Research Using Plants Explains Why Animal Models Are Not Predictive


A press release from the University of California - Davis reports the results of a genome wide association study in the plant Arabidopsis thaliana. It appears that complex traits are influenced by thousands of gene and the environment. The article, in the online journal PLoS Biology, is available here. Kliebenstein states: “We showed that both external and internal environments altered the identified genes so significantly that using plant tissues from different developmental stages, or that were treated with the silver nitrate, led to the identification of very different gene sets for particular traits.” According to the press release: “The group noted that the developmental stage of the plant had three times as much influence as the environment on the genes they identified.”

The above is why replacing one gene from a mouse with another from a human is not likely to inform scientists about human disease. In addition, living organisms are complex systems with different levels of organization where initial conditions are very important. Further, there are modifier genes and other environmental factors that affect the system. In part, this is why individual humans respond differently to drugs and disease. There is an excellent article on human variation to drugs and how genomics is informing clinicians in the July 20, 2011 issue of JAMA. I am not sure if you can access it for free, but the url is

Animal modelers replicate one or two of the symptoms of human disease and assume they have replicated the cause of the symptoms. An understanding of complexity science and studies like the above expose the fallacy of this position. Nevertheless, consider the following from Baker writing in Nature 2011: “Researchers evaluating animal models consider three kinds of validity. Construct validity means that a test measures what it claims to. In animal models, that means that whatever causes symptoms in the animal is also what contributes to disease in humans. Such validity is relatively easy to achieve when a condition is caused by a single gene, but most are more complicated.

In reality, even single gene disorders are influenced by background genes and modifier genes. Thein in Haematologica 2005: "As the defective genes for more and more genetic disorders become unravelled, it is clear that patients with apparently identical genotypes can have many different clinical conditions even in simple monogenic disorders. b-thalassemia occurs when there is a deficiency in the synthesis of b-globin chains. The clinical manifestations of b-thalassemia are extremely diverse, spanning a broad spectrum from severe anemia and transfusion-dependency to the asymptomatic state of thalassemia trait. The remarkable phenotypic diversity of the b-thalassemias is prototypical of how a wide spectrum of disease severity can be generated in single gene disorders. The most reliable and predictive factor of disease phenotype is the nature of the mutation at the b-globin locus itself. However, relating phenotype to genotype is complicated by the complex interaction of the environment and other genetic factors at the secondary and tertiary levels, some implicated from family studies, and others, as yet unidentified. This article reviews the clinical and hematologic diversity encountered in b-thalassemia with an overview of the modifier genes that moderate their disease expression."

Agarwal and Moorchung in J Nippon Med Sch 2005: "It is now increasingly apparent that modifier genes have a considerable role to play in phenotypic variations of single-gene disorders. Intrafamilial variations, altered penetrance, and altered severity are now common features of single gene disorders because of the involvement of several genes in the expression of the disease phenotype. Oligogenic disorders occur because of a second gene modifying the action of a dominant gene. It is now certain that cancer occurs due to the action of the environment acting in combination with several genes. Although modifier genes make it impossible to predict phenotype from the genotype and cause considerable difficulties in genetic counseling, they have their uses. In the future, it is hoped that modifier genes will allow us to understand cell and protein interactions and thus allow us to understand the pathogenesis of disease. . . . in recent times it has been seen that the number of diseases that can be explained with the classical Mendelian genetics model is gradually diminishing."

Baker continues: “Mikhail Pletnikov, a neurobiologist at Johns Hopkins University in Baltimore, Maryland, models schizophrenia by combining genes and environmental stressors. For complex disorders, he says, “we’ve passed that period where we manipulate one gene to try to understand the whole disease”.” Yet the number of genetically modified mice continues to grow.

Baker: “Face validity means that a test seems to measure what it needs to, for example that the symptoms in an animal model mirror those in a human. For heart rate or tumour growth, such measures may be straightforward, but for diseases assessed by behaviour, it is considerably more complicated. BTBR mice, a strain used to study autism, avoid interacting with other mice and groom themselves excessively. When BTBR males are exposed to female urine, they do not vocalize and scent-mark as males from other strains do. These traits and others map well onto the diagnostic criteria for autism in humans, which include deficits in interaction and communication, along with repetitive behaviour.” Once again, we see animal modelers ignoring the fact that animals and human are complex systems that cannot be studied by reductionism on some levels of organization or for some traits. Gene expression studies consistently reveal that the human brain is the organ most different from other species. Autism in humans is not what some call autism in animal models.

Baker: “Predictive validity is the extent to which a test predicts a future outcome. In an animal model, the animal should respond to drugs in a way that corresponds to human reactions. For example, antidepressants are sometime evaluated by their effects on the forced-swim test, which measures how long a mouse will try to climb out of a tank of water before giving up. For disorders such as autism, however, there are no effective drugs to serve as positive controls. Even when drugs do exist, the symptoms or mechanisms captured by a single behavioural assay are unlikely to capture everything that is important. . . .” Given our previous books, articles, and blogs, I will assume the reader is familiar with the prediction argument.

Baker: “But whether testing rats or mice, says Crawley [chief of behavioural neuroscience at the US National Institute of Mental Health], researchers must remember that the goal of an animal model is not perfection but utility. “An animal model may not be 100% translatable, but maybe 80% is good enough to test for possible treatments”.”

80% is good enough if what the scientist wants is a heuristic device, or to study basic physiological processes in the species being examined, or to simply seek knowledge for knowledge sake. But it is not enough for predictive models and prediction is how scientists sell their research to society.

Before attempting complicated science using animal models, scientists should be required to pass a test including material from the philosophy of science and the fundamentals of plant genetics. I doubt most current recipients of NIH grants that involve animal models would pass.


Agarwal, S., and N. Moorchung. 2005. Modifier genes and oligogenic disease. J Nippon Med Sch 72 (6):326-34.

Baker, M. 2011. Animal models: inside the minds of mice and men. Nature 475 (7354):123-8.

Thein, S. L. 2005. Genetic modifiers of beta-thalassemia. Haematologica 90 (5):649-60.


Popular Video