Much of my junior year at Georgetown University was spent in an animal research facility. Along with my undergraduate thesis mentor and several fellow students, I studied the impact of a high-fat (ketogenic) diet in Sprague-Dawley rats. We had read reports in which human children with epilepsy who were fed this ketogenic diet experienced fewer seizures. Now we were attempting to ascertain whether rats eating a ketogenic chow would experience seizures at a different rate than those eating a chow rich in carbohydrates.

I graduated before the research project was complete, but later learned that some differences in seizure incidence between the two groups of rats were identified. Yet the team was never able to figure out the root cause underlying these differences.

I do know that my mentor at the time is one of the most intelligent and supportive academics I have ever known. Also, I believe our study made sense on a number of levels, and I’m thankful for the experience I gained while working on the project.

Nevertheless, after leaving the animal testing facility, my continued science education led me further and further away from research on mice (murine research). A major change in my thinking occurred around 2004, when I discovered (you guessed it!) the human microbiome – the existence of which I had not known about during my murine research days. I quickly became fascinated by early reports on the body’s microbial ecosystems, the composition of which are fairly unique to humans.

In 2005, I also began to work with the non-profit Autoimmunity Research Foundation (ARF). Part of our work centered on studying the Vitamin D Nuclear Receptor (VDR) – a receptor that plays an important role in regulating the human immune response. Our molecular data, and dozens of preliminary patient case reports, suggested that a medication already used to treat low blood pressure – olmesartan medoxomil – could activate the receptor.

We knew that the expected next step would be to study how olmesartan affects the VDR in mice. But from the get go, the prospect of using mice as research subjects presented us with many serious problems. For one thing, the human VDR and mouse VDR have very little in common.

As described here, the human VDR controls genes necessary for a robust immune response, as well as dozens more connected to autoimmune conditions and cancers. It regulates expression of several families of antimicrobial peptides (AMPs). The AMPs are natural antibiotics that participate in every facet of innate immunity. When activated, they have widespread antibacterial, antiviral, and antifungal activity.

However, the mouse/rat VDR does not similarly control the murine innate immune system. Although mice have VDRs, they express vastly different genes than the human VDR. For example, the gene encoding the calcium-binding protein osteocalcin is “robustly” transcribed by the VDR in humans, but not in mice. Furthermore, the rat VDR does not express the cathelicidin antimicrobial peptides – the AMPs that most aggressively help the body target microbes and other infectious agents.

In fact, Gombart and team have clarified the time period during which evolutionary forces caused the mouse and human VDRs to perform different functions. Approximately 55-60 million years ago, a “divergence in steroid hormone nuclear receptor gene regulation” placed the cathelicidin pathway under VDR control in only humans and closely related primates. Even then, cathelicidin in primates is still not identical to that in man.

Even more problems

After noting these differences, I did more background research on the manner in which the murine and human immune systems differ. A 2004 paper by Javier Mestas at University of California Irvine caught my attention. Metas discusses major differences between murine and human biology, particularly when it comes to the immune response. For example, Toll-Like Receptors are important proteins that allow immune cells to actively respond to microbes. Thirteen of these receptors have currently been discovered but, according to Mestas, their actions vary widely between mice and men. For example, TLR8 detects RNA in humans but has no known function in mice. Conversely, TLR10 exists in humans only.

Mestas ends his paper by stating, “There has been a tendency to ignore differences and in many cases, perhaps, make the assumption that what is true in mice is necessarily true in humans. By making such assumptions we run the risk of overlooking aspects of human immunology that do not occur, or cannot be modeled, in mice.”

At this point, we had forged a partnership with the University of California San Francisco ALS clinic who who were prepared to trial olmesartan in an ALS patient. Yet without a murine trial under our belt, we knew the FDA might not grant this patient access to the medication.

This led to a second major problem. We could not find a mouse model of ALS that seemed applicable to humans. Our use of olmesartan in ALS hinged on the hypothesis that the immune response, and subsequently the microbiome, might be dysregulated in ALS. However, the murine microbiome has developed under very different evolutionary and environmental conditions than the human microbiome. Also, dysregulation of the human microbiome happens gradually, over the course or years or even decades. But the average mouse lives for just weeks. So while we could artificially create the symptoms of ALS in a mouse, such symptoms would not result from the same molecular processes that may drive the actual human disease.

The results of other studies using murine models of ALS are not encouraging. For example, the journal Nature published a comment by Steve Perrin, chief scientific officer of the ALS Therapy Development Institute in Massachusetts. Perrin’s group has tested over 100 compounds identified as candidate drugs for ALS in mice with symptoms similar to the disease. After a massive expenditure of time and money, not one of these compounds was found to help human patients with ALS. Eight of the compounds had shown promise in studies on mice, but had later failed in human trials. These trials involved thousands of patients, all of whom were not allowed to take other medications for their ALS during the study period.

Perrin emphasizes that in most of the above trials, the murine models of ALS used simply did not correlate with human biology. For example, some trials used mice that express a mutant form of the protein TDP43. At first, mice with this mutation appeared to develop ALS symptoms. But later, Perrin and team realized that these TDP43 mice were dying of bowel obstructions and not the muscle-wasting and breathing disturbances that make the disease fatal in humans. They further realized that although the first generation of TDP43 mice had been reported to die within 200 days, later generations bred from the original animals lived for up to 400 days without signs of the illness.

Application to the FDA

Around 2009, I began discussing murine and human biological differences in my speeches and presentations, hoping for a reaction from the research community. I coined the phrase “Men are not tall mice without tails” in an attempt to clarify the problem in more simple terms. At face value, the phrase seemed obvious. Even a non-scientist can come up with an extensive list of ways in which humans differ drastically from mice. Yet few researchers commented on my feedback. Instead, I became increasingly aware of a growing reality: a huge number of trials aimed at understanding human inflammatory disease are performed in mice.

Einstein once said that “Insanity is doing the same thing over and over and expecting a different result.” Our team had become aware of too many failed murine drug trials. As described here, “approximately 90% of the drugs reported to have efficacy in high profile journals based on animal models fail in (human) clinical trials.” We decided not to study the effects of olmesartan in rodents. Doing so seemed unfair to our donors and to the already overworked members of our research team.

We applied to the FDA with molecular data and patient case histories as evidence for our request. Our application included a long section explaining our rationale for excluding mice as research subjects. This included computer-derived data suggesting that that the mouse VDR might not even respond to olmesartan.

Despite this, the FDA responded that we could not move forward without testing the drug in animals. We were required to repeat the animal safety testing already performed for olmesartan’s original indication, hypertension. From that point the drug would again need to traverse Phase 1, Phase 2, Phase 3 (and perhaps even Phase 4) clinical trials before gaining approval in the new indication, a process which takes, on average, around 12 years.

Frustratingly, the FDA did not comment on the rest of the scientific information in our application (the application was almost 3 inches high). In one response to the FDA, we wrote, “We have proposed that a safe, widely used, blood-pressure drug (olmesartan) could be retargeted to a different indication, ALS. Until we can figure out how to satisfy the FDA animal studies requirement, our research is at a standstill, and ALS remains a deadly disease without an effective therapy.”

Seok and team change the game

Then, in 2013, a seminal paper was published that I thought might push the FDA to change its position. A team led by Seok at Stanford conducted the first study to compare the human and murine immune response at the molecular level. In other words, they took humans and mice suffering from the “same” medical conditions and studied their gene expression rather than their external physical signs and symptoms.

When an animal becomes ill, its body adjusts the activity of genes upwards or downwards in an attempt to coordinate pathways that best promote healing. These genomic responses typically involve coordinated changes in the expression (activity) of thousands of genes.

Seok and team looked at human patients with three acute inflammatory conditions – trauma, burns, and sepsis. The severe inflammatory stress associated with these events produces a “genomic storm” that impacts all major cellular functions and pathways. The team subsequently identified 5,544 human genes whose activity changes when people suffer these conditions.

They looked for equivalents to these human genes, or murine orthologs, in three different murine models of trauma, burns, and sepsis. Human genes and their murine orthologs are assumed to have similar functions because, long ago, humans and mice arose from a common ancestor. Of the 5,544 human genes identified by the team as altered by trauma, burns, or sepsis, 4,918 had corresponding orthologs in the three murine models under study.

The team then compared how the activity of each human gene changed in relation to its murine ortholog under conditions of burns, trauma, or sepsis. What they found was eye-opening. In the case of all three inflammatory disease states, activity of the human genes under study changed in a manner that was completely different than that of their related murine orthologs. In fact, the genomic responses observed in the two species were so different that random chance would have provided the team with the same results.

When the researchers looked more closely at the 100 genes that changed the most in the inflammatory disease states, they saw a small increase in correlation between the species but it remained low. Furthermore, the genomic responses observed in the three different murine models of trauma, burns, and sepsis hardly correlated with one another.

In contrast, despite marked differences in the two disease states, the human burn and trauma study subjects displayed patterns of gene expression that correlated better with each other than with the corresponding mouse models. This was true despite the fact that these human subjects had sustained different levels of illness, and were often treated with very different medications.

The team describes multiple considerations that might explain the tremendous discrepancies. These include the complexity of human disease, and the evolutionary distance between humans and mice (it has been at least 75 million years since humans and mice shared a common ancestor). They also point out differences in the cells that form mouse and human tissue. In addition, many murine models suffer from problems with inbreeding.

They also discuss more straightforward problems. Mice tend to be much more resilient than humans when confronted with an inflammatory challenge. For example, in most strains of mice, the lethal dose of endotoxin is 5–25 mg/kg. By comparison, a dose that is 1,000,000-fold less (30 ng/kg) has been reported to cause shock in humans. Also, (and I realize this sounds cruel), it’s hard to burn mice to the same extent as humans. The amount of skin injured in a burn victim is measured by a metric called Total Burn Surface Area (TBSA). Mice with a TBSA of over 25% tend to die, while humans can withstand a higher TBSA and still live.

In addition, Soek and team found that the timing of the genomic responses associated with trauma, burns, and sepsis varied widely between the humans and mice. For example, in the murine models, genomic disturbances returned to baseline within hours to four days, but lasted for 1-6 months or more in human subjects. This means that “late events related to the clinical care of [human] patients (such as fluids, drugs, surgery, and life support) likely alter genomic responses that are not captured in murine models.”

All in all, the study questions how accurately research on acute inflammatory disease can be translated from mice to humans. Researchers using mice to study chronic inflammatory disease face this same problem, possibly to a greater degree. In chronic inflammatory disease, the immune response changes over years or even decades, so genomic differences associated with time span might be even more pronounced. Chronic inflammatory diseases are also increasingly tied to dysbiosis of the human microbiome – a state that is extremely difficult to model in mice.

Suggested policy changes

Seok and team propose several measures that might improve murine research. Every drug or drug candidate functions at the molecular level. They argue then that research teams should be required to obtain molecular data on any rodents under study. This information could clarify whether the mouse model might “fail to mimic the molecular behavior of key genes, key pathways, or the genome-wide level thought to be important for the relevant human disease.”

They also argue that more studies should be performed in actual human patients with the goal of clarifying the molecular mechanisms driving different disease states. Data from such studies could then be used to develop mouse models that more accurately mimic human illness. For example, in a New York Times article about the study, Dr. Richard Hotchkiss, a sepsis researcher at Washington University comments: “This is a very important paper… It argues strongly — go to the patients. Get their cells. Get their tissues whenever you can. Get cells from airways. To understand sepsis, you have to go to the patients.”

This feedback pushes the scientific community in a new direction. Today a drug must be shown to work in mice before it can be tested in other settings. Seok and team argue for the reverse. Before a study in rodents moves forward, researchers should be required to provide mathematical data, or molecular data from human patients, which proves that their murine model can be used with accuracy.

Interestingly, one of the former heads of the FDA has pushed for similar measures. Andrew von Eschenbach commissioned the FDA from 2006 to 2009. He now chairs “Project FDA” at a think tank called the Manhattan Institute. The Project encourages the FDA to allow molecular science and patient-based outcomes to take a larger role in drug approval. According to the Project’s website:

“Project FDA believes the FDA can become a bridge for innovation, rather than a barrier to it, and that this can be achieved without sacrificing patient safety. For instance, advances in molecular medicine that allow companies to target specific sub-groups of patients, combined with electronic health records, should allow the FDA to streamline and improve time consuming and expensive pre-market product testing that can take a decade or more, and implement vigorous post-market surveillance of “real world” patients after drugs or devices demonstrate safety and efficacy in early testing. This approach will not only accelerate access to innovative products; it should enhance efforts to safeguard public health.”

Reaction to the study

Since its publication, several research teams have published comments in response to the Seok paper. Most of these comments are written by researchers with long track records for using murine models. They largely attempt to point out shortcomings in the study’s methodology. For example, in the comment “Abandon the mouse research ship? Not just yet!” researchers from a number of institutions point out that the  Seok study used only male mice, and that perhaps female mice should have been included.

However, Seok and team were only following standard practice by using male mice. According to New Scientist magazine, “Despite decades of research demonstrating that gender matters in animal research, many studies continue to be done in only one gender, because doing otherwise would increase complexity and cost.” Jeffrey S. Mogil, a McGill University psychology researcher, writes, “Everyone uses only males, across the board.”

Seok and team were also criticized for using only one strain of mouse in their analysis. The authors of the “Abandon ship” comment suggest that the team might have used multiple animal models. But few research teams have the funding to test a drug in one animal model, let alone several. This means that Seok’s choice to perform their research in only one strain of mouse fairly reflects what happens in a typical study.

Other comments written by researchers in the stroke community are characterized by denial. In one of these comments, Dr. Ulrich Dirnagl writes that, “Stroke research has a dismal record when it comes to translating its findings into novel and effective therapies. Several hundred clinical trials have been unable to replicate neuroprotection, which was highly effective in rodents.” Yet after describing this reality, he still concludes that “despite the current nihilism, preclinical stroke research (on mice) can successfully predict human pathophysiology, clinical phenotypes, and therapeutic outcomes.”

Another comment points out that mice have been used, “very effectively in many cases to study basic biology in conditions of health and disease.” This is true. For example, in 2007 the Nobel Prize in Medicine was awarded to researchers who developed what are known as “knockout mice.” Studies in these mice test the consequences of deleting a single gene in the animal, an exercise that has indeed clarified biological processes. However, Seok and team examine murine research in a totally different context. Their study tests the ability of murine models to translate to complex medical conditions – conditions characterized by long-term changes in thousands of genes.

Returning to the the first “Abandon ship” comment, the authors miss the boat (haha) by stating, “it remains uncertain whether and/or to what extent the results of gene expression profiling should be used to judge the biological validity of animal models for human disease.” To the contrary, there should be no debate on this topic. Why not use the latest technologies, in concert with the most current understanding of genomics, to evaluate murine models? Of course doing so raises the level of complexity associated with murine research, but marks the only way to study mice in the context of 21st century science. This is particularly true since the drugs tested in murine models function at the molecular level.

In fact, it is this last point that makes the Seok and team paper so important. One can nitpick the methods of the study ad nauseam, but the researchers weren’t obsessed with conducting the most perfect murine trial of all time (if that’s even possible). Rather, the study clarifies how a typical mouse study can fail if it lacks essential knowledge derived from the field of molecular biology.

Nothing changes

This begs the questions: has the Seok and team study pushed murine research into the 21st century, or at least started a serious discussion on the topic?

After the study was published, sepsis researcher Mitchel Fink told the New York Times, “This is a game changer.” He continued, “When I read the paper, I was stunned by just how bad the mouse data are. It’s really amazing — no correlation at all. These data are so persuasive and so robust that I think funding agencies are going to take note. Until now, to get funding, you had to propose experiments using the mouse model.”

Similarly, in one of the comments to the paper, researcher Andrew Drake asserts that Seok and team have “provided a new standard for testing models and interpreting drug efficacy results.” He continues, “the basic template has now been drafted: regulatory agencies should begin asking for gene expression data validating the relevance of in vivo clinical studies…”

However, as far as I can tell, the study has not prompted any policy changes, or been formally address by funding agencies. No large-scale meetings or initiatives have been created to discuss the research considerations raised by Seok and team. The FDA has not actually implemented any of Eschenbach’s proposals. Murine data remains the cornerstone of scientific research and is required by regulatory agencies before other types of trials can move forward.

Under these conditions everyone loses. Research teams lose the ability to obtain accurate data. Drug companies lose millions/billions of dollars that could otherwise be spent on innovation. Patients lose the ability to participate in trials that might truly improve their health. Taxpayers lose the opportunity to fund trials that might better help friends or family. And of course, mice lose their lives (in often cruel ways) for no good reason.

Again, I am not suggesting that all murine research is useless. My concern is one of priority – I do not think that murine data should regularly be valued over mathematical data or patient case histories. Murine research can still play a role in 21st century science. But we must create a novel framework in which mice are used under only the most fruitful circumstances.