The role of adolescent lifestyle habits in biological aging: A prospective twin study

  1. Anna Kankaanpää  Is a corresponding author
  2. Asko Tolvanen
  3. Aino Heikkinen
  4. Jaakko Kaprio
  5. Miina Ollikainen
  6. Elina Sillanpää
  1. Gerontology Research Center (GEREC), Faculty of Sport and Health Sciences, University of Jyväskylä, Finland
  2. Methodology Center for Human Sciences, University of Jyväskylä, Finland
  3. Institute for Molecular Medicine Finland (FIMM), HiLife, University of Helsinki, Finland

Abstract

Background:

Adolescence is a stage of fast growth and development. Exposures during puberty can have long-term effects on health in later life. This study aims to investigate the role of adolescent lifestyle in biological aging.

Methods:

The study participants originated from the longitudinal FinnTwin12 study (n = 5114). Adolescent lifestyle-related factors, including body mass index (BMI), leisure-time physical activity, smoking, and alcohol use, were based on self-reports and measured at ages 12, 14, and 17 years. For a subsample, blood-based DNA methylation (DNAm) was used to assess biological aging with six epigenetic aging measures in young adulthood (21–25 years, n = 824). A latent class analysis was conducted to identify patterns of lifestyle behaviors in adolescence, and differences between the subgroups in later biological aging were studied. Genetic and environmental influences on biological aging shared with lifestyle behavior patterns were estimated using quantitative genetic modeling.

Results:

We identified five subgroups of participants with different adolescent lifestyle behavior patterns. When DNAm GrimAge, DunedinPoAm, and DunedinPACE estimators were used, the class with the unhealthiest lifestyle and the class of participants with high BMI were biologically older than the classes with healthier lifestyle habits. The differences in lifestyle-related factors were maintained into young adulthood. Most of the variation in biological aging shared with adolescent lifestyle was explained by common genetic factors.

Conclusions:

These findings suggest that an unhealthy lifestyle during pubertal years is associated with accelerated biological aging in young adulthood. Genetic pleiotropy may largely explain the observed associations.

Funding:

This work was supported by the Academy of Finland (213506, 265240, 263278, 312073 to J.K., 297908 to M.O. and 341750, 346509 to E.S.), EC FP5 GenomEUtwin (J.K.), National Institutes of Health/National Heart, Lung, and Blood Institute (grant HL104125), EC MC ITN Project EPITRAIN (J.K. and M.O.), the University of Helsinki Research Funds (M.O.), Sigrid Juselius Foundation (J.K. and M.O.), Yrjö Jahnsson Foundation (6868), Juho Vainio Foundation (E.S.) and Päivikki and Sakari Sohlberg foundation (E.S.).

Editor's evaluation

This is an important article that is methodologically compelling that provides evidence that an unhealthy lifestyle during adolescence accelerates epigenetic age in adulthood and that these associations are largely explained by the effect of shared genetic influences. The main strengths of this article are the relatively large sample size, longitudinal assessment of lifestyle factors, and sophisticated statistical analyses. The article will be of interest to a broad audience, including individuals working on methylation, epidemiology, and/or aging.

https://doi.org/10.7554/eLife.80729.sa0

eLife digest

For most animals, events that occur early in life can have a lasting impact on individuals’ health. In humans, adolescence is a particularly vulnerable time when rapid growth and development collide with growing independence and experimentation. An unhealthy lifestyle during this period of rapid cell growth can contribute to later health problems like heart disease, lung disease, and premature death. This is due partly to accelerated biological aging, where the body deteriorates faster than what would be expected for an individual’s chronological age.

One way to track the effects of lifestyle on biological aging is by measuring epigenetic changes. Epigenetic changes consist on adding or removing chemical ‘tags’ on genes. These tags can switch the genes on or off without changing their sequences. Scientists can measure certain epigenetic changes by measuring the levels of methylated DNA – DNA with a chemical ‘tag’ known as a methyl group – in blood samples. Several algorithms – known as ‘epigenetic clocks’ – are available that estimate how fast an individual is aging biologically based on DNA methylation.

Kankaanpää et al. show that unhealthy lifestyles during adolescence may lead to accelerated aging in early adulthood. For their analysis, Kankaanpää et al. used data on the levels of DNA methylation in blood samples from 824 twins between 21 and 25 years old. The twins were participants in the FinnTwin12 study and had completed a survey about their lifestyles at ages 12, 14, and 17.

Kankaanpää et al. classified individuals into five groups depending on their lifestyles. The first three groups, which included most of the twins, contained individuals that led relatively healthy lives. The fourth group contained individuals with a higher body mass index based on their height and weight. Finally, the last group included individuals with unhealthy lifestyles who binge drank, smoked and did not exercise.

After estimating the biological ages for all of the participants, Kankaanpää et al. found that both the individuals with higher body mass indices and those in the group with unhealthy lifestyles aged faster than those who reported healthier lifestyles. However, the results varied depending on which epigenetic clock Kankaanpää et al. used to measure biological aging: clocks that had been developed earlier showed fewer differences in aging between groups; while newer clocks consistently found that individuals in the higher body mass index and unhealthy groups were older. Kankaanpää et al. also showed that shared genetic factors explained both unhealthy lifestyles and accelerated biological aging.

The experiments performed by Kankaanpää et al. provide new insights into the vital role of an individual’s genetics in unhealthy lifestyles and cellular aging. These insights might help scientists identify at risk individuals early in life and try to prevent accelerated aging.

Introduction

Epidemiological studies of life course have indicated that exposures during early life have long-term effects on later health (Kuh et al., 2003). Unhealthy environments and lifestyle habits during rapid cell division can affect the structure or functions of organs, tissues, or body systems, and these changes can subsequently affect health and disease in later life (Biro and Deardorff, 2013; Power et al., 2013). For example, lower birth weight and fast growth during childhood predispose individuals to coronary heart disease and increased blood pressure in adulthood (Osmond and Barker, 2000). In addition to infancy and childhood, adolescence is also a critical period of growth.

Adolescence is characterized by pubertal maturation and growth spurts. Early pubertal development is linked to worse health conditions, such as obesity and cardiometabolic risk factors in adulthood (Prentice and Viner, 2013). However, childhood obesity can lead to early onset of puberty, especially among girls (Li et al., 2017; Richardson et al., 2020), and, therefore, can confound the observed associations between early pubertal development and worse later health. Moreover, early pubertal development is linked to substance use and other risky behaviors in adolescence (Hartman et al., 2017; Savage et al., 2018), but the associations are partly explained by familial factors (Savage et al., 2018).

Many unhealthy lifestyle choices, such as smoking initiation, alcohol use, and a physically inactive lifestyle, are already made in adolescence and increase the risk of developing several noncommunicable diseases over the following decades (Lopez et al., 2006). Once initiated, unhealthy habits are likely to persist into adulthood (Latvala et al., 2014; Maggs and Schulenberg, 2005; Rovio et al., 2018; Salin et al., 2019). A recent systematic review showed that healthy habits tend to cluster during childhood and adolescence, and typically, about half of the adolescents fall into subgroups characterized by healthy lifestyle habits (Whitaker et al., 2021). However, small minorities of adolescents are classified as heavy substance users or as having multiple other risk behaviors (Whitaker et al., 2021). The long-term consequences of the accumulation of unhealthy adolescent behaviors on health in later life have been rarely studied.

An unhealthy lifestyle in adolescence can affect biological mechanisms of aging at the molecular level and, subsequently, morbidity. Epigenetic alterations, including age-related changes in DNA methylation (DNAm), constitute a primary hallmark of biological aging (López-Otín et al., 2013). Epigenetic clocks are algorithms that aim to quantify biological aging using DNAm levels within specific CpG sites. The first-generation clocks, Horvath’s and Hannum’s clocks, were trained to predict chronological age (Hannum et al., 2013; Horvath, 2013), whereas the second-generation clocks, such as DNAm PhenoAge and GrimAge, are better predictors of health span and lifespan (Levine et al., 2018; Lu et al., 2019). For epigenetic clocks, the difference between an individual’s epigenetic age estimate and chronological age provides a measure of age acceleration (AA). The DunedinPoAm estimator differs from its predecessors in that it has been developed to predict the pace of aging (Belsky et al., 2020). The pace of aging describes longitudinal changes over 12 years in several biomarkers of organ-system integrity among same-aged individuals. Recently, the DunedinPACE estimator, which constitutes an advance on the original DunedinPoAm, was published (Belsky et al., 2022). DunedinPACE was trained to predict pace of aging measured over 20-year follow-up, and only the reliable probes were used in the prediction. From the life-course perspective, epigenetic aging measures are useful tools to assess biological aging at all ages and detect changes induced by lifetime exposures.

Previous studies have linked several lifestyle-related factors, such as higher body mass index (BMI), smoking, alcohol use, and lower leisure-time physical activity (LTPA), with accelerated biological aging measured using epigenetic clocks (Oblak et al., 2021; Quach et al., 2017). However, most of these studies were based on cross-sectional data on older adults. The first studies on the associations of adolescent lifestyle-related exposures with biological aging assessed with epigenetic aging measures indicated that advanced pubertal development, higher BMI, and smoking are associated with accelerated biological aging in adolescence (Etzel et al., 2022; Raffington et al., 2021; Simpkin et al., 2017).

The few previous studies conducted on this topic have focused on single lifestyle factors, and a comprehensive understanding of the role of adolescent lifestyle in later biological aging remains unclear. Our first aim is to define the types of lifestyle behavior patterns that can be identified in adolescence using data-driven latent class analysis (LCA). The second aim is to investigate whether the identified behavioral subgroups differ in biological aging in young adulthood and whether the associations are independent of baseline pubertal development. The third aim is to assess the genetic and environmental influences shared between biological aging and adolescent lifestyle behavior patterns.

Methods

The participants were Finnish twins and members of the longitudinal FinnTwin12 study (born during 1983–1987) (Kaprio, 2013; Rose et al., 2019). A total of 5600 twins and their families initially enrolled in the study. At the baseline, the twins filled out the questionnaires regarding their lifestyle-related habits at 11–12 years of age, and follow-up assessments were conducted at ages 14 and 17.5 years. The response rates were high for each assessment (85–90%). In young adulthood, at an average age of 22 years, blood samples for DNA analyses were collected during in-person clinical studies after written informed consent was signed. The data on health-related behaviors were collected with questionnaires and interviews. A total of 1295 twins of the FinnTwin12 cohort were examined and measured either in-person or through telephonic interviews. DNAm was determined and biological aging was assessed for 847 twins, out of which 824 twins had also information on lifestyle-related habits in adolescence. Data collection was conducted in accordance with the Declaration of Helsinki. The Indiana University IRB and the ethics committees of the University of Helsinki and Helsinki University Central Hospital approved the study protocol (113/E3/2001and 346/E0/05).

DNAm and assessment of biological age

Genomic DNA was extracted from peripheral blood samples using commercial kits. High molecular weight DNA samples (1 μg) were bisulfite converted using EZ-96 DNA methylation-Gold Kit (Zymo Research, Irvine, CA) according to the manufacturer’s protocol. The twins and co-twins were randomly distributed across plates, with both twins from a pair on the same plate. DNAm profiles were obtained using Illumina’s Infinium HumanMethylation450 BeadChip or the Infinium MethylationEPIC BeadChip (Illumina, San Diego, CA). The Illumina BeadChips measure single-CpG resolution DNAm levels across the human genome. With these assays, it is possible to interrogate over 450,000 (450k) or 850,000 (EPIC) methylation sites quantitatively across the genome at single-nucleotide resolution. Of the samples included in this study, 744 were assayed using 450k and 80 samples using EPIC arrays. Methylation data from different platforms was combined and preprocessed together using R package minfi (Aryee et al., 2014). We calculated detection p-values comparing total signal for each probe to the background signal level to evaluate the quality of the samples (Maksimovic et al., 2016). Samples of poor quality (mean detection p>0.01) were excluded from further analysis. Data were normalized by using the single-sample Noob normalization method, which is suitable for datasets originating from different platforms (Fortin et al., 2017). We also used Beta-Mixture Quantile (BMIQ) normalization (Teschendorff et al., 2013). Beta values representing CpG methylation levels were calculated as the ratio of methylated intensities (M) to the overall intensities (beta value = M/(M + U + 100), where U is unmethylated probe intensity). These preprocessed beta values were used as input in the calculations of the estimates of epigenetic aging.

We utilized six epigenetic clocks. The first four clocks, namely, Horvath’s and Hannum’s epigenetic clocks (Hannum et al., 2013; Horvath, 2013) and DNAm PhenoAge and DNAm GrimAge estimators (Levine et al., 2018; Lu et al., 2019), produced DNAm-based epigenetic age estimates in years by using a publicly available online calculator (https://dnamage.genetics.ucla.edu/new) (normalization method implemented in the calculator was utilized, as well). For these measures, AA was defined as the residual obtained from regressing the estimated epigenetic age on chronological age (AAHorvath, AAHannum, AAPheno, and AAGrim, respectively). The fifth and sixth clocks, namely, DunedinPoAm and DunedinPACE estimators, provided an estimate for the pace of biological aging in years per calendar year (Belsky et al., 2020; Belsky et al., 2022). DunedinPoAm and DunedinPACE were calculated using publicly available R packages (https://github.com/danbelsky/DunedinPoAm38; Belsky et al., 2020 and https://github.com/danbelsky/DunedinPACE; Belsky et al., 2022, respectively). The epigenetic aging measures were screened for outliers (>5 standard deviations away from mean). One outlier was detected according to DunedinPACE and was recoded as a missing value.

The components of DNAm GrimAge (adjusted for age) were also obtained, including DNAm-based smoking pack-years and the surrogates for plasma proteins (DNAm-based plasma proteins): DNAm adrenomedullin (ADM), DNAm beta-2-microglobulin (B2M), DNAm cystatin C, DNAm growth differentiation factor 15 (GDF15), DNAm leptin, DNAm plasminogen activator inhibitor 1 (PAI-1), and DNAm tissue inhibitor metalloproteinases 1 (TIMP-1).

Lifestyle-related factors in adolescence

BMI at ages 12, 14, and 17 years

BMI (kg/m2) was calculated based on self-reported height and weight.

LTPA at ages 12, 14, and 17 years

The frequency of LTPA at the age of 12 years was assessed with the question ‘How often do you engage in sports (i.e., team sports and training)?’ The answers were classified as 0 = less than once a week, 1 = once a week, and 2 = every day. At ages 14 and 17 years, the question differed slightly: ‘How often do you engage in physical activity or sports during your leisure time (excluding physical education)?’ The answers were classified as 0 = less than once a week, 1 = once a week, 2 = 2–5 times a week, and 3 = every day.

Smoking status at ages 14 and 17 years

Smoking status was determined using the self-reported frequency of smoking and classified as 0 = never smoker, 1 = former smoker, 2 = occasional smoker, and 3 = daily smoker.

Alcohol use (binge drinking) at ages 14 and 17 years

The frequency of drinking to intoxication had the following classes: ‘How often do you get really drunk?’ 0 = never, 1 = less than once a month, 2 = approximately once or twice a month, and 3 = once a week or more.

Pubertal development at age 12 years

Baseline pubertal development was assessed using a modified five-item Pubertal Development Scale (PDS) questionnaire (Petersen et al., 1988; Wehkalampi et al., 2008). Both sexes answered three questions each concerning growth in height, body hair, and skin changes. Moreover, boys were asked questions about the development of facial hair and voice change, while girls were asked about breast development and menarche. Each question had response categories 1 = growth/change has not begun, 2 = growth/change has barely started, and 3 = growth/change is definitely underway, except for menarche, which was dichotomous, 1 = has not occurred or 3 = has occurred (see also Wehkalampi et al., 2008). PDS was calculated as the mean score of the five items, and higher values indicated more advanced pubertal development at age 12 years.

Lifestyle-related factors in young adulthood at age 21–25 years

BMI (kg/m2) was calculated based on the measured height and weight.

LTPA was assessed using the Baecke questionnaire (Baecke et al., 1982). A sport index was based on the mean scores of four questions on sports activity described by Baecke et al., 1982 and Mustelin et al., 2012 for the FinnTwin12 study. The sport index is a reliable and valid instrument to measure high-intensity physical activity (Richardson et al., 1995).

Smoking was self-reported and classified as never, former, or current smoker.

Alcohol use (100% alcohol grams/day) was derived from the Semi-Structured Assessment for the Genetics of Alcoholism (Bucholz et al., 1994) and based on quantity and frequency of use and the content of alcoholic beverages, assessed by trained interviewers.

Statistical analysis

Patterns of lifestyle behaviors in adolescence

To identify the patterns of lifestyle behaviors in adolescence, an LCA was conducted, which is a data-driven approach to identify homogenous subgroups in a heterogeneous population. The classification was based on BMI and LTPA at ages 12, 14, and 17 years and smoking status and alcohol use at ages 14 and 17 years (10 indicator variables). All variables were treated as ordinal variables, except for continuous BMI. The classification was based on the thresholds of the ordinal variables and the means and variances of BMI.

An LCA model with 1–8 classes was fitted. The following fit indices were used to evaluate the goodness of fit: Akaike’s information criterion, Bayesian information criterion (BIC), and sample size-adjusted BIC. The lower values of the information criteria indicated a better fit for the model. Moreover, we used the Vuong–Lo–Mendell–Rubin likelihood ratio (VLMR) test and the Lo–Mendell–Rubin (LMR) test to determine the optimal number of classes. The estimated model was compared with the model with one class less, and the low p-value suggested that the model with one class less should be rejected. At each step, the classification quality was assessed using the average posterior probabilities for most likely latent class membership (AvePP). AvePP values close to 1 indicate a clear classification. In addition to the model fit, the final model for further analyses was chosen based on the parsimony and interpretability of the classes.

Differences in biological aging

The mean differences in biological aging between the lifestyle behavior patterns were studied using the Bolck–Croon–Hagenaars approach (Asparouhov and Muthén, 2021). The class-specific weights for each participant were computed and saved during the latent class model estimation. After that, a secondary model conditional on the latent lifestyle behavior patterns was specified using weights as training data: Epigenetic aging measures were treated as distal outcome one at a time, and the mean differences across classes were studied while adjusting for sex, age and baseline pubertal development. Similarly, the mean differences in the components of DNAm GrimAge and lifestyle-related factors in young adulthood were studied. The models of epigenetic aging measures were additionally adjusted for BMI in adulthood. To evaluate the effect sizes, standardized mean differences (SMDs) were calculated.

Genetic and environmental influences

Genetic and environmental influences on biological aging in common with lifestyle behavior patterns were studied using quantitative genetic modeling. For simplicity, we adjusted the epigenetic aging variables for sex, age, and baseline pubertal development prior to the analysis.

We first carried out univariate modeling to study genetic and environmental influences on epigenetic aging measures (Neale and Cardon, 1992). The variance in the epigenetic aging measures was decomposed into the latent variables representing additive genetic (A), dominant genetic (D), or shared environmental (C) and non-shared environmental (E) components (ACE model or ADE model). The sequences of the models were fitted (ACE, ADE, AE, CE, and E). Because dominance in the absence of additive effects is rare, the model including D and E components (DE-model) was omitted. We used Satorra–Bentler scaled chi-squared (χ2) test, comparative fit index (CFI), Tucker–Lewis index (TLI), root mean square error of approximation (RMSEA), and standardized root-mean-square residual (SRMR) to evaluate the goodness of fit of the models. The model fits the data well if the χ2 test is not statistically significant (p>0.05), CFI and TLI values are close to 0.95, the RMSEA value is below 0.06, and the SRMR value is below 0.08 (Hu and Bentler, 1999). Moreover, BIC was used to compare non-nested models. A lower BIC value indicates a better model fit. The most parsimonious model with the sufficient fit to the data was considered optimal.

On the one hand, as described above, total variance in biological aging was decomposed in the components explained by genetic, shared, and unshared environmental factors (aTot2+ cTot2+eTot2)(=VarTot) (Figure 1A). On the other hand, we can use the secondary model to study the differences in biological aging between the adolescent lifestyle behavior patterns, as described above, and decompose the variance in biological aging into the variance explained by the adolescent lifestyle behavior patterns Var(Model) and the variance of the residual term Var(Res). We also conducted univariate modeling for the residual term of biological aging, which corresponds to the variation in biological aging not explained by the adolescent lifestyle behavior patterns (aRes2+ cRes2+eRes2) (= VarRes) (Figure 1B). The residual terms were obtained by specifying a latent variable corresponding to the residuals of the secondary model described above (without including covariates), and the factor scores were saved. Finally, the proportion of variation in biological aging explained by the genetic factors shared with adolescent lifestyle patterns was evaluated as follows: (aTot2-aRes2)/ VarTot. The proportion of variation in epigenetic aging explained by the environmental factors was evaluated similarly. These proportions reflect the extent to which the same genetic/environmental factors contribute to the association between the adolescent lifestyle patterns and biological aging (i.e., size of the genetic and environmental correlations between the phenotypes).

Decomposition of (A) total variation in biological aging and (B) the variation of the residual term.

Missing data were assumed to be missing at random (MAR). The model parameters were estimated using the full information maximum likelihood (FIML) method with robust standard errors. Under the MAR assumption, the FIML method produced unbiased parameter estimates. The standard errors of the latent class models and secondary models were corrected for nested sampling (TYPE = COMPLEX). Descriptive statistics were calculated using IBM SPSS Statistics for Windows, version 20.0 (IBM Corp, Armonk, NY), and further modeling was conducted using Mplus, version 8.2 (Muthén and Muthén, 1998).

Results

The descriptive statistics of the study variables are presented in Table 1. A total of 5114 twins answered questionnaires on lifestyle-related behaviors during their adolescent years at least once. For 824 twins, epigenetic aging estimates were obtained. The mean age (SD) of the twins having information on biological aging was 22.4 (0.7) years. The means of the epigenetic age estimates were estimated as follows: Horvath’s clock 28.9 (3.6), Hannum’s clock 18.2 (3.3), DNAm PhenoAge 13.0 (5.3), and DNAm GrimAge 25.2 (3.3) years. The intraclass correlation coefficients (ICCs) of epigenetic aging measures were consistently higher in MZ twin pairs than in DZ twin pairs (Table 2). This suggests an underlying genetic component in biological aging. The correlations between the different epigenetic aging measures ranged from –0.12 to 0.73. The lowest correlation was observed between AAHorvath and DunedinPoAm and between AAHorvath and DunedinPACE. All other correlations were positive. The highest correlations (>0.5) were observed between AAHannum and AAPheno, AAGrim and DunedinPoAm, and DunedinPoAm and DunedinPACE.

Table 1
Descriptive statistics of the adolescent lifestyle-related variables in all twins and in the subsample of twins with information on biological aging.
All twins (n = 5114)Subsample (n = 824)
nMean (SD) or %nMean (SD) or %
Zygosity4852824
 MZ165034.033540.7
 Same-sex DZ160333.026231.8
 Opposite-sex DZ159933.022727.5
Sex5114824
 Female258450.547057.0
 Male253049.535443.0
At age 12
 Pubertal development (1–3)51111.6 (0.5)8231.6 (0.5)
 Body mass index491317.6 (2.6)79317.7 (2.6)
 Leisure-time physical activity5038813
 Less than once a week187737.329535.3
 Once a week249949.641651.2
 Every day66213.110212.5
At age 14
 Body mass index447319.3 (2.7)78719.5 (2.6)
 Leisure-time physical activity4590799
 Less than once a week68815.011013.8
 Once a week79617.314918.6
 2–5 times a week218247.537046.3
 Every day92420.117021.3
 Smoking status4570800
 Never395486.568785.9
 Former2966.5577.1
 Occasional1222.7243.0
 Daily smoker1984.3324.0
 Alcohol use (binge drinking)4565796
 Never350176.760275.6
 Less than once a month75616.613517
 Once or twice a month2756.0506.3
 Once a week or more330.791.1
At age 17
 Body mass index415821.4 (3.0)76021.4 (2.7)
 Leisure-time physical activity4208766
 Less than once a week74817.813217.2
 Once a week68616.313017.0
 2–5 times a week197747.036347.4
 Every day79718.914118.4
 Smoking status4190762
 Never241957.745459.7
 Former49311.88310.9
 Occasional2135.1486.3
 Daily smoker106525.417623.1
 Alcohol use (binge drinking)4217766
 Never88120.915219.8
 Less than once a month180742.934044.4
 Once or twice a month124029.422229.0
 Once a week or more2896.9526.8
  1. MZ, monozygotic twins; DZ, dizygotic twins; SD, standard deviation.

Table 2
The intraclass correlation coefficients (ICCs) of epigenetic aging measures by zygosity and correlation coefficients between the measures (n = 824).
ICCs (95% CI)Correlation coefficients (95% CI) off-diagonal and means (standard deviations) on the diagonal
MZ twin pairsDZ twin pairsAAHorvathAAHannumAAPhenoAAGrimDunedinPoAmDunedinPACE
AAHorvath0.71 (0.63, 0.79)0.40 (0.24, 0.55)0.00 (3.51)
AAHannum0.66 (0.56, 0.76)0.32 (0.16, 0.48)0.40 (0.33, 0.48)0.00 (3.27)
AAPheno0.69 (0.60, 0.78)0.16 (0.00, 0.33)0.36 (0.29, 0.44)0.61 (0.56, 0.66)0.00 (5.25)
AAGrim0.72 (0.63, 0.80)0.35 (0.15, 0.55)0.08 (0.01, 0.16)0.32 (0.24, 0.40)0.39 (0.33, 0.46)0.00 (3.24)
DunedinPoAm0.62 (0.52, 0.71)0.42 (0.24, 0.60)–0.05 (-0.12, 0.03)0.20 (0.13, 0.27)0.41 (0.35, 0.47)0.57 (0.52, 0.63)1.00 (0.07)
DunedinPACE0.71 (0.64, 0.78)0.46 (0.31, 0.61)–0.04 (–0.11, 0.04)0.30 (0.22, 0.38)0.49 (0.43, 0.55)0.55 (0.49, 0.61)0.62 (0.57, 0.67)0.88 (0.10)
  1. CIs were corrected for nested sampling.

  2. CI, confidence interval; AA, age acceleration; MZ, monozygotic; DZ, dizygotic.

Patterns of lifestyle behaviors

Increasing the number of classes continued to improve AIC, BIC, and ABIC (Table 3). However, the VLMR and LMR tests indicated that even a solution with four classes would be sufficient. In the fifth step, a class of participants with high BMI was extracted. Previous studies have shown the role of being overweight or obese in biological aging (Lundgren et al., 2022). After including the sixth class, the information criteria still showed considerable improvement, but the AvePPs for several classes were below 0.8. For these reasons, and to have adequate statistical power for subsequent analyses, a five-class solution was considered optimal. The AvePPs ranged from 0.78 to 0.91 for the five-class solution, indicating reasonable classification quality.

Table 3
Model fit of the latent class models (n = 5114).
AICBICABICVLMRLMRClass sizesAvePP
128842129012128929
122533122880122711<0.001<0.00174.0%, 26.0%0.95, 0.92
119937120460120206<0.001<0.00144.9%, 40.5%, 14.6%0.88, 0.89, 0.93
118030118729118389<0.001<0.00136.4%, 32.7%, 16.7%, 14.2%0.83, 0.86, 0.87, 0.92
1171671180431176170.5290.53032.0%, 22.8%, 19.9%, 15.9%, 9.5%0.78, 0.82, 0.85, 0.88, 0.91
1165261175781170760.1690.17031.5%, 18.5%, 15.7%, 14.0%, 12.7%, 7.7%0.77, 0.84, 0.83, 0.78, 0.78, 0.90
1160991173281167310.0430.04421.0%, 17.5%, 15.2%, 13.8%, 12.9%, 12.8%, 6.9%0.73, 0.82, 0.70, 0.77, 0.83, 0.83, 0.91
1156951171011164180.4070.40820.3%, 16.2%, 13.6%, 13.5%, 12.3%, 11.3%, 9.3%, 3.4%0.72, 0.75, 0.82, 0.71, 0.83, 0.80, 0.82, 0.89
  1. AIC, Akaike’s information criterion; BIC, Bayesian information criterion; ABIC, sample size-adjusted Bayesian information criterion; VLMR, Vuong–Lo–Mendell–Rubin likelihood ratio test; LMR, Lo–Mendell–Rubin-adjusted likelihood ratio test; AvePP, average posterior probabilities for most likely latent class membership.

Of the participants, 32% fell into the class of healthiest lifestyle habits (C1) (see Figure 2, and the distributions of indicator variables according to the adolescent lifestyle behavior patterns in Table 4). They had normal weight, on average, and were more likely to engage in regular LTPA compared to the other groups; most of them were non-smokers and did not use alcohol regularly. Every fifth (19.9%) participant belonged to the second class (C2), characterized by the low mean level of BMI in the range of normal weight for children (low-normal BMI) (Cole et al., 2007). They also had healthy lifestyle habits, but they were not as physically active as the participants in class C1. The participants placed in the third class (C3, 22.8%) had lifestyle habits similar to those of the participants in class C1; however, they had a higher level of BMI in the range of normal weight for children (high-normal BMI). About every tenth (9.5%) of the participants belonged to the fourth class (C4), with the highest level of BMI (high BMI). At each measurement point, the mean BMI level exceeded the cutoff points for overweight in children (Cole et al., 2007). The prevalence of daily smoking was slightly higher in C4 compared to classes C1, C2, and C3. Of the participants, 15.9% were classified into the subgroup characterized by the unhealthiest lifestyle behaviors (C5). Most of them were daily smokers and used alcohol regularly at the age of 17. They also had a lower probability of engaging in regular LTPA compared to the other groups; however, they were of normal weight, on average.

Figure 2 with 1 supplement see all
Classes with different lifestyle patterns (n = 5114).

Mean and probability profiles (95% confidence intervals) of the indicator variables utilized in the classification: (A) body mass index, (B) regular leisure-time physical activity (LTPA) (several times a week), (C) daily smoking, and (D) regular alcohol use (once a month or more). For categorical variables, the probabilities of belonging to the highest categories are presented.

Figure 2—source data 1

The estimation results of a latent class analysis (LCA) model with five classes.

https://cdn.elifesciences.org/articles/80729/elife-80729-fig2-data1-v1.xlsx
Table 4
The classes with different adolescent lifestyle behavior patterns (n = 5114).
C1 (32.0%)C2 (19.9%)C3 (22.8%)C4 (9.5%)C5 (15.9%)
Est95% CIEst95% CIEst95% CIEst95% CIEst95% CI
Body mass index
 At age of 12 years16.815.7, 17.915.214.7, 15.719.117.5, 20.722.721.7, 23.817.216.9, 17.5
 At age of 14 years18.617.6, 19.516.716.0, 17.320.919.2, 22.624.823.3, 26.218.918.6, 19.2
 At age of 17 years20.819.7, 22.018.818.1, 19.422.621.6, 23.727.125.0, 29.220.620.3, 20.9
Leisure-time physical activity
 At age of 12 years
 Less than once a week0.290.22, 0.370.450.39, 0.510.350.26, 0.430.440.37, 0.500.440.39, 0.48
 Once a week0.540.48, 0.590.460.41, 0.500.520.47, 0.560.470.39, 0.540.460.41, 0.50
 Every day0.170.14, 0.210.090.04, 0.140.140.07, 0.210.100.06, 0.130.110.08, 0.14
 At age of 14 years
 Less than once a week0.080.05, 0.110.170.12, 0.220.140.07, 0.220.180.13, 0.230.270.22, 0.31
 Once a week0.140.07, 0.200.200.17, 0.240.160.10, 0.230.230.17, 0.280.200.16, 0.23
 2‒5 times a week0.520.45, 0.590.450.41, 0.490.510.43, 0.590.430.37, 0.490.400.35, 0.45
 Every day0.270.23, 0.300.180.13, 0.230.190.12, 0.250.170.12, 0.210.140.10, 0.17
 At age of 17 years
 Less than once a week0.100.05, 0.140.190.14, 0.230.130.06, 0.200.270.19, 0.350.350.29, 0.40
 Once a week0.150.11, 0.180.180.15, 0.210.150.11, 0.190.180.14, 0.230.190.15, 0.23
 2‒5 times a week0.500.44, 0.560.450.41, 0.490.530.48, 0.570.440.36, 0.520.360.32, 0.41
 Every day0.260.22, 0.290.180.13, 0.230.200.12, 0.270.110.07, 0.150.100.07, 0.13
Smoking status
 At age of 14 years
 Never0.990.98, 1.000.980.95, 1.000.970.95, 1.000.830.74, 0.930.330.24, 0.43
 Former0.010.00, 0.020.020.00, 0.030.020.00, 0.040.090.04, 0.140.290.24, 0.34
 Occasional0.000.01–0.01, 0.020.000.00, 0.010.040.01, 0.070.130.10, 0.16
 Daily smoker0.000.000.00, 0.010.000.040.00, 0.070.250.19, 0.31
 At age of 17 years
 Never0.690.61, 0.770.730.65, 0.810.680.59, 0.780.500.41, 0.590.030.00, 0.06
 Former0.120.09, 0.150.090.05, 0.130.120.07, 0.160.110.06, 0.160.150.12, 0.19
 Occasional0.060.04, 0.070.040.02, 0.060.040.01, 0.060.050.02, 0.070.070.05, 0.10
 Daily smoker0.130.08, 0.180.140.09, 0.180.170.09, 0.240.340.24, 0.440.740.69, 0.79
Alcohol use (binge drinking)
 At age of 14 years
 Never0.880.85, 0.910.940.90, 0.970.840.79, 0.890.760.69, 0.830.230.15, 0.31
 Less than once a month0.110.08, 0.140.050.02, 0.080.130.09, 0.170.180.12, 0.240.460.41, 0.51
 Once or twice a month0.010.00, 0.020.020.00, 0.030.030.01, 0.040.050.02, 0.080.270.22, 0.32
 Once a week or more0.000.000.000.000.00, 0.010.040.02, 0.06
 At age of 17 years
 Never0.210.18, 0.250.330.26, 0.410.220.16, 0.280.230.15, 0.300.010.00, 0.02
 Less than once a month0.480.43, 0.520.450.40, 0.490.460.41, 0.520.410.35, 0.470.260.22, 0.31
 Once or twice a month0.280.24, 0.320.180.12, 0.240.280.23, 0.330.290.23, 0.350.510.46, 0.55
 Once a week or more0.030.00, 0.060.040.02, 0.060.030.00, 0.060.080.04, 0.110.220.18, 0.26
  1. Mean and probability profiles of the indicator variables utilized in the classification.

  2. BMI, body mass index; Est, estimated mean or probability; CI, confidence interval; C1, the class with the healthiest lifestyle pattern; C2, the class with low-normal BMI; C3, the class with healthy lifestyle and high-normal BMI; C4, the class with high BMI; C5, the class with the unhealthiest lifestyle pattern.

Boys were slightly over-represented in the classes that were most physically active (C1, C3) and had the highest levels of BMI (C3, C4) (percentage of boys: C1: 57.2%; C3: 51.5%; and C4: 52.7%), and under-represented in the classes with lowest levels of BMI (C2) and the unhealthiest lifestyle behavior pattern (C5) (C2: 42.7%; C5: 44.1%). There were also differences in pubertal development at baseline between the groups. The subgroups with the highest levels of BMI (C3, C4) and the class with unhealthiest lifestyle habits (C5) were, on average, the most advanced in pubertal development (mean PDS, C3: 1.67 95% CI: [1.63–1.71], C4: 1.69 [1.64–1.74]; and C5: 1.68 [1.63–1.72]), while the class with the healthiest lifestyle pattern (C1) and that with the lowest level of BMI (C2) were less advanced in pubertal development (C1: 1.53 [1.50–1.56]; C2: 1.44 [1.41–1.47]).

The distribution of lifestyle behavior patterns in the subsample of participants having information on biological aging was very similar to that in the large cohort data (C1: 33.0%; C2: 16.6%; C3: 20.6%; C4: 10.1%; C5: 19.7%). In the subsample, the differences in lifestyle-related factors were maintained well over the transition from adolescence to young adulthood (Figure 2—figure supplement 1).

Differences in biological aging

There were differences among the classes in AAPheno (Wald test: p=0.006), AAGrim (p=2.3e-11), DunedinPoAm (p=3.1e-9), and DunedinPACE (p=5.5e-7) in the models adjusted for sex, age, and baseline pubertal development. There were no differences in biological aging when Horvath’s clock (p=0.550) and Hannum’s clock (p=0.487) were used. The overall results considering AAGrim, DunedinPoAm, and DunedinPACE were very similar (Figure 3 and Table 5).

Figure 3 with 1 supplement see all
Mean differences between the adolescent lifestyle behavior patterns in biological aging measured with (A) DNAm PhenoAge, (B) DNAm GrimAge, (C) DunedinPoAm, and (D) DunedinPACE estimators (n = 824).

The analysis was adjusted for sex (female), standardized age, and baseline pubertal development. Means and 95% confidence intervals are presented. C1, the class with the healthiest lifestyle pattern; C2, the class with low-normal body mass index (BMI); C3, the class with a healthy lifestyle and high-normal BMI; C4, the class with high BMI; C5, the class with the unhealthiest lifestyle pattern; AA, age acceleration.

Figure 3—source data 1

Means and 95% confidence intervals of biological aging according to the adolescent lifestyle behavior patterns (BCH approach).

https://cdn.elifesciences.org/articles/80729/elife-80729-fig3-data1-v1.xlsx
Table 5
Differences in biological aging between classes with different adolescent lifestyle behavior patterns.
AAPhenoAAGrimDunedinPoAmDunedinPACE
Diff95% CISMDDiff95% CISMDDiff95% CISMDDiff95% CISMD
C2 vs. C1
 M1–0.55–2.15, 1.06–0.10–0.57–1.37, 0.23–0.18–0.01–0.03, 0.01–0.14–0.03–0.05, 0.00–0.30
 M2–0.13–1.79, 1.54–0.02–0.54–1.38, 0.29–0.17–0.01–0.03, 0.01–0.14–0.01–0.04, 0.02–0.10
C3 vs. C1
 M11.04–0.54, 2.630.200.97–0.01, 1.950.300.00–0.02, 0.020.000.02–0.01, 0.040.20
 M20.60–1.01, 2.210.110.94–0.10, 1.970.290.00–0.02, 0.020.000.00–0.03, 0.030.00
C4 vs. C1
 M11.970.44, 3.500.381.830.74, 2.91*0.560.050.03, 0.07*0.710.070.04, 0.11*0.70
 M20.66–1.31, 2.630.131.730.26, 3.210.530.040.01, 0.07*0.570.02–0.02, 0.070.20
C5 vs. C1
 M1–0.36–1.76, 1.04–0.072.701.74, 3.66*0.830.040.02, 0.07*0.570.030.00, 0.060.30
 M2–0.45–1.82, 0.93–0.092.691.73, 3.66*0.830.040.02, 0.06*0.570.030.00, 0.060.30
C3 vs. C2
 M11.59–0.07, 3.250.301.540.58, 2.50*0.480.01–0.01, 0.040.140.040.01, 0.07*0.50
 M20.73–1.10, 2.550.141.480.36, 2.60*0.460.01–0.02, 0.030.140.01–0.03, 0.040.10
C4 vs. C2
 M12.520.85, 4.18*0.482.401.28, 3.51*0.740.070.04, 0.09*1.000.100.06, 0.14*1.00
 M20.79–1.59, 3.160.152.270.59, 3.95*0.700.050.02, 0.09*0.710.03–0.02, 0.080.30
C5 vs. C2
 M10.19–1.40, 1.770.043.272.32, 4.23*1.010.060.03, 0.08*0.860.060.03, 0.09*0.60
 M2–0.32–1.97, 1.33–0.063.242.21, 4.27*1.000.050.03, 0.08*0.710.040.01, 0.070.40
C4 vs. C3
 M10.93–0.82, 2.670.180.85–0.45, 2.160.260.050.03, 0.08*0.710.060.02, 0.10*0.60
 M20.06–1.91, 2.030.010.79–0.68, 2.260.240.050.02, 0.08*0.710.02–0.02, 0.070.20
C5 vs. C3
 M1–1.40–2.99, 0.18–0.271.730.62, 2.84*0.530.040.02, 0.07*0.570.02–0.02, 0.050.20
 M2–1.05–2.63, 0.54–0.201.760.63, 2.88*0.540.050.02, 0.07*0.710.030.00, 0.060.30
C5 vs. C4
 M1–2.33−3.84, –0.82*–0.440.88–0.32, 2.070.27–0.01–0.03, 0.02–0.14–0.04–0.08, 0.00–0.40
 M2–1.10–3.01, 0.80–0.210.96–0.51, 2.440.300.00–0.03, 0.030.000.01–0.04, 0.050.10
  1. AA, age acceleration; BMI, body mass index; Diff, difference; CI, confidence interval; SMD, standardized mean difference; C1, the class with the healthiest lifestyle pattern; C2, the class with low-normal BMI; C3, the class with healthy lifestyle and high-normal BMI; C4, the class with high BMI; C5, the class with the unhealthiest lifestyle pattern; M1, model was adjusted for sex, age, and pubertal status at age 12; M2, model was additionally adjusted for BMI in adulthood.

  2. *

    The corresponding 99% confidence interval did not overlap zero.

The group with the unhealthiest lifestyle pattern (C5) was, on average, 1.7–3.3 years biologically older than the groups with healthier lifestyle patterns and normal weight (C1–C3) when DNAm GrimAge was used to assess biological aging (Table 5, M1). Moreover, the unhealthiest group had, an average, 2–3 weeks/calendar year faster pace of biological aging, as measured with DunedinPoAm. The differences in DunedinPACE were very similar to those observed in DunedinPoAm, but there was no difference between the unhealthiest class (C5) and the class with a healthy lifestyle and high-normal BMI (C3) and, moreover, the difference between the healthiest class (C1) was not significant at 0.01 level.

When DNAm GrimAge was used, the group with a high BMI (C4) was, on average, 1.8–2.4 years biologically older than the two groups with healthier lifestyle patterns (C1 and C2) (Table 5, M1). When measured with the DunedinPoAm estimator, the class had, on average, 3–4 weeks/calendar year faster pace of aging, and when measured with the DunedinPACE estimator, it had 4–5 weeks/calendar year faster pace of aging. Moreover, when DunedinPoAm and DunedinPACE were used, the class had approximately 3 weeks/calendar year faster pace of aging compared to the group with healthy lifestyle with normal-high BMI (C3), and when DunedinPACE was used, the class had 2 weeks/calendar year faster pace of aging compared to the group with unhealthiest lifestyle pattern (C5). When DNAm PhenoAge was used to assess biological aging, only the group with a high BMI stood out. The group was biologically 2.0–2.5 years older than the groups with lower mean levels of BMI (C1–C2, C5). Based on the estimation results of the models, baseline pubertal development was associated with advanced biological aging only when Hannum’s clock was used to derive biological AA (standardized regression coefficient B = 0.10 [0.01–0.18]).

According to the previous literature, it is controversial whether childhood obesity has a direct effect on later health or whether the association is fully mediated by BMI in adulthood (Park et al., 2012). The role of adult BMI may depend on which disease outcome is studied (Richardson et al., 2020). After additionally adjusting the model for BMI in adulthood, the differences in AAPheno and DunedinPACE between the class of participants with high BMI (C4) and those with lower BMI (C1, C2, C5) were attenuated (Table 5, M2). This finding suggests that the observed differences in biological aging probably are fully mediated by BMI in adulthood. However, the differences in biological aging were only slightly attenuated when the DNAm GrimAge and DunedinPoAm estimators were used, suggesting that childhood overweight may leave permanent imprint on biological aging assessed with these measures. However, when DNAm GrimAge was used, the difference between the classes C4 and C1 was not significant at 0.01 level.

In our study, high standard deviations of epigenetic age estimates were observed. Therefore, variation in AA measures may largely be attributable to technical variation, which is not biologically meaningful. Recently developed principal component (PC)-based clocks are shown to improve the reliability and validity of epigenetic clocks (Higgins-Chen et al., 2022). We therefore replicated our main analyses using PC-based epigenetic clocks (data not shown). The standard deviations of epigenetic age estimates were similar or even higher compared with those obtained with the original clocks, but the correlations between AA measures assessed with different clocks were consistently higher when PC-based epigenetic clocks were used. Importantly, the observed associations with the adolescent lifestyle behavior patterns did not substantially change.

Differences in DNAm-based plasma proteins and smoking pack-years

Overall, after controlling for sex, age, and baseline pubertal development, there were differences in DNAm-based ADM (Wald test: p=0.010), B2M (p=0.014), and Packyrs (p=1.3e-5), but not in DNAm-based cystatin C (p=0.140), GDF15 (p=0.228), Leptin (p=0.228), PAI-1 (p=0.055), and TIMP-1 (p=0.089) between the adolescent lifestyle behavior patterns. The class with the unhealthiest lifestyle habits (C5) differed unfavorably from the other classes only by DNAm smoking pack-years while the class of participants with high BMI (C4) stood out by several DNAm-based plasma proteins including DNAm ADM, PAI-1, and TIMP-1 (Figure 3—figure supplement 1).

Genetic and environmental effects

Twin pairs with biological aging data on both members of the pair were used in the quantitative genetic modeling to estimate the genetic and environmental components of variance for biological aging (n = 154 monozygotic and 211 dizygotic pairs). The model including additive genetic and non-shared environmental component (AE model) was considered optimal for all the epigenetic aging measures (Table 6). Generally, ACE and ADE fit the data about as well, and models without genetic component (CE model) provided significantly worse fit. Based on these results, AE model was also chosen for the further modeling of the residual term of biological aging. Genetic factors explained 62–73% of the total variation in biological aging depending on the estimator. The rest of the variation (27–38%) was explained by unshared environmental factors.

Table 6
The estimation results of the univariate model for biological aging among young adult twin pairs (MZ n = 154, DZ n = 211).
Model fitParameter estimates and their 95% confidence intervals
Χ2dfSCpCFITLIRMSEASRMRBICa2/totalc2 or d2/totale2/totalTotal
AAPheno
 ACE5.231.270.1550.980.990.060.0620090.650.56, 0.740.000.350.26, 0.451.000.89, 1.12
 ADE0.630.990.9041.001.020.000.0220030.03–0.46, 0.510.650.15, 1.150.330.25, 0.410.990.88, 1.09
 AE7.040.960.1360.970.990.060.0620030.650.56, 0.74-0.350.26, 0.451.000.89, 1.12
 CE43.540.96<0.0010.600.800.230.112038-0.390.30, 0.480.610.52, 0.700.990.88, 1.10
 E10750.96<0.0010.000.590.330.212093--1.000.990.88, 1.10
AAGrim
 ACE4.332.050.2310.990.990.050.0919890.730.66, 0.800.000.270.20, 0.341.030.87, 1.20
 ADE5.631.550.1330.980.980.070.0919890.640.09, 1.190.09–0.48, 0.660.270.20, 0.341.030.87, 1.19
 AE5.741.540.2200.980.990.050.0919830.730.66, 0.80-0.270.20, 0.341.030.87, 1.19
 CE33.040.87<0.0010.720.860.200.122018-0.500.40, 0.600.500.41, 0.601.020.87, 1.17
 E10451.41<0.0010.060.620.330.262115--1.001.020.87, 1.17
DunedinPoAm
 ACE1.331.120.7221.001.020.000.0420030.520.20, 0.850.09–0.20, 0.370.390.30, 0.480.980.86, 1.11
 ADE1.231.600.7461.001.020.000.0420030.620.53, 0.700.000.380.30, 0.470.980.86, 1.10
 AE1.641.200.8021.001.020.000.0419970.620.53, 0.70-0.380.30, 0.470.980.86, 1.10
 CE12.741.100.0130.880.940.110.072009-0.450.36, 0.550.550.45, 0.640.980.86, 1.10
 E85.151.15<0.0010.000.550.300.222087--1.000.980.86, 1.10
DunedinPACE
 ACE2.031.080.5821.001.000.000.0519980.540.20, 0.870.08–0.21, 0.370.390.30, 0.480.990.87, 1.11
 ADE1.331.680.7400.991.000.000.0519810.42–0.13, 0.970.27–0.31, 0.840.320.24, 0.390.980.84, 1.13
 AE2.141.580.7241.001.100.000.0519760.680.52, 0.82-0.320.24, 0.400.990.84, 1.15
 CE23.741.45<0.0010.760.880.160.102007-0.450.35, 0.540.550.46, 0.650.980.84, 1.13
 E78.551.47<0.0010.090.640.280.232118--1.000.980.84, 1.13
  1. The epigenetic aging measures were adjusted for sex, age, and baseline pubertal development prior to analysis.

  2. SC, scaling correction; CFI, comparative fit index; RMSEA, root mean square error of approximation; SRMR, standardized root-mean-square residual; BIC, Bayesian information criterion; MZ, monozygotic; DZ, dizygotic.

The proportion of the total variation in biological aging in early adulthood explained by adolescent lifestyle behavior patterns was 3.7% for AAPheno, 16.8% for AAGrim, 15.4% for DunedinPoAm, and 10.5% for DunedinPACE (Figure 4). The association between adolescent lifestyle patterns and biological aging in early adulthood was largely explained by shared genetic influences; the genetic factors shared with adolescent lifestyle explained 3.7, 13.1, 12.6, and 10.5%, respectively, of the total variation in biological aging. Depending on the biological aging estimate, only 0–3.7% of the total variation in biological aging was explained by (unshared) environmental factors shared with adolescent lifestyle patterns. The rest of the total variation in biological aging was explained by genetic and (unshared) environmental factors unique to biological aging.

Proportions of the total variation in biological aging explained by genetic and (unshared) environmental factors shared with adolescent lifestyle patterns among young adult twin pairs (MZ n = 154, DZ n = 211).

The results are based on the model including additive genetic and non-shared environmental component (AE model). AA, age acceleration.

Figure 4—source data 1

Genetic and environmental factors underlying the association between adolescent lifestyle patterns and biological aging.

https://cdn.elifesciences.org/articles/80729/elife-80729-fig4-data1-v1.xlsx

Discussion

We conducted a twin study with a longitudinal lifestyle follow-up during the adolescent years and measured biological aging from genome-wide DNAm data using the most recent epigenetic aging clocks. Our findings supported previous studies, which showed that lifestyle-related behaviors tend to cluster in adolescence. In our study, most participants generally followed healthy lifestyle patterns, but we could also identify a group of young adults characterized by higher BMI (10% of all participants) in adolescence, as well as a group (16% of all participants) with more frequent co-occurrence of smoking, binge drinking, and low levels of physical activity in adolescence. We observed differences in biological aging between the classes characterized by adolescent lifestyle patterns in young adulthood, but the differences depended on the utilized epigenetic aging measure. Both the class with the overall unhealthiest lifestyle and that with a high BMI were biologically 1.7–3.3 years older than the classes with healthier lifestyle patterns when DNAm GrimAge was used to assess biological aging (AAGrim). Moreover, they had 2–5 weeks/calendar year faster pace of biological aging (DunedinPoAm). The class with high BMI was biologically the oldest one when and DNAm PhenoAge and DunedinPACE were used. There were no differences when Horvath’s and Hannum’s clocks were used to estimate biological aging. The differences in lifestyle-related factors were maintained well over the transition from adolescence to young adulthood. However, genetic factors shared with adolescent lifestyle explained most of the observed differences in biological aging.

In our study, when the most recently published epigenetic aging measures were used, the class with the unhealthiest lifestyle was biologically 1.7–3.3 years older (AAGrim) and had 2–3 weeks/calendar year faster pace of biological aging (DunedinPoAm) than the classes with healthier patterns. These measures can predict mortality and morbidity, especially cardiometabolic and lung diseases (Belsky et al., 2020; Belsky et al., 2022; Lu et al., 2019). A previous meta-analysis focusing on adults in a wide age range (17–99 years) showed that the number of healthy lifestyle behaviors is inversely associated with all-cause mortality risk (Loef and Walach, 2012). The mortality risk was up to 66% lower for individuals having multiple healthy behaviors compared to those adhering to an unhealthy lifestyle (smoking, low or high levels of alcohol use, unhealthy diet, no physical activity, and overweight). The accumulation of multiple unhealthy lifestyle habits during lifetime probably has a more detrimental effect on biological aging as well than any single lifestyle habit. However, our approach did not allow us to disentangle the effects of single lifestyle habits on biological aging. Our results suggest that the unhealthy lifestyle-induced changes in biological aging begin to accumulate in early life. These changes might predispose individuals to premature death in later life.

To the best of our knowledge, this is the first study to investigate common genetic influences underlying lifestyle clusters and biological aging. Our results suggest genetic correlation between adolescent lifestyle and biological aging; individuals who are genetically prone to unhealthy lifestyles or overweight in adolescence are also susceptible to faster biological aging later in young adulthood. The shared genetic influences on two phenotypes may be due to several scenarios (Solovieff et al., 2013). They may arise from genetic pleiotropy; in this case, the genes may be a common cause for both adolescent lifestyle and biological aging. Another possible reason is causal relation between the phenotypes. In this case, genetic factors may affect adolescent lifestyle, which lies on the causal path to biological aging (or vice versa). However, for the relationship to be causal, it is necessary that there are shared environmental influences on the phenotypes (De Moor et al., 2008). In our study, environmental influences shared with adolescent lifestyle on biological aging were observed only when DNAm GrimAge and DunedinPoAm estimators were used. In line with our study, McCartney et al., 2021 showed that there are shared underlying genetic contributions between single lifestyle factors and biological aging (AAGrim, AAPheno) using polygenetic risk scores for epigenetic AA. Their Mendelian randomization analysis also suggested causal influences of BMI and smoking on biological aging, but only when DNAm GrimAge was used.

To the best of our knowledge, this is also the first study reporting the association between adolescent BMI (relative weight) and biological aging in later life. Previous systematic reviews have concluded that being overweight or obese in childhood and adolescence has a consistent impact on mortality and morbidity in later life (Park et al., 2012; Reilly and Kelly, 2011). In particular, the associations with cardiometabolic morbidity are well-established, but the results of the studies investigating the associations independent of adult BMI are inconclusive (Park et al., 2012). A more recent study showed that early-life body size indirectly predisposes coronary artery disease and type 2 diabetes through body size in adulthood rather than having a direct effect (Richardson et al., 2020). Our results considering biological aging are in line with the existing literature but depend on the epigenetic clock utilized. In our study, the participants assigned to the class that was, on average, overweight in adolescence were biologically older (based on AAPheno, AAGrim, DunedinPoAm, and DunedinPACE) in young adulthood compared to the classes of normal weight and healthy lifestyle habits. The group stood out, especially when AAPheno and DunedinPACE were used to measure biological aging, but adult BMI explained the observed differences in these measures. Practically all variance of AAPheno and DunedinPACE shared with adolescent lifestyle was explained by shared genetic factors. Therefore, these measures probably capture aspects of biological aging that are attributed to genetic factors shared with BMI. Mainly, the differences in AAGrim and DunedinPoAm did not attenuate after additionally controlling for adult BMI, suggesting that higher BMI in adolescence has a direct long-term effect on biological aging measured with these epigenetic clocks.

LTPA is associated with a lower risk of mortality and cardiovascular diseases (Li et al., 2013; Löllgen et al., 2009). Twin studies and genetically informed studies have suggested that genetic pleiotropy can partly explain these frequently observed associations (Karvinen et al., 2015; Sillanpää et al., 2022). Previous studies have shown that LTPA is also associated with slower biological aging (Kankaanpää et al., 2021). In this study, lower levels of physical activity in adolescence were closely intertwined with other unhealthy behaviors. To fully understand the role of adolescence physical activity in later biological aging would require a more comprehensive analysis of activity patterns, intensities, and modes, as well as subgroup analyses that account for other lifestyle factors, such as diet.

Adolescent smoking behavior and alcohol use appeared to be strongly clustered, in line with the findings of a recent systematic review (Whitaker et al., 2021). For this reason, the associations of smoking and alcohol use with biological aging might be difficult to disentangle. Smoking is the most detrimental lifestyle factor, and its association with accelerated biological aging has been frequently reported (Oblak et al., 2021). However, the results obtained for the association between alcohol use and biological aging remain unclear (Oblak et al., 2021). A recent study showed that smoking has a causal effect on AAGrim, whereas alcohol use did not exhibit such effect (McCartney et al., 2021). Epigenetic methylation changes due to alcohol seem to be much fewer in number and magnitude compared to smoking exposure (Stephenson et al., 2021). In our study, the unhealthiest lifestyle class, in which smoking and alcohol use co-occurred, exhibited accelerated biological aging, especially when GrimAge and DunedinPoAm were used. These epigenetic aging measures are highly sensitive to tobacco exposure (Belsky et al., 2020; Lu et al., 2019). DNAm GrimAge is a composite biomarker comprising seven DNAm surrogates for plasma markers and smoking pack-years, which can predict the time to death (Lu et al., 2019). DunedinPoAm utilizes a specific CpG site (located within the gene AHRR), the methylation of which is strongly affected by tobacco exposure (Belsky et al., 2020). For these reasons, most of the variation in biological aging, which is explained by environmental factors shared with adolescent lifestyle, is probably due to smoking exposure.

To better understand the observed differences in biological aging, we also studied differences in DNAm-based surrogates included in the DNAm GrimAge estimator (Figure 3—figure supplement 1). Surprisingly, the class with the unhealthiest lifestyle pattern differed unfavorably from those with healthier habits only in DNAm-based smoking pack-years. The class with a high BMI had increased levels of several DNAm-based plasma markers, including DNAm PAI-1 and TIMP-1, which are associated with markers of inflammation and metabolic conditions (Lu et al., 2019). These findings support the suggestions that AAGrim is a useful biomarker for cardiovascular health and a potential predictor of cardiovascular disease already in young adulthood (Joyce et al., 2021).

Recent studies have yielded inconsistent results regarding the association between pubertal timing and biological aging (Hamlat et al., 2021; Maddock et al., 2021). In our models studying the differences in biological aging across adolescent lifestyle patterns, pubertal development at the age of 12 was not associated with accelerated biological aging in young adulthood (except for AAHannum). Moreover, the class with a high BMI included participants with advanced pubertal development, which might reflect the common genetic background underlying BMI and age at menarche (Kaprio et al., 1995). All these findings support the studies showing that childhood obesity, which tracks forward into adulthood, explains the observed associations between advanced pubertal status and worse cardiovascular health (Bell et al., 2018) and can further reflect the genetic architecture underlying BMI, pubertal development, and worse health (Day et al., 2015).

Our study has the following major strengths. Adolescent lifestyle-related patterns were identified using population-based large cohort data (N ~ 5000), with longitudinal measurements of lifestyle-related factors assessed using validated questionnaires. Response rates were high and the distribution of the lifestyle-related patterns in the subsample of twins with information on biological aging was similar to the distribution in large cohort data, supporting the generalizability of our findings. Moreover, adolescent lifestyle behavior patterns were identified using data-driven LCA. This approach enabled us to use all available data on adolescent lifestyle-related behaviors and identify the patterns without using artificial cutoff points for the variables. The reciprocal associations between different lifestyle-related factors, as well as their joint association with biological aging, are complex, and individual associations are difficult to interpret. However, our approach produced results with easy interpretation. The data were prospective, and biological aging was assessed with novel epigenetic aging measures, including a recently published DunedinPACE estimator. Furthermore, for the first time, we could evaluate the proportions of genetic and environmental influences underlying adolescent lifestyle as a whole in relation to biological aging by using quantitative genetic modeling. However, our study also has some limitations. Adolescent lifestyle-related behaviors were self-reported and, therefore, might be susceptible to recall bias and bias through social desirability.

In conclusion, later biological aging reflects adolescent lifestyle behavior. Our findings advance research on biological aging by showing that a shared genetic background can underlie both adolescent lifestyle and biological aging measured with epigenetic clocks.

Data availability

A subsample of the FTC with DNA methylation age estimates, phenotypes, and information on the adolescent lifestyle behaviour patterns (BCH weights) will be located in the Biobank of the National Institute for Health and Welfare. All these data will be publicly available for use by qualified researchers following a standardised application procedure (see the website https://thl.fi/en/web/thl-biobank/for-researchers for details on the application process). Because of the consent given by study participants and the high degree of identifiability of the twin siblings in Finland, the full cohort data cannot be made publicly available. The full cohort data are available through the Institute for Molecular Medicine Finland (FIMM) Data Access Committee (DAC) for authorized researchers who have IRB/ethics approval and an institutionally approved study plan. For more details, please contact the FIMM DAC (fimm-dac@helsinki.fi). The codes used to analyse the data are provided in the supplementary file 1 and the processed data used to generate figures have been uploaded as the source data files.

References

    1. Kaprio J
    2. Rimpelä A
    3. Winter T
    4. Viken RJ
    5. Rimpelä M
    6. Rose RJ
    (1995)
    Common genetic influences on BMI and age at menarche
    Human Biology 67:739–753.
    1. Maggs JL
    2. Schulenberg JE
    (2005)
    Trajectories of alcohol use during the transition to adulthood
    Alcohol Research and Health 28:195–201.
  1. Website
    1. Muthén LK
    2. Muthén BO
    (1998) Mplus User’s Guide
    Accessed October 7, 2022.

Decision letter

  1. Joris Deelen
    Reviewing Editor; Max Planck Institute for Biology of Ageing, Germany
  2. Eduardo Franco
    Senior Editor; McGill University, Canada
  3. Jenny van Dongen
    Reviewer; VU Amsterdam, Netherlands
  4. Esther Walton
    Reviewer; University of Bath, United Kingdom

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "The role of adolescent lifestyle habits in biological aging: A prospective twin study" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. The following individuals involved in the review of your submission have agreed to reveal their identity: Jenny van Dongen (Reviewer #1); Esther Walton (Reviewer #3).

As is customary in eLife, the reviewers have discussed their critiques with one another. What follows below is the Reviewing Editor's edited compilation of the essential and ancillary points provided by reviewers in their critiques and in their interaction post-review. Please submit a revised version that addresses these concerns directly. Although we expect that you will address these comments in your response letter, we also need to see the corresponding revision clearly marked in the text of the manuscript. Some of the reviewers' comments may seem to be simple queries or challenges that do not prompt revisions to the text. Please keep in mind, however, that readers may have the same perspective as the reviewers. Therefore, it is essential that you attempt to amend or expand the text to clarify the narrative accordingly.

Essential revisions:

The reviewers provided a number of comments, mostly for providing clarity, which can be found below. One additional point that came up during the discussion is the use of the newly generated PCA-based epigenetic clocks created by Morgan Levine's group (https://www.nature.com/articles/s43587-022-00248-2), because this has been shown to improve the performance of some of the clocks used by the authors. However, given that it will likely take quite some additional work to recalculate the clocks based on this method, we decided to leave it up to you to reanalyze the data using the PCA-based clocks or just address this point in your Discussion section.

Reviewer #1 (Recommendations for the authors):

Introduction

Would it be possible to provide references for the following sentences?

Typically, about half of the adolescents fall into subgroups characterised by healthy lifestyle habits

However, small minorities of adolescents are classified as heavy substance users or as having multiple other risk behaviors

Methods

• It is mentioned that illumina 450k and illumina EPIC arrays were used. It might be informative to add the number of samples? Do I understand correctly that the raw data from the 2 platforms were QC'd and normalized together? That could be more explicitly mentioned.

• It is mentioned that the online age calculator was used. Could the authors please clarify what exactly was uploaded to the calculator? Do I understand correctly that pre-QC'd methylation β-values were used? Isn't the preferred method to upload raw methylation data because the online calculator does its own internal QC and normalization?

• In the introduction, it is mentioned that DunedinPACE is essentially an improvement of DunedinPoAm, but in the methods it is described that both measures are tested. Could the authors please clarify why they tested both measures? If these 2 clocks are supposed to measure the same thing, but DunedinPoAm is just outdated, why it still considered?

• The question on alcohol use at age 14 and 17 measures binge drinking, which is a different measure from the alcohol use measure at age 21-25. It is not clear why this measure was used. Was it the only measure of alcohol use available at the age of 14 and 17? Could the authors comment on how suited this measure is for quantifying alcohol use in this age group?

• On page 10, it is mentioned "the highest response category of the original questionnaire (development completed) was omitted for all items, except for menarche…". Could the authors please clarify what exactly was done in such cases, i.e. were these participants discarded from the analysis, or did they receive an NA (but was a sum score then still computed?), or were they recoded to the second-highest response category?

• How did the analysis that tested for mean differences in biological ageing between the lifestyle behavior patterns take account for the clustering of observations within families?

• The procedure to derive the proportion of variation in biological aging explained by genetic factors shared with adolescent lifestyle patterns was not entirely clear to me. For example, it is described that "On the other hand, it comprised the variance explained by the adolescent lifestyle patterns …". Is this sentence still referring to the same model as in the previous sentence? Perhaps it would help to clarify how many models were fitted.

• Could the authors clarify the rationale of the analysis that additionally corrected for adult BMI. I think the understandability of the paper could be improved if this would be explicitly explained in the methods and/or Results section.

• (How) did the authors correct for multiple testing?

Results/Discussion/interpretation

• My impression upon reading the Results section is that it seems to me (but maybe I'm wrong) that BMI is the main discriminative factor driving the 5 lifestyle classes. Could the authors comment on how they think about his, i.e. how much do the other variables; exercise, smoking and alcohol contribute? Is it possible to say something about this? Although I recognize that a major strength of the current analysis is that it make use of multiple lifestyle measures that are collectively summarized in one overall measure of lifestyle, I am left wondering how much each of the individual lifestyle measures is contributing and if we are mainly looking at the effect of BMI here. How different do the authors expect that the results would have been if the analysis would simply have been performed on BMI only?

• It is mentioned that group 4 shows differences in several DNAm-based plasma proteins, but group 5 only a difference in smoking pack-years, which seems a bit counter-intuitive, but no interpretation is given. Is the interpretation that group 5 appears to have the most unhealthy lifestyle but is biologically not the unhealthiest group? Or could this be a power issue (although group 5 I believe was larger than group 4)?

• It is concluded that (e.g. discussion p 19 and in the abstract): "Our findings suggest pleiotropic genetics effects; that is, the same genes affect both adolescent lifestyle and the pace of biological aging". However, pleiotropic effects are only one possible interpretation for shared genetic influences on two traits. Causal effects of lifestyle on biological aging, or vice versa of biological aging on lifestyle can also be captured in the shared genetic component (i.e. lifestyle in particular BMI is highly heritable, and if BMI has a causal effect on biological aging, this would give rise to a genetic correlation between these traits, i.e. rg includes both causal effects in both directions, as well as pleiotropy). The current analysis did not model or test causal paths. So I feel that the explanation of shared genetic influences should be more nuanced here.

• The discussion, on page 20 cites "previous meta-analysis showed that the number of healthy lifestyle behaviors is associated with all-cause mortality risk". Could information be added on the age of the populations in this study (is it also based on adolescents, as the current study is?)

• The discussion, on page 20, mentions "Our results suggested that the accumulation of multiple unhealthy lifestyle habits in adolescence has a more detrimental effect on biological aging that any single lifestyle habit." I have a hard time deducing on which results this conclusion is based. Did they also examine the effects of the single lifestyle habits? Or is this referring to previous work?

Reviewer #3 (Recommendations for the authors):

1) Can the authors also add (to table 2 or elsewhere) the correlations between chronological age and each epigenetic clock?

2) It wasn't fully clear why the authors adjust for adult BMI. What's the model / DAG here? Is adult BMI assumed to be a mediator or a downstream effect of epigenetic age? Line 481 suggests the authors consider mediation. However, I would assume that adolescent and adult BMI probably lie on a continuum and are largely correlated. So, why control for this association? I.e. what does it mean if adolescent BMI has a direct effect on adult ageing that is independent of adult BMI? What are the clinical implications? Another sentence on this or a DAG would help here.

3) 62-73% of explained variance through genetic factors seems large and 0-4% for (unshared) environmental factors seems very low. How does this compare to heritability reports in other studies? Is it likely that all variation is explained by genetics? Maybe this just needs some slight rephrasing also acknowledging the large amount of unexplained variation.

4) The authors write 'However, genetic factors shared with adolescent lifestyle explained most of the observed differences in biological aging'. I wasn't sure about the 'shared with adolescent lifestyle' phrase? Shouldn't this phrase be removed – or alternatively, please explain how exactly the link between genetics and lifestyle factors was assessed.

Related to this point, I wasn’t sure about the interpretation around pleiotropic genetic effects? The authors argue that they identified pleiotropic effects underlying BMI and epigenetic ageing. However, other scenarios are also possible, correct? E.g., even though any suggestions around causality should be done with extreme caution, it would be possible that genetic predictors for one trait (BMI, ageing) are on the causal pathway for the other trait. A brief discussion around this point might be helpful.

5) Independent lifestyle effects: there are two somewhat conflicting statements. In line 447, the authors argue that their ‘results suggested that the accumulation of multiple unhealthy lifestyle habits in adolescence has a more detrimental effect on biological aging than any single lifestyle habit’. However, in line 487+, they state that ‘lower levels of physical activity in adolescence were closely intertwined with other unhealthy behaviors. To fully understand the role of adolescence physical activity in later biological aging would require a more comprehensive analysis of activity patterns, intensities and modes, as well as subgroup analyses that account for other lifestyle factors, such as diet’. Maybe, these two separate statements could be more aligned.

6) Did the authors observe the same classes in the subgroup with DNAm data? Maybe Figure 2-suppl could be more in line with Figure 2 (same style). Or alternatively, combine a dot- and boxplot in Figure 2-suppl (and Figure 3). Also, please add to Figure 2-suppl the age at which each lifestyle factor was measured.

https://doi.org/10.7554/eLife.80729.sa1

Author response

Essential revisions:

The reviewers provided a number of comments, mostly for providing clarity, which can be found below. One additional point that came up during the discussion is the use of the newly generated PCA-based epigenetic clocks created by Morgan Levine’s group (https://www.nature.com/articles/s43587-022-00248-2), because this has been shown to improve the performance of some of the clocks used by the authors. However, given that it will likely take quite some additional work to recalculate the clocks based on this method, we decided to leave it up to you to reanalyze the data using the PCA-based clocks or just address this point in your Discussion section.

Thank you for this interesting suggestion. We have now calculated the epigenetic age estimates using principal component (PC)-based epigenetic clocks (PCHorvath, PCHannum, PCPhenoAge and PCGrimAge) created by Levine’s group (Higgins-Chen et al., 2022). The means of the epigenetic age estimates derived from the PC-based clocks were higher compared with those obtained with the original clocks, and the standard deviations were at the same level or even higher (Author response table 1). Also, several outliers (> 5 standard deviations away from mean) were detected according to PCA-based clocks (according to PCHannum 3 outliers, PCPhenoAge 4 and PCGrimAge 1) whereas no outliers were observed when original clocks were used. Recoding these observations to missing values did not result in considerably smaller standard deviations (Author response table 1). Moreover, the correlations between PCA-based age estimates and chronological age were even weaker than obtained with the original clocks (based on Horvath 0.13 vs. 0.17, Hannum 0.06 vs. 0.10, PhenoAge -0.07 vs. 0.03, and GrimAge 0.06 vs. 0.14). However, the correlations between epigenetic age acceleration (AA) measures assessed with different clocks were consistently higher when PC-based epigenetic clocks were used (Author response image 1).

Author response table 1
Means (standard deviations) of the epigenetic age estimates obtained using original and PC-based clocks.
Original clockPC clockPC clock, outliers excluded
Horvath's clock28.9 (3.6)30.8 (3.8)-
Hannum's clock18.2 (3.3)34.4 (4.0)34.4 (3.8)
DNAm PhenoAge13.0 (5.3)16.8 (6.4)16.6 (5.8)
DNAm GrimAge25.2 (3.3)38.8 (3.2)38.8 (3.2)
Author response image 1
Correlations between epigenetic age acceleration (AA) measures assessed with different clocks.

We also tested the differences between the adolescent lifestyle behavior patterns in biological aging assessed with PC-based epigenetic clocks and the results were very similar to the original ones. There were no differences in biological aging when PCHorvath (Wald test: p = 0.550) and PCHannum (p = 0.104) were used to assess biological aging, but differences were observed when PCPhenoAge (p = 0.031) PCGrimAge (p = 8.3e-8) were used (Author response image 2).

Author response image 2
Mean differences between the adolescent lifestyle behavior patterns in biological aging measured with original and PC-based (A) DNAm PhenoAge and (B) DNAm GrimAge.

The analysis was adjusted for sex (female), standardized age and baseline pubertal development. Means and 95% confidence intervals are presented. C1 = the class with the healthiest lifestyle pattern, C2 = the class with low–normal BMI, C3 = the class with a healthy lifestyle and high–normal BMI, C4 = the class with high BMI, C5 = the class with the unhealthiest lifestyle pattern. AA, age acceleration.

We have now mentioned this in the Results section (lines 409-418).

“In our study, high standard deviations of epigenetic age estimates were observed. Therefore, variation in AA measures may largely be attributable to technical variation, which is not biologically meaningful. Recently developed principal component (PC)-based clocks are shown to improve the reliability and validity of epigenetic clocks (Higgins-Chen et al., 2022). We therefore replicated our main analyses using PC-based epigenetic clocks (data not shown). The standard deviations of epigenetic age estimates were similar or even higher compared with those obtained with the original clocks, but the correlations between AA measures assessed with different clocks were consistently higher when PC-based epigenetic clocks were used. Importantly, the observed associations with the adolescent lifestyle behavior patterns did not substantially change.”

References:

Higgins-Chen, A. T., Thrush, K. L., Wang, Y., Minteer, C. J., Kuo, P.-L., Wang, M., Niimi, P., Sturm, G., Lin, J., Moore, A. Z., Bandinelli, S., Vinkers, C. H., Vermetten, E., Rutten, B. P. F., Geuze, E., Okhuijsen-Pfeifer, C., van der Horst, M. Z., Schreiter, S., Gutwinski, S., … Levine, M. E. (2022). A computational solution for bolstering reliability of epigenetic clocks: implications for clinical trials and longitudinal tracking. Nature Aging, 2(7), 644–661. https://doi.org/10.1038/s43587-022-00248-2

Reviewer #1 (Recommendations for the authors):

Introduction

Would it be possible to provide references for the following sentences?

Typically, about half of the adolescents fall into subgroups characterised by healthy lifestyle habits

However, small minorities of adolescents are classified as heavy substance users or as having multiple other risk behaviors

We refer to a systematic review conducted by Whitaker and colleagues in the three consecutive sentences. We have now clarified this (lines 89-93).

Methods

• It is mentioned that illumina 450k and illumina EPIC arrays were used. It might be informative to add the number of samples? Do I understand correctly that the raw data from the 2 platforms were QC'd and normalized together? That could be more explicitly mentioned.

We have now added the number of samples accordingly (lines 158-159)

“Of the samples included in this study, 744 samples were assayed using 450k and 80 samples using EPIC arrays.”

The raw data from different platforms was combined and preprocessed together. We have now mentioned this (lines 159-161).

• It is mentioned that the online age calculator was used. Could the authors please clarify what exactly was uploaded to the calculator? Do I understand correctly that pre-QC'd methylation β-values were used? Isn't the preferred method to upload raw methylation data because the online calculator does its own internal QC and normalization?

We normalized and conducted quality control for the methylation data prior to uploading data to the calculator, but we used normalization method implemented in the calculator, as well. The different normalization methods have different purposes. Our single-sample Noob normalization (ssNoob) is the best performing normalization method when data from the EPIC and 450k-arrays is integrated (Fortin et al., 2017). It is not affected by whether samples are combined or not before normalization. We also used Β-mixture quantile (BMIQ) normalization for correcting probe design bias (Teschendorff et al., 2013). According to the calculator tutorial, the purpose of the normalization method implemented in the calculator is to make data comparable to the training data of the epigenetic clock.

The function implemented in the calculator also makes some quality checks (produces predicted sex and tissue) that help to identify suspicious samples. However, own quality control is needed, as well.

References:

Fortin, J. P., Triche, T. J., & Hansen, K. D. (2017). Preprocessing, normalization and integration of the Illumina HumanMethylationEPIC array with minfi. Bioinformatics, 33(4), 558–560. https://doi.org/10.1093/bioinformatics/btw691

Teschendorff, A. E., Marabita, F., Lechner, M., Bartlett, T., Tegner, J., Gomez-Cabrero, D., & Beck, S. (2013). A β-mixture quantile normalization method for correcting probe design bias in Illumina Infinium 450 k DNA methylation data. Bioinformatics, 29(2), 189–196. https://doi.org/10.1093/bioinformatics/bts680

Calculator tutorial: https://horvath.genetics.ucla.edu/html/dnamage/TUTORIALonlineCalculator.pdf

• In the introduction, it is mentioned that DunedinPACE is essentially an improvement of DunedinPoAm, but in the methods it is described that both measures are tested. Could the authors please clarify why they tested both measures? If these 2 clocks are supposed to measure the same thing, but DunedinPoAm is just outdated, why it still considered?

We agree that DunedinPoAm estimator is outdated. However, both DunedinPoAm and DunedinPACE are very recently published. There are some dozens of studies applying DunedinPoAm while few have reported results based on DunedinPACE. We believe that information on DunedinPoAm is useful for some readers.

• The question on alcohol use at age 14 and 17 measures binge drinking, which is a different measure from the alcohol use measure at age 21-25. It is not clear why this measure was used. Was it the only measure of alcohol use available at the age of 14 and 17? Could the authors comment on how suited this measure is for quantifying alcohol use in this age group?

We also had information on the frequency of alcohol use at the age of 14 and 17 assessed with question ‘How often do you drink alcohol at all?’. However, this question does not take into account quantity of the alcohol used. Both measures have been widely used in studies that utilize data from the FinnTwin12 cohort (Rose et al., 2019). In Finland in the late 1990s, when members of the FinnTwin12 cohort were adolescents, alcohol use was increasingly drunkenness-oriented among 14-18 year-olds (Lintonen et al., 2000). Therefore, we think that the question on binge drinking is suitable measure for quantifying alcohol use in this population.

References:

Rose, R. J., Salvatore, J. E., Aaltonen, S., Barr, P. B., Bogl, L. H., Byers, H. A., Heikkilä, K., Korhonen, T., Latvala, A., Palviainen, T., Ranjit, A., Whipp, A. M., Pulkkinen, L., Dick, D. M., & Kaprio, J. (2019). FinnTwin12 Cohort: An Updated Review. Twin Research and Human Genetics, 22(5), 302–311. https://doi.org/10.1017/thg.2019.83

Lintonen, T., Rimpelä, M., Ahlström, S., Rimpelä, A., & Vikat, A. (2000). Trends in drinking habits among Finnish adolescents from 1977 to 1999. Addiction, 95(8), 1255–1263. https://doi.org/10.1046/j.1360-0443.2000.958125512.x

• On page 10, it is mentioned "the highest response category of the original questionnaire (development completed) was omitted for all items, except for menarche…". Could the authors please clarify what exactly was done in such cases, i.e. were these participants discarded from the analysis, or did they receive an NA (but was a sum score then still computed?), or were they recoded to the second-highest response category?

In the original questionnaire which included Pubertal developmental scale (PDS) developed by Petersen et al., there are four response categories: 1 = growth/change has not begun, 2 = growth/change has barely started and 3 = growth/change is definitely underway, and 4 = development complete (Petersen et al., 1988). The highest response option was removed already when designing the questionnaire, because completing the development was assumed to be very rare by the age of 12. Therefore, there were no such problematic cases in the data as described by the reviewer.

To clarify the text, we have now removed the following sentence:

“The highest response category of the original questionnaire (development completed) was omitted for all items, except for menarche.”

Interested readers may find further information from (Wehkalampi et al., 2008).

References:

Petersen, A. C., Crockett, L., Richards, M., & Boxer, A. (1988). A self-report measure of pubertal status: Reliability, validity, and initial norms. Journal of Youth and Adolescence, 17, 117–133.

Wehkalampi, K., Silventoinen, K., Kaprio, J., Dick, D. M., Rose, R. J., Pulkkinen, L., & Dunkel, L. (2008). Genetic and environmental influences on pubertal timing assessed by height growth. American Journal of Human Biology, 20(4), 417–423. https://doi.org/10.1002/ajhb.20748

• How did the analysis that tested for mean differences in biological ageing between the lifestyle behavior patterns take account for the clustering of observations within families?

We used TYPE=MIXTURE COMPLEX option of the ANALYSIS command in conjunction with the CLUSTER=family option of the VARIABLE command in Mplus (Muthén & Muthén, 1998-2018). This approach corrects the standard errors for non-independence of observations within families. We have briefly mentioned this in the end of the statistical analysis section (lines 304-305).

• The procedure to derive the proportion of variation in biological aging explained by genetic factors shared with adolescent lifestyle patterns was not entirely clear to me. For example, it is described that "On the other hand, it comprised the variance explained by the adolescent lifestyle patterns …". Is this sentence still referring to the same model as in the previous sentence? Perhaps it would help to clarify how many models were fitted.

Thank you for raising this question. We noticed that the Figure 1 was unintentionally dropped out from the full submission. We have now included it in the manuscript, and hopefully, that makes the description of the analysis easier to follow. We have also clarified the paragraph which describes the estimation of genetic and environmental influences (lines 283-300).

• Could the authors clarify the rationale of the analysis that additionally corrected for adult BMI. I think the understandability of the paper could be improved if this would be explicitly explained in the methods and/or Results section.

Thank you for this excellent suggestion. We have now briefly clarified the rationale in the Results section (lines 397-408).

“According to the previous literature, it is controversial whether childhood obesity has a direct effect on later health, or whether the association is fully mediated by BMI in adulthood (Park et al., 2012). The role of adult BMI may depend on which disease outcome is studied (Richardson et al., 2020). After additionally adjusting the model for BMI in adulthood, the differences in AAPheno and DunedinPACE between the class of participants with high BMI (C4) and those with lower BMI (C1, C2, C5) were attenuated (Table 5, M2). This finding suggests that the observed differences in biological aging probably are fully mediated by BMI in adulthood. However, the differences in biological aging were only slightly attenuated when the DNAm GrimAge and DunedinPoAm estimators were used, suggesting that childhood overweight may leave permanent imprint on biological aging assessed with these measures.”

• (How) did the authors correct for multiple testing?

To control for Type I error rate due to multiple testing we have now calculated the 99% confidence intervals for the mean differences presented in the Table 2 and pointed out the differences that exist after controlling for multiple testing. Mainly, the interpretation of the results remained the same but the difference in biological aging between the unhealthiest and the healthiest pattern (C5 vs C1) was not significant at 0.01 level when DunedinPACE was used.

Results/Discussion/interpretation

• My impression upon reading the Results section is that it seems to me (but maybe I'm wrong) that BMI is the main discriminative factor driving the 5 lifestyle classes. Could the authors comment on how they think about his, i.e. how much do the other variables; exercise, smoking and alcohol contribute? Is it possible to say something about this? Although I recognize that a major strength of the current analysis is that it make use of multiple lifestyle measures that are collectively summarized in one overall measure of lifestyle, I am left wondering how much each of the individual lifestyle measures is contributing and if we are mainly looking at the effect of BMI here. How different do the authors expect that the results would have been if the analysis would simply have been performed on BMI only?

We agree that BMI may have a dominant role in the extraction of the patterns. However, if the classification was based only on BMI, we would have missed the separate classes of participants with generally healthy lifestyle (C1) and participants with unhealthiest lifestyle pattern (C5). The participants in both classes were of normal weight on average, but there were remarkable differences in lifestyle habits between the classes. There were also clear differences between the classes in biological aging measured with AAGrim and DunedinPoAm estimators.

Based on the reported analyses, we cannot say how much individual lifestyle habits, including leisure-time physical activity, smoking and alcohol use contribute to the classification. Probably physical activity and alcohol use have only a minor role, but interestingly, we found that low levels of physical activity and binge drinking co-occured with smoking.

• It is mentioned that group 4 shows differences in several DNAm-based plasma proteins, but group 5 only a difference in smoking pack-years, which seems a bit counter-intuitive, but no interpretation is given. Is the interpretation that group 5 appears to have the most unhealthy lifestyle but is biologically not the unhealthiest group? Or could this be a power issue (although group 5 I believe was larger than group 4)?

Surprisingly, we did not observe any increased levels of DNAm-based plasma proteins in young adulthood in the class with the unhealthiest lifestyle pattern compared to the classes with healthier habits (Figure 3—figure supplement 1). Therefore, we think that lack of differences is not due to power issue. Probably, the consequences of the unhealthy lifestyle will be seen also in the DNAm-based plasma proteins in later adulthood if unhealthy habits persist.

• It is concluded that (e.g. discussion p 19 and in the abstract): "Our findings suggest pleiotropic genetics effects; that is, the same genes affect both adolescent lifestyle and the pace of biological aging". However, pleiotropic effects are only one possible interpretation for shared genetic influences on two traits. Causal effects of lifestyle on biological aging, or vice versa of biological aging on lifestyle can also be captured in the shared genetic component (i.e. lifestyle in particular BMI is highly heritable, and if BMI has a causal effect on biological aging, this would give rise to a genetic correlation between these traits, i.e. rg includes both causal effects in both directions, as well as pleiotropy). The current analysis did not model or test causal paths. So I feel that the explanation of shared genetic influences should be more nuanced here.

Thank you for this important note. We agree that our interpretation of the shared genetic influences was too straightforward. We have now included more in-depth discussion on the possible reasons for the shared genetic influences (lines 490-496).

“The shared genetic influences on two phenotypes may be due to several scenarios (Solovieff et al., 2013). They may arise from genetic pleiotropy; in this case, the genes may be a common cause for both adolescent lifestyle and biological aging. Another possible reason is causal relation between the phenotypes. In this case, genetic factors may affect adolescent lifestyle, which lies on the causal path to biological aging (or vice versa). However, for the relationship to be causal, it is necessary that there are shared environmental influences on the phenotypes (de Moor et al., 2008).”

• The discussion, on page 20 cites "previous meta-analysis showed that the number of healthy lifestyle behaviors is associated with all-cause mortality risk". Could information be added on the age of the populations in this study (is it also based on adolescents, as the current study is?)

The age range of the populations in this meta-analysis was wide ranging from 17 to 99 years, but majority were older adult populations. Probably due to the lack of data, adolescents were not included in the study. We have now added the age range of the target populations in the manuscript (lines 475-476).

• The discussion, on page 20, mentions "Our results suggested that the accumulation of multiple unhealthy lifestyle habits in adolescence has a more detrimental effect on biological aging that any single lifestyle habit." I have a hard time deducing on which results this conclusion is based. Did they also examine the effects of the single lifestyle habits? Or is this referring to previous work?

We agree that based on our results, we cannot make this kind of conclusion. We did not conduct comparisons of the effects of the single lifestyle habits vs. the effect of overall lifestyle on biological aging. Therefore, we have now rephrased (lines 480-485):

“The accumulation of multiple unhealthy lifestyle habits during lifetime probably has a more detrimental effect on biological aging, as well, than any single lifestyle habit. However, our approach did not allow us to disentangle the effects of single lifestyle habits on biological aging. Our results suggest that the unhealthy lifestyle-induced changes in biological aging begin to accumulate in early life. These changes might predispose individuals to premature death in later life.”

Reviewer #3 (Recommendations for the authors):

1) Can the authors also add (to table 2 or elsewhere) the correlations between chronological age and each epigenetic clock?

Due to the narrow age range of this study (21 to 25 years), the correlations were low. The correlations between epigenetic age and chronological age ranged from 0.03 to 0.17. Considerably higher estimates (ranging from 0.54 to 0.76) were observed in our previous study among participants in a wider age range (21 to 42 years) (Kankaanpää et al., 2021), which also included the participants of the current study. In the current study, chronological age did not correlate with pace of aging measured with DunedinPoAm and even slightly negative correlation (-0.09) was observed with DunedinPACE.

References:

Kankaanpää, A., Tolvanen, A., Saikkonen, P., Heikkinen, A., Laakkonen, E. K., Kaprio, J., Ollikainen, M., & Sillanpää, E. (2022). Do Epigenetic Clocks Provide Explanations for Sex Differences in Life Span? A Cross-Sectional Twin Study. The Journals of Gerontology: Series A, 77(9): 1898-1906. https://doi.org/10.1093/gerona/glab337

2) It wasn't fully clear why the authors adjust for adult BMI. What's the model / DAG here? Is adult BMI assumed to be a mediator or a downstream effect of epigenetic age? Line 481 suggests the authors consider mediation. However, I would assume that adolescent and adult BMI probably lie on a continuum and are largely correlated. So, why control for this association? I.e. what does it mean if adolescent BMI has a direct effect on adult ageing that is independent of adult BMI? What are the clinical implications? Another sentence on this or a DAG would help here.

Thank you for raising this question. We agree that this was unclear. We have now provided rationale for the adjustment in the result section (lines 397-408).

“According to the previous literature, it is controversial whether childhood obesity has a direct effect on later health, or whether the association is fully mediated by BMI in adulthood (Park et al., 2012). The role of adult BMI may depend on which disease outcome is studied (Richardson et al., 2020). After additionally adjusting the model for BMI in adulthood, the differences in AAPheno and DunedinPACE between the class of participants with high BMI (C4) and those with lower BMI (C1, C2, C5) were attenuated (Table 5, M2). This finding suggests that the observed differences in biological aging probably are fully mediated by BMI in adulthood. The differences in biological aging were only slightly attenuated when the DNAm GrimAge and DunedinPoAm estimators were used, suggesting that childhood overweight may leave permanent imprint on biological aging assessed with these measures.”

3) 62-73% of explained variance through genetic factors seems large and 0-4% for (unshared) environmental factors seems very low. How does this compare to heritability reports in other studies? Is it likely that all variation is explained by genetics? Maybe this just needs some slight rephrasing also acknowledging the large amount of unexplained variation.

In our study, genetic factors explained 62-73% of the total variation in biological aging depending on the utilized estimator, and unshared environmental factors accounted the rest of the variation (27-38%). Previous studies using twin and pedigree data have obtained moderate to high heritability estimates for age acceleration (AA) measures (Horvath & Raj, 2018). Our heritability estimates are somewhat higher than previously reported for AA measures based on DNAm PhenoAge (0.33-0.41) (Jylhävä et al., 2019; Levine et al., 2018) and DNAm GrimAge (0.30-0.58) (Lu et al., 2019; Lundgren et al., 2022). We are not aware of any studies reporting on the heritability of pace of aging measured with DunedinPoAm and DunedinPACE.

The reported proportions 0-4% reflect the size of environmental correlation between adolescent lifestyle and biological aging i.e. to which extent shared environmental factors explain both adolescent lifestyle and biological aging (see also answers above). Because of the nominal scale of the latent class variable, we were not able to directly estimate the genetic and environmental correlations using traditional bivariate twin modeling. Therefore, we evaluated these proportions indirectly. We have now clarified the description of the estimation of genetic and environmental influences (lines 283-300) and the presentation of the results (lines 437-449).

References:

Horvath, S., & Raj, K. (2018). DNA methylation-based biomarkers and the epigenetic clock theory of ageing. Nature Reviews Genetics, 19(6), 371–384. https://doi.org/10.1038/s41576-018-0004-3

Levine, M. E., Lu, A. T., Quach, A., Chen, B. H., Assimes, T. L., Bandinelli, S., Hou, L., Baccarelli, A. A., Stewart, J. D., Li, Y., Whitsel, E. A., Wilson, J. G., Reiner, A. P., Aviv, A., Lohman, K., Liu, Y., Ferrucci, L., & Horvath, S. (2018). An epigenetic biomarker of aging for lifespan and healthspan. Aging, 10(4), 573–591. https://doi.org/10.18632/aging.101414

Jylhävä, J., Hjelmborg, J., Soerensen, M., Munoz, E., Tan, Q., Kuja-Halkola, R., Mengel-From, J., Christensen, K., Christiansen, L., Hägg, S., Pedersen, N. L., & Reynolds, C. A. (2019). Longitudinal changes in the genetic and environmental influences on the epigenetic clocks across old age: Evidence from two twin cohorts. EBioMedicine, 40, 710–716. https://doi.org/10.1016/j.ebiom.2019.01.040

Lu, A. T., Quach, A., Wilson, J. G., Reiner, A. P., Aviv, A., Raj, K., Hou, L., Baccarelli, A. A., Li, Y., Stewart, J. D., Whitsel, E. A., Assimes, T. L., Ferrucci, L., & Horvath, S. (2019). DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging, 11(2), 303–327. https://doi.org/10.18632/aging.101684

Lundgren, S., Kuitunen, S., Pietiläinen, K. H., Hurme, M., Kähönen, M., Männistö, S., Perola, M., Lehtimäki, T., Raitakari, O., Kaprio, J., & Ollikainen, M. (2022). BMI is positively associated with accelerated epigenetic aging in twin pairs discordant for BMI. Journal of Internal Medicine. https://doi.org/10.36684/33-2020-1-680-686

4) The authors write 'However, genetic factors shared with adolescent lifestyle explained most of the observed differences in biological aging'. I wasn't sure about the 'shared with adolescent lifestyle' phrase? Shouldn't this phrase be removed – or alternatively, please explain how exactly the link between genetics and lifestyle factors was assessed.

Related to this point, I wasn't sure about the interpretation around pleiotropic genetic effects? The authors argue that they identified pleiotropic effects underlying BMI and epigenetic ageing. However, other scenarios are also possible, correct? E.g., even though any suggestions around causality should be done with extreme caution, it would be possible that genetic predictors for one trait (BMI, ageing) are on the causal pathway for the other trait. A brief discussion around this point might be helpful.

Thank you for these important notes. We have now removed the phrase as suggested. We have also clarified the description of the estimation of genetic and environmental influences (lines 283-300). We also noticed that the Figure 1 was unintentionally dropped out from the full submission. We have now included it in the manuscript, and hopefully, that makes the description of the analysis easier to follow.

We must admit that our interpretation around pleiotropic genetic effects was too straightforward. We have now briefly discussed this topic (lines 490-496):

“The shared genetic influences on two phenotypes may be due to several scenarios (Solovieff et al., 2013). It may arise from genetic pleiotropy; in this case, the genes may be a common cause for both adolescent lifestyle and biological aging. Another possible reason is causal relation between the phenotypes. In this case, genetic factors may affect adolescent lifestyle, which lies on the causal path to biological aging (or vice versa). However, for the relationship to be causal, it is necessary that there are shared environmental influences on the phenotypes (de Moor et al., 2008).“

5) Independent lifestyle effects: there are two somewhat conflicting statements. In line 447, the authors argue that their 'results suggested that the accumulation of multiple unhealthy lifestyle habits in adolescence has a more detrimental effect on biological aging than any single lifestyle habit'. However, in line 487+, they state that 'lower levels of physical activity in adolescence were closely intertwined with other unhealthy behaviors. To fully understand the role of adolescence physical activity in later biological aging would require a more comprehensive analysis of activity patterns, intensities and modes, as well as subgroup analyses that account for other lifestyle factors, such as diet'. Maybe, these two separate statements could be more aligned.

We thank you for raising this point. We agree that these statements are somewhat conflicting. We did not conduct comparisons of the effects of the single lifestyle habits vs. the effect of overall lifestyle on biological aging. For that reason, we have modified the first statement (lines 480-485):

“The accumulation of multiple unhealthy lifestyle habits during lifetime may have a more detrimental effect on biological aging, as well, than any single lifestyle habit. However, our approach did not allow us to disentangle the effects of single lifestyle habits on biological aging. Our results suggest that the unhealthy lifestyle-induced changes in biological aging begin to accumulate in early life. These changes might predispose individuals to premature death in later life.”

6) Did the authors observe the same classes in the subgroup with DNAm data? Maybe Figure 2-suppl could be more in line with Figure 2 (same style). Or alternatively, combine a dot- and boxplot in Figure 2-suppl (and Figure 3). Also, please add to Figure 2-suppl the age at which each lifestyle factor was measured.

Initially, we conducted LCA in the subgroup of participants with DNAm data. The BIC value reached its minimum value at the 4-class-solution, and the interesting class of participants with high BMI (C4) was extracted at fifth step. Using large cohort data, we were able to confirm that the pattern C4 actually exists. The mean and probability profiles of the 5-class solution were very similar to those obtained with large cohort data. The class-sizes were C1: 34.5%, C2: 20.2%, C3: 19.2%, C4: 6.6%, and C5: 19.5% (the class-sizes in the subsample based on the LCA-solution using large cohort data: C1: 33.0%, C2: 16.6%, C3: 20.6%, C4: 10.1%, C5: 19.7%).

We have now modified the Figure 2 —figure supplement 1 to be more in line with Figure 2 and added the age at which each lifestyle factor was measured.

https://doi.org/10.7554/eLife.80729.sa2

Article and author information

Author details

  1. Anna Kankaanpää

    Gerontology Research Center (GEREC), Faculty of Sport and Health Sciences, University of Jyväskylä, Jyväskylä, Finland
    Contribution
    Conceptualization, Formal analysis, Methodology, Writing – original draft
    For correspondence
    anna.k.kankaanpaa@jyu.fi
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6973-5385
  2. Asko Tolvanen

    Methodology Center for Human Sciences, University of Jyväskylä, Jyväskylä, Finland
    Contribution
    Supervision, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6430-8897
  3. Aino Heikkinen

    Institute for Molecular Medicine Finland (FIMM), HiLife, University of Helsinki, Helsinki, Finland
    Contribution
    Data curation, Writing – review and editing, Preprocessing of the DNAm data
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5770-6475
  4. Jaakko Kaprio

    Institute for Molecular Medicine Finland (FIMM), HiLife, University of Helsinki, Helsinki, Finland
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Project administration, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3716-2455
  5. Miina Ollikainen

    Institute for Molecular Medicine Finland (FIMM), HiLife, University of Helsinki, Helsinki, Finland
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3661-7400
  6. Elina Sillanpää

    1. Gerontology Research Center (GEREC), Faculty of Sport and Health Sciences, University of Jyväskylä, Jyväskylä, Finland
    2. Institute for Molecular Medicine Finland (FIMM), HiLife, University of Helsinki, Helsinki, Finland
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Project administration, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6375-959X

Funding

Academy of Finland (213506)

  • Jaakko Kaprio

Academy of Finland (297908)

  • Miina Ollikainen

Academy of Finland (341750)

  • Elina Sillanpää

EC FP5 GenomEUtwin

  • Jaakko Kaprio

National Institutes of Health (HL104125)

  • Jaakko Kaprio

EC MC ITN Project EPITRAIN

  • Jaakko Kaprio

University of Helsinki Research Funds

  • Miina Ollikainen

Sigrid Juselius Foundation

  • Jaakko Kaprio

Yrjö Jahnsson Foundation (6868)

  • Elina Sillanpää

Juho Vainio Foundation

  • Elina Sillanpää

Päivikki and Sakari Sohlberg foundation

  • Elina Sillanpää

Academy of Finland (312073)

  • Jaakko Kaprio

Academy of Finland (263278)

  • Jaakko Kaprio

Academy of Finland (265240)

  • Jaakko Kaprio

Academy of Finland (346509)

  • Elina Sillanpää

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This work was supported by the Academy of Finland (213506, 265240, 263278, 312073 to JK, 297908 to MO, and 341750, 346509 to ES), EC FP5 GenomEUtwin (JK), National Institutes of Health/National Heart, Lung, and Blood Institute (grant HL104125), EC MC ITN Project EPITRAIN (JK and MO), the University of Helsinki Research Funds (MO), Sigrid Juselius Foundation (JK and MO), Yrjö Jahnsson Foundation (6868), Juho Vainio Foundation (ES), and Päivikki and Sakari Sohlberg foundation (ES).

Ethics

Human subjects: Data collection was conducted in accordance with the Declaration of Helsinki. The Indiana University IRB and the ethics committees of the University of Helsinki and Helsinki University Central Hospital approved the study protocol (113/E3/2001and 346/E0/05). The blood samples for DNA analyses were collected during in-person clinical studies after written informed consent was signed.

Senior Editor

  1. Eduardo Franco, McGill University, Canada

Reviewing Editor

  1. Joris Deelen, Max Planck Institute for Biology of Ageing, Germany

Reviewers

  1. Jenny van Dongen, VU Amsterdam, Netherlands
  2. Esther Walton, University of Bath, United Kingdom

Publication history

  1. Preprint posted: May 31, 2022 (view preprint)
  2. Received: June 1, 2022
  3. Accepted: October 1, 2022
  4. Version of Record published: November 8, 2022 (version 1)

Copyright

© 2022, Kankaanpää et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 874
    Page views
  • 86
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Anna Kankaanpää
  2. Asko Tolvanen
  3. Aino Heikkinen
  4. Jaakko Kaprio
  5. Miina Ollikainen
  6. Elina Sillanpää
(2022)
The role of adolescent lifestyle habits in biological aging: A prospective twin study
eLife 11:e80729.
https://doi.org/10.7554/eLife.80729

Further reading

    1. Epidemiology and Global Health
    2. Medicine
    Qing Shen, Huan Song ... Unnur Valdimarsdóttir
    Research Article Updated

    Background:

    The association between cardiovascular disease (CVD) and selected psychiatric disorders has frequently been suggested while the potential role of familial factors and comorbidities in such association has rarely been investigated.

    Methods:

    We identified 869,056 patients newly diagnosed with CVD from 1987 to 2016 in Sweden with no history of psychiatric disorders, and 910,178 full siblings of these patients as well as 10 individually age- and sex-matched unrelated population controls (N = 8,690,560). Adjusting for multiple comorbid conditions, we used flexible parametric models and Cox models to estimate the association of CVD with risk of all subsequent psychiatric disorders, comparing rates of first incident psychiatric disorder among CVD patients with rates among unaffected full siblings and population controls.

    Results:

    The median age at diagnosis was 60 years for patients with CVD and 59.2% were male. During up to 30 years of follow-up, the crude incidence rates of psychiatric disorder were 7.1, 4.6, and 4.0 per 1000 person-years for patients with CVD, their siblings and population controls. In the sibling comparison, we observed an increased risk of psychiatric disorder during the first year after CVD diagnosis (hazard ratio [HR], 2.74; 95% confidence interval [CI], 2.62–2.87) and thereafter (1.45; 95% CI, 1.42–1.48). Increased risks were observed for all types of psychiatric disorders and among all diagnoses of CVD. We observed similar associations in the population comparison. CVD patients who developed a comorbid psychiatric disorder during the first year after diagnosis were at elevated risk of subsequent CVD death compared to patients without such comorbidity (HR, 1.55; 95% CI, 1.44–1.67).

    Conclusions:

    Patients diagnosed with CVD are at an elevated risk for subsequent psychiatric disorders independent of shared familial factors and comorbid conditions. Comorbid psychiatric disorders in patients with CVD are associated with higher risk of cardiovascular mortality suggesting that surveillance and treatment of psychiatric comorbidities should be considered as an integral part of clinical management of newly diagnosed CVD patients.

    Funding:

    This work was supported by the EU Horizon 2020 Research and Innovation Action Grant (CoMorMent, grant no. 847776 to UV, PFS, and FF), Grant of Excellence, Icelandic Research Fund (grant no. 163362-051 to UV), ERC Consolidator Grant (StressGene, grant no. 726413 to UV), Swedish Research Council (grant no. D0886501 to PFS), and US NIMH R01 MH123724 (to PFS).

    1. Epidemiology and Global Health
    Bingyi Yang, Bernardo García-Carreras ... Derek A Cummings
    Research Article

    Background: Over a life-course, human adaptive immunity to antigenically mutable pathogens exhibits competitive and facilitative interactions. We hypothesize that such interactions may lead to cyclic dynamics in immune responses over a lifetime.

    Methods: To investigate the cyclic behavior, we analyzed hemagglutination inhibition titers against 21 historical influenza A(H3N2) strains spanning 47 years from a cohort in Guangzhou, China and applied Fourier spectrum analysis. To investigate possible biological mechanisms, we simulated individual antibody profiles encompassing known feedbacks and interactions due to generally recognized immunological mechanisms.

    Results: We demonstrated a long-term periodicity (about 24 years) in individual antibody responses. The reported cycles were robust to analytic and sampling approaches. Simulations suggested that individual-level cross-reaction between antigenically similar strains likely explain the reported cycle. We showed that the reported cycles are predictable at both individual and birth-cohort level and that cohorts show a diversity of phases of these cycles. Phase of cycle was associated with the risk of seroconversion to circulating strains, after accounting for age and pre-existing titers of the circulating strains.

    Conclusions: Our findings reveal the existence of long-term periodicities in individual antibody responses to A(H3N2). We hypothesize that these cycles are driven by pre-existing antibody responses blunting responses to antigenically similar pathogens (by preventing infection and/or robust antibody responses upon infection), leading to reductions in antigen specific responses over time until individual's increasing risk leads to an infection with an antigenically distant enough virus to generate a robust immune response. These findings could help disentangle cohort-effects from individual-level exposure histories, improve our understanding of observed heterogeneous antibody responses to immunizations, and inform targeted vaccine strategy.

    Funding: This study was supported by grants from the NIH R56AG048075 (D.A.T.C., J.L.), NIH R01AI114703 (D.A.T.C., B.Y.), the Wellcome Trust 200861/Z/16/Z (S.R.) and 200187/Z/15/Z (S.R.). This work was also supported by research grants from Guangdong Government HZQB-KCZYZ-2021014 and 2019B121205009 (Y.G. and H.Z.). D.A.T.C., J.M.R. and S.R. acknowledge support from the National Institutes of Health Fogarty Institute (R01TW0008246). J.M.R. acknowledges support from the Medical Research Council (MR/S004793/1) and the Engineering and Physical Sciences Research Council (EP/N014499/1). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.