Debate rages over mouse studies’ relevance to humans

Last year, researchers concluded rodents are not good mimics of human inflammation; new study says the reverse

In direct conflict with a controversial 2013 study, researchers say that mice may be better mimics of human inflammation than previously suggested.

The earlier study found that inflammation brought on by burns, sepsis and trauma changed gene activity very differently in mice and humans (SN: 3/23/13, p.10). But two scientists have now reanalyzed data from that study using different statistical methods. Mice and humans actually have very similar gene activity changes, the researchers report August 4 in the Proceedings of the National Academy of Sciences.

In addition to using distinct statistical methods, the two teams analyzed different subsets of data. Each group accuses the other of bias in their data choices.

The original study’s conclusion jeopardized funding for studies that use mice as human stand-ins, says Klaus Schughart, a geneticist and infectious disease researcher at the Helmholtz Centre for Infection Research in Braunschweig, Germany, who was not involved in either study. He thinks the new study better represents how useful mice are as imitations of human diseases.

Animal mimics — or models, as scientists call them — are never perfect representations of people, but mice have been shown time and again to react as humans do to many diseases and injuries, Schughart says.

In last year’s report, though, researchers examined patterns of gene activity changes in response to traumas, such as a burns and sepsis, in mice and humans and measured how well those patterns matched up. The answer appeared in the paper’s title: poorly.

In humans, more than 13,000 genes altered activity when a person was burned, for instance. But in mice, activity changed in only few more than 4,000 genes. Of the genes with altered activity, only a fraction — not quite 2,300 genes — changed in both species, and even fewer of those reacted in the same way.

Tsuyoshi Miyakawa, a neuroscientist at Fujita Health University in Toyoake, Japan, and Keizo Takao at the National Institute for Physiological Sciences in Okazaki, Japan, thought that approach used in that study stacked the deck against mice. They reanalyzed the data, looking at the subset of genes that change activity in both species.

For example among the 2,283 genes for which burns changed activity in both species, 940 became more active in both mice and humans, 852 decreased activity in both, and 664 increased activity in one species but decreased in the other, Miyakawa and Takao found. The fact that so many genes changed expression in the same direction, up or down, is meaningful, Miyakawa said. The title of his and Takao’s paper differs by only one word from the 2013 report’s title: Mice “greatly” instead of “poorly” mimic humans.

But Ronald Tompkins, a surgeon at Massachusetts General Hospital in Boston who coauthored last year’s study, says Miyakawa and Takao “cherry-picked” their data, giving a view biased toward mice. The 1,500 or so genes with activity changes in the same direction represent only about 11 percent of all changes in humans, Tompkins says. That’s a poor mimic, he says.

Assuming that mice and humans are essentially the same and examining only shared responses hasn’t led to effective therapies for trauma, burn and sepsis victims, Tompkins says. More research should focus on people. “We’re confident about our data set and our interpretation,” he says.

The difficulty is that the two groups are asking fundamentally different biological questions, says George Tseng, a biostatistician at the University of Pittsburgh. The first group asked whether mice can fully represent the human disease. The second group asked whether mice could partially mimic the gene activity changes. “Both papers were right, but the second paper, in my opinion, is scientifically more meaningful for animal model research,” Tseng says.

That’s because Miyakawa and Takao’s used a different statistical method, one that is better because it makes fewer assumptions, he says. He would have liked the Japanese researchers to apply their improved analytical method to the biological question asked in the first study. He suspects that the method would detect greater overlap between human and mouse reaction to trauma than the 2013 paper reported.

If there is a moral to the story, Tseng says it’s that different biological questions and data analyses can produce wildly different conclusions. “Mathematics cannot lie, but problematic study design…and interpretation can create unintentionally misleading conclusions.”

Editor’s note: This story was updated on August 27, 2014, to clarify in the first sentence that the study is of the relevance of mouse models to human inflammation. It was further updated on September 16, 2014, to correct the number of genes whose activity decreased in both species — 852, not 854.

Tina Hesman Saey is the senior staff writer and reports on molecular biology. She has a Ph.D. in molecular genetics from Washington University in St. Louis and a master’s degree in science journalism from Boston University.

More Stories from Science News on Genetics