Human Blood Thanatotranscriptome Dynamics and Cross-Donor Generalisability of Postmortem Interval Prediction: A Longitudinal RNA-seq Reanalysis

Abstract

Background. Thanatotranscriptomic profiles may encode time-dependent molecular changes after death, but forensic utility requires temporal signals that generalise across individuals.

Objective. To characterise the genes and molecular pathways that change consistently with absolute postmortem interval (PMI) in serial human postmortem blood samples.

Methods. We reanalysed processed counts from 54 serial postmortem blood samples of seven individuals in GSE163207, spanning an absolute PMI of 2.35-37.83 h. After expression filtering and trimmed mean of M-values normalisation, limma-voom models included individual fixed effects and continuous absolute PMI. Eligible genes required a Benjamini-Hochberg false-discovery rate <0.05, an absolute log2 change >=0.10 per 10 h, and concordant direction in at least five individuals. Enrichment used the expressed-gene background. Nested leave-one-individual-out elastic-net validation kept feature selection and tuning within training donors.

Results. Among 12,633 expressed genes, 1,976 were eligible (954 increasing; 1,022 decreasing). Increasing genes were enriched for cytoskeletal, extracellular-matrix, oxidative-stress, and mitochondrial processes; decreasing genes were enriched for adaptive immune and antigen-presentation pathways. Sensitivity analysis produced an effect correlation of 0.987, with 1,645 genes retaining significance and direction. Cross-donor prediction yielded an RMSE of 10.03 h (95% CI 8.12-11.32), MAE of 8.63 h (95% CI 6.33-10.40), and R2 of -0.219, underperforming the training-mean baseline.

Conclusion. Human blood undergoes reproducible early postmortem molecular remodelling, but this seven-individual cohort does not support generalisable individual PMI prediction.