Simultaneous analysis of miRNA and 16S rDNA by a multiplex droplet digital PCR system for body fluid source...

Forensic body fluid identification, crucial for crime scene reconstruction, is often limited by sample scarcity and degradation. Hence, there is growing interest in more stable biomarkers like miRNA and microbial markers. While each has its limitations, they effectively complement one another. Specifically, miRNA requires complex multi-marker panels to distinguish fluids like saliva and vaginal secretion, whereas microbial markers can identify these readily due to their distinctive microbiomes. Conversely, when dealing with body fluids with low microbial biomass or similar microbial composition, microbial markers exhibit limitations, whereas miRNA can accurately identify them. Capitalising on this complementary relationship, we aimed to accurately discriminate among the five common body fluids (peripheral blood, menstrual blood, saliva, semen, and vaginal secretion). To this end, we optimised a co-extraction protocol for RNA and DNA to maximise information yield from trace samples, and verified its feasibility using laboratory-prepared substrate samples and aged samples (six years). Subsequently, we developed a single-tube, 5-plex droplet digital PCR system. This system targets five specific markers, comprising miRNA markers (miR-451a, miR-891a-5p, the internal reference miR-320a-3p) and microbial markers (16S rDNA of Streptococcus salivarius and Lactobacillus crispatus). Then, based on a dataset of 115 samples, five machine learning models were developed, including random forest (RF), eXtreme gradient boosting, k‑nearest neighbours, logistic regression and support vector machine. The results showed that the RF model demonstrated the optimal identification performance, achieving 100% accuracy on a validation set (30% of the total 115 samples). The method was further validated for robustness using 25 external test samples, 45 substrate samples (e.g., toilet paper, cotton swabs, underwear) and 15 aged samples, achieving prediction accuracies of 100%, 100%, and 93%, respectively. Sensitivity assessment established the detection limits for each body fluid, ranging from 0.3 pg (vaginal secretion DNA) to 0.3 ng (semen miRNA). Subsequently, three samples per body fluid were selected for sensitivity stability verification, achieving 100% prediction accuracy. These results demonstrate that this integrated method is a robust and promising tool for forensic practice, highlighting the significant potential of strategies that combine multiple markers to address complex identification challenges.