When an STR profile does not produce a match in a DNA database, donor age estimation can provide law enforcement agencies with investigative leads to narrow down the pool of suspects. A promising biomarker for this purpose is DNA methylation. Predictable changes in methylation patterns at specific CpG sites in the human genome provide a basis for estimating a person’s age from biological stains at the time of deposition. Many age estimation tools have already been built and validated for forensically relevant tissues, including blood, saliva, cartilage, etc. Some of them perform highly accurately, achieving mean absolute errors of as little as 3 years. 

Despite many efforts to advance and implement methylation-based age estimation in forensic casework worldwide, there is one major drawback. Since its conception, this method has only been applicable to single source stains. Mixture samples have not been acceptable as input material so far, because it was impossible to disentangle which reads belong to which contributor of the mixture, thereby making it impossible to calculate person-specific methylation values. One proof-of-concept study has recently demonstrated that bioinformatic mixture deconvolution for age prediction is possible. However, further efforts are needed to evaluate how, with which technology and for which scenarios mixture deconvolution for methylation-based age prediction is feasible. 

Our study aims to approach mixture deconvolution for age prediction by using long-read sequencing technology from Oxford Nanopore Technologies (ONT). We seek to demonstrate that 2-person mixture samples can be deconvoluted bioinformatically, allowing the age of each contributor to be estimated separately. 

We sequenced a reference cohort of 50 single source blood samples with ONT to create a preliminary age estimation model that works on this reference data set. We then generated 2-person mixtures from different combinations of the 50 single source samples. We assume that one person in each mixture is known and we bioinformatically removed all reads belonging to the known contributor based on the SNP profile. We then used the remaining reads to calculate the age of the unknown contributor with the previously generated age estimation model. 

This study shall serve as a proof-of-concept whether mixture deconvolution for donor age estimation is possible using ONT in a scenario where one contributor is known. If this is the case, it will be interesting to see whether other scenarios, e.g. when both contributors are unknown, will also be accessible for donor age estimation through new analysis strategies in the future.

Authors

• Charlotte Sutter (Zurich Institute of Forensic Medicine, University of Zurich, Switzerland)
• Cordula Haas (Zurich Institute of Forensic Medicine, University of Zurich, Switzerland)
• Jacqueline Neubauer (Zurich Institute of Forensic Medicine, University of Zurich, Switzerland)