Highlights
A set of 11 CpGs for age prediction was investigated in 148 Danish blood samples.
Assay optimization improved coverage and reliability in DNA methylation quantification.
The original model developed for UK individuals showed a bias in age predictions in Danes, underestimating their ages.
Retraining the model on Danish data gave high accuracy (MAE = 3.35 years) and eliminated the bias.
The new model maintained accurate age predictions with DNA inputs as low as 10ng
Abstract
DNA methylation patterns have emerged as reliable markers for age estimation, offering potential applications in forensic investigations, namely, in cases where there is no information about a possible suspect, in the identification of victims of mass disasters, or in immigration cases when assessing the age of individuals seeking asylum.
This study aimed to evaluate the 11-CpG panel proposed by Aliferi et al. (2022) for age estimation. During the implementation phase, the ELOVL2 amplicon from the original work was replaced with a shorter fragment, and the two PCR multiplexes were optimized by changing the amplicons and primer conditions of each multiplex. The technical performance of the optimised assay was assessed using artificially methylated DNA standards. Robust quantification of the methylation levels at the 11 CpG sites was observed. Sensitivity tests demonstrated that DNA inputs down to 10ng could produce reliable methylation quantification.
Using the optimised panel, 148 Danish blood samples (18 – 68 years of age) were typed for their methylation status at the 11 CpG sites. Results showed that the DNA methylation at the 11 CpG loci was significantly correlated with age (0.68 ≤ r ≤ 0.88) in the Danish sample set, confirming the potential of the 11 CpGs in age prediction.
A Danish age prediction model was constructed using 108 of the Danish blood samples and a support vector machine with polynomial function (SVMp). The performances of the new model and the original model based on UK individuals were compared using the remaining 40 Danish blood samples. Comparing the published model to the one developed in this study gave similar results with mean absolute errors (MAE) of 3.28 and 3.35, respectively. However, the original model showed a bias in the age predictions, underestimating the age by an average of 1.53 years in the Danish samples. This bias towards underestimation was not observed in the newly developed age prediction model based on Danish individuals.
In summary, this assay provides a reasonably accurate age estimation of a single-source donor, if the sample material is blood and more than 10ng of nuclear DNA can be extracted from the sample.
