The PedBE clock accurately estimates DNA methylation age in pediatric buccal cells

Abstract

The development of biological markers of aging has primarily focused on adult samples. Epigenetic clocks are a promising tool for measuring biological age that show impressive accuracy across most tissues and age ranges. In adults, deviations from the DNA methylation (DNAm) age prediction are correlated with several age-related phenotypes, such as mortality and frailty. In children, however, fewer such associations have been made, possibly because DNAm changes are more dynamic in pediatric populations as compared to adults. To address this gap, we aimed to develop a highly accurate, noninvasive, biological measure of age specific to pediatric samples using buccal epithelial cell DNAm. We gathered 1,721 genome-wide DNAm profiles from 11 different cohorts of typically developing individuals aged 0 to 20 y old. Elastic net penalized regression was used to select 94 CpG sites from a training dataset (n = 1,032), with performance assessed in a separate test dataset (n = 689). DNAm at these 94 CpG sites was highly predictive of age in the test cohort (median absolute error = 0.35 y). The Pediatric-Buccal-Epigenetic (PedBE) clock was characterized in additional cohorts, showcasing the accuracy in longitudinal data, the performance in nonbuccal tissues and adult age ranges, and the association with obstetric outcomes. The PedBE tool for measuring biological age in children might help in understanding the environmental and contextual factors that shape the DNA methylome during child development, and how it, in turn, might relate to child health and disease.

Keywords: DNA methylation; adolescence; age; development; epigenetic clock.

Fig. 1.

Pediatric buccal DNA methylation age accurately predicted chronological age. (A) The training data (datasets 1 through 7) was used in an elastic net regression for selection of 94 age-related CpG sites. Estimated age (PedBE age) (y axis) versus chronological age (x axis). (B) The test data (datasets 8 through 11) was to independently assess the PedBE clock (r = 0.98, test error = 0.35 y). Each data point represents an individual, with the color indicating the corresponding dataset. Solid black and dashed lines denote the linear regression and perfect correlation lines, respectively.

Fig. 2.

Longitudinal data demonstrated higher accuracy of the PedBE clock as compared to the pan-tissue Horvath DNAm clock. (A, Left) Dataset 10 PedBE clock versus chronological age across 2 time points (follow-up ranged from 6 mo to 2 y, depending on the individual). Each point color represents an individual separated by a line indicating the time between sampling. (Right) Dataset 10 pan-tissue Horvath DNAm clock versus chronological age. (B, Left) Dataset 11 PedBE clock versus age across 2 times points separated by 1 y between sampling. (Right) Dataset 11 pan-tissue Horvath DNAm clock versus chronological age. Each colored line represents an individual and the time between sampling is denoted by the beginning and end of each line. Time gaps between sample collection varied across individuals ranging from 6 mo to 2 y. Additionally, for dataset 10, individuals at time point one varied in age between 4 and 12 y. PedBE age was calculated for each individual at both time points.

Fig. 3.

PedBE deviation was associated with gestational age at 3 and 9 mo and individuals diagnosed with ASD in independent cohorts. (A) Dataset 16 is a longitudinal cohort with sampling at 3 mo (Left) and 9 mo (Right) of age in the same individuals. (B) In dataset 9B (n = 81), PedBE age deviation equals PedBE regressed onto chronological age, while controlling for sex, batch, predicted buccal proportion, and ethnicity. A nonparametric propensity score-matching method was applied to ensure the groups were balanced regarding covariate measures.