Early life predictors of body composition trajectories from adolescence to mid‐adulthood

Guatemala has experienced rapid increases in adult obesity. We characterized body composition trajectories from adolescence to mid‐adulthood and determined the predictive role of parental characteristics, early life factors, and a nutrition intervention.

mortality globally ("Global Burden of Disease Study.GBD Compare" 2019) -is a public health concern across all populations and age groups (NCD Risk Factor Collaboration (NCD-RisC), 2017).In the last four decades, data from low-and middle-income countries (LMICs) have shown rapid and steady increments in excess weight with various degrees of child stunting, facing a paradoxical coexistence of malnutrition at individual, household, and population levels (Popkin et al., 2020).Guatemala exemplifies this nutrition transition given its increasing prevalence of adult obesity (17.6% of men and 29.6% of women in 2019) ("Global Nutrition Report.Country Nutrition Profiles: Guatemala" 2021) while having the highest prevalence of child stunting in the Americas (42.8%) (FAO, IFAD, PAHO, WFP, & UNICEF, 2021).Thus, it is of great relevance to study the early life predictors of adult body weight and composition in the context of the double burden of malnutrition in order to develop effective interventions for obesity prevention in LMICs.
Despite historical debates on the best approach to tackle child growth faltering, it is recognized that protein quantity and quality should be considered (Semba, 2016).However, aspects of nutrition in early life, including high-protein and protein quality, low-fat diets, infant formula composition, and breastfeeding and complementary feeding practices, might be responsible also for the metabolic programming of adult body weight and composition and obesity risk (Druet & Ong, 2008;Larnkjaer et al., 2020;Rolland-Cachera et al., 2016;Wells et al., 2007;Yang & Huffman, 2013).Trials and observational studies have shown that a poor diet in utero is associated with a higher risk of obesity in adulthood (Ravelli et al., 1976;Ravelli et al., 1999;Smith et al., 2009;Stein et al., 2007;Stewart et al., 2009;Yang & Huffman, 2013).Infants fed with high-protein infant formulas have higher adiposity and higher risk of obesity later in life (childhood and adulthood) than breastfed infants or those who receive low-protein formulas (Druet & Ong, 2008;Rolland-Cachera et al., 2016;Yang & Huffman, 2013).But, evidence relating early life nutrition to excess weight and adiposity is moderate or weak, and evidence on other early life nutrition practices remains limited and inconclusive (Patro-Gołąb et al., 2016).For instance, most findings of a recent systematic review and meta-analysis of clinical trials that studied the effects of nutrition interventions in early life on adult cardiometabolic biomarkers were null (He & Stein, 2021).Pooled estimates of this meta-analysis showed only modest beneficial effects on glucose and lipids, and an increase in body mass index (BMI) (He & Stein, 2021).Indeed, most studies on life course patterns of body composition and the role of early life factors on these patterns and cardiometabolic outcomes are limited to BMI (Li et al., 2007;Mattsson et al., 2019;Munthali et al., 2017;Woo, 2019), a measure of body size that ignores body composition.Moreover, most of the evidence comes from studies in high-income countries (HICs).However, differences between HICs and LIMCs in environments and stages of the demographic, epidemiologic and nutritional transitions, may limit generalization of findings from HICs to the context of LMICs.For example, a history of stunting in early life (a prevalent scenario of low-and middle-income settings) has been related to a higher risk of adult obesity (Hoffman et al., 2000;Leonard et al., 2009;Schroeder et al., 1999).The mechanisms behind excess weight in adults living in LMICs might differ from individuals who grew up in industrialized societies.Thus, the characterization of individual patterns of body composition over time and the long-term role of nutrition interventions on adult weight and body composition needs further study.
Findings from the Institute of Nutrition of Central America and Panama (INCAP) Longitudinal Study (Ramirez-Zea & Mazariegos, 2020) in Guatemala, showed that the exposure to atole (a nutrition supplement high in energy and protein) in early life reduced the prevalence of stunting (Martorell, 2020).Children who received atole during the first 2 and 3 years of life had faster linear growth and greater weight during this period (Schroeder et al., 1995), and were taller, had higher weight and higher fat-free mass (FFM) in adolescence (Rivera et al., 1995).However, exposure to this nutrition supplement was also significantly associated with higher BMI, fat mass, and obesity in adult life (Ford et al., 2018).Therefore, it is important to further characterize associations of this early life nutrition intervention with higher adiposity in mid-adulthood.
In this sense, discrimination between body composition compartments (e.g., fat mass and fat-free mass), which have distinctive metabolic functions, might be more relevant to understand the cardiometabolic risk of nutrition interventions.From a life course perspective, body composition in adolescence and young adulthood predicts body composition and cardiometabolic risk in mid-adulthood because adolescence is an stage of rapid growth, and of hormonal and body composition changes (Siervogel et al., 2003), in which BMI and fat mass have a moderate to strong tracking into adult adiposity (Singh et al., 2008;Umer et al., 2017).Thus, we aim to extend on previous analysis of the INCAP Longitudinal Study on body size across the life course by (1) describing the longitudinal changes in fat mass index (FMI), fat-free mass index (FFMI), and BMI from adolescence to midadulthood to identify groups of individuals that follow similar growth trajectories; (2) determining early life factors that predict the latent class membership of body composition trajectories (FMI, FFMI, and BMI); and (3) examining whether or not exposure to improved nutrition in early life is a predictor of body composition latent class membership.

| Study design and study population
We conducted a secondary analysis of data from the INCAP Nutrition Supplementation Trial Longitudinal Study.This community randomized controlled trial took place in four rural villages (two sets of matched villages) of Guatemala between January 1969 and February 1977, to evaluate the effect of a nutrition intervention on cognitive and physical development (Ramirez-Zea & Mazariegos, 2020).The study enrolled 2392 children who were under 7 years of age at study launch or were born during the study period.The treatment villages received atole, a hot beverage providing energy (163 kcal per serving) and protein (11.5 g per serving) whose base was a vegetable protein mixture (Incaparina), dry skim milk and some sugar, whereas the control villages received fresco, a cold beverage providing only energy (59 kcal per serving), whose ingredients were sugar, some flavor, and color (Martorell et al., 1995).Some micronutrients were added to both drinks.Participants have been followed prospectively since the end of the original trial.Body composition measures were collected in 1988-89 (ages 10.8-27.1 years, median 17.4), 1997-99 (ages 20-35.7 years, median 26.7), 2002-04 (ages 25.2-41.3 years, median 31.8), and 2015-17 (ages 37.0-55.0 years, median 44.0).A detailed description of the original trial, its follow-up waves, and attrition information is published elsewhere (Ramirez-Zea et al., 2010;Ramirez-Zea & Mazariegos, 2020).

| Outcomes
Trained study staff followed standard methodologies to collect anthropometric (weight and height) and body composition measures (fat mass [FM] and fat-free mass [FFM]).Body composition was estimated through indirect methods.In the first three waves of data collection, the study team used body composition prediction equations developed under a two-compartment model with the use of body density determined by hydrostatic weighing and sex-and age-specific anthropometric measures.Hydrodensitometry was used to generate predictive equations that were validated among adolescents and young adults.Conlisk et al. used hydrodensitometry in a group of 201 individuals living in Guatemala City (103 males, and 98 females; ages 11-25 years).The predictive error was 4.9% and the predictors accounted for 92%-96% of the variance in fat-free mass (Conlisk et al., 1992).These body composition equations were used to estimate body composition of study participants in the 1988-89 and 1997-1999 follow-up waves (ages 10.8-27.1 and 10-35.7 years, respectively).Ramirez-Zea et al. used hydrodensitometry in 211 healthy Guatemalan individuals (114 males and 123 females; ages 18-56 years).The predictive error was 3.7% and the group of predictors accounted for 74%-80% of variance in percent body fat (Ramirez-Zea et al., 2006).The derived equations were used to estimate body composition of study participants in the 2002-2004 follow-up wave (ages 25.2-41.3years).In the 2015-7 wave, deuterium oxide (D 2 O) dilution technique was used to measure body composition (van Marken Lichtenbelt et al., 1994).The study team collected saliva samples before and after participants drank a specific volume of water labeled with deuterium to estimate total body water (Kroker-Lobos et al., 2020), and then estimated FFM with the use of a hydration constant for adults (0.732) (van Marken Lichtenbelt et al., 1994).FM was calculated as the difference between total body mass and FFM (Kroker-Lobos et al., 2020).We calculated BMI, FMI, and FFMI as weight, fat mass and fat-free mass in kg divided by height in squared meters, respectively.We note that FMI + FFMI = BMI.

| Exposure to the nutrition intervention
Study participants had different exposures to the intervention due to the open nature of enrollment.We estimated age of exposure based on date of birth (assuming that all participants were born at term [pregnancy duration of 266 days]), the start of the trial (large villages: January 1, 1969; small villages: May 1, 1969) and the end of the trial (February 28, 1977) (Ford et al., 2018;Ramirez-Zea et al., 2010).We defined three exposure categories: (1) full exposure to atole in the 'first 1000 days' (the period from conception to the second birthday) if children were born between September 24, 1969 (large villages) or January 22, 1970 (small villages) andFebruary 28, 1975; (2) partial exposure to atole in the first 1000 days if children were born before September 24, 1969 (large villages), before January 22, 1970(small villages), or after February 18, 1975;and (3) no exposure to atole in the first 1000 days if children were born before 1966 or were born in a village assigned to fresco.We used a difference-in-difference approach to simultaneously capture supplement type (atole versus fresco) and timing of exposure (full or partial exposure in early life versus no exposure in early life), as previously used in analyses of data from this cohort (Dimick & Ryan, 2014;Hoddinott et al., 2008).

| Early life predictors
Baseline characteristics were ascertained through interviews at the trial start by trained field personnel.Data on maternal factors (height, age at delivery, schooling attainment), paternal factors (age, school attainment), child characteristics (year of birth, birth order), and household socioeconomic status (SES) at birth were available.Height in centimeters, birth year, and schooling attainment in years were included as continuous variables in our models.We created tertiles of SES score, which was obtained from information about household assets, services, and consumer durable goods using principal component analysis.We created four categories of birth order: first, second, third, fourth or more.In this cohort, individuals reached schooling attainment before adolescence, thus we also considered participants' schooling as a predictor of body composition trajectories.

| Adult characteristics
Sociodemographic and lifestyle characteristics were obtained through interviews at the time of the most recent wave of data collection (2002)(2003)(2004) with median age of 32 years and/or 2015-2017 median age of 44 years).Data on participants' SES (tertiles created from a cumulative SES score obtained through a principal component analysis), place of residence (urban vs rural), number of children (women, number of children born; men, number of children alive), physical activity (hours/ day), time spent on sedentary activities (hours/day), alcohol consumption, and smoking were included in the characterization of individuals.Physical activity level was estimated using the short form of the International Physical Activity Questionnaire (IPAQ).

| Estimation of body composition trajectories
Figure 1 shows the selection of our analytic sample (n = 1364 participants; 742 women and 622 men).From among all children enrolled in the trial (n = 2392), we excluded 1028 study participants, 981 of whom had fewer than two body composition measurements in adulthood (Table S1), and 47 who had died or migrated from Guatemala between the third and fourth follow-up waves and hence for whom data from the 2015-7 wave were not available.No large differences were observed between individuals in the analytic sample and those excluded for which data were available.Excluded participants were older, belonged to middle or high SES at birth, had lower schooling (men), were more likely to drink alcohol and smoke (only men), and a higher proportion lived in urban areas (Table S2).
We used latent class growth analysis (LCGA) to identify groups of individuals (latent classes) that share similar trajectories of body composition (FMI, FFMI) and size (BMI) from adolescence to mid-adulthood.We derived these trajectories for men and women separately.We used body composition and size measures at four time points to determine the latent class model; a minimum of two measurements are needed to identify a linear model (intercept and slope), while three or more measurements are needed to identify a quadratic model (Table S3).First, we specified a single-class latent growth curve model for the whole data to verify a longitudinal change of body composition (a single mean and variance population).Second, we tested three functions of change: linear, quadratic, and cubic.To select the most appropriate pattern of change, we examined model fit statistics, beta estimates, and a visual examination of trajectories.Third, we specified a conditional LCGA model with no within-class variances (variances were set to zero) to determine distinct classes.We used age at each follow-up to derive the conditional latent class trajectories because of the wide differences in age among study participants.We determined the final number of classes by a combination of approaches such as theoretical framework, fit indices, model parsimony, and model interpretability (Jung & Wickrama, 2008).
We used Bayesian Information Criteria (BIC), Akaike Information Criteria (AIC), and adjusted BIC to check the best model fit.Then, we examined entropy values to assess separation of latent classes and accuracy in the classification (Ram & Grimm, 2009).We used the Bootstrap Likelihood Ratio Test and the Lo, Mendell, and Rubin Likelihood Ratio Test (LMR-LRT) to compare a series of subsequent models (k classes versus k À 1 classes, for models with 2, 3, and 4 classes) until a model did not show statistical significance relative to a prior model (p ≥ .05).Finally, to select the best solution that fitted the data well, we evaluated the number of individuals in each class (selected only models where all classes had >1% of the study sample), examined the posterior probabilities, and evaluated graphically estimated versus observed trajectories for each latent class.

| Early life predictors of body composition trajectories
We built a first set of binomial or multinomial logistic regression models (depending on the number of trajectories identified) to estimate the association of maternal factors (age, height, schooling attainment), paternal factors (age, schooling attainment), birth order and household SES at child's birth with body composition trajectories.In a second set of models, we tested for the association of participants' schooling attainment with body composition trajectories.In both sets of models, we controlled for structural variables like year of birth and village of birth to account for the study design.We applied multiple degrees of freedom tests (multivariate Wald test for models with two latent classes and joint F test for models with three latent classes) to determine the statistical significance of categorical variables.In a third set of models, we estimated the association between the nutrition intervention during the first 1000 days of life and body composition trajectories.Further, we tested for the interaction between supplement type and the timing of exposure.We ran three subsequent models to adjust for potential confounders: model 1 to control only for birth year, model 2 with additional adjustment for early life covariates, and model 3 to control for participants' schooling attainment.Given the high proportion of participants with at least one sibling in the study (86%), we adjusted for within-family correlations using generalized estimating equation models (GEE) in R and the cluster option in Stata.

| Multiple imputation
Before performing the main analyses, we built sexspecific multiple imputation models to deal with missing data (early life covariates).Data missing patterns are presented in Table S4.We assumed missing at random (MAR) as the underlying pattern, and used the R package MICE (Multivariate Imputation by Chained Equations) to generate 15 multiple imputed datasets with a maximum of 50 iterations.In our imputation models, we included all covariates used in the main analyses except the outcome variables.We assessed graphically the convergence of the multiple imputed datasets, evaluating the distribution of the continuous variables through density plots, and the frequency of categorical variables through bar graphs.We used the "with and pool" functions in R and "mi estimate" in Stata to run the main analyses in each of the imputed datasets and to combine the estimates into a single result with their corresponding standard errors and confidence intervals.
We used Mplus version 7.4 to identify the body composition trajectories.We used R version 4.0 and Stata version 16 to conduct the descriptive and regression analyses.Statistical significance was established at a p value <.05.

Ethics approval and consent to participate
The original trial was conducted before the establishment of formal ethical review committees.Subsequent data collection waves followed protocols approved by committees in Guatemala and the US.Participants, or their parents as appropriate, provided written informed consent (in Spanish) for all study procedures.The present analysis was approved by the Emory University IRB (protocol 71 041).

| RESULTS
Descriptive characteristics of study participants at baseline are presented in Table 1.Around 24% of women and men were exposed to the nutrition intervention during the full first 1000 days of life.Baseline characteristics were similar when comparing participants fully exposed to the nutrition supplement in early life versus those  non-exposed or with partial exposure; except birth length (both sexes), and gestational age, paternal schooling, and household SES at birth among men.

| Body composition and size latent class trajectories
Based on model fit statistics and a visual examination of trajectories at median ages of 17, 27, 32, and 44 years, we concluded that linear patterns in trajectories of FMI (women) and BMI (women and men) and quadratic patterns in trajectories of FMI (men) and FFMI (women and men) best fitted the data (Table S5).We selected the final number of latent classes based on models with the lowest AIC and BIC, significant LMR-LRT, the highest values of entropy (>0.60 for FFMI, and >0.70 for FMI and BMI), and high posterior probabilities (>0.80) (Tables S6-S8).Models with two classes are referred to as "low" and "high" trajectories, and models with three classes as "low", "middle", and "high" trajectories.Among women, we identified two latent class trajectories of FMI (low: 79.6% of study sample; high: 20.4%) and BMI (low: 73.0%; high: 27.0%), and three latent class trajectories of FFMI (low: 20.2%; middle: 55.9%; high: 23.9%).Among men, we identified two latent class trajectories of FMI (low: 79.6% of study sample; high: 20.4%) and FFMI (low: 62.4%; high: 37.6%), and three latent classes of BMI (low: 43.1%; middle: 46.9%; high: 10.0%). Figure 2 shows a graphic representation of loess-smoothed trajectories of body size and composition over age for each latent class.
We observed that all classes had different intercepts and slopes that did not overlap from the period of adolescence to mid-adulthood.Table S9 shows the changes in mean FMI, FFMI, and BMI over time.In the period of adolescence to mid-adulthood, FMI of women in high and low trajectories increased by 11.2 and 6.8 kg/m 2 units (two to three times the adolescence values).Among men, FMI increased by 8.1 and 4.6 kg/m 2 units in high and low trajectories, respectively (three to four times the adolescence values).FFMI of participants in high and middle trajectories increased by 1-2 units between ages 17 and 32 years, and then dropped slightly or remained flat until midadulthood.BMI of women in high and low trajectories increased by 12 and 7 kg/m 2 units, moving from a normal weight in adolescence to obesity and overweight in adult life, respectively.Men in the low trajectory had a normal BMI across time whereas men in middle and high trajectories moved from a normal BMI in adolescence to overweight and obesity in adult life, respectively (changes of 10 units).
Tables 2 and 3 present characteristics of study participants within latent class trajectories of FMI and FFMI, respectively.Women in the high FMI class membership had a higher mean birth weight, had achieved fewer years of schooling attainment, and had a higher BMI and waist circumference than those women in the low trajectory (Table 2).Men in the high FMI class trajectory had mothers with higher schooling attainment, were wealthier, had achieved more years of schooling attainment, almost half resided in urban areas, had higher BMI and waist circumference, consumed alcohol in less proportion, were less physically active, and were more sedentary than individuals classified in the low trajectory (Table 2).We observed statistically significant differences in birth weight, paternal age, schooling attainment, place of residence, number of children, BMI, waist circumference, and sedentary behaviors between classes of FFMI in women (Table 3).Men in the high FFMI class trajectory had mothers with higher schooling attainment, a higher proportion lived in urban areas, had a higher number of children, had overweight and a higher waist circumference compared to men in the low trajectory (Table 3).Characteristics of BMI membership showed similar patterns to those observed in FMI trajectories (Table S10).

| Early life predictors of body composition trajectories membership
Among men, maternal schooling attainment, paternal age, and individual's schooling attainment were each positively associated with odds of high FMI latent class membership relative to the low latent class trajectory (OR: 1.21 [95%CI: 1.07, 1.38]; OR: 1.04 [95%CI: 1.00, 1.07]; and OR:1.10 [95%CI: 1.04, 1.17], respectively) (Table 4).Among women, there was a 9% decrease in the odds of high FMI latent class membership per year of individual's schooling attainment (OR: 0.91 [95%CI: 0.85, 0.97]) (Table 4).Results for BMI were similar to those for FMI (Table S11).The odds of high FFMI latent class membership increased in both women (OR: 1.16 [95%CI: 0.97, 1.39]) and men (OR: 1.17 [95%CI: 1.05, 1.31]) per year of maternal schooling attainment (Table 5).Maternal age and paternal schooling attainment in men were associated with lower odds of being in the high FFMI latent class.For FFMI, none of the estimates for women reached statistical significance (Table 5).

| Exposure to the nutrition intervention and body composition trajectories membership
Neither full nor partial exposure to the nutrition supplement during the first 1000 days of life were significant predictors of class membership for FMI, FFMI, or BMI (Table 6 and Table S12).Estimates were on both sides of the null and confidence intervals were imprecise.We observed similar trends when we combined full and partial exposure and compared it to non-exposure (Table S13).Minimally and fully adjusted models showed consistent results.

| DISCUSSION
We identified two latent class trajectories of BMI in women and three in men.These latent classes had different intercepts and slopes that did not overlap over time.This characterization of body size latent class trajectories from adolescence to mid-adulthood (ages 10.8-55.0years) is consistent with an earlier analysis that identified two latent BMI classes among women and three among men for BMI in the period from birth to age 42 years (Ford et al., 2016).We further characterized FMI and FFMI trajectories.We identified two latent class trajectories of FMI in women and men, and three FFMI latent classes in women and two in men.Similar to BMI trajectories, body composition trajectories showed heterogeneity and a separation of distinct classes from adolescence to midadulthood.Ford et al. had previously identified the separation of BMI classes since the first month after birth suggesting that prenatal and early life factors might have a role in determining specific trajectories of body size (Ford et al., 2016).In our exploration of early life determinants of body size and body composition trajectories, we found that women's schooling was inversely associated with high latent class membership for both FMI and BMI, whereas a higher schooling of their mothers was a predictor of high class membership for FFMI.In men, maternal schooling, paternal age, and participant's schooling were associated with a higher probability of FMI high latent class.Maternal schooling also predicted a FFMI high

Childhood characteristics
Birth weight (kg) 2.9 (0.4) 3.0 (0.4) 3.2 (0.4) .0083.1 (0.5) 3.1 (0.4) .615 Birth latent class, whereas maternal age and paternal schooling were associated with lower probability of FFMI high class membership.Finally, we found that neither full nor partial exposure to improved nutrition in early life was a significant predictor of latent class membership for any of the three outcomes.Adult FFMI and FMI might be the result of the programing role of genetic factors, hormonal regulations, environmental conditions in sensitive periods of development, and the tracking of secular trends across the life course (Wells, 2006(Wells, , 2007)).In the present characterization of FMI and FFMI trajectories, we observed differences among men and women, and large increments in adiposity since early adulthood (especially among women).Sex-dimorphism of FMI and FFMI since early life, and changes as individuals grow older have been described previously.At birth, boys are heavier and longer than girls, but, girls have more fat mass than boys (Rodríguez et al., 2004).These sex-patterns track into infancy, childhood, and adolescence as boys are heavier given a higher accretion in fat-free mass, while girls have more fat mass and display higher increments in adiposity.In adult life, men and women have different body size, shape and composition (Wells, 2006(Wells, , 2007)).Sexdifferences that are consistent across human populations, but with varying proportions of FMI and FFMI across ethnic groups (Hull et al., 2011;Jin et al., 2019;Xiong et al., 2012) and even geographical regions within a country (Jin et al., 2019).Thus, there have been proposed sexand age-specific references values for different population groups (Kasovi c et al., 2021;Kelly et al., 2009;Lu et al., 2012;Schutz et al., 2002;Wells, 2014).It is hypothesized that genetic factors and environmental conditions such as periods of deprivation (famines) (Ravelli et al., 1976;Ravelli et al., 1999;Stein et al., 2007;Zhou et al., 2018) and the obesogenic environment (diets high in calories and sedentary behaviors) might be the mechanisms underlying the observed variability in FMI and FFMI among populations (Wells, 2006(Wells, , 2007)).However, limited evidence exists on the early life determinants of FMI and FFMI, and most of the literature available is on the determinants of body size (BMI).
In our examination of the role of early life factors on body composition trajectories, we found that participant's schooling attainment was associated with a lower odds of FMI high latent class among women and a higher odds among men.These sex differences in the direction of associations are consistent with other analyses in Latin America that have found that women with higher education have lower risk of obesity whereas men with higher education and living in cities with lower economic development have higher risk (Mazariegos et al., 2021).Maternal schooling was a positive predictor of FMI high latent class among men and of FFMI in both sexes.Paternal age was a positive predictor of FMI high latent class among men, and maternal age and paternal schooling were inversely associated with FFMI high latent class in men.Effect sizes were small but significant indicating that parental characteristics such as age and schooling attainment might have a long-term role on the establishment of body composition trajectories from adolescence to mid-adulthood.We infer that parental characteristics such as older age and higher schooling attainment might be related with sedentary economic activities, a higher socioeconomic status, and greater access to foods that  together might explain the higher risk of adiposity of their children.In contrast, a lower paternal schooling attainment might indicate that fathers started working at a younger age and in agricultural tasks (economic activity of 97% of men in these villages in 1967) with a consequent higher own and progeny lean mass.In addition to early life factors, multiple environmental factors in adulthood, such as lifestyle (physical activity, alcohol consumption, and sedentary behaviors), place of residence, socioeconomic status, and number of children might be underlying mechanisms driving the changes in body composition in adulthood as we observed in our crosssectional characterization of latent classes.Thus, there is a need to determine the temporal role (longitudinal) of these and other adult factors on body composition trajectories, especially from young adulthood to midadulthood, ages where we observed the rapid increase in adiposity.
Our findings also showed that full and/or partial exposure to atole during the first 1000 days of life did not predict latent class membership in any of the outcomes studied.It is possible that the effect of atole on body composition might have been already incorporated in earlier stages of life.For instance, Rivera et al. found that participants who received atole in early life were taller, had higher weight, and FFM in adolescence.However, they did not observe differences in height or weight in adolescence after controlling for weight or height at age 3, concluding that these differences were already established in infancy or early childhood (Rivera et al., 1995).The medium to strong tracking of child BMI into adolescence and adult body composition has also been noted (Siervogel et al., 2003;Singh et al., 2008).An earlier study of this study population found that BMI at ages 15 days to 5 years were moderate predictors of adiposity (skinfold thickness, fat percentage, and waist-to-hip ratio) in adolescence and early adulthood (Schroeder & Martorell, 1999).It is likely that any long-term role of the supplement on body composition might be indirect, for example, through the improvement of human capital dimensions such as height, cognitive skills, and schooling attainment that have positively impacted productivity  (economic and home) of study participants in adult life (Behrman et al., 2020;Martorell, 2017), but at the same time might have influenced individuals' lifestyle.
There is no accepted standard for FMI and FFMI changes over adulthood.Thus, to assess the magnitude of the changes in body composition, we compared our findings to sex and race-specific normative references of FMI and FFMI (Shypailo & Wong, 2020).In our study, the mean FMI of individuals in the low latent class was lower than references for Hispanic adolescents and young adults, but the mean FMI of participants in high classes was closer to these reference values.Individuals in high class trajectories had adiposity in normal ranges in adolescence and young adulthood whereas study participants in low trajectories had lower adiposity but normal BMI.In our longitudinal characterization, we observed that in mid-adulthood, the mean FMI in women and men in the high trajectories was almost two and three times the reference values of young adulthood, respectively.Even study participants in low trajectories had higher FMI values than reference values in early adulthood.In our study, we also found that FFMI of participants in high and middle classes increased slightly from adolescence to young adulthood, and then dropped slightly or remained flat until mid-adulthood.Women in high trajectories had higher values than reference levels for adolescence and young adulthood whereas men had similar values to reference levels (Shypailo & Wong, 2020).Women in low trajectories showed a steady decrease in FFMI levels from age 35 until age 55, reaching similar values as in their early adolescence.Men in low trajectories also showed decreases in FFMI from 30 years of age.This suggests that study participants in high and middle trajectories have maintained normal values of FFMI from adolescence to mid-adulthood, and have not started to experience losses in muscle and bone mass as a result of aging.However, individuals in low trajectories probably are undergoing through this process.These changes in body composition, especially the large increments in adiposity in the period from young adulthood to mid-adulthood, exemplify the nutrition transition that Guatemala and many other LMICs have experienced.This transition is the result of changes in the food environment, dietary patterns, and physical activity, paired with a history of high prevalence of child chronic undernutrition (Ford et al., 2017;Miranda et al., 2019).Study participants have experienced this double burden of malnutrition across their life span since they had moderate or severe stunting in infancy and childhood (Schroeder & Martorell, 1999), and in adulthood they are experiencing a marked increase in adiposity as a consequence of the nutrition transition of the last 20-30 years in the communities where they live.Indeed, developmental links between a short stature and adiposity have been described previously.An earlier study in Brazil found that children with stunting (ages 8-11 years) had a higher respiratory quotient, higher carbohydrate oxidation, and lower fatty acids oxidation during fasting than children with normal height (Hoffman et al., 2000).In a study with a group of adult Siberians, results were consistent but only among women.Women short in stature had higher respiratory quotients, higher BMI, higher percent body fat, and higher serum lipids than women with normal stature (Leonard et al., 2009).These findings indicate that individuals with history of childhood stunting have less capacity to oxidize fatty acids that in turn are stored, and might be associated with higher adiposity later in life.This physiologic change in fat metabolism among children with history of growth faltering has been suggested to be the result of lower levels of insulin-like growth factor I (IGF-I), a hormone with a role in growth and lipolysis (Laron, 2001;Marcus et al., 1994).
Some limitations of our study were: (1) potential selection bias due to attrition in a cohort that has been followed for about 50 years.However, an analysis of the most recent data from the INCAP showed that attrition did not alter the internal validity of findings (Ford et al., 2018); (2) missing covariates, for which we used multiple imputation techniques assuming MAR; (3) in the original trial design, the treatment allocation was done at the village level (only four villages).Thus, we treated individuals as the unit of analysis, and to account for any imbalance between the treatment and control groups, we controlled for covariates at baseline and used the double difference approach that controls for village fixed effects; (4) the 1997 follow-up included only a subset of the total cohort, specifically younger participants who had been fully exposed to the nutrition intervention; this might be a source of systematic bias; (5) methods to assess body composition differed across study waves.In the first three waves, prediction equations derived from body density measures were used and in the 2015-7 wave deuterium oxide dilution technique was the method of choice.However, all equations were derived for this specific population using established and validated techniques (Conlisk et al., 1992;Ramirez-Zea et al., 2006).This cohort is not representative of the Guatemalan population.However, this is a well-characterized study population with one of the longest follow-ups in LMICs and a reasonable sample size that allowed us to examine sex-stratified trajectories of body composition compartments beyond body size.We also used a LCGA approach through which we proved that there was heterogeneity in the data, and we were able to identify distinct classes of body composition trajectories from adolescence to midadulthood.Our findings might be generalizable to contexts with similar characteristics, with a high prevalence of child stunting, and a subsequent transition to an adult obesogenic environment, a double burden of malnutrition that might be the case of other LMICs.

| CONCLUSIONS
Our findings showed that individuals followed distinct trajectories of body composition that were already established by the time they reached adolescence.FFMI remained stable in the period of adolescence to midadulthood, while in this same period of almost 30 years, FMI increased by two to three times the reference values of early adulthood.Early life characteristics such as parental characteristics (age and schooling), and participant's schooling attainment had a small but significant long-term role in predicting body composition trajectories whereas the nutrition supplement in the first 1000 days of life did not have a direct role.Thus, there is a need to first, further characterize the patterns of body composition compartments starting in early life, and second, to identify and determine the temporal (longitudinal) role of adulthood factors on body composition changes, especially from young adulthood to mid-adulthood, life stages of a rapid and exacerbated rise in adiposity.administration (equal), resources (equal), supervision (equal), formal analysis (supporting), writing -review and editing (equal).Aryeh D. Stein: conceptualization (equal), funding acquisition (equal), project administration (equal), resources (equal), supervision (equal), formal analysis (supporting), writing -review and editing (equal).
T A B L E 1 Baseline characteristics of study participants by sex and nutritional supplement exposure (atole), INCAP Longitudinal Study.
specific body composition latent class trajectories from adolescence to mid-adulthood in participants of the INCAP Longitudinal Study.Loess-smoothed trajectories over time (we plotted these trajectories using the exact age at which each measured was collected, ages 10.8-55 years).BMI: body mass index; FFMI: fat-free mass index; FMI: fat mass index.
T A B L E 2 Descriptive characteristics of study participants, by FMI latent class trajectory and sex, INCAP Longitudinal Study.

T A B L E 3
Descriptive characteristics of study participants, by FFMI latent class trajectory and sex, INCAP Longitudinal Study.
T A B L E 4 Early life predictors of fat-free mass index latent class trajectories from adolescence to mid-adulthood among individuals who participated in the INCAP Longitudinal Study.
T A B L E 5 Early life predictors of fat-free mass indices latent class trajectories from adolescence to mid-adulthood among individuals who participated in the INCAP Longitudinal Study.
T A B L E 6 Association between nutritional supplement (atole) exposure in early life and body composition latent class trajectories from adolescence to mid-adulthood among individuals who participated in the INCAP Longitudinal Study.Women (n = 742) Men (n = 622) For women, joint test (F test) with 6 degrees of freedom, p value = .81;for men, multivariate Wald test with 3 degrees of freedom, p value = .39. b For women, joint test (F-test) with 4 degrees of freedom, p value = .65;for men, multivariate Wald test with 2 degrees of freedom, p value = .42. a