The SUN study is a prospective and multipurpose cohort composed of Spanish graduates and conducted in a Mediterranean setting: Spain.22 The recruitment of participants started in 1999 and the enrollment in the study is permanently open under the identifier NCT02669602 at ClinicalTrials.gov. A detailed description of the methods is available elsewhere.22 Briefly, baseline and follow-up data are collected through self-reported postal or web-based questionnaires every 2 years.22 The SUN cohort gathers information on a wide array of self-reported, previously validated characteristics, including anthropometric variables (such as body mass index),22 clinical history (such as total cholesterol,23 blood pressure24 and glucose levels25), sociodemographic and lifestyle-related factors (such as a 136-item semiquantitative Food Frequency Questionare,26 smoking status22 and physical activity27). The research protocol was written in accordance with the principles of the Declaration of Helsinki and was approved by the University of Navarre institutional review board. Specific written informed consent was requested to participate in this study.

In May 2008, 1921 adults older than 55 years at the baseline questionnaire, were invited to participate in a genetic study.28 A total of 1085 accepted to participate and 986 returned saliva samples,28 although only 953 samples could be correctly analyzed. Moreover, 67 participants were excluded because they reported a total energy intake outside Willett's predefined values (< 800 or> 4000kcal/d for men and <500 or> 3500kcal/d for women).29 The final sample available for this cross-sectional study was 886 (figure 1 of the supplementary data).

The primary endpoint was shortening of TL. Saliva samples from participants were collected with specialized kits (DNA Collection-Oragene OG-250 Saliva Kit). DNA was extracted according to the manufacturer's instructions and frozen at −80∘C. The measurement of TL was made with the monochrome multiplex real-time quantitative polymerase chain reaction (RT-qPCR) following the method of Cawthon described elsewhere.30 The T/S ratio, which is proportional to average TL, is obtained by the comparison of these 2 parameters compared with a reference DNA. The master mix contained 10 ng genomic DNA, a QuantiTect CYBR Green PCR kit (Qiagen), telomere and albumin primers, and nuclease-free water to complete the final volume (10μL). Telomere primers (telc and teg) and albumin primers (albu and albd) were combined (final concentration 900nM each). The primer sequences (Sigma Aldrich St Louis, MO, USA; purified by high-performance liquid chromatography) were telg (5′-ACACTAAGGTTTGGGTTTGGGTTTGGGTTTGGGTT AGTGT-3′), telc (5′ TGTTAGGTATCCCTATCCCTATCCCTA TCCCTATCCCTAACA-3′), albu (5′ -CGGCGGCGGGCGG CGCGGGCTGGGCGGAAATGCTGCACAGAATCCTTG-3′), and albd (5′-GCCCGGCCCGCCGCGCCCGTCCCGCCGGA AAAGCATGGTCGCCTGTT-3′). The qPCR used was the CFX 384TM Real-Time System (BioRad) with the following protocol: 15minutes at 95∘C for enzyme activation followed by 2 cycles of 95∘C at 15seconds and 49∘C at 15seconds, and 35 cycles of 15s at 95∘C, 10seconds at 63∘C, 15seconds at 74∘C (first signal acquisition), and 15seconds at 88∘C (second signal acquisition). For each sample, we generated a melting curve from 45 to 95∘C ramped at 0.2∘C/s. We used 384 well plates and for quality control the samples were run in triplicate.

A calibration curve of reference DNA samples (150–2.34 ng/mL in 2-fold dilutions; linearity agreement R2> 0.99) was included on each high-throughput plate.

CVH score was assessed based on 7 key risk factors: smoking status, physical activity, diet, body mass index, total cholesterol, fasting blood glucose, and blood pressure. Each metric was categorized into 3 levels of “poor”, “intermediate” and “ideal” and assigned scores of 0, 1 and 2, respectively, according to the AHA definitions3 (table 1). Total score therefore ranged from 0 to 14, and our variable was categorized in 3 groups: poor (0 to 9 points), intermediate (10 to 11 points) and ideal (12 to 14 points). Variables were taken from the questionnaire immediately prior to the saliva collection, except for physical activity and Healthy Eating Index index, which have only data in the baseline questionnaire.

Smoking status was categorized as poor (current smokers), intermediate (former smokers) and ideal (never smokers). We classified physical activity according to frequency and duration of moderate and vigorous intensity of each activity as poor (no exercise), intermediate (1 to 149minutes per week of moderate exercise or 1 to 74minutes of vigorous exercise) and ideal (more than 150minutes of moderate exercise or more than 75minutes of vigorous exercise per week).

Diet was assessed with the Alternative Healthy Eating Index (AHEI-2010). For this index, we used information supplied by Chiuve et al..31 It ranged from 50 to 110. This index is based on food and nutrients predictive of chronic disease risk.31 Vegetables, fruit, sugar-sweetened beverages and fruit juice, nuts and legumes, red/processed meat, transfatty acids, long-chain omega-3 fatty acids, polyunsaturated fatty acids, sodium, and alcohol intake were scored from 0 to 10. In this analysis, AHEI-2010 was divided into poor diet quality (< 50), intermediate diet quality (≥ 50 to≤80) and ideal diet quality (> 81).32

Body mass index was categorized as poor (≥ 30kg/m2), intermediate (≥ 25 to <30kg/m2) and ideal (< 25kg/m2). Blood pressure was classified using systolic blood pressure (SBP) (mmHg) and diastolic blood pressure (DBP) (mmHg) as poor (SBP≥140 or DBP≥90), intermediate (SBP≥120 to <140 or DBP≥80 to <90) and ideal (SBP <120 or DBP <80). For participants with only 1 blood pressure measurement recorded, that single reading was used. For the missing data, we considered poor SBP for participants taking any antihypertensive medication and ideal SBP and DBP for those reporting not having prevalent hypertension. Cholesterol was classified as poor (≥ 240mg/dL), intermediate (≥ 200 to <240mg/dL) and ideal (< 200mg/dL). Fasting glucose was classified as poor (≥ 126mg/dL), intermediate (≥ 100 to≤125mg/dL) and ideal (< 100mg/dL). For the missing data, we considered poor cholesterol for participants with hypercholesterolemia diagnosis or taking any anticholesteremic agents, and ideal cholesterol for those reporting not having prevalent cholesterol.

The information about covariates was collected at the baseline questionnaire of the SUN project, which includes age at saliva collection and age at inclusion (years); sex; marital status; years of university education; family history of hypertension, diabetes, obesity and CVD; and prevalence of cancer and depression. Other covariates included were sedentary lifestyle, special diets, alcohol consumption, sodium intake, and ultra-processed food consumption.

For the main analysis, participants were divided into tertiles of poor (0-9), intermediate (10-11) and ideal (12-14) CVH score. We examined the characteristics of the study sample according to the CVH categories using inverse probability weighting with adjustment for age at saliva collection and sex. For continuous variables, we used means±standard deviation, and percentages for categorical variables. As a quality control in our study, the Spearman correlation was performed to assess the association between TL and age at saliva collection. We also used the Spearman correlation to assess the association between TL and CVH score.

We calculated the number and percentage of participants by number of metrics. Furthermore, we calculated the number of participants by ideal CVH metrics.

As it was previously defined,21,33,34 short telomeres were defined as a TL bellow the 20th percentile. To assess the association between CVH score or CHV metrics and having short telomeres, logistic regression models were fitted to estimate the odds ratio (OR) and 95% confidence intervals (95%CI) considering the first tertile (poor CVH) as the reference category. We ran 2 adjusted models: the first one adjusted for age at saliva collection and sex, and the second one additionally adjusted for age at inclusion, marital status (single, married, widowed, divorced); years of university education; family history of hypertension, diabetes, obesity, and CVD; and prevalence of cancer and depression. Furthermore, tests of linear trend across successive tertiles were performed by assigning to each participant the value of the median in his/her tertile and considering these variables as continuous.

ANCOVA analysis was carried out to evaluate the mean adjusted TL according to tertiles of CVH score. To evaluate the association between each of the upper tertiles (intermediate and ideal CVH) compared with the lower tertile (poor CVH), multivariable linear regression adjusted models were carried out.

Multivariable linear regression models for TL with CVH score were done to explore the linear relationship between these 2 variables.

The P value for interaction between sex and CVH score was calculated using the likelihood ratio test for each scenario.

We also conducted sensitivity analyses to verify the robustness of our findings by rerunning the model with additional adjustments: using the 5th and 95th percentiles as limits for allowable total energy intake, ultra-processed food and alcohol consumption, following a special diet at baseline, sodium intake, sedentary lifestyle and weight gain; and excluding participants with a personal history of obesity, CVD, diabetes or hypertension.

Logistic regression analyses were used to calculate the predicted and observed probability, with the odds of having short TL (≤ 20th percentile) as the dependent variable, and the number of metrics of CVH score as the independent variable. Then, a graph was obtained using the observed and predicted marginal probabilities.

Statistical analyses were performed using STATA version 14.0 (StataCorp). We considered 2-tailed P values <.05 to be statistically significant.

The strengths of this study include the technique used for the measurement of TL (RT-qPCR) which allows quantification of telomeres and the single copy gene in the same well in a single reaction, reducing potential measurement errors.

However, this study also has some limitations. First, participants of the SUN cohort had a healthier profile probably because they were graduates and thus they were more aware of the importance of a good diet and lifestyle; therefore, this study may not represent the general population. Second, the low number of women in our study may account for the differences found between the 2 sexes. Third, DNA was isolated from saliva samples containing leukocytes and epithelial cells at varying proportions.43 Even though the measurement of TL differs in cells and tissues, salivary TL and leukocyte TL are positively correlated.43,44 Furthermore, the method is less expensive and less invasive for participants. Fourth, the study design is cross-sectional and thus causal effects cannot be inferred. Fifth, our ability to detect small associations, especially in the stratified analyses, may be limited due to the relatively small sample size. Sixth, it is not the same to study TL as a continuous variable in the whole population and to assess the effect of CVH score, especially in those participants with shorter TL (below the 20th percentile), which is an outcome of greater magnitude. These differences may explain the failure to find significant linear associations between TL and CVH score. Possiby, an ideal CVH score may only exert a protective association for more powerful outcomes than for small changes in TL in the whole population.