Summary

A bioRxiv preprint reports that repeated measurements of circulating alanine aminotransferase, combined with time since acetaminophen exposure, estimated liver necrosis in mice. The method could let researchers follow liver injury in the same animals instead of using separate groups at every timepoint.

A preprint from researchers at the University of Edinburgh reports a blood-based method for estimating the amount of liver tissue that has undergone necrosis in acetaminophen-treated mice. The approach uses repeated measurements of circulating alanine aminotransferase (ALT), rather than relying only on liver tissue collected when an animal is euthanised.

The study, posted on bioRxiv on September 16, 2026, is a preclinical mouse study and has not been presented as a human diagnostic method. Its central finding is that the cumulative pattern of ALT release over time predicted the percentage of liver area showing centrilobular necrosis, the tissue injury used as the reference measurement.

Contents

How the estimation method works

ALT is an enzyme released into the bloodstream when liver cells are injured. In this study, the researchers measured plasma ALT repeatedly from before acetaminophen exposure through 48 hours afterwards. Instead of using a single ALT value, they calculated the cumulative area under the concentration-time curve up to the point at which each mouse was examined.

The prediction model also included the time elapsed since dosing. A generalised additive model then related those measurements to histological necrosis: the proportion of liver tissue showing damage on haematoxylin and eosin-stained sections.

Histology remains the reference measurement for this type of tissue injury, but it requires a terminal sample. A time course based only on histology therefore generally requires different animals to be euthanised at different timepoints. Repeated plasma sampling offers a way to follow injury in the same animal over time.

The researchers also measured microRNA-122, a liver-associated circulating molecule used in liver-injury research, and considered information related to regeneration. Neither addition improved the model's prediction in this experiment.

What the mouse study found

Forty-five fasted, 12-week-old male C57BL/6J mice received one intraperitoneal dose of 350 mg/kg acetaminophen, also known as paracetamol. Plasma ALT and microRNA-122 were measured serially from baseline to 48 hours.

The reference histology showed centrilobular necrosis ranging from 0% to 59% of the assessed area, with a mean of 32%. Using cumulative ALT exposure and time since dosing, the model produced a leave-one-animal-out cross-validated mean absolute error of 7.7% necrotic area, with a 95% confidence interval of 5.6% to 10.0%.

Leave-one-animal-out validation means that the model was repeatedly fitted using all but one animal and then tested on the animal left out. This tests how closely the estimated necrosis matched the tissue measurement for animals not used in that particular model fit.

The preprint reports that traditional regression and machine-learning models produced equal or larger errors. It also reports that adding another biomarker or regeneration information did not improve the estimates.

Why serial measurements could change study design

The practical value of the method is not simply replacing a microscope-based measurement with a blood measurement. It could allow researchers to measure the progression of liver injury repeatedly in one cohort, then compare the resulting estimates with terminal histology. According to the researchers, that would provide a more stable estimate of variation within an experimental group and could reduce the need for separate cohorts at every timepoint.

The result applies specifically to the experimental setting described in the preprint: male mice exposed to one high acetaminophen dose and followed for 48 hours. Whether the same ALT kinetics can estimate necrosis reliably across different mouse strains, sexes, exposure levels, causes of liver injury or human patients will require further research. The abstract also reports 45 dosed mice but gives the cross-validated model result as n=49; the details supplied with the abstract do not explain that difference.

Sources