Professor of Medicine, Director of the MDS Center, Columbia University
Mouse Models

An obvious truth, which is either being ignored or going unaddressed in cancer research, is that mouse models do not mimic human disease well and are essentially worthless for drug development. We cured acute leukemia in mice in 1977 with drugs that we are still using in exactly the same dosage and duration today in humans—with dreadful results. Imagine the artificiality of taking human tumor cells, growing them in lab dishes, then transferring them to mice whose immune systems have been compromised so they cannot reject the implanted tumors, and then exposing these "xenografts" to drugs whose killing efficiency and toxicity profiles will then be applied to treat human cancers. The inherent pitfalls of such an entirely synthesized unnatural model system have also plagued other disciplines.

A recent scientific paper showed that nearly 150 drugs tested, at the cost of billions of dollars, in human trials of sepsis failed because the drugs had been developed using mice. Unfortunately, what looks like sepsis in mice turned out to be very different from what sepsis is in humans. Coverage of this study by Gina Kolata in the New York Times brought heated response from within the biomedical research community: "There is no basis for leveraging a niche piece of research to imply that mice are useless models for all human diseases. . . . The key is to construct the appropriate mouse models and design the experimental conditions that mirror the human situation."

The problem is that there are no appropriate mouse models which can "mirror the human situation." So why is the cancer research community still dominated by the dysfunctional tradition of employing mouse models to test hypotheses for development of new drugs?

Robert Weinberg of the Whitehead Institute at MIT has provided the best answer. In an interview, he offered two reasons. First, there's no other model with which to replace that poor mouse, and second, the FDA "has created inertia because it continues to recognize these [models] as the gold standard for predicting the utility of drugs."

There is a third reason related more to the frailties of human nature. Too many eminent laboratories and illustrious researchers have devoted entire lives to studying malignant diseases in mouse models, and they are the ones reviewing one another's grants and deciding where the NIH money gets spent. They are not prepared to concede that mouse models are basically valueless for most cancer therapeutics.

One of the main reasons we continue to stick to this archaic ethos is to obtain funding. Here is one example: I decided to study a bone marrow malignant disease called myelodysplastic syndromes (MDS), which frequently evolves to acute leukemia, back in the early 1980s. One early decision I made was to concentrate my research on freshly obtained human cells and not to rely on mice or petri dishes alone. In the last three decades, I have collected over 50,000 bone-marrow biopsies, normal control buccal smear cells, and blood, serum, and plasma samples in a well-annotated tissue repository backed by a computerized bank of clinical, pathologic, and morphologic data. Using these samples, we have identified novel genes involved in causing certain types of MDS, as well as sets of genes related to survival, natural history of the disease, and response to therapy. But when I used bone-marrow cells from treated MDS patients to develop a genomic-expression profile that was startlingly predictive of response and applied for an NIH grant to validate the signature, the main criticism was that before confirming it through a prospective trial in humans, I should first reproduce it in mice.

It's time to let go of the mouse models—at least, as surrogates for bringing drugs to the bedside. Remember what Mark Twain said: "What gets us into trouble is not what we don't know; it's what we know for sure that just ain't so."