Illustration of mice holding X-rays of their microbiome.

Davide Bonazzi/@Salzmanart

Mouse microbes may make scientific studies harder to replicate

In the first experiment, Laura McCabe’s lab seemed to hit a home run. The physiologist and her team at Michigan State University (MSU) in East Lansing were testing how a certain drug affects bone density, and they found that treated lab mice lost bone compared with controls. “I was thinking, ‘Hey, great! Let’s repeat it one more time to be certain,’” McCabe recalls.

They ordered a seemingly identical batch of mice—same strain, same vendor—and kept them under the same conditions: same type of cage, same bedding, same room. This time, however, treated mice gained bone density. “Maybe one was a fluke,” 
McCabe thought. They did a third run—and saw no effect at all. She was baffled.

I think what this [reproducibility] conversation is doing is expanding the variables for investigators to think about.

J. R. Haywood, Michigan State University's assistant vice president of regulatory affairs

She knew that signals from the gut can affect how bone forms and gets reabsorbed, so her team took fecal samples from control mice in each of the three experiments and analyzed their gut microbes. They found something unexpected: Each group had a different microbial makeup to begin with.

McCabe has no idea where the mice acquired their distinct gut bacteria—from the containers that ferried them from the vendor? From a technician’s clothing? But how the drug affected her subjects clearly depended on what already lived inside them. 

It’s easy to see how such effects could make it difficult to replicate experiments, a concern that has roiled fields from psycho
logy to cancer. A few years ago, two pharmaceutical companies reported that they could not replicate the vast majority of academic findings in preclinical experiments. Pressure to publish and a bias against negative results account for some replication problems. But other failures to replicate likely have “practical explanations: different animal strains, different lab environments or subtle changes in protocol,” as Francis Collins and 
Lawrence Tabak, director and principal deputy director, respectively, of the U.S. National Institutes of Health (NIH) in Bethesda, Maryland, wrote in Nature in 2014. In other words, sometimes a study doesn’t hold up because the replicator is unknowingly performing a slightly different experiment.  

Increasingly, experimenters are questioning the potential research impact of the microbiome—a term often used to refer to commensal gut bacteria, but which also includes resident viruses, fungi, protozoa, and single-celled archaea species. Rarely even discussed a few years ago, this potential source of variability attracts growing attention at lab animal care conferences, says MSU’s attending veterinarian, Claire Hankenson. “We didn’t know to look for it before,” she says.

Yet a mouse’s microbes can be maddeningly hard to pin down. The species living in a mouse are always changing, impossible to fully standardize, and for the most part unmeasured. Adding to the challenge, researchers are realizing that it may be a bad idea to simply wipe out lab mice’s microbial guests, some of which are critical for health and immune responses and help make the animals into robust, meaningful research subjects. How, many researchers wonder, can this variable ever be controlled? 

An invisible variable

Mice are stirring on a quiet summer after
noon at Stanford University in Palo Alto, California. Roused from their daytime sleep by human visitors, they fill their room at the Veterinary Service Center with the faint rustle of shredded paper. Their clear, 
shoebox-sized cages, lining ceiling-high racks arranged like library shelves, represent investigators’ best efforts to control every variable that might skew the outcome of studies. These little worlds are meticulously standardized: equal volumes of sterile bedding, steady cycles of light and dark, even a consistent flow of temperature-controlled air. Every cage is attached to two pressurized vents, as if these mice occupied the cabins of their own personal airplanes.

But are these environments as identical as they look? Researchers surprised by inconsistent results wonder what hidden variables may lurk in these cages. “I think what this [reproducibility] conversation is doing is expanding the variables for investigators to think about,” says J.R. Haywood, MSU’s assistant vice president of regulatory affairs.

And mice’s resident microbes are an emerging concern. The zoo of organisms on and inside each animal can shift for all kinds of reasons, including a change in the formulation of mouse chow, or in the sources of grain or protein within a brand. Cagemates share microbes, thanks to their penchant for eating one another’s feces. Some researchers suspect that even stress, such as an early separation from the mother, can also change a mouse’s microbial ecosystem.

An explosion of recent studies in both animals and people suggests that resident microbes can influence susceptibility to diseases from HIV to asthma, predispose to obesity across generations, and tinker with how the body responds to drugs. Tying such effects to experimental results is challenging, but some hints have cropped up. In one early example, more than a decade ago, a research team at Pfizer detected an odd change in rats’ urine: a sudden shift in the relative concentrations of two compounds produced when the body breaks down food. The change could muddy toxicology studies that rely on urine metabolites to measure how a drug gets broken down in the body. Researchers traced the unusual rat colony to a single room at the vendor’s facility, and they restored the original urine composition in a few weeks by cohousing the rats with animals from other rooms. Although back then no one was sequencing rodent microbiomes, the Pfizer team suspected that microbes were responsible.  

Housing for lab animals is meticulously standardized, but even so, variations in their resident microbes persist.

Housing for lab animals is meticulously standardized, but even so, variations in their resident microbes persist.

Annedde / iStockphoto

In a more recent example, last year scientists at the University of Missouri (MU), Columbia, working with a mouse model of multiple sclerosis accidentally reversed the symptoms by adding a common antibiotic to the animals’ water. They restored symptoms simply by cohousing their mice with a microbially richer strain, suggesting that the traits they had come to rely on in their research hinged on a delicate balance of mouse microbes.

Veterinary pathologists Craig Franklin and Aaron Ericsson, also at MU Columbia, are trying to account for such effects by measuring and manipulating those microbes. “Even 5 years ago, most people considered doing microbiota analysis untouchable, unless that was the expertise in their lab,” Franklin says. But today, more labs are sequencing fecal samples in search of bacterial genes or paying others to do so, he says. His own team offers such analysis for $125 a sample through the NIH-funded Mutant Mouse Resource and Research Center.

Last year, as a first step in defining a “normal” lab mouse microbiome, they analyzed feces from mice from two major vendors. Like all mammal poop, the samples were dominated by bacteria of two phyla: Firmicutes, thought to play a role in the absorption of dietary fats, and Bacteroidetes, associated with high-fiber diets. But the richness of species in a mouse varied between the vendors, as did the abundance of certain microbes.

Mice from one vendor, for example, were notably lacking segmented filamentous bacteria (SFB), a commensal lineage recently shown to help mice produce key antibodies and immune cells in their intestines. This is “a normal symbiont that can have dramatic effects,” Franklin says, and its presence or absence could alter studies of inflammatory response. “We don’t think that SFB is the lone soldier out there,” he adds.

SFB is the dominant microbe of concern for researchers, says Jennifer Phelan, product manager at the commercial mouse supplier Taconic Biosciences in Germantown, New York. But she adds that she fields all kinds of study-specific questions about the contents of mouse guts. “It went from one query per month to … sometimes several per week.” In 2014, the company began including the presence of SFB in health reports to customers, and mice are now available with or without it. Phelan also advises scientists to request mice from the same room when they reorder for an ongoing study.

Rethinking clean

A solitary white mouse sits at the bottom of every rack of cages at the Stanford mouse facility. This “sentinel” is another window into the mouse microbial world. Its sole job is to scratch around in its neighbors’ filth. Caretakers periodically dump in a tablespoon of soiled bedding from other cages on the rack, then test the sentinel for disease.

A mainstay of lab animal facilities, sentinels can pick up the major pathogens already known to sicken mice or skew results. When infectious agents are detected, facilities like Stanford’s scramble to sterilize them away. But Franklin and others suspect that in their zeal to clean up, facilities may have wiped out some of the microbial complexity that makes mice useful models for human disease. Variations in the 
microbiome may skew results, but a diverse 
microbiome and exposure to microbes may be critical for some studies.

Earlier this year, a team led by immuno
logist David Masopust of the University of Minnesota, Twin Cities, tried cohousing lab mice with mice purchased from pet stores. The “dirty” visitors harbored diseases long eradicated from most labs, such as hepatitis and pneumonia. The sudden exposure to these disease-ridden cagemates killed nearly a quarter of the colony, but the survivors began producing a subset of the memory T cells key to fighting infection. They became, the authors argued, a more realistic model of the human immune system. 

V. Altounian/Science

For immunology studies, the approach may be catching on. “My lab is incredibly excited about this,” says Stephen McSorley, an immunologist at the University of California, Davis, School of Veterinary Medicine. Earlier this year, his lab brought in its own “dirty” colony from a company that sells mice as food for zoo animals. He hopes the animals will help him create more realistic mouse models of human chlamydia and salmonella infections.

In ongoing work, Franklin and Ericsson are finding that pet store mice also have highly rich intestinal flora. The prospect that they may offer a better approximation of the human gut than standard lab mice is enticing to some investigators. But introducing untold diseases into animal facilities runs counter to a hard-won culture of cleanliness and to the notion that tightly controlled mice make for more reproducible research. MSU’s Hankenson, for one, isn’t ready to embrace the pet store model. “To have this proposal, ‘Hey don’t keep them quite as clean, it’s okay if you add a dabble of this and a bit of that back into the mouse.’ … It’s a little alarming,” she says. “We’ve worked so hard to have very healthy animals for everyone to use.”

Short of reimagining the lab mouse, others say, investigators need strategies for monitoring how microbes might be influencing their small research subjects. A first step is to design studies that separate the effects of microbial genes from the genes of their animal host, says Herbert Virgin, an immunologist at the Washington University School of Medicine in St. Louis in Missouri. In June, he and his colleague Thaddeus Stappenbeck argued in Nature that all mouse studies should use littermate controls: When studying the activity of a particular gene, breed mice with and without that gene by starting with heterozygous parents. The two groups of offspring will differ only in whether they carry that gene, not in their resident microbes, allowing the influence of the host gene and the microbes to be disentangled.

Virgin also argues that papers should note important details that could affect the microbiome, such as diet and exposure to antibiotics. Franklin predicts that authors will be asked to include fecal microbiome analysis in the material and methods sections of their papers in the next 20 years. But so far, genetic analysis can identify only a fraction of the species in an animal.

For now, adding multiple types of 
microbiomes to a study—more complexity, more unknown variables—may be the best way to know whether results are likely to hold up across diverse microbial makeups. Franklin’s group has created four mouse colonies with different complex microbiota, based on the compositions found in four major vendors. For a few thousand dollars per procedure, they can transfer mouse embryos from other labs into females from each colony, so that the mouse pups pick up the new microbiomes and can be compared to the original strains. So far, no one has taken them up on the offer.

Teasing out messy, microbe-confounded results can not only make a result more reproducible; it may also end up yielding scientific spoils. For MSU’s McCabe, the conflicting findings have led her into a whole new study of how microbes help regulate bone density. The mice that lost bone density in her experiment had started with more bacteria associated with inflammation, a clue she’s following up. “What seemed a very negative result,” she says, “we’re going to use as a powerful finding.”


For more of our coverage on microbes visit our Microbiome topic page.