Plant and fungal eDNA metabarcoding for geographic sourcing of forensic palynological evidence across urban...

Plant pollen and spores are promising forensic indicators for geographic sourcing because their DNA is persistent and their community composition is shaped by local vegetation and environmental conditions. However, reference datasets designed specifically for forensic palynological applications remain limited. Here, we evaluated the feasibility of environmental DNA (eDNA) metabarcoding for geographic sourcing of palynological trace evidence across urban functional zones in Xi'an, China. Airborne palynomorphs were collected from nine city-wide sites representing seven functional zones in spring and summer. High-throughput amplicon sequencing was performed using the plant rbcL and matK barcodes and the fungal ITS1 region, followed by diversity analysis, LEfSe biomarker screening, random forest classification and SourceTracker-based source attribution. Plant and fungal palynomorph communities differed significantly among urban functional zones. Plant assemblages showed marked seasonal succession, with wind-pollinated trees dominant in spring and insect-pollinated plants, ferns and vines increasing in summer. Fungal assemblages contained a stable core community together with season-specific taxa. Random forest models achieved site-level accuracies of 84.38% for spring rbcL and nearly 87.5% for summer matK in plant datasets, and > 95% for fungal ITS1 datasets in both seasons. In the forensic validation analysis, SourceTracker assigned 30%-56% of plant-derived and 41%-48% of fungal-derived components in summer dust samples to their true sampling site. These results demonstrate that airborne palynomorph eDNA contains strong spatial and seasonal information. Plants and fungi provide complementary forensic signals: plants are informative seasonal indicators, whereas fungi offer higher spatial resolution. The rbcL barcode showed greater amplification stability in airborne samples, and the integration of metabarcoding with machine-learning classification provides a practical framework for forensic geographic sourcing.