Energy intake and leukocyte telomere length in young adults

Abstract

Background: Dietary energy restriction in mammals, particularly at a young age, extends the life span. Leukocyte telomere length (LTL) is thought to be a bioindicator of aging in humans. High n−6 (omega-6) PUFA intake may accelerate LTL attrition.

Objective: We determined whether lower energy and higher PUFA intakes in young adulthood are associated with shorter LTL in cross-sectional and longitudinal analyses.

Design: In a longitudinal observational study (405 men, 204 women), diet was determined at baseline by a semiquantitative food-frequency questionnaire, and LTL was determined by Southern blots at mean ages of 30.1 y (baseline) and 43.2 y (follow-up). Spearman correlations and multivariable linear regression were used.

Results: Baseline energy intake was inversely associated with follow-up LTL in men (standardized β = −0.171, P = 0.0005) but not in women (P = 0.039 for sex interaction). The difference in men between the highest and lowest quintiles of energy was 244 base pairs (bp) (95% CI: 59, 429 bp) and between extreme quintiles of LTL was 440 kcal (95% CI: 180, 700 kcal). Multivariable adjustment modestly attenuated the association (β = −0.157, P = 0.002). Inverse associations, which were noted for all macronutrients, were strongest for the unsaturated fatty acids. In multivariable models including energy and the macronutrients (as percentage of energy), the significant inverse energy-LTL association (but not the PUFA-LTL association) persisted. The energy-LTL association was restricted to never smokers (standardized β = −0.259, P = 0.0008; P = 0.050 for the smoking × calorie interaction).

Conclusions: The inverse calorie intake–LTL association is consistent with trial data showing beneficial effects of calorie restriction on aging biomarkers. Further exploration of energy intake and LTL dynamics in the young is needed.

INTRODUCTION

LTL⁴ at any age reflects LTL at birth and its attrition thereafter, which is defined by the replicative history of hematopoietic stem cells (1, 2). Because inflammation and oxidative stress, which accelerate LTL shortening (3, 4), are the hallmarks of aging, LTL may serve as a bioindicator of aging (5). Short LTL has been associated with cardiovascular risk (6, 7), shorter health span (8), and diminished longevity (9–11).

A seminal finding in the 1930s that calorie restriction extends the life span of rodents was confirmed in other species (12). The effect on maximum life span in rodents was stronger when restriction was implemented at a young age (13). This effect cannot simply be attributed to the leanness caused by calorie restriction but is probably the result of the energy restriction itself (14, 15). There is a suggestive trend for life span extension in energy-restricted nonhuman primates (12) and evidence for a reduction in age-related mortality (16). In humans, caloric restriction has beneficial effects on metabolic, hormonal, and functional changes associated with aging and diminished cardiovascular and cancer risks (12, 15, 17). To date, little work has been reported on the association of energy intake with LTL in humans. Studies have been cross-sectional in nature (18–20), and data for young adults are absent.

Israelis have an unusually high intake of n−6 PUFAs, in the range of 8–10% of total energy (21, 22), with a mean linoleic acid (18:2n−6) content of 25% in subcutaneous adipose tissue (21, 23, 24), which is largely attributable to linoleic acid–rich vegetable oils, margarine, and seeds (21). These figures compare with an estimated intake of 6.7% in the United States (25) and an adipose tissue linoleic acid content of 10–12% in Europeans (24, 26). Concern has been expressed [and incorporated into upper limits of PUFA intake in some national recommendations (eg, references 27–29)] that high linoleic acid intakes may promote atherosclerosis by a putative proinflammatory effect resulting from 1) the metabolic conversion of linoleic acid to arachidonic acid and/or 2) the susceptibility of its double bonds to oxidation, thereby enhancing oxidative modification of LDL (eg, reference 25 for review; 30, 31). Concern has also been raised about deleterious effects of high ratios of n−6 to n−3 fatty acids (32). An American Heart Association statement has dismissed these concerns and supported linoleic acid intakes of ≥5–10% energy as safe and beneficial (25). This issue, however, may not have been entirely laid to rest. For example, a recent cross-sectional study reported that energy-adjusted linoleic acid intake was inversely associated with LTL in women, suggesting a potential deleterious effect (19).

For these reasons, we examined the association of LTL with total energy intake and with dietary macronutrient intake, with emphasis on the fatty acids, among young adults in Jerusalem.

SUBJECTS AND METHODS

Study population

In 1976–1979 the Jerusalem LRC Prevalence Study initially examined 8646 predominantly 17-y-old Jewish residents, representing full age cohorts in Jerusalem, and a sample comprising 6923 of their parents (33). In 1989–1991 (the baseline of the current study), 2 overlapping samples of cohort members residing in Jerusalem were reinvited at age 28–32 y for a standardized interview and examination (34–36). The first was a sex-stratified random sample consisting of 884 young adults who met the eligibility requirements (570 men, 73% response; 314 women, 68% response). The second sample, which contributed an additional 168 individuals, comprised offspring of parents who had a documented acute myocardial infarction or sudden cardiac death over a 10-y follow-up period on the basis of WHO-MONICA (Monitoring of Trends in Cardiovascular Disease) diagnostic criteria (116 men, 74% response; 52 women, 68% response) (35). A total of 1052 eligible subjects (686 men, 366 women) were examined in 1989–1991. In 2003–2006 (the follow-up period of the present study), 631 of the baseline subjects were reexamined [71.3% response rate, after exclusion of 162 ineligible participants who were not current Jerusalem residents, were homebound, had a serious illness (eg, metastatic cancer, end-stage renal disease), were pregnant, or were within 3 mo of delivery]. LTL was assayed in baseline and follow-up samples in 620 participants (413 men and 207 women); of these, 609 were judged to have satisfactory dietary data (see below).

The Jerusalem LRC follow-up study was approved by the Hadassah-Hebrew University Institutional Review Board, and the procedures were in accordance with its ethical standards. Participants provided signed informed consent.

Data collection

Eligible participants examined at baseline (mean age: 30.1 y) and follow-up (mean age: 43.2 y) reported themselves to be free of a febrile or infectious disorder or antibiotic use over the preceding 14 d. Information on sociodemographic characteristics, personal and family medical history, reproductive history, and health-relevant behaviors including smoking, alcohol intake, exercise, and diet was collected at baseline. Standardized measurements of blood pressure, heart rate, and anthropometric variables (weight; height; waist, hip, and arm circumferences; and subscapular, triceps, and supra-iliac skinfold thicknesses) were performed by trained technicians at baseline and at follow-up. Baseline and follow-up blood samples, drawn after a 12-h overnight fast, were immediately placed on ice, and the buffy coat and plasma were stored at −80°C.

Dietary intake was assessed at baseline by a self-administered, semiquantitative food-frequency questionnaire (140 food items tailored to Israeli food preferences) that was adapted from a previously used interviewer-administered instrument (21). The nutrient content of these items was derived from Israeli and US food tables, including the National Heart, Lung, and Blood Institute Nutrition Data System tables for composition of fatty acids (except for locally produced margarines), supplemented by special laboratory analysis of selected items (22). The source instrument had shown a satisfactory association of dietary fatty acid intake with biomarker data based on subcutaneous adipose tissue aspirations [eg, the Spearman correlation for dietary PUFAs, expressed as a percentage of total fat intake, with adipose tissue linoleic acid (the sole source of which is dietary) was 0.42 and for dietary SFAs with adipose tissue myristic acid (14:0) was 0.36 (21)]. In our study population, the correlation of dietary n–3 fatty acids (attributable to fish intake computed from 12 different varieties of fresh and preserved fish in the self-administered questionnaire and expressed as percentage of total fat) with erythrocyte membrane EPA (20:5n−3) was 0.46 in men and 0.45 in women. We excluded 8 men and 3 women who reported energy intakes <600 kcal (men), <500 kcal (women), or >5000 kcal, which left 405 men and 204 women for analysis.

LTL

Before Southern blot analysis of the mean length of the TRFs, DNA was extracted by using Genomic DNA purification kits (Gentra Systems) from the buffy coat of samples stored since 1989–1991 (baseline) and 2003–2006 (follow-up). The integrity of the DNA was verified in all samples. An HinfI/RsaI restriction enzyme combination was used to generate the TRFs, which were resolved in 5% agarose gels (37). The baseline and follow-up TRFs from the same individual were resolved in adjacent lanes of each gel, with duplicate samples run on different gels. The laboratory received coded samples, and laboratory staff were blinded as to the identity of the subjects and the order of the baseline and follow-up samples, which varied. The CV for the between-gel duplicates [computed as: SQRT(Σ(x₁ – x₂)²/2k)/X, where the xs are the replicate pair values, k is the number of subjects, and X is the overall mean of the k pairs of observations] was similar for the baseline (2.1%) and follow-up (2.2%) samples.

Statistical methods

LTLs at baseline and follow-up were reasonably normally distributed, with a slight skewing to the right (see Table 1). To assess sex-specific associations with LTL and LTL attrition (the latter expressed as the mean yearly change in telomere bp over the follow-up period), we initially used nonparametric Spearman correlations. Pearson correlations provided similar estimates. Applying linear regression, we treated LTL as a continuous dependent variable and restricted the analysis to the follow-up measure of LTL, to allow for the passage of time (mean: 13.1 y) between the baseline dietary assessment and the follow-up LTL determination. An ln transformation of LTL did not appreciably affect the associations and was not reported. Regression coefficients were reported as standardized βs (β). The independent variables, ln calorie intake and the nutrients expressed as ln grams per day (transformed to normalize their distributions; model 1) or ln calories and the nutrients expressed as a percentage of calories (which strongly attenuates the correlation with energy intake; model 2) were introduced into the regression models as bivariate pairs and included calories and a single nutrient each time.

TABLE 1.

Characteristics of the study sample: the Jerusalem LRC longitudinal study¹

Characteristics	Men	Women	P for sex difference
n	405	204
Age at baseline (y)	30.1 ± 0.82	30.1 ± 0.8	0.4
Age at follow-up (y)	43.2 ± 0.9	43.1 ± 0.9	0.6
LTL
Baseline LTL (bp)	7251 ± 668	7475 ± 655	<0.001
Median (IQR)	7230 (6740–7715)	7453 (6975–7890)
Follow-up LTL (bp)	6930 ± 625	7126 ± 630	<0.001
Median (IQR)	6865 (6487–7370)	7077 (6675–7508)
LTL attrition (bp/y)	24.5 ± 14.9	26.9 ± 14.3	0.0813
Median (IQR)	24.9 (15.6–33.6)	27.1 (18.3–34.6)
Diet (at baseline)
Total energy (kcal/d)	2322 ± 810	2036 ± 731	<0.001
Carbohydrate (% of energy)	48.6 ± 8.0	48.6 ± 7.0	0.9
Protein (% of energy)	16.6 ± 3.4	15.8 ± 3.3	0.008
Fat (% of energy)	34.8 ± 7.2	35.6 ± 6.4	0.3
SFAs (% of energy)	11.3 ± 3.7	11.2 ± 3.1	0.7
MUFAs (% of energy)	13.0 ± 3.3	12.9 ± 3.0	0.7
PUFAs (% of energy)	8.4 ± 2.5	9.3 ± 2.7	<0.001
PS ratio [median (IQR)]	0.77 (0.57–1.04)	0.86 (0.61–1.10)	0.041
Lifestyle variables (at baseline)
Current smokers (%)	44.9	29.9	<0.001
Past smokers (%)	13.8	10.3
Never smokers (%)	41.2	59.8
Pack-years
Whole sample	5.7 ± 7.0	2.7 ± 5.0	<0.0013
Ever smoked	10.2 ± 6.4	7.1 ± 5.9	<0.0013
Alcohol frequency of ≥3–4 times/wk (%)	6.7	0.5	0.001
Vigorous exercise4 (%)			0.17
≥2 times/wk	13.6	8.3
<2 times/wk	7.9	8.3
None	78.5	83.4
Anthropometric variables (at baseline)
BMI (kg/m2)	25.2 ± 3.5	24.0 ± 3.7	<0.001
Follow-up BMI	27.7 ± 4.3	26.2 ± 4.6	<0.001
Waist-hip ratio	0.88 ± 0.05	0.77 ± 0.06	<0.001
Triceps skinfold thickness (cm)	1.41 ± 0.55	2.17 ± 0.60	<0.001
Subscapular skinfold (cm)	1.61 ± 0.62	1.56 ± 0.62	0.28
Supra-iliac skinfold (cm)	2.42 ± 0.88	2.07 ± 0.79	<0.001
Weight gain, baseline to follow-up (kg)	7.9 ± 7.0	6.0 ± 6.3	0.001
Plasma lipids at baseline (mg/dL)
Total cholesterol	172.5 ± 32.5	162.3 ± 30.9	<0.001
HDL cholesterol	36.5 ± 10.3	44.9 ± 11.0	<0.001
LDL cholesterol5	109.3 ± 28.8	99.0 ± 27.5	<0.001
Triglycerides	135.9 ± 84.8	92.1 ± 40.8	<0.0013

¹bp, base pairs; LRC, Lipid Research Clinic; LTL, leukocyte telomere length; PS, PUFA:SFA ratio.

²Mean ± SD (all such values).

³Mann-Whitney U test; otherwise, Student t tests were used for continuous data and chi-square tests for categorical data.

⁴Exercise causing heavy breathing and sweating.

⁵Computed by the Friedewald method; not computed for 8 men with triglycerides >400 mg/dL.

We included possible baseline confounders associated with energy intake: BMI (in kg/m²), vigorous exercise causing sweating and breathlessness (none, less than twice/wk, twice or more/wk), and cigarette smoking (current, past, never). Country of origin (defined as country of birth of the father and grouped by continent—Europe, Asia, Africa, and Israel) was also included (models 1 and 2). We repeated the analyses to also include weight change between baseline and follow-up and with substitution of pack-years of smoking for the smoking variables. To further account for body size, we redefined energy intake as calories per kilogram body weight (ln-transformed) and included height in the models. Subjects’ age, which was restricted in range, did not affect estimates. Although we did not consider plasma lipids and lipoproteins as confounders, but rather as potential mediating variables, we repeated the models with the inclusion of the lipoproteins to assess their possible effects. Subsequently, we introduced all of the macronutrients in a backward stepwise regression procedure with energy forced into the model, using a P value of ≥0.15 for variable elimination (model 3, nutrients in ln g/d; model 4, nutrients as a percentage of calories). Effect modification of the energy-LTL association by sex, smoking, and BMI was tested in the regression models by using multiplicative terms.

Power

Given the numbers of men and women in this study, and setting an α value of 0.05, this sample had a power of 0.92 and 0.69 (one-tailed) and 0.86 and 0.58 (2-tailed) in men and women, respectively, to detect a correlation of 0.15 and a power of 0.94 and 0.70 in men and women, respectively, to detect a 3% increment in R² in a linear regression model.

RESULTS

Relevant characteristics of the study population are presented in Table 1. Participants were aged 28–32 y at baseline and 41–47 y at follow-up. Mean LTL was longer in women than in men (by 224 bp at baseline and 196 bp at follow-up, P < 0.001 at both time points). LTL shortening over the 13.1-y mean follow-up period was evident in 590 of the 609 participants. Eighteen individuals (3.0%) showed elongation, and one showed no change, probably reflecting misclassification due to the combined effects of measurement error of LTL at baseline and follow-up (38). The average shortening of LTL was 24.5 ± 14.9 bp/y in men and 26.9 ± 14.3 bp/y in women (P = 0.081 for sex difference).

Mean daily energy intakes of 2322 kcal in men and 2036 kcal in women at baseline comprised 16–17% protein, 49% carbohydrates, and 35–36% fats. The proportions of energy intake attributable to PUFAs (8.4% and 9.3% in men and women, respectively), MUFAs (13.0% and 12.9%, respectively), and SFAs (11.3% and 11.2%, respectively) were as expected for this population. Alcohol intake in this population was characteristically extremely low in both sexes, more so for women, as was the level of exercise. Smoking prevalence was high in men.

Significant inverse Spearman correlations of baseline dietary energy and macronutrient intake with both baseline and follow-up LTL (but not with attrition of LTL) were shown in men only, whereas associations in women were very weak and not significant (Table 2). At follow-up, correlations in men were −0.17 for total energy, −0.18 for total fat, −0.14 for protein, −0.13 for carbohydrate, −0.18 for both PUFAs and MUFAs and −0.14 for SFAs. These negative correlations were attenuated when nutrient intake was expressed as a proportion of total energy intake, and only those for total fat (r = −0.11), MUFAs (r = −0.14), and PUFAs (r = −0.11) remained nominally significant.

TABLE 2.

Nonparametric (Spearman) correlations of diet measured at baseline with LTL at ages ∼30 y (baseline) and ∼43 y (follow-up) and with average annual shortening of telomere length in men and women: the Jerusalem LRC longitudinal study¹

	Men (n = 405)2				Women (n = 204)2
Baseline variables	Mean ± SD	Baseline LTL	Follow-up LTL	LTL change/y	Mean ± SD	Baseline LTL	Follow-up LTL	LTL change/y
Total energy intake (kcal)	2322 ± 810	−0.150 (0.002)	−0.169 (0.001)	0.060 (0.23)	2036 ± 731	0.029 (0.7)	−0.005 (0.9)	0.113 (0.11)
(kcal/kg body weight)	30.6 ± 11.6	−0.154 (0.002)	−0.167 (0.001)	0.036 (0.5)	33.5 ± 13.5	0.016 (0.8)	−0.019 (0.8)	0.101 (0.15)
Total protein (g/d)	95.2 ± 36.0	−0.117 (0.019)	−0.135 (0.007)	0.066 (0.18)	79.4 ± 27.8	0.027 (0.7)	−0.005 (0.9)	0.093 (0.19)
Total fat (g/d)	92.1 ± 42.8	−0.157 (0.002)	−0.176 (<0.001)	0.044 (0.4)	82.0 ± 36.2	0.057 (0.4)	0.029 (0.7)	0.096 (0.17)
Total carbohydrates (g/d)	282 ± 102	−0.119 (0.016)	−0.133 (0.007)	0.053 (0.29)	248 ± 94	0.011 (0.9)	−0.022 (0.7)	0.113 (0.11)
SFAs (g/d)	29.7 ± 15.9	−0.119 (0.017)	−0.135 (0.007)	0.046 (0.4)	26.0 ± 13.3	0.087 (0.22)	0.057 (0.4)	0.093 (0.19)
MUFAs (g/d)	34.2 ± 17.6	−0.156 (0.002)	−0.178 (<0.001)	0.050 (0.3)	29.6 ± 13.8	0.051 (0.5)	0.027 (0.7)	0.084 (0.23)
PUFAs (g/d)	22.2 ± 11.4	−0.162 (0.001)	−0.175 (<0.001)	0.011 (0.8)	21.0 ± 10.3	0.003 (0.97)	−0.023 (0.7)	0.106 (0.13)
Protein (% of energy)	16.6 ± 3.4	0.097 (0.5)	0.095 (0.057)	0.036 (0.5)	15.8 ± 3.3	−0.044 (0.5)	−0.054 (0.4)	−0.018 (0.8)
Carbohydrates (% of energy)	48.6 ± 8.0	0.058 (0.24)	0.063 (0.21)	0.002 (0.97)	48.6 ± 7.0	−0.023 (0.7)	−0.022 (0.8)	0.008 (0.9)
Fat (% of energy)	34.8 ± 7.2	−0.105 (0.035)	−0.109 (0.028)	−0.017 (0.7)	35.5 ± 6.4	0.074 (0.3)	0.070 (0.3)	0.013 (0.9)
SFAs (% of energy)	11.3 ± 3.7	−0.017 (0.7)	−0.025 (0.6)	0.018 (0.7)	11.2 ± 3.1	0.099 (0.16)	0.081 (0.25)	0.042 (0.5)
MUFAs (% of energy)	13.0 ± 3.3	−0.128 (0.010)	−0.137 (0.006)	−0.013 (0.8)	12.9 ± 3.0	0.056 (0.4)	0.067 (0.3)	−0.034 (0.6)
PUFAs (% of energy)	8.4 ± 2.5	−0.101 (0.042)	−0.106 (0.034)	−0.031 (0.5)	9.3 ± 2.7	−0.043 (0.5)	−0.044 (0.5)	0.029 (0.7)
BMI (kg/m2)	25.2 ± 3.5	−0.018 (0.7)	−0.020 (0.7)	0.028 (0.6)	24.0 ± 3.7	−0.045 (0.5)	−0.049 (0.5)	0.011 (0.9)
Waist-hip ratio	0.88 ± 0.05	−0.009 (0.9)	−0.029 (0.6)	0.068 (0.17)	0.77 ± 0.06	−0.026 (0.7)	−0.036 (0.6)	0.076 (0.28)
Skinfold thickness3 (cm)	5.45 ± 1.87	−0.059 (0.24)	−0.061 (0.22)	0.016 (0.8)	5.80 ± 1.82	−0.069 (0.3)	−0.088 (0.21)	0.050 (0.5)
Weight change from age 30 to 43 y (kg)	7.9 ± 7.0	0.031 (0.5)	−0.047 (0.4)	−0.010 (0.8)	6.0 ± 6.3	−0.025 (0.7)	−0.053 (0.5)	0.165 (0.018)
Pack-years of smoking (baseline)	5.7 ± 7.0	−0.032 (0.5)	−0.065 (0.19)	0.133 (0.007)	2.7 ± 5.0	−0.005 (0.9)	−0.002 (0.97)	0.015 (0.8)
Plasma LDL cholesterol (mg/dL)	109 ± 29	−0.105 (0.036)	−0.102 (0.043)	0.008 (0.87)	99 ± 27	0.089 (0.20)	0.066 (0.35)	0.093 (0.19)
Plasma HDL cholesterol (mg/dL)	36.5 ± 10.3	0.013 (0.79)	0.044 (0.38)	−0.076 (0.13)	44.9 ± 11.0	0.131 (0.061)	0.084 (0.23)	0.195 (0.005)
Plasma triglycerides (mg/dL)	136 ± 85	0.018 (0.72)	−0.013 (0.80)	0.132 (0.008)	92 ± 41	−0.038 (0.59)	0.004 (0.96)	−0.074 (0.29)

¹LTL, leukocyte telomere length; LRC, Lipid Research Clinic.

²P values in parentheses.

³Sum of triceps, subscapular, and supra-iliac skinfold thicknesses.

Men, but not women, who were current smokers at baseline tended to have shorter LTL than did past and never smokers (mean ± SD baseline LTL: 7190 ± 643, 7462 ± 617, and 7248 ± 700 bp, respectively; P = 0.028; and follow-up LTL: 6854 ± 589, 7074 ± 574, and 6965 ± 670 bp, respectively; P = 0.044). Pack-years of smoking, exercise, and anthropometric measures at baseline were not significantly associated with LTL measured at either time point in either sex, and BMI at follow-up remained unrelated to LTL. Baseline plasma LDL cholesterol was inversely related to both baseline and follow-up LTL in men (r = −0.11, P = 0.036, and r = −0.10, P = 0.043) but not in women. Baseline HDL cholesterol and triglycerides were not associated with baseline or follow-up LTL. There were no significant macronutrient correlations with shortening of LTL over time in either sex over the mean 13.1 y follow-up period. Correlations of energy intake with LTL attrition were weak and not significant [0.06 (P = 0.23) in men and 0.11 (P = 0.11) in women]. Associations noted for change in LTL during follow-up were positive correlations of pack-years of smoking with LTL shortening in men (r = 0.13, P = 0.007) and of weight change from baseline to follow-up with LTL attrition in women (r = 0.17, P = 0.018). Plasma triglycerides in men (r = 0.13, P = 0.008) and, unexpectedly, HDL cholesterol in women (r = 0.19, P = 0.005) were associated with increased LTL shortening.

We next assessed the diet-LTL association by using the baseline diet to predict LTL at follow-up. Regression analyses of LTL on baseline calorie intake showed an inverse sex-adjusted association (β = −0.112, P = 0.006) that was entirely attributable to the association in men (β = −0.171, P = 0.0005, compared with β <0.001, P = 0.997, in women; P = 0.039 for sex interaction). This inverse association that was restricted to men was evident when calories were grouped by sex-specific quintiles, which reflected a wide distribution of energy intake (with ascending quintile cutoffs of 1650, 2008, 2351, and 3002 kcal in men and 1387, 1779, 2142, and 2610 kcal in women) (Figure 1). In men there was a generally graded inverse association with a difference of 244 bp (95% CI: 59, 429 bp; P = 0.01) between the mean TRFs of the extreme quintiles of calorie intake. Because there also was no evidence of significant macronutrient associations with LTL in women, and because of the apparent sex interactions (for PUFAs, P = 0.073, and for MUFAs, P = 0.025), we restricted our data presentation in Table 3 to men. The inverse association of calorie intake with LTL was modestly attenuated when adjusted for possible confounders, none of which made significant independent contributions to the multivariable model (β = −0.157, P = 0.002) or when follow-up smoking was introduced instead of baseline smoking (β = −0.163, P = 0.001). The unadjusted association persisted when caloric intake per kilogram of body weight was used (β = −0.167, P = 0.0007) or when height, skinfold thicknesses, waist circumference, or waist-hip ratio were added to the model (not shown in Table 3). The association was unaffected by adjustment for plasma LDL and HDL-cholesterol, for non–HDL cholesterol, or for ln triglycerides. The significant sex × energy interaction was unaffected when adjusted for plasma lipids and lipoproteins.

FIGURE 1.

LTL measured at a mean age of 43.2 y by quintile of calorie intake determined at a mean age of 30.1 y in 405 men and 204 women (unadjusted). Quintile cutoffs were as follows: 1650, 2008, 2351, and 3002 kcal for men and 1387, 1779, 2142, and 2610 kcal for women. Vertical bars represent 95% CIs. Test for trend in quintiles (coded 1–5) was performed by linear regression. P = 0.075 for sex × energy interaction. 1 kb = 1000 base pairs. kb, kilobase; LTL, leukocyte telomere length.

TABLE 3.

Regression models in 405 men of single macronutrient predictors of leukocyte telomere length at follow-up, adjusted for energy intake and covariates¹

	Model 12		Model 23
Models	Standardized β (g/d)	Adjusted for covariates4	Standardized β (% of energy)	Adjusted for covariates5
1. Energy (ln kcal)	−0.171 (0.0005)	−0.157 (0.002)
2. Energy	−0.207 (0.021)	−0.198 (0.035)	−0.168 (0.001)	−0.153 (0.004)
Protein	0.043 (0.63)	0.049 (0.60)	0.012 (0.82)	0.012 (0.76)
3. Energy	−0.276 (0.010)	−0.268 (0.016)	−0.165 (0.001)	−0.149 (0.003)
Carbohydrates	0.118 (0.27)	0.119 (0.28)	0.053 (0.28)	0.057 (0.26)
4. Energy	−0.050 (0.63)	−0.020 (0.86)	−0.155 (0.002)	−0.138 (0.009)
Total fat	−0.137 (0.19)	−0.153 (0.15)	−0.067 (0.19)	−0.072 (0.16)
5. Energy	−0.178 (0.021)	−0.157 (0.048)	−0.174 (<0.001)	−0.160 (0.002)
SFAs	0.009 (0.91)	0.000 (1.0)	0.022 (0.65)	0.020 (0.69)
6. Energy	−0.021 (0.83)	−0.003 (0.98)	−0.144 (0.005)	−0.129 (0.014)
MUFAs	−0.175 (0.066)	−0.178 (0.065)	−0.104 (0.042)	−0.100 (0.050)
7. Energy	−0.087 (0.30)	−0.058 (0.50)	−0.156 (0.002)	−0.138 (0.007)
PUFAs	−0.103 (0.22)	−0.120 (0.16)	−0.087 (0.080)	−0.098 (0.053)
8. Energy	−0.172 (0.001)	−0.158 (0.002)	−0.172 (0.001)	−0.158 (0.002)
Marine n−3	0.016 (0.75)	0.026 (0.61)	−0.012 (0.81)	−0.008 (0.87)

¹R² values for models 1 and 2 were between 2.9% and 3.9% (unadjusted) and 4.5% and 5.7% (adjusted). P values are shown in parentheses.

²Model 1 nutrients were expressed as grams per day and were ln-transformed.

³Model 2 nutrients were expressed as percentage of energy and were not ln-transformed.

⁴Adjusted for exercise (2 dummy variables), smoking (2 dummy variables), and BMI at baseline and for origin (Europe, North Africa, Asia, Israel; 3 dummy variables).

⁵Adjusted for covariates as for model 1. Additional adjustment of all models for person-years of smoking (instead of the smoking variables) and weight change during follow-up slightly augmented the associations.

The association of calorie intake with LTL persisted in bivariate models that included energy intake together with either protein, carbohydrates, or saturated fat expressed in grams per day (Table 3, model 1). However, in models that included energy intake together with total fat, MUFAs, or PUFAs, also expressed in grams per day, neither energy intake nor the dietary fat variables remained significantly associated with LTL, indicating that their strong covariance had cancelled out the independent contribution of the other (model 1). On introduction of the nutrients expressed as a percentage of total energy (Table 3, model 2), the inverse energy-LTL association persisted in all bivariate models (with protein included in the model: β for energy = −0.168, P = 0.001; with carbohydrates: β for energy = −0.165, P = 0.0008; with total fat: β for energy = −0.155, P = 0.002) and in the covariate-adjusted models. PUFAs and MUFAs showed marginally significant inverse associations with LTL. These analyses were unaffected by inclusion of plasma lipoproteins in the models. The explained variance (R²) was modest: for the unadjusted models 1 and 2, it was between 2.9% and 3.9% and for the adjusted models, 4.5% and 5.7%.

We assessed the effect of smoking status on the energy-LTL relation. There was no significant difference in the mean (±SD) BMI of men by smoking status (current, past, and never smoker: 25.4 ± 3.6, 25.1 ± 2.8, and 25.0 ± 3.7, respectively; P = 0.58). However, male current smokers reported a significantly higher mean energy intake (2464 ± 841, 2033 ± 686, and 2264 ± 784 kcal, respectively; P = 0.0008 for ln energy) and had a higher intake per kilogram of body weight (P = 0.006 for ln energy/kg). An interaction of smoking status and energy intake with LTL was noted in men. The energy-LTL association was largely restricted to men who had never smoked (covariate-adjusted β = −0.259, P = 0.0008, compared with β = −0.103, P = 0.166, for ever smokers; P = 0.050 for the smoking × calorie intake interaction) and persisted when adjusted also for height. The difference between the LTL means of the upper and lower quintiles of energy intake among the never smokers was 434 bp (95% CI: 116, 752 bp). There was no evidence for modification of the energy-LTL association by BMI (P > 0.7).

In multivariable analyses that allowed all of the nutrients (expressed in g/d) to enter in a backward stepwise elimination procedure, with energy intake forced into the model (Table 4, model 3), energy was no longer associated with LTL, and MUFA was significantly inversely related to LTL with SFA retained in the model but was attenuated when adjusted for covariates. In a multivariable model of nutrients expressed as a proportion of calories (Table 4, model 4), only MUFAs as a percentage of total energy intake was retained in the model (providing coefficients identical to model 2) and energy remained significantly associated with LTL.

TABLE 4.

Backward stepwise regression in 405 men of leukocyte telomere length on all macronutrient predictors introduced into the models, with adjustment for energy intake and covariates¹

	Model 323		Model 424
Variables	Standardized β (g/d)	Adjusted for covariates5	Standardized β (% of energy)	Adjusted for covariates5
Energy	−0.039 (0.69)	−0.003 (0.98)	−0.144 (0.005)	−0.129 (0.014)
SFAs	0.154 (0.114)
MUFAs	−0.293 (0.015)	−0.178 (0.065)	−0.104 (0.042)	−0.100 (0.050)

¹R² values for model 3 were 4.3% (unadjusted) and 5.6% (adjusted) and for model 4 were 3.9% (unadjusted) and 5.7% (adjusted). P values are shown in parentheses.

²P > 0.15 for variable elimination.

³Model 3 nutrients were expressed as grams per day and were ln-transformed.

⁴Model 4 nutrients were expressed as percentage of energy (and were not ln-transformed).

⁵Adjusted for exercise (2 dummy variables), smoking (2 dummy variables), and BMI at baseline and for origin (Europe, North Africa, Asia, Israel; 3 dummy variables).

DISCUSSION

Few studies have assessed dietary associations with LTL, and fewer, if any, have used the more laborious Southern blot method of measuring telomere length, the current gold standard (37). We found calorie intake at a mean age of 30 y to be significantly inversely associated with LTL at the mean ages of 30 and 43 y but not with LTL change during the relatively brief 13-y window of follow-up. This finding was confined to men and was evident for all of the macronutrients (although more so for unsaturated fatty acids), suggesting that energy per se, and not specific nutrients, might play a role. Adjustment of each nutrient (g/d) for caloric intake attenuated or eliminated the nutrient-LTL associations. With the nutrients expressed as a percentage of calories (whereby the nutrients are far less strongly associated with caloric intake), energy remained inversely related to LTL. PUFAs as a percentage of total energy intake and MUFAs as a percentage of total energy intake showed marginally significant inverse associations with LTL in bivariate models, but only MUFAs as a percentage of total energy intake remained in the multivariable backward stepwise procedure. Associations were largely unaffected by covariate adjustment.

The energy-LTL finding was mainly confined to men who had never smoked. This arguably reflects the “natural” state not tampered with by smoking effects on metabolism and energy consumption, as for example in our study in which smokers reported greater energy intake according to body weight. The literature is inconsistent with regard to these effects (eg, references 39–44), whereas weight gain after smoking cessation is frequent and has been recently attributed to increased caloric consumption rather than decreased energy expenditure (44). Smoking, which is also associated with increased oxidative stress and inflammation (45), appeared to abrogate the protective effects of lower caloric intake on LTL.

In animal models, energy restriction has been associated with increased life span and appears to retard the aging process (12). In humans, caloric restriction of 20–25% under controlled trial conditions was associated with biomarkers of aging and longevity (15, 17). In our observational study of free-living men, there was a broad distribution of reported energy intake, with >2-fold differences between the extreme quintiles. We note that there was a difference of 440 kcal (95% CI: 180,700 kcal; P = 0.001) in the baseline mean calorie intakes of the first and fifth quintiles of follow-up LTL. Our findings of an inverse association between energy intake and LTL in young men appear to be consistent with a beneficial effect of lower energy intake on biomarkers of longevity, which has been shown in energy-restriction trials (15, 17).

A cross-sectional analysis in men and women aged 45–84 y in the MESA (Multiethnic Study of Atherosclerosis) showed a nonsignificant (P = 0.11) inverse association of energy intake with LTL. Age- and sex-specific analyses were not reported (18). In cross-sectional data from the Nurses’ Health Study, there was no association of caloric intake with LTL in women (19). A similar absence of association was seen in elderly Chinese (20). A possible explanation for the discrepancy in the findings of our study with the few others reported might be the age difference, because our study sample was younger. In rodent models the association of caloric restriction with longevity was age dependent, with the effect being greater for those in which the intervention was implemented early in life (13). We point out in this regard that the attrition of LTL is greatest during the first years of life in humans, when growth is most rapid (1). Furthermore, diets may change with time, and consequently cross-sectional comparisons, particularly at older ages, would not reflect past dietary exposures that may have affected LTL change. Last, we note that previous studies that examined this association used qPCR. The lower precision of LTL measurement by qPCR than by Southern blots (37, 46) might play a role.

We do not have a compelling explanation for the sex difference in our findings—ie, the absence of an association in women. This could reflect the lower power of the smaller sample size in women. Alternatively, the absence of consistency between the sexes in the age range of our study participants, in which the women were premenopausal, might be due to biological (hormonal) sex differences that may modulate the association. Others faced with sex differences in dietary-LTL associations (20) invoked an explanation of a possibly greater resistance of women to effects of oxidative stress on telomeres, a resistance not evident in our cohort in which premenopausal women showed, in fact, a slightly higher rate of LTL attrition than did men. Another explanation could relate to possible sex differences in the validity of dietary reporting, which might result in differential misclassification. A possible hint in this direction is the higher reported energy intake per kilogram of body weight in women than men in our sample (P = 0.002). Although self-reported dietary intake is prone to error, and energy intake tends to be generally underestimated (eg, references 47 and 48), a review of studies that used doubly labeled water did not show a sex difference in underreporting (48). Last, we must also consider the possibility that the inverse association evident in men is a chance finding, notwithstanding the high statistical significance.

A multivariable-adjusted inverse association of qPCR-determined LTL with energy-adjusted PUFAs (g/d) that was noted among women in the Nurses’ Health Study was ascribed predominantly to linoleic acid, the intake of which has increased considerably in the United States. The authors surmised that this putatively negative effect of PUFAs was counterbalanced by its protective effects, such as on plasma lipids (19). Among elderly Hong Kong Chinese residents (aged ≥65 y), energy-adjusted fats and oils used for cooking (fatty acid composition not given) were associated with shorter LTL (measured by qPCR) in women but not in men (20). In our study, the unadjusted inverse associations of PUFAs and MUFAs (g/d) with LTL observed in men (but not in women) were attenuated by adjustment for energy. MUFAs and PUFAs, when expressed as a percentage of energy, were also marginally inversely associated with LTL with energy in the model. In multivariable nutrient stepwise models that included energy, the inverse association of MUFAs as a percentage of total energy intake persisted. We note that although MUFAs are conventionally regarded as protective and are recommended (together with PUFAs) to replace SFAs in the diet, reservations have also been raised about high MUFA intakes (49).

We observed no significant associations of LTL at baseline or follow-up with pack-years of cigarette smoking, vigorous exercise, BMI, other measures of obesity at baseline, and BMI at follow-up in either sex, or with weight change during follow-up in men. Current smoking was associated with shorter LTL in men, pack-years of smoking with longitudinal LTL shortening in men, and weight gain with LTL attrition in women. Findings across studies, most of which have used qPCR (eg, references 18–20 and 50), and some Southern blots (eg, references 7, 51, and 52), are inconsistent.

In conclusion, our findings require confirmation. In light of the inverse energy-LTL association observed, we suggest that replicative studies be undertaken, particularly in childhood and adolescence. The possible interaction with smoking status should be evaluated. Unraveling the independent relations of energy and macronutrients with LTL remains a challenge. There is some support in the literature for inverse associations of total fat and unsaturated fatty acids with LTL. This question, too, warrants further investigation.