimination of the full model, the only variables eliminated were SNP variables. A full model always performs better than a reduced model, but a simple 1-tailed Z test (http://www.socscistatistics.com/tests/ztest/) for comparing 2 proportions (accuracy metric from the reduced and full confusion matrices) is able to reveal whether the additional RPTOR SNP variables statistically improved the ability to correctly assign individuals to defined groups. Two additional methodological issues were: (i) all model fitting was performed on a randomly selected 2/3 partition of the available data (model training data), with the remaining one third of the data (validation data) being used to determine how well the trained model performs when applied to novel data, and (ii) the natural frequency of the response variable (e.g., 37% in the case of EHT) was used to establish the probability cut point that must be exceeded in order to assign an individual to, e.g., the EHT phenotype. We declared that the frequency of the phenotype reflected its natural frequency when we constructed our confusion matrices.

RESULTS

Characteristics of participants

The physiological and clinical phenotypes of the normal lifespan and long-lived subjects in the present study have been detailed previously.^2,22,38 Table 1 shows mean ± SD BMI, systolic BP, and diastolic BP of each group of subjects in the present study.

Table 1.

BMI and blood pressure for each group of subjects (mean ± SD) at examination 1

Parameter	Normal lifespan	Long-lived	P value
BMI (kg/m2)
Overweight	25.7±2.2 (n = 259)	25.4±1.8 (n = 244)	0.075
Lean/normal weight	20.9±1.4 (n = 115)	21.1±1.5 (n = 195)	0.26
Systolic BP (mm Hg)
EHT	149.6±14.2 (n = 153)	149.5±16.4 (n = 149)	0.94
ISH	154.3±12 (n = 119)	155.6±13.7 (n = 111)	0.46
NT	120.8±10.9 (n = 221)	121.2±10.7 (n = 290)	0.73
Diastolic BP (mm Hg)
EHT	93.2±8.9 (n = 153)	91.1±9.2 (n = 149)	0.046
ISH	94±9.7 (n = 119)	92.6±8.7 (n = 111)	0.27
NT	76.1±7.9 (n = 221)	76.2±7 (n = 290)	0.87

Abbreviations: BMI, body mass index; BP, blood pressure; EHT, essential hypertension; ISH, isolated systolic hypertension; NT, normotensive.

Genetic association study findings

For each cohort, the genotype frequencies of each RPTOR SNP in the respective case and control groups together with results of chi-square analyses are presented in Supplementary Table S1 (overweight/obese vs. normal weight/lean), Supplementary Table S2 (EHT vs. NT), and Supplementary Table S3 (ISH vs. NT).

Supplementary Tables S4–S6 show the OR for the effect of the minor allele of each SNP on overweight/obesity, EHT and ISH, respectively, after correction for longevity status by conditional logistic regression analysis adjusted for age at examination 1. In all of these analyses, after correction for multiple comparisons, no statistically significant association was found between RPTOR genotype and overweight/obesity, EHT or ISH in either the normal lifespan cohort or the long-lived cohort.

Post hoc power analyses were performed. For each study, conditional logistic regression models for overweight, EHT, and ISH used 2 variables to compare heterozygotes (Mm) against major allele homozygotes (MM) and to compare minor allele homozygotes (mm) against MM individuals. Power to detect an association between EHT and a genetic variable depended on both the proportion of the sample with the specified genotype and the number of subjects successfully genotyped for the SNP (where sample size varied for different SNPs). The expected proportion of heterozygotes = 2p × (1 – p), where p = minor allele frequency. This was fairly stable across a broad range of values of p, implying that the minimal detectable OR for 80% power would not vary much across SNPs. In contrast, the proportion of homozygotes, p ², varied considerably across loci, which meant there was also substantial variation in the minimal detectable OR. In brief, results for the minimal detectable OR with 80% power to detect associations for the SNPs rs4889856 (n = 801, minor allele frequency of 0.34) and rs6565494 (n = 720, minor allele frequency of 0.24) are as follows. When comparing heterozygotes with major allele homozygotes, the minimal detectable ORs for the 3 outcomes—overweight/obesity, EHT, and ISH—ranged from 1.5 to 1.6, a very small spread of values. For comparisons of the 2 homozygous genotypes, the minimal detectable OR was 2.0 or less for rs4889856 but ranged from 2.6 to 3.1 for rs6565494. Therefore, while almost all comparisons of heterozygotes with major allele homozygotes having ORs ≥ 1.6 were expected to be significant at the 0.05 level in our study, there were actually slightly fewer significant results than expected if genetic variants had no effect. Minimal detectable ORs when comparing homozygotes would be 2.0 or less for the variants having minor alleles that were more common, but 3.0 or greater for ISH in the case of less common alleles.

Recursive partitioning analyses

Overweight/obesity.

The reduced model for overweight/obesity excluded weight, height, and BMI variables because these were used to create the response variable. Keeping them in the analysis would artificially improve the model. All RP analyses used the “early stopping” option (described earlier) to reduce the chance of over fitting the model. The “reduced model” for overweight/obesity contained 55 trees with 6 of the 16 potential predictor variables sampled per split, 10 minimum splits per tree, and a minimum split size of 5. The “full model” contained 40 trees with 16 of the 36 potential variables sampled per split, 10 minimum splits per tree, and a minimum split of 5. The “reduced” model produced an overall accuracy of 90% in assigning overweight/obese and lean/normal weight individuals to their correct membership group (Table 2a), as opposed to 94% in the full model (Table 2b) containing 21 of the 61 possible RPTOR SNPs. A 1-tailed Z test for comparing 2 proportions produced a P value of 0.008, suggesting that the SNPs significantly enhanced the ability of the RP model to accurately assign individuals to either the overweight or the non-overweight groups. It should be noted, however, that none of the SNPs individually contributed as much as the quantitative laboratory and examination variables (see Figure 2A). A final word of caution is that the ROC curves for both the reduced and full models (Figure 3A) reveal a sizable degradation of assignment efficiency in the validation data. An area greater than 0.69, as achieved in this analysis, is good evidence but must be interpreted relative to the much higher value in the training data (0.97) and to the 0.5 baseline area of a ROC curve that is equivalent to making assignments based on a fair coin toss.

Table 2.

Confusion matrix produced during RP analysis for overweight/obesity

Observed	Training data		Validation data
	Predicted		Predicted
	0	1	0	1
(a) Reduced RP model for overweight/obesity (0 = no, 1 = yes)
0	190	34	48	38
1	19	302	39	143
	Accuracy = 90%		Accuracy = 71%
	Specificity = 85%		Specificity = 56%
	Sensitivity = 94%		Sensitivity = 79%
(b) Full RP model for overweight/obesity (0 = no, 1 = yes)
0	199	21	44	40
1	10	305	37	137
	Accuracy = 94%		Accuracy = 70%
	Specificity = 90%		Specificity = 52%
	Sensitivity = 97%		Sensitivity = 79%

Contained in the model were 15 laboratory/examination, overweight/obesity predictor variables augmented by 21 SNP predictors. This revealed the accuracy, specificity, and sensitivity of (a) the reduced model and (b) the full model in assigning overweight or not individuals to their correct membership group.

Abbreviations: RP, recursive partitioning; SNP, single nucleotide polymorphism.

Figure 2.

Results from recursive partitioning analysis for (A) overweight/obesity, (B) essential hypertension, and (C) isolated systolic hypertension. Shown is contribution to fit for each of the variables in the full model generated from RP analysis. In (A), 19 SNPs failed to improve assignment efficiency; in (B), 6 SNPs failed to improve assignment efficiency; and in (C), 4 SNPs failed to improve assignment efficiency. In (A), (B), and (C), all SNP contributions were less than laboratory/examination variables. The most promising of the 61 SNPs genotyped are shown: 19 for overweight/obesity, 6 for EHT, and 4 for ISH. Abbreviations: BMI_1, body mass index at examination 1 (1965–1968); BMI_4, BMI at examination 4 (1991–1993); Overweight/obese, overweight or obese at examination 1 (1965–1968); CASI, Cognitive Abilities Screening Instrument (score range: 0–100) at examination 4 (1991–1993); CHOL_4, fasting plasma total cholesterol at examination 4 (1991–1993); GLUC_4, fasting glucose at examination 4 (1991–1993); EHT, essential hypertension; GRIP_4, grip strength at examination 4 (1991–1993); HDL_4, fasting plasma high density lipoprotein cholesterol at examination 4 (1991–1993); Height_4, height at examination 4 (1991–1993); INS_4, fasting insulin at examination 4 (1991–1993); ISH, isolated systolic hypertension; log INS_4, logarithm of fasting insulin at examination 4 (1991–1993); RP, recursive partitioning; SNP, single nucleotide polymorphism; Waist:Hip, waist:hip ratio at examination 4 (1991–1993); Weight_1, body weight at examination 1 (1965–1968).

Figure 3.

Receiver operating characteristic curves showing sensitivity and specificity for training data (left panel) and validation data (right panel) that formed the final output of recursive partitioning analysis testing for influence of the variables on (A) overweight/obesity, (B) essential hypertension, and (C) isolated systolic hypertension.

Essential hypertension.

As before, all RP analyses used the “early stopping” option to reduce the chance of over fitting the model. The “reduced model” for EHT contained 100 trees with 3 of the 13 potential predictor variables sampled per split, 10 minimum splits per tree, and a minimum split size of 5. The “full model” contained 21 trees with 6 of the 19 potential variables sampled per split, 10 minimum splits per tree, and a minimum split of 5. The “reduced” model produced an overall accuracy of 88% in assigning EHT individuals to their correct membership group (Table 3a), as opposed to 90% in the full model (Table 3b) containing 6 of the 61 possible RPTOR SNPs. A 1-tailed Z test for comparing 2 proportions produced a P value of 0.22, suggesting that even the most predictive SNPs (rs4969322 and rs4890052), while suggestive, did not statistically enhance the ability of the RP model to accurately assign individuals to either the EHT or NT group (Table 3a). As before, the laboratory/examination variables (especially BMI, weight, and waist to hip ratio) stood out as the dominant predictors (Figure 2B). Grip strength and HDL cholesterol were lower tier predictors. The overall accuracy of the full model was 90% (Table 3b), but that accuracy declined to 61% in the validation data. Similarly, the area of the ROC curve decreased from 98% to 66% (Figure 3B), which suggested caution when applying this model to new data. Taken together, the results of this analysis confirm and strengthen the conclusions derived from the logistic regression analyses.

Table 3.

Confusion matrix produced during RP analysis for EHT

Observed	Training data		Validation data
	Predicted		Predicted
	0	1	0	1
(a) Reduced RP model for EHT (0 = NT, 1 = EHT)
0	290	53	111	58
1	10	192	44	56
	Accuracy = 88%		Accuracy = 62%
	Specificity = 86%		Specificity = 66%
	Sensitivity = 95%		Sensitivity = 56%
(b) Full RP model for EHT (0 = NT, 1 = EHT)
0	298	45	101	68
1	10	192	37	63
	Accuracy = 90%		Accuracy = 61%
	Specificity = 87%		Specificity = 60%
	Sensitivity = 95%		Sensitivity = 63%

Contained in the model were 13 laboratory/examination predictor variables. This revealed the accuracy, specificity and sensitivity of (a) the reduced model and (b) the full model in assigning EHT and NT to their correct membership group.

Abbreviations: EHT, essential hypertension; NT, normotensive; RP, recursive partitioning.

Isolated systolic hypertension.

The “early stopping” option was once again used to reduce the chance of over fitting the ISH model. The “reduced model” contained 28 trees with 4 of the 11 potential predictor variables sampled per split, 10 minimum splits per tree, and a minimum split size of 5. The “full model” contained 28 trees with 5 of the 15 potential variables sampled per split, 10 minimum splits per tree, and a minimum split of 5. The “reduced” model produced an overall accuracy of 87% in assigning ISH individuals to their correct membership group (Table 4a), as opposed to 88% in the full model (Table 4b) containing only 4 of the 61 possible RPTOR SNPs. A 1-tailed Z test for comparing 2 proportions produced a P value of 0.58, suggesting that even the most predictive SNP (rs2589118) did not significantly enhance the ability of the RP model to accurately assign individuals to either the ISH or NT group (Table 4). As seen before, the overall assignment accuracy of either the reduced (87%) or full (88%) model is high in the training data, but the area of the validation ROC curve (0.69) and the overall accuracy of the assignment (66%) drops off significantly in the validation data (Figure 3C). As before, the RP results for ISH are consistent with and strengthen the conclusions derived from the logistic regression analyses.

Table 4.

Confusion matrix produced during RP analysis for ISH

Observed	Training data		Validation data
	Predicted		Predicted
	0	1	0	1
(a) Reduced RP model for ISH (0 = NT, 1 = ISH)
0	306	67	114	69
1	5	167	24	62
	Accuracy = 87%		Accuracy = 65%
	Specificity = 82%		Specificity = 62%
	Sensitivity = 97%		Sensitivity = 72%
(b) Full RP model for ISH (0 = NT, 1 = ISH)
0	312	61	116	67
1	5	167	25	61
	Accuracy = 88%		Accuracy = 66%
	Specificity = 84%		Specificity = 63%
	Sensitivity = 97%		Sensitivity = 71%

Contained in the model were 13 laboratory/examination predictor variables. This revealed the accuracy, specificity and sensitivity of (a) the reduced model and (b) the full model in assigning ISH and NT to their correct membership group.

Abbreviations: ISH, isolated systolic hypertension; NT, normotensive; RP, recursive partitioning.

DISCUSSION

After extensive analysis of SNPs in the mTOR pathway gene RPTOR, the present study of American men of Japanese ancestry found no significant association of any of these SNPs with either EHT or ISH after correction for multiple testing. The prediction of overweight/obesity was statistically improved, however, by the inclusion of RPTOR SNPs. The study involved contingency table analysis, logistic regression, and recursive partitioning, the latter being regarded as among the most powerful methods for statistical analysis of large complex sets of genetic information.³⁵ The findings applied to 2 different cohorts emanating from a longitudinal study of aging and its associated phenotypes. Consistency of the findings in the different cohorts strengthens our conclusions.

The results obtained from conventional association analyses of EHT and ISH were reinforced by the findings from the more powerful method of interrogation of data from such studies, namely recursive partitioning analysis. A reason that prompted us to use RP was the apparent excess of P < 0.05 values for phenotypes relating to body size and BP or hypertension in an earlier study.²² RP analysis was, however, able to reveal an association of overweight/obesity with genetic variation in RPTOR. Nevertheless, RP also showed that the most important predictors of overweight/obesity, as well as EHT and ISH, were from expected variables such as BMI, HDL, and insulin sensitivity.

Since we used tagging SNPs our analyses provided maximal coverage of genetic variability across the gene plus 5kb of 5′ and 5kb of 3′ DNA, taking advantage of LD. We cannot, however, exclude the possibility that one or more polymorphisms not in LD with the SNPs tested, or of population-specific variants not present in the current population, could be associated with EHT or ISH. Moreover, our exhaustive interrogation of RPTOR for association with body weight and hypertension is instructive in deciphering the molecular genetics of these conditions, especially as it relates to the possible role of growth-promoting pathways in their etiology.

A study in the Netherlands found significantly lower expression of RPTOR mRNA in subjects who had attained the age of 89 years or more, as well as in their offspring.³⁹ This is consistent with a role for raptor in the hyperfunction theory of aging. Lower intracellular raptor concentration, and thus lower TOR complex activity, would be expected to contribute to lower BMI and thereby protect against overweight/obesity, so reducing the prevalence of conditions associated with elevated BMI. Since EHT is more common in overweight and obese individuals, lower RPTOR expression might be hypothesized to protect against EHT. Given the negative findings for RPTOR SNPs and hypertension in the present study, our results would be consistent with indirect effect(s) on RPTOR gene expression, post-transcriptional regulation, or involvement of distal elements not in LD with the SNPs we have tested in our research.

Our long-lived cohort had a lower body weight in middle age (examination 1), namely 62.0±7.9kg, compared with 64.1±9.4kg in the normal lifespan cohort (P = 0.009). The long-lived cohort also had a lower prevalence of EHT at examination 1 (32%) compared with the normal lifespan group (43%; P = 0.018). Moreover, severe EHT (BP ≥ 160/≥100mm Hg) was less common in the long-lived cohort (14%) compared with the normal lifespan cohort (26%; P = 0.0028). ISH too was less common in the long-lived group: 22% vs. 36% (P = 0.016) and severe ISH was 3.6% vs. 12% (P = 0.0006) for each respective group.

Certain features of our case and control groups merit comment. Since we only studied males, it remains to be seen whether any of the SNPs tested might be associated with overweight/obesity, EHT, or ISH in females.

Our study had several strengths. First, the Honolulu Heart Program study cohort can be considered to resemble a nested case–control design in that cases and controls were selected from an ongoing cohort study, with longitudinally collected data. This has advantages over a regular case–control study in that several phenotypes of interest, such as biological/physiological phenotypes, disease prevalence, and functional status, were obtained by direct clinical examination when the participants were younger. Some of the information was collected from subjects before determination of their status as normal lifespan vs. long-lived. Major clinical conditions such as hypertension underwent adjudication by a morbidity committee. A second strength was that the candidate gene RPTOR selected for analysis was chosen a priori based on hypothesis-driven criteria that included data from studies of model organisms that have shown the importance of the evolutionarily conserved mTOR pathway in phenotypes of body growth and aging.⁴⁰ Thirdly, a prior study of RPTOR in the Honolulu Heart Program cohort had provided preliminary suggestive evidence that genetic variation in RPTOR might be worthy of exploration in overweight/obesity, EHT, and ISH.²² Fourthly, the Honolulu Heart Program cohort is a highly homogeneous group of individuals and no population stratification was detected in our study participants.²

In conclusion, the present study suggests that common genetic variation in RPTOR may be associated with overweight/obesity but does not discernibly contribute to either EHT or ISH in American men of Japanese ancestry. Our findings should be replicated in other populations, especially younger cohorts in which genetic factors play a larger role, before being widely accepted.

SUPPLEMENTARY MATERIAL

Supplementary materials are available at American Journal of Hypertension (http://ajh.oxfordjournals.org).

DISCLOSURE

The authors declared no conflict of interest.