Gene activity profiles may one day help solve murders.
PA Images/Alamy Stock Photo
By Elizabeth Pennisi
Sherlock Holmes should have been a geneticist. Thanks to an extensive new survey of gene activity in human tissue after death, computational biologists have taken the first steps toward predicting when someone died based on those patterns.
“One can imagine a time where labs will be equipped with [artificial intelligence programs] that use gene expression together with other contextual information to determine time and cause of death, among other things,” says Ilias Tagkopoulos, a computer scientist at the University of California, Davis, who was not involved with the work.
Computational biologist Roderic Guigó didn’t start out as a death detective. Guigó, of the Centre for Genomic Regulation in Barcelona, is also part of the Genotype-Tissue Expression pilot (GTEx), a large consortium of geneticists and molecular biologists that has been measuring gene activity in tissues from hundreds of people, living and dead. The goal is to determine how the body makes different cells do different things, given that they all carry the same DNA instructions. It also seeks to determine how slight variations in DNA from person to person changes what cells do. Other researchers have already shown that some genes stay active up to 4 days after death. Guigó wanted to find out how gene activity changes as the time to preservation is extended.
He and his colleagues looked at 9000 samples of 36 tissues, “an impressive dataset,” Tagkopoulos says. Each sample included data on the time between the death of the donor and the preservation of the sample. Each tissue has a distinct pattern of increases and decreases in gene activity over time, and these changes can be used to backtrack to the time of death, the team reports today in Nature Communications.
“The response to the death of the organism is quite tissue specific,” Guigó explains. For example, there was very little change over time in the brain’s or spleen’s gene activity, but more than 600 muscle genes either quickly increased or decreased activity after the loss of life.
Guigó and his colleagues developed software that “learned” the patterns of 399 people. They then tested how well the machine learning software did predicting the time of death of 129 other people. The software discovered, for example, that in blood, decreased activity of genes involved in DNA production, immune response, and metabolism—but an increase in those involved with stress responses—signaled the person had died about 6 hours before preservation. The majority of gene activity changes, both increases and decreases, occur between 7 and 14 hours after death. Then after 14 hours, gene activity seems to stabilize, they report.
The findings make sense, Tagkopoulos says. “At a cellular level, death is a cascade of events affecting biological processes at different timescales,” he says, and genes control that cascade.
This software is the first step toward harnessing gene activity for forensics. “At this point, our program is an academic exercise” to show that signatures in gene activity may relay time of death information, Guigó says. And cost and efficiency are issues as well, adds Tagkopoulos. While Guigó’s group has shown that the time of death can be estimated just as well using gene activity levels from two tissues—the lung and the thyroid—his team has not yet been able to reduce the number of genes needed to make the prediction. The more genes analyzed, the more expensive the work, Tagkopoulos says.
Even so, Guigó is eager to see what else he can learn from these patterns. “Changes in gene expression may also carry the signatures of the cause of death,” he says. But they didn’t have enough detailed information to investigate the hypothesis—perhaps, they are saving that for future studies.
Source: Science Mag