DNA methylation-based age estimation for adults and minors: considering sex-specific differences and non-linear correlations.

DNA methylation patterns change during human lifetime; thus, they can be used to estimate an individual's age. It is known, however, that correlation between DNA methylation and aging might not be linear and that the sex might influence the methylation status. In this study, we conducted a comparative evaluation of linear and several non-linear regressions, as well as sex-specific versus unisex models. Buccal swab samples from 230 donors aged 1 to 88 years were analyzed using a minisequencing multiplex array. Samples were divided into a training set (n = 161) and a validation set (n = 69). The training set was used for a sequential replacement regression and a simultaneous 10-fold cross-validation. The resulting model was improved by including a cut-off of 20 years, dividing the younger individuals with non-linear from the older individuals with linear dependence between age and methylation status. Sex-specific models were developed and improved prediction accuracy in females but not in males, which might be explained by a small sample set. We finally established a non-linear, unisex model combining the markers EDARADD, KLF14, ELOVL2, FHL2, C1orf132, and TRIM59. While age- and sex-adjustments did not generally improve the performance of our model, we discuss how other models and large cohorts might benefit from such adjustments. Our model showed a cross-validated MAD and RMSE of 4.680 and 6.436 years in the training set and of 4.695 and 6.602 years in the validation set, respectively. We briefly explain how to apply the model for age prediction.

Development of two age estimation models for buccal swab samples based on 3 CpG sites analyzed with pyrosequencing and minisequencing.

The analysis of DNA methylation levels of specific CpG sites is one of the most promising molecular techniques to estimate an individual's age. Numerous studies were published recently presenting age estimation models based on DNA methylation patterns from blood samples, with only a few using saliva or buccal swabs. The aim of this study was to identify age-dependent methylation of 88 CpG sites in eight different marker regions (PDE4C, ELOVL2, ITGA2B, ASPA, EDARADD, SST, KLF14 and SLC12A5) in buccal swab samples. A total of 141 buccal swabs from individuals with age ranging from 21 to 69 years were split into a training set (n = 95) and a validation set (n = 46). Samples of the training set were analyzed by pyrosequencing and markers with best age correlation were identified. Stepwise linear regression analysis was performed resulting in an age estimation model including three of the examined CpG sites and showing a mean absolute deviation of estimated from chronological age of 5.11 years. To allow easy implementation into forensic laboratories without the need for pyrosequencing equipment, a multiplex minisequencing reaction was developed, including the same CpG sites previously identified by pyrosequencing. An adjusted age estimation model was evaluated with a mean absolute deviation of estimated from chronological age of 5.16 years. The independent validation set of 46 buccal swab samples was used to test model performances. Mean absolute deviation of estimated from chronological age was 5.33 years and 6.44 years for the pyrosequencing model and the minisequencing model, respectively. Comparison of the two methods showed a high concordance of results, both, qualitatively and quantitatively. In conclusion, buccal swabs offer a suitable alternative to blood samples for molecular age estimation with the additional advantage of being collected non-invasively. Furthermore we showed that minisequencing offers a cost-effective and easy-to-integrate alternative to pyrosequencing for the analysis of methylation status of individual CpG sites.