Why Are We Still Testing On Animals - Part I
In This Article
- An Imperfect Bridge
- Cancer: How Many Times Can We Cure a Mouse?
- A Model Is Not the Disease
- Alzheimer’s Disease: Treating a Model Is Not Treating a Person
- The Human Cost of Repeated Hope
- Sepsis: When Controlled Experiments Meet Human Complexity
- What Might We Have Found Instead?
- HIV: Valuable Knowledge Without Reliable Prediction
- Why Similarity Is Not Identity
- Animal Research Can Fail in Both Directions
- Failure Is Part of Science
- The Standard Should Be Human Outcomes
- What Failure Costs Animals
- The Alternative Is Not Less Science
When Animal Research Fails to Help People
By John Lieberman, MD, MSL
Summary
Animal research is often described as an imperfect but necessary bridge between the laboratory and the patient. But a bridge should reliably carry us where we need to go.
In cancer, Alzheimer’s disease, sepsis, and HIV, animal models have repeatedly produced encouraging findings that failed to deliver the same benefits in people. These failures do not mean animal research has taught us nothing. They do mean its ability to predict human outcomes has often been overstated.
The cost is measured not only in research dollars, but also in lost time, disappointed patients, delayed treatments, and the lives of animals used in experiments.
What You’ll Learn
- Why curing disease in an animal model may not help a patient
- How laboratory models simplify complex human illnesses
- What cancer research reveals about animal-to-human translation
- Why Alzheimer’s treatments successful in mice often fail in people
- How sepsis exposes the limits of highly controlled experiments
- Why nonhuman primates cannot fully reproduce HIV infection
- How researchers should judge the value of any experimental model
Estimated reading time: 11 minutes
An Imperfect Bridge
Animal testing is often defended as an imperfect but necessary bridge between basic science and human medicine.
That description contains an important truth. No ethical research system should expose people to unknown risks without gathering preliminary evidence.
But a bridge should carry us toward the intended destination. When a model repeatedly leads researchers away from successful human treatment, its role must be reconsidered.
The problem begins when useful knowledge is confused with reliable prediction. A model may help researchers understand a biological mechanism without reliably predicting whether a treatment will help human patients.
Cancer: How Many Times Can We Cure a Mouse?
Cancer is not one disease. It is a large group of diseases shaped by genetics, aging, environmental exposures, immune function, metabolism, inflammation, and chance.
Even two people with tumors in the same organ may have cancers driven by different biological mechanisms.
Laboratory animals can help researchers isolate part of that complexity. Scientists may implant human tumor cells into a mouse, alter a gene so the animal develops a particular kind of tumor, expose the animal to a cancer-causing substance, or use animals with weakened immune systems so foreign cells are not rejected.
These models can answer focused questions. They can help researchers study tumor growth, test a biological hypothesis, or compare treatments under controlled conditions.
But the control that makes an experiment easier to interpret can also make it less like human disease.
A tumor implanted in a young mouse is not the same as a cancer that developed over decades in an older person. A mouse bred to have a nearly uniform genetic background does not represent the diversity of human patients. An immunodeficient animal cannot reproduce every interaction between a tumor and an intact human immune system.
A laboratory tumor may grow quickly and predictably. Human cancers evolve. They become resistant. They interact with other illnesses and medications. They may change after every round of treatment.
The late Richard Klausner, who served as director of the National Cancer Institute, was widely quoted as observing that researchers had cured cancer in mice for decades without achieving the same success in humans.
The point was not that mouse studies had produced no knowledge. The point was that experimental success in a mouse should never be presented as though it were success in a patient.
A Model Is Not the Disease
Scientists use models because they cannot reproduce every part of reality. A map leaves out details. A computer simulation simplifies physical conditions. A laboratory model isolates certain biological features.
The value of a model depends on whether the features it preserves are the ones that matter.
This becomes especially difficult in medicine. Diseases do not exist independently of the bodies experiencing them.
A person with cancer may also have heart disease, diabetes, kidney problems, chronic inflammation, or a weakened immune system. A person’s age, sex, genetics, ancestry, previous treatments, diet, microbiome, and environment may all affect the response to a drug.
Laboratory animals are often young, relatively healthy, genetically similar, and raised under controlled conditions. Those features reduce experimental variability. They also distance the model from the patients medicine must serve.
The question is not whether animal models resemble human disease in some way. Most do.
The question is whether they reproduce the biology needed to predict the human outcome being studied.
Alzheimer’s Disease: Treating a Model Is Not Treating a Person
Alzheimer’s disease offers one of the clearest examples of the difference between modeling a feature and reproducing a disease.
Researchers can genetically alter mice to produce amyloid plaques, tau abnormalities, memory problems, or other characteristics associated with Alzheimer’s.
Experimental treatments may reduce plaques or improve performance on memory-related tasks. The results can appear highly promising.
Then the treatment reaches people and fails to produce the expected benefit.
The review Why Study Mice in Alzheimer’s Disease? explains both the value and limitations of mouse models. No animal model reproduces the full human disease.
Human Alzheimer’s disease develops over many years. It is influenced by aging, blood vessels, inflammation, metabolism, genetics, immunity, and other factors researchers are still working to understand.
A genetically engineered mouse may reproduce selected features. It does not experience decades of human aging. It does not share the complete human brain structure, lifespan, vascular system, immune response, or life history.
Reducing amyloid in a mouse is therefore not the same as preserving memory, personality, independence, and recognition in a human being.
Recent Alzheimer’s medicines represent meaningful scientific progress for some patients with early disease. But slowing decline is not the same as restoring what the disease has taken.
Families still need treatments that prevent Alzheimer’s, stop it, or reverse its damage. Every model should be judged by how well it helps achieve those human goals.
The Human Cost of Repeated Hope
Laboratory breakthroughs do not remain in laboratories. They become press releases. They become headlines. They become stories shared by families desperate for hope.
A spouse caring for a husband with Alzheimer’s may read that scientists reversed memory loss in mice. An adult child may believe a cure is close. Years later, the treatment may fail in clinical trials or never reach them.
Scientific findings should be communicated. But researchers, universities, journalists, and advocates must distinguish early animal results from evidence that a treatment helps people.
Overstatement carries a cost. It can distort public understanding, affect donations and policy, and create cycles of hope and disappointment.
Most importantly, it can hide the limitations of the system producing those results.
Sepsis: When Controlled Experiments Meet Human Complexity
Sepsis occurs when the body’s response to infection becomes dangerously dysregulated. It can cause tissue damage, organ failure, and death.
Sepsis is not a single, uniform condition. It can begin with different infections, affect patients of very different ages and health histories, and progress differently from one person to another.
A laboratory model cannot easily reproduce that diversity.
Researchers may induce a standardized infection or inflammatory response in animals of similar age and genetic background. The experiment may begin at a known time, and treatment may be administered according to a precise schedule.
Those controls help isolate a biological mechanism. They also create a situation unlike the emergency department or intensive care unit.
Human patients do not arrive as standardized subjects.
The review Rethinking Animal Models of Sepsis: Working Towards Improved Clinical Translation Whilst Integrating the 3Rs describes the challenge of translating results from controlled animal models into the complex reality of human sepsis.
Many treatments that appeared promising in animal models failed to improve human outcomes.
This does not prove the underlying research was worthless. It shows that a model can successfully answer a narrow biological question while failing to predict whether an intervention will save patients.
What Might We Have Found Instead?
No one can know what discoveries would have emerged if funding had followed a different path. We cannot rerun decades of history.
But we can ask whether greater investment in human biology might have produced better results.
- What if researchers had earlier access to sophisticated human immune-cell systems?
- What if patient samples, organoids, clinical data, and computational models had received the same institutional support as animal facilities?
- What if sepsis research had focused more heavily on identifying distinct patient subgroups rather than treating sepsis as a single condition?
- What if researchers had been rewarded for showing that a familiar model did not translate?
These questions are not attacks on past scientists. Researchers used the tools available to them.
The questions are about the future. Once better tools become available, continuing the old allocation of funding is a choice.
HIV: Valuable Knowledge Without Reliable Prediction
Nonhuman primates have played an important role in HIV research. They allow scientists to study immune responses, viral transmission, tissues, and disease processes that would be difficult or unethical to examine in people.
That work has contributed to scientific knowledge.
But nonhuman primates do not naturally reproduce every feature of human HIV infection.
Scientific contribution and predictive reliability are not the same thing.
Researchers often study simian immunodeficiency virus or engineered combinations of simian and human viruses in macaques. These models can reproduce important aspects of infection and help researchers examine immune mechanisms and test ideas.
They cannot fully reproduce the interaction between HIV and the human immune system.
A review titled Non-Human Primate Models for AIDS Vaccine Research described the strengths and limitations of these models, including the fact that no single primate model captures the entire human disease.
Decades of HIV research have produced extraordinary advances. Antiretroviral treatment can suppress the virus and allow many people with HIV to live long lives. Pre-exposure prophylaxis can dramatically reduce the risk of acquiring HIV. Treatment that maintains an undetectable viral load prevents sexual transmission.
Yet a broadly protective, approved preventive HIV vaccine has remained elusive.
This does not mean every primate experiment failed. It means a model may deepen understanding without telling us whether a vaccine will protect people.
Why Similarity Is Not Identity
Humans share many biological features with other animals. That is why animal research can produce useful knowledge.
But shared features do not eliminate important differences.
Two species may share a receptor while regulating it differently. They may possess similar organs while metabolizing a drug differently. They may mount immune responses that appear similar but lead to different outcomes.
Small biological differences can have enormous consequences.
A treatment’s success may depend on when a gene turns on, how long a protein remains active, which cells are present, or how several organs interact.
The closer the scientific question gets to a specific human outcome, the more human relevance matters.
Animal Research Can Fail in Both Directions
Much of the debate focuses on false hope: a treatment looks safe or effective in animals and then fails in people.
But animal testing can also produce false warnings.
A useful drug may appear toxic in a species that responds differently from humans. A compound may be abandoned before researchers fully examine its human relevance.
This failure is harder to measure. When a drug fails in a clinical trial, researchers can document the result. When a compound is abandoned because of an animal finding, no human trial may ever occur.
We may never know whether the warning was relevant.
That uncertainty weakens the argument that animal testing is always the cautious choice. A method can harm patients by failing to identify danger. It can also harm patients by preventing a useful treatment from moving forward.
Failure Is Part of Science
Biomedical research will never eliminate failure. That should not be the goal. A system that never fails would probably be a system unwilling to test bold ideas.
The goal is to fail intelligently.
Researchers should learn which models predict human outcomes. They should abandon those that repeatedly do not. They should distinguish a model that explains a mechanism from one that predicts treatment success.
They should publish negative findings, compare animal results with later human results, and build databases that reveal where translation succeeds and where it breaks down.
Most importantly, no method should be protected from scrutiny because it is traditional.
The Standard Should Be Human Outcomes
New technologies should not be accepted merely because they appear innovative. Organoids can be poorly designed. Computer models can contain flawed assumptions. Artificial intelligence can reproduce bias. Human cell systems can lack important whole-body interactions.
Every method has limitations.
But animal tests must be held to the same standard.
Too often, a new human-based method is evaluated by asking whether it reproduces the animal result. That comparison may be useful, but it should not be the final measure.
An animal result is not automatically the truth.
The final benchmark should be what happens in humans.
- Does the method predict human toxicity?
- Does it predict metabolism?
- Does it identify a dose that helps patients?
- Does it distinguish people likely to benefit from those likely to be harmed?
- Does it improve clinical success?
A method that performs better against human outcomes should receive priority, whether it is old or new.
What Failure Costs Animals
When a model repeatedly fails, the cost is not only scientific.
Animals may be bred specifically for experiments. Some may be genetically altered to develop disease. Some may undergo surgery, infection, poisoning, restraint, isolation, repeated blood draws, invasive testing, or euthanasia.
The ethical defense for those experiences is usually based on expected human benefit. That benefit cannot simply be assumed.
The more severe the experiment, the stronger the scientific justification should be. The more uncertain the model, the harder it becomes to justify the harm.
Respect for animals requires more than reducing pain during an experiment. It requires questioning whether the experiment should happen at all.
The Alternative Is Not Less Science
For much of the last century, the limitations of animal models presented an agonizing problem. Researchers knew that animals were not humans, but they lacked tools capable of reproducing important features of human biology.
That is no longer entirely true.
Scientists can now grow human organoids. They can place living human cells in microengineered devices. They can study patient-specific tissue. They can analyze real-world clinical data. They can build computational models of metabolism and toxicity. They can use artificial intelligence to identify patterns across enormous datasets.
None of these tools can reproduce an entire human being by itself.
Neither can an animal.
The future will not depend on finding one perfect replacement. It will depend on combining several human-relevant methods to answer specific questions.
The alternative to animal testing is not the absence of testing.
It is better testing.
That is the subject of Part 3.
Continue the Series
Part 3: The Future of Medical Research Does Not Have to Depend on Animals examines human-based technologies and the policy changes needed to accelerate their adoption.
References
- Drummond E, Wisniewski T. Why Study Mice in Alzheimer’s Disease? Journal of Internal Medicine. 2017;281(5):503–515.
- Nandi M, Jackson SK, Macrae D, et al. Rethinking Animal Models of Sepsis: Working Towards Improved Clinical Translation Whilst Integrating the 3Rs. Clinical Science. 2020;134(13):1715–1734.
- Hu SL. Non-Human Primate Models for AIDS Vaccine Research. Current Drug Targets: Infectious Disorders. 2005;5(2):193–201.
- Ineichen BV, Furrer E, Grüninger SL, Zürrer WE, Macleod MR. Analysis of Animal-to-Human Translation Shows That Only 5% of Animal-Tested Therapeutic Interventions Obtain Regulatory Approval for Human Applications. PLOS Biology. 2024;22(6):e3002667.
