PMI estimation through metabolomics and potassium analysis on animal vitreous humour

Abstract

Introduction: The estimation of post-mortem interval (PMI) remains a major challenge in forensic science. Most of the proposed approaches lack the reliability required to meet the rigorous forensic standards.

Objectives: We applied ¹H NMR metabolomics to estimate PMI on ovine vitreous humour comparing the results with the actual scientific gold standard, namely vitreous potassium concentrations.

Methods: Vitreous humour samples were collected in a time frame ranging from 6 to 86 h after death. Experiments were performed by using ¹H NMR metabolomics and ion capillary analysis. Data were submitted to multivariate statistical data analysis.

Results: A multivariate calibration model was built to estimate PMI based on 47 vitreous humour samples. The model was validated with an independent test set of 24 samples, obtaining a prediction error on the entire range of 6.9 h for PMI < 24 h, 7.4 h for PMI between 24 and 48 h, and 10.3 h for PMI > 48 h. Time-related modifications of the ¹H NMR vitreous metabolomic profile could predict PMI better than potassium up to 48 h after death, whilst a combination of the two is better than the single approach for higher PMI estimation.

Conclusion: The present study, although in a proof-of-concept animal model, shows that vitreous metabolomics can be a powerful tool to predict PMI providing a more accurate estimation compared to the widely studied approach based on vitreous potassium concentrations.

Keywords: 1H NMR metabolomics; Animal model; CIA; PMI; Potassium concentration; Vitreous humour.

Fig. 1

PCA model: PMI increases from the upper right-hand corner to the bottom left-hand corner (panel A). Investigating the loading plot (panel B), the decreasing of glucose, pyruvate, and 3-OH-butyrate with the increasing of PMI is observed, whilst late PMI samples (48–84 h) are characterised by higher levels of taurine, choline, creatine, hypoxanthine, ethanolamine, and succinate. Samples of the training set are indicated as “training” whereas samples of the test set as “test”

Fig. 2

Regression model for ordinal data: confusion matrices obtained calculating the training set (A), performing cross-validation (B) and predicting the test set (C)

Fig. 3

The SR obtained for the ordinal regression model (SRordinal) and that of the regression model (SR) are reported in the same plot; both the ordinal and the regression models discovered 3-hydroxybutyrate, negatively correlated to PMI, and alanine, glutamate, and glycine, positively correlated to PMI, as significantly relevant in predicting PMI (the profiles of these metabolites are reported in Fig. S2 of Supplementary Materials). The SR values have been multiplied by the sign of the Pearson correlation coefficient calculated between PMI and metabolite concentration; dashed red lines indicate the thresholds of SR at level α = 0.05

Fig. 4

Potassium concentration ([K⁺]) vs. PMI; the dashed line indicates the regression line estimated by linear regression

Fig. 5

Potassium concentration vs. quantified metabolites: SR plot; threonine, choline, alanine, hypoxanthine, taurine, creatine, glutamate, and glycine, positively correlated to [K⁺], and glucose and 3-hydroxybutyrate, negatively correlated to [K⁺], were discovered as relevant. The SR values have been multiplied by the sign of the Pearson correlation coefficient calculated between [K⁺] and metabolite concentration; dashed red lines indicate the thresholds of SR at level α = 0.05