Independent validation of DNA-based approaches for age prediction in blood

- Age association of sjTREC and DNA methylation in the ELOVL2, C1orf132, TRIM59, KLF14, and FHL2 genes were reanalyzed in 100 Korean blood samples.
- Our study successfully replicated the age association test for sjTREC and DNA methylation in the 5 genes.
- The age predictive performance of previously reported models is relatively consistent in different population groups.
- The extent of age association at certain CpG loci differed between the Korean and Polish populations.
- Combination of DNA methylation and sjTREC improved the prediction accuracy in the eldest group.

Numerous molecular biomarkers have been proposed as predictors of chronological age. Among them, T-cell specific DNA rearrangement and DNA methylation markers have been introduced as forensic age predictors in blood because of their high prediction accuracy. These markers appear highly promising, but for better application to forensic casework sample analysis the proposed markers and genotyping methods must be tested further. In the current study, signal-joint T-cell receptor excision circles (sjTRECs) and DNA methylation markers located in the ELOVL2, C1orf132, TRIM59, KLF14, and FHL2 genes were reanalyzed in 100 Korean blood samples to test their associations with chronological age, using the same analysis platform used in previous reports. Our study replicated the age association test for sjTREC and DNA methylation markers in the 5 genes in an independent validation set of 100 Koreans, and proved that the age predictive performance of the previous models is relatively consistent across different population groups. However, the extent of age association at certain CpG loci was not identical in the Korean and Polish populations; therefore, several age predictive models were retrained with the data obtained here. All of the 3 models retrained with DNA methylation and/or sjTREC data have a CpG site each from the ELOVL2 and FHL2 genes in common, and produced better prediction accuracy than previously reported models. This is attributable to the fact that the retrained model better fits the existing data and that the calculated prediction accuracy could be higher when the training data and the test data are the same. However, it is notable that the combination of different types of markers, i.e., sjTREC and DNA methylation, improved prediction accuracy in the eldest group. Our study demonstrates the usefulness of the proposed markers and the genotyping method in an independent dataset, and suggests the possibility of combining different types of DNA markers to improve prediction accuracy.