Perspectives On Medical Research
Volume 4, 1993
Scientific Problems with Animal Models
Stephen R. Kaufman
Historian and philosopher of science Thomas Kuhn notes that every scientific age has its "paradigms," theories nearly universally regarded as true that form the framework for ongoing scientific investigations.(1) Paradigms are rarely challenged until overwhelming contradictory evidence forces their revision or rejection. A currently dominant paradigm is that animal "models" are necessary for medical progress.(2-4) For most members of the current scientific establishment, the issue is not whether animal models should be used but which models are most useful. However, critics of animal models argue that they are inherently flawed (5-7) and point out the frequency with which animal models provide misleading information.(8-12)
All species differ; animal-model conditions never exactly mimic human ones. Animal models are only analogues of human conditions because they share certain characteristics. Philosophers Hugh LaFollette and Niall Shanks observe that animal models are used primarily for two functions--to predict human responses to stimuli (such as infectious, traumatic, or toxic conditions and therapeutic drugs or devices) and to offer new ways of conceptualizing human anatomy, physiology, or pathology. Researchers who use animal models employ the following reasoning: a given animal model resembles an analogous human condition in some of its features (say, A, B, and C); therefore, it is reasonable to proceed as if an additional feature (D) found in the animal model--for example, a physiological function or a drug response--can be expected to be a feature of the human condition as well. As LaFollette and Shanks point out, this assertion is logical only if feature D is causally related to A, B, and C--in both the animal model and the human condition. That is, A, B, and C must be causal factors of feature D. The following reasoning illustrates a failure to recognize the importance of causal relationships:
Two dogs bark, love bones, and wag their tafls when their human companions arrive home; because the two dogs are similar in these respects, they can also be expected to be of the same breed. If we know the first dog's breed, we can reliably predict the second's.
Breed, however, is not causally related to the three features that the two dogs are already known to share. If we know a dog's breed and we also know that a second dog has the same parents as the first, then we can reliably predict the second dog's breed--even if the two dogs differ in many other respects, such as coat color or temperament.
LaFollette and Shanks distinguish between weak and strong models. Strong animal models are identical to the analogous human features in all causally relevant respects, and research using such models can be confidently applied to humans. Although many animal research advocates assert that animal models faithfully reproduce human conditions, LaFollette and Shanks argue that most animal models are weak models of little direct applicability to humans. Neverthe- less, LaFollette and Shanks do not reject animal research's value. They maintain that animal models may be helpful but are probably not necessary for medical progress.(5)
In public, animal research proponents often suggest that weak causal models are in fact strong. For example, the Stanford Committee on Ethics states, "Cancer kills humans and animals alike.(13) At any buf the most simplistic level, the comparison immediately begins to break down. For example, malignancies that are experimentally induced in nonhuman animals and malignancies that occur spontaneously in humans significantly differ in their causes.(14-16) Other important differences include the greater virulence of most experimental cancer strains and differing mechanisms of tumor growth and metastasis. Even nonhuman cancers that apparently share many characteristics with human cancers make unreliable research models, since human and nonhuman cancers inevitably differ in some relevant causal factors. Viewed in this light, an animal model such as the mouse-leukemia model is a poor means of attempting to identify potential anti-cancer drugs, and this model has, in fact, proved grossly inadequate.(17)
Even if a disease's main causal factors were well understood--and were alike--in both humans and other animals, animal models would still be undermined by systemic differences between animal models and human conditions. Because of evolutionary divergence, species show differences in virtually every aspect of organ and tissue function. All organ subsystems interact, so every physiological difference between a given "laboratory"-animal species and the human species necessarily affect every causal factor. Consequently, all tissues of an animal model will tend to react to an experimental manipulation differently from a supposedly analogous human condition. Animal models of human conditions tend to provide only the most obvious and general information, such as that cancers kill; in order for them to provide reliable and specific information, the model and the human condition must have identical causal factors and have no significant systemic differences that affect these causal factors. This is impossible, since there are always differences in causal factors between the model and the human condition and because systemic differences are an inevitable consequence of evolutionary divergence.
In theory, then, animal modeling is unreliable in predicting human responses to stimuli; and it has proved so in practice. Animal tests of acute lethal toxicity,(18) eye irritancy,(19-21) skin irritancy,(22-24) teratogenesis (birth defects),(25-27) and carcinogenesis (28,29) have generally provided inconsistent results and failed to correspond to human experience. R. Heywood has estimated that only about 5-25 % of toxic effects found in animal experiments occur in humans.(30) Of course, animal models can serve as strong models when researchers attempt to predict gross toxicological effects, such as the ability of strong acids to burn the eye's surface; however, such effects could readily be predicted from the most rudimentary knowledge of chemistry. Most animal tests are supposedly intended to identify subtle effects, and they perform poorly in this regard.
Animal tests have also proved inadequate as a means of identifying potentially useful drugs. U.S. law requires that drugs be found effective and safe in animal testing before they are tested on humans. This law fails to reflect animal tests' poor predictive value: Ronald Hansen found that only about 12% of drugs that passed Phase I animal tests and entered human testing reached the market;(31) earlier, Samuel Irwin had found that only 2.3% of drugs selected for clinical trial were eventually marketed.(32) Most new drugs are similar to existing drugs, and so their clinical effect can be at least partially predicted based on structural analogy. Also, modern biochemical methods can help characterize specific drug-receptor interactions, and these interactions can suggest specific drug effects. Therefore, it is debatable whether animal tests help identify which drugs are most suitable for human clinical trials (the critical step in determining human safety and efficacy).
In addition to having failed to accurately predict drugs' efficacy and toxic side-effects, animal tests have, no doubt, prompted researchers to abandon numerous drugs and therapies that proved ineffective or toxic in nonhuman animals but would have benefitted humans. It is impossible to determine how many valuable therapies were discarded on the basis of misleading animal studies.
Are animal models worthless, then? Although causal dissimilarities and systemic differences undermine animal models, they are not necessarily useless. For example, animal data need not accord perfectly with human data to be relevant. For example an animal test that correctly identified carcinogens 90% of the time could help formulate reasonable public health guidelines. However, as noted above, most animal tests do not accurately identify subtle toxic effects. Therefore, animal toxicity data may be valuable in theory, but in practice it is generally inconsistent and misleading.
Although most animal models are weak models, certain strong ones can reliably predict gross toxicological effects. For example canaries were once used to test for carbon monoxide in coal mines because canaries are much more sensitive to this toxic gas than humans are. Although animal models cannot reliably elucidate mechanisms of disease induction and spread in humans, they have, in the past, afforded strong models for research on the organisms themselves. To illustrate, rats infected with the syphilis spirochete yield little insight into human syphilis infection. Nevertheless, Erhlich discovered arsenobenzol as a treatment for syphilis by infecting rats with the spirochete and then trying different compounds for possible anti-syphilis effect. In Ehrlich's studies, rats served primarily as reservoirs to harbor the organism, facilitating research on the organism itself. Today, in vitro cultures have replaced animals as mere reservoirs for almost all infectious agents.
Also, animal models may provide information about the species under investigation, because there are generally few major differences in physiological parameters among individuals of the same species. Most animal experimenters, however, claim to address human health issues.
Many philosophers of science have distinguished between validating (or disproving) hypotheses and formulating them. (33-35) An animal model cannot be used to test a hypothesis about humans because differences in causal factors between the animal model and the human condition render the animal model invalid as a predictor. The only way to support or disprove a hypothesis about human anatomy, physiology, or pathology is by studying human beings. Animal-model conditions are analogues, and it is impossible to validate or disprove any hypothesis by analogy. Therefore, logically animal models cannot directly contribute to medical discovery. Medical historian Brandon Reines maintains that animal models primarily "dramatize" hypotheses about humans without actually validating or disproving them.
Although animal models cannot validate or disprove hypotheses, they may function as heuristic devices that assist the process of discovery.(5-7) That is, they may suggest different ways of conceptualizing problems and thereby help generate new hypotheses. In this regard weak models have potential value. An unexpected finding during animal experimentation (including experimentation that was poorly conducted or that failed to accomplish its original objectives) may lead to an insight.
Such insights, however, can also arise via other research approaches, such as observing human patients, conducting epidemiological studies, performing in vitro tests, or engaging in computer or mechanical modeling. Once again, then, animal models do not appear to be necessary for medical progress. In fact, medical historian Brandon Reines (36,37) and physician Paul Beeson (38) consider the role of animal models as heuristic aids very limited.
In a review of hepatitis research, Beeson writes: "progress in the understanding and management of human disease must begin, and end, with studies of man."(38) Although much hepatitis research has used animals, Beeson has found that hypotheses about hepatitis have derived from clinical observations, and that clinical studies have been necessary to test their validity.(38)
Reines observes that nearly all hypotheses about human conditions derive from human clinical research.(36,37,39) Animal experimenters, he contends, perform the, superfluous and irrelevant function of experimenting with different animal models until they find one that accords with the clinical findings; typically they then claim that their model has "validated" the clinically derived hypothesis. Often, Reines observes, animal modelers highlight confirmatory animal data while discounting animal data that contradict their findings.
Although Beeson doesn't share Reines' conclusion that animal experimenta- tion is largely irrelevant to medical discovery, he agrees that most insights derive from human studies. Beeson writes, "The initial observations of manifestations and courses of human disease must be made in human beings. The important contributions of epidemiology depend on accurate clinical definitions. The occurrence of rare sequels or late manifestations is beyond any feasible approach through experiments on other species."(38) Beeson cites progress in understanding hepatitis, appendicitis, rheumatic fever, typhoid fever, ulcerative colitis, and hyperparathyroidism as representative of most medical progress in having occurred almost exclusively through the study of humans.(38) Nevertheless, like other researchers who have acknowledged the primary importance of clinical investigation yet remain lodged in the animal-model paradigm, Beeson continues to maintain the importance of animal experimentation.(40-45)
The history of polio research illustrates many of animal experimentation's strengths and limitations. Proponents of animal research frequently claim that animal experiments were crucial in controlling polio.(4,46) John R. Paul's review of polio research indicates that animal experimentation facilitated some insights but delayed others.(47)
In the 1800s, polio's clinical presentation and natural history were deduced from bedside observation and postmortem studies of human victims. Ivar Wickman's detailed epidemiological analyses of two Swedish epidemics in the early 1900s revealed that mild or even subclinical cases contributed to contagious spread of the disease. Most investigators, focusing on polio's life-threatening paralysis, considered polio a central nervous system disease. But Wickman found that polio affects the alimentary tract (throat, stomach, and intestines) and suggested that the gastrointestinal system may be the initial site of infection. (48,49)
By contradicting Wickman's observations, animal data delayed understanding of polio's true pathogenesis and natural history. The first animal model of polio was developed by Simon Flexner, who induced polio-like paralysis in rhesus monkeys after placing infected human tissue into their noses. Convinced that his animal model precisely paralleled the human disease, he concluded that human polio was introduced to the brain via the nose and confined to the central nervous system. For decades, most scientists adhered to this erroneous theory, and this led to misguided therapeutic measures.(47)
While animal studies remained the principal focus of polio research in the United States, Swedish clinical investigators continued to make important contributions. They tested for the presence of polio in tissues of human polio victims and family members by inoculating monkeys with test samples. If a test monkey contracted polio, the sample was determined to be infected. The investigators found that polio carriers could have polio virus present in their throats and intestines up to seven months after exposure. (Here, monkeys were used as bioassays, in which researchers sought a gross effect. Today, few animals are used as bioassays, because more reliable non-animal bioassays exist.)
Meanwhile, Flexner and other animal researchers continued to study rhesus monkeys infected with viruses obtained from other rhesus monkeys. This process selected for more virulent polio strains that tended to infect nervous tissue. Consequently, the animal model increasingly diverged from human polio in pathogenesis and natural history. Systemic differences between humans and rhesus monkeys undermined Flexner's animal model as a causal model of human polio.
In the 1940s researchers found that polio infection in chimpanzees accords more closely to the human disease. Like humans and unlike rhesus monkeys, chimpanzees were found to harbor the polio virus in their alimentary, tracts. Researchers were now more willing to accept the clinically derived hypothesis that polio infects the human alimentary tract. But this response merely demonstrates the research establishment's reluctance to accept clinical findings in humans until parallel findings have been produced--however artificially--in the laboratory in another species.
Animal models of polio were not very helpful as causal models, and they significantly delayed development of an effective vaccine. After clinical studies showed that polio virus infects gastrointestinal tissue, decades of experimentation on rhesus monkeys suggested that the virus infects only neural tissue. Vaccine researchers mistakenly believed that polio would only grow in neural tissue, but vaccines derived from these cultures were too dangerous. In 1948, John Enders, Thomas Weller, and Frederick Robbins grew polio on human intestinal tissue, which led to a safe vaccine. Albert Sabin, who developed the Sabin oral polio vaccine, has written, "the work on prevention was long delayed by an erroneous conception of the nature of the human disease based on misleading experimental models of the disease in monkeys."(50)
Nevertheless, animal experiments may have served a heuristic function by inspiring new ways of thinking about polio. For example, studies of TO virus encephalitis in mice revealed that animals infected early in life tend to have a more benign course: after initial exposure to TO virus, mice become immune.(51) According to Paul, this may have helped researchers derive a theory to explain the clinical observation that major epidemics tended to occur in remote areas. Because sparse or isolated populations did not permit an endemic state of polio infection, few children were exposed earlier in life, when the disease tended to have a more benign course. This insight did not require animal studies; it could have been derived entirely from clinical investigations, including population studies. The TO virus model illustrates the limited utility of weak models.
Some strong models were also used in the fight against polio. Swedish investigators used monkeys to test for the presence of polio virus in tissue samples; later, researchers used the mouse neutralization test for similar purposes. In addition, monkeys were used in immunological studies that demonstrated multiple distinct viral strains. In these cases, however, researchers merely assessed whether or not the animals became infected under different conditions. Today the absence or presence of a virus in human tissue can be more reliably determined using in vitro methods.
In theory and practice, animal models generally fail to reliably predict human responses to stimuli. While some strong animal models exist, most are weak models of human responses to stimuli. Weak animal models may serve as heuristic devices and help to inspire new ways of conceptualizing clinically relevant issues, but they are not indispensable analogues that are directly applicable to humans. As merely heuristic devices, animal models are not necessary for progress in human medicine.
References
1. Kuhn T. The Structure of Scientific Revolutions. Chicago, University of Chicago Press, 1969.
2. Bernard C. An Introduction to the Study of Experimental Medicine. Paris, Henry Schuman, 1949.
3. American Medical Association. Use ofanimals in Biomedical Research: The Challenge and Response [White Paper]. Chicago, AMA, 1988.
4. Gay WI (ed). Health Benefits of Animal Research. Washington DC, Foundation for Biomedical Research, 1986.
5. LaFollette H, Shanks N. Animal models in biomedical research: Some epistemological worries. Public Affairs Quarterly 1992;7(2):113-130.
6. LaFollette H, Shanks N. The intact systems argument: Problems with the standard defense of animal experimentation. Southern Journal of Philosophy 1993;31:323-333.
7. LaFollette H, Shanks N. Animal modelling in psychopharmacological contexts. Behav Br Sci, Dec. 1993, in press.
8. Peller S. Quantitative Research in Human Biology and Medicine. Bristol, John Wright & Sons, 1981.
9. Rowan AN. Of Mice, Models, & Men: A Critical Evaluation of Animal Research. Albany, SUNY Press, 1984.
10. Sharpe R. The Cruel Deception. Wellingborough, Thorsons Publishers Limited, 1988.
11. Bross ID. Scientific Fraud vs. Scientific Truth. Buffalo, Biomedical Metatechnology Press, 1991.
12. Wiebers DO, Adams HP, Whisnant JP. Animal models of stroke: Are they relevant to human disease? Stroke 1990;21:1-3.
13. Thomas JA, Hamm TE, Perkins PL, Raffin TA, the Stanford University Medical Center Committee on Ethics. Animal research at Stanford University: Principles, policies, and practices. N Engl J Med 1988;318:1630-1632.
14. Kaufman SR, Reines BP, Casele H, Lawson L, Lurie J. Model No. 42: Adenocarcinoma (Rat). Perspec An Res 1989; 1 (suppl): 15-26.
15. Bross ID. Crimes of Official Science: A Casebook. Buffalo, Biomedical Metatechnology Press, 1987.
16. Reines BP. Cancer Research on Animals: Impact and Alternatives. Chicago, National Anti-Vivisection Society, 1986.
17. Pihl A. UICC Study Group on chemosensitivity testing of human tumors. Problems-- applications--future prospects. Int J Canc 1986;37:1-5.
18. Zbinden G, Flury-Roversi M. Significance of the LD50 test for toxicological evaluation of chemical substances. Arch Toxicol 1981;47:77-99.
19. Weil CS, Scala RA. Study of intra- and interlaboratory variability in the results of rabbit eye and skin irritancy tests. Toxicol Appl Pharmacol 1971;19:276-360.
20. Swanson DW. Eye initancy testing, in Balls M, Riddell RJ, Worden AN (eds). Animals and Alternatives in Toxicity Testing. New York, Academic Press, 1983, pp 337-367.
21. Freeberg FE, Hooker DT, Griffith JF. Correlation of animal eye test data with human experience for household products: an update. J Toxicol-Cutaneous Ocular Toxicol 1986;5:115-123.
22. Davies RE, Harper KH, Kynoch SR. Inter-species variation in dermal reactivity. J Soc Cosmetic Chemists 1972;23:371-381.
23. Kligman AM. Assessment of mild irritants, in Frost P, Horwitz SN (eds). Principles of Cosmetics for the Dermatologist St. Louis, Mosby, 1982, pp 265-273.
24. Phillips L. A comparison of rabbit and human skin response to certain irritants. Toxicol Appl Pharmacol 1972;21:369-382.
25. Lasagna L. Regulatory agencies, drugs, and the pregnant patient, in Stern L (ed). Drug Use in Pregnancy. ADES Health, Science Press, 1984.
26. Lewis PJ. Animal tests for teratogenicity, their relevance to clinical practice, in Hawkins DF (ed). Drugs and Pregnancy: Human Teratogenesis and Related Problems. Edinburgh, Churchill Livingston, 1983, pp 17-21.
27. Smithells RW. Drug teratogenesis, in Inman WH (ed). Monitoring for Drug Safety. Philadelphia, JB Lippincott, 1980.
28. Salsburg D. The lifetime feeding study in mice and rats--an examination of its validity as a bioassay for human carcinogens. Fundament Appl Toxicol 1983;3:63-67.
29. Freedman DA, Zeisel H. From mouse to man: The quantitative assessment of cancer risks. Stat Sci 1988;3:3-28.
30. Heywood R. Clinical toxicity--could it have been predicted? Post-market experience, in Lumley CE, Walker SR (eds). Animal Toxicity Studies: Their Relevance for Man. Lancaster, Quay Pub, 1989.
31. Hansen RW. The pharmaceutical development process: Estimates of development costs and times and the effects of proposed regulatory changes, in Chien R (ed). Issues in Pharmaceutical Economics. DC Health, 1979:151-181.
32. Irwin S. Drug screening and evaluative procedures. Science 1962;136:123-128.
33. Hempel CG. Philosophy of Natural Science. Englewood Cliffs, NJ, Prentice-Hall, 1966.
34. Popper KR. The Logic of Scientific Discovery. New York, Harper, 1968.
35. Root-Bernstein RS. Discovering. Cambridge, Harvard Univ Pr, 1989.
36. Reines BP. On the locus of medical discovery. J Medicine Philosophy 1991;16:183-209.
37. Reines BP. Psychopharmacologic discovery: The relative contributions of clinical and laboratory studies. Perspec Med Res 1990;2:13-26.
38. Beeson PB. The growth of knowledge about a disease: Hepatitis. Am J Med 1979;67: 366-370.
39. Reines BP. On the role of anomaly in Harvey's discovery of the mechanism of the pulse. Perspec Biol Med 1990;34:128-133.
40. Beeson PB. Animals in research: Context of a quote [Letter]. Science 1989;245:1437.
41. Good RA. Runestones in immunology. J Immunol 1976; 117:1413-1428.
42. Good RA. Keystones. J Clin Invest 1968;47:1466-1471.
43. Good RA. The value of animal research [letter]. Science 1991.
44. Sitaram N, Gershon S. Animal models to clinical testing--promises and pitfalls. Prog Neuro-psychopharmacol Biol Psychiat 1983;7:227-228.
45. Sitaram N, Gershon S. Gershon ES. Use of animals in research [letter]. Am J Psychiat 1988; 145:653.
46. Committee on the Use of Laboratory Animals in Biomedical and Behavioral Research. Use of Laboratory Animals in Biomedical and Behavioral Research. Washington, DC, National Academy Press, 1988.
47. Paul JR. History of Poliomyelitis. New Haven, Yale University Press, 1971.
48. Wickman I. Studien über Poliomyelitis acuta; zugleich ein Beitrag zur Kenntnis der Myelitis acuta. Berlin, Karger, 1905.
49. Wickman I. Beiträge zur Kenntnis der Heine-Medinschen Krankheit. Berlin, Karger, 1907.
50. Sabin AB. Statement of Albert B. Sabin, M.D. [before the Subcommittee on Hospitals and Health Care of the Committee on Veterans' Affairs, House of Representatives]. Serial no. 98-48, April 26, 1984.
51. Von Magnus H. Studies on mouse encephalitis virus (TO strain). Acta Pathologica 1950;27:605-624, 1951;28:234-257.