Nestler and Hyman present a sobering but valuable addition to the literature on developing animal models of neuropsychiatric disorders. Quite rightly, they highlight the need for developing such animal models because there are limitations on investigating specific functions of the living human brain. Although the authors cover a broad range of possibilities given the title, they focus on the challenges ahead, astutely warning of complacency when attempting to develop or use animal models.
While warnings of the difficulty of creating animal models based on DSM-IV criteria are not unknown, reminding researchers remains useful. The key to this review was, however, to put animal modeling into context. Nestler and Hyman call for researchers from all areas to consider whether they are working/trying to publish a good model of disease or a tool to investigate the neurobiology of behavior. As I have pointed out to many who have asked, I have used α7 nicotinic acetylcholine receptor knockout (KO) mice as a tool to understand the contribution of this receptor to behavior, and while I am interested in behaviors that relate to schizophrenia, I do not argue they are an animal model of schizophrenia. The same can be said, of course, for DISC1 KO mice, for unless evidence turns up suggesting that polymorphisms in this gene result in a complete loss of function of this gene (that is highly penetrant), such mice are a tool to understand the role of DISC1 in behavior.
Using these tools is important to understand the neurobiological underpinnings of these behaviors. Behaviors such as hallucinations, obsessive behaviors, and grandiosity are more difficult to model in animals. Quantifiable behaviors such as forms of cognition are more readily testable, however. Indeed, the authors state that assessing cognition in these models in schizophrenia is a promising target. I would suggest this is also true for bipolar disorder, given the link between cognition and functional outcome in this disease as well (Green, 2006). When assessing cognition in models of schizophrenia and bipolar disorder, we have recently provided an extensive review on the tools available to do so (Young et al., 2009). One of the clearest needs has still been for clinicians to better characterize the cognitive dysfunction in patients and how they may relate to genetic underpinnings. Thus, animal models can be more specifically targeted with greater construct validity.
Another point made by Nestler and Hyman is for authors to make reference to face, predictive, and construct validities wherever possible. These different forms of validity are not novel, but as the authors point out, there are still contentions as to their meaning. Much effort went into describing examples of construct validity, unsurprisingly given the importance of this aspect of validity when developing novel treatments that treat the disorder as opposed to symptoms only. Predictive and face validity were given far less prominence, though it perhaps should have been noticed that face validity is no doubt the weakest of all validities, yet the one most reported when describing a model's validity for the disease. An example given is of excessive grooming in rodents that can result in self-injury being used as a model of obsessive-compulsive disorder. Nestler and Hyman rightly point out that while this model has face validity, the hypothesis remains an intellectual leap with related cognitive and emotional context undetermined.
The strength of this review—and its focus—lies in Nestler and Hyman’s identification that a genetic model based on a gene with limited penetrance is still not a model with sufficient construct validity. This is because if the gene is low penetrant for the disease, then that gene would not cause the extent of neurobiological changes that occur in the disease.
Thus, I applaud the authors' call for sober discussion on the merits and demerits on the validity of the animal model from researchers wanting to publish their findings. It would have been worth adding one final note of caution on the nature of the beast we serve, however. We all wish to publish in high-impact journals, to capture the reader’s imagination about the merits of our work, to increase the impact of our work, and provide relevance. For publishing in some journals, these topics are held to be essential. I would therefore go beyond the recommendation of Nestler and Hyman and say that it is down to all of us, not only authors upon submission, but editors and reviewers also, to critically evaluate the claims made of an animal model described, or evaluate a particular cognitive function, and publish only if these claims are substantiated.
Green, M. F., 2006. Cognitive impairment and functional outcome in schizophrenia and bipolar disorder. J Clin Psychiatry 67 Suppl 9, 3-8; discussion 36-42. Abstract
Young, J. W., Powell, S. B., Risbrough, V., Marston, H. M., Geyer, M. A., 2009. Using the MATRICS to guide development of a preclinical cognitive test battery for research in schizophrenia. Pharmacol Ther 122, 150-202. Abstract
View all comments by Jared Young
This is a welcome review article on an important topic. The key point being addressed is that the lack of improved therapy for psychiatric illness is due to the lack of carefully validated animal models. As described by the authors, testing a drug in the appropriate animal model is a critical stage in the development of new therapies for illnesses, particularly those defined by altered emotional states and marked behavioral changes, as described by the DSM (current version IV).
The review describes the challenges and limitations faced by scientists in this field, such as the difficulty identifying the pathology of human brain disorders. Importantly, the review focuses on the necessity for development of better animal models for identifying improved therapies, and discusses limitations of the current models. These include incomplete validation and poor predictive validity, both of which present considerable difficulties when looking for novel targets and treatments. However, there are two key reasons for the limitations of the current models that are not fully explored by this review.
The first concerns the historical issue of how animal models have been developed from the serendipitous discovery of efficacy in patients with drugs such as diazepam and chlorpromazine. Following their discovery, the pharmacology of these agents was identified and the animal models derived from a reverse translational approach. Thus, chlorpromazine was subsequently found to have high affinity for the dopamine D2 receptor and good efficacy for alleviation of psychosis. Animal models for psychosis then evolved around enhanced mesolimbic dopamine activity, such as amphetamine-induced hyperactivity. Clearly, such a model will only detect D2 antagonists and not therapeutic agents with a novel mechanism of action. Adherence to such a conservative strategy has clearly hampered any real progress at identifying new therapies. This issue is elegantly covered by Holly Moore in her recent review article (Moore, 2010). These models are not derived from the etiology of the illness or ethological analysis of behavior exhibited by the species used, and will not detect therapies with a novel mechanism of action. The same problem persists for development of new treatments for anxiety disorders. Animal models for this disorder have progressed in an attempt to at least consider what makes an animal fearful, and to apply an ethological approach (see recent review by Bailey and Crawley, 2009). However, this approach must be developed further to translate into improved therapy. One advance in the field of schizophrenia is the acknowledgement that current therapies do not improve cognitive deficit symptoms, which are a key feature of the disorder, are present in all patients, and predict outcome. The MATRICS initiative has attempted to define and describe in detail the cognitive deficits in patients, to suggest both a clinical and preclinical measure for each of the domains affected, and to identify novel targets for new therapies.
One explanation for the shortage of improved animal models is lack of appropriate investment by the relevant parties (e.g., by pharmaceutical companies, medical charities) in time, effort, and financial resource. This, combined with the lack of appropriately trained behavioral pharmacologists, has compounded the problem. These issues have been reviewed recently by de Graaf (2006) and Hendrie (2010).
The second issue, perhaps even more fundamental to the search for improved pharmacotherapy, is apparent disregard for the species used in the current models. Rats and mice have traditionally been used due to their relative ease of reproduction in captivity, small size, and easy adaptation to laboratory conditions. Their use clearly has provided much benefit in terms of developing certain therapies; however, for complex social illnesses such as depression, they may not be appropriate. Models, to date, have in general lacked good understanding of the importance of ethology. Firstly, ethology of the human disease state should be considered (i.e., what are the real changes observed and why?) and secondly, ethology of the species to be used to mimic this. Detailed understanding (and description) of the behavior of the species being used is paramount to the development of an appropriate animal model. Taking depression as an example, using the DSM to construct the model is problematic. The DSM is an imprecise diagnostic tool only (that has changed more than five times in its own lifetime), not a definitive description of the disorder. Description of the disorder comes from ethology, which gives different, easier, and clearly defined targets at which to aim. This point is raised by Hendrie and Pickles (2010). In their review, they argue that the lack of advancement in our treatment for depression over the past 50 years (comparing imipramine with fluoxetine) can be attributed to the fact that the wrong species have been used to model depression in terms of social rather than physiological characteristics, depression being an evolutionary adaptation to an altered social situation. It is only by dissecting the illness in this way, taking an ethological approach to define the nature of the illness, and then selecting the species which displays similar characteristics (which might well be rodents, depending on the illness being studied), that we can make any further progress in this area. Therefore, despite great advances in our technical capability, this alone will not produce better outcomes for people with debilitating illnesses such as depression, anxiety, and schizophrenia.
Bailey and Crawley J. (2009) Anxiety related behaviours in mice. In: Methods of Behaviour Analysis in Neuroscience.
De Graaf, J (2006) Fall and rise of behavioural pharmacology. Drug Discovery Today: technologies 3: 181-185.
Hendrie CA (2010) The funding crisis in psychopharmacology: an historical perspective. J. Psychopharmacology. 24(3):439-40. Abstract
Hendrie CA and Pickles AR (2010) Depression as an Evolutionary Adaptation: Anatomical Organisation Around the Third Ventricle. Medical Hypotheses 74: 735-740. Abstract
Moore H (2010) The role of rodent models in the discovery of new treatments for schizophrenia: up-dating our strategy. Schizophr Bull. 2010 Nov;36(6):1066-72. Abstract
View all comments by Jo Neill
The excellent review paper by Nestler and Heyman on animal models in psychiatric disorders illustrates the current problems with finding new and innovative treatments for patients. The current perception is that psychiatric diseases are so difficult to model that more and more pharmaceutical companies are leaving the psychiatry domain.
Although animal models are great to elucidate the biology of physiological processes, on top of their limited predictivity for the clinical situation, they have a number of underestimated limitations in drug discovery and development (Geerts, 2009). These limitations include 1) differences in neurotransmitter circuitry (i.e., receptor distribution), 2) the incomplete representation of the full human pathology, 3) the absence in animal models of important functional genotypes that might interfere with the primary pharmacology, 4) the pharmacodynamic interference of allowed co-medications in clinical trials, 5) the difference in drug affinities between rat and human subtype receptors (sometimes the “best” rodent profile is not always the best “human” profile), and 6) drug metabolism (not only pharmacokinetics but also the nature of unique human metabolites). In such cases, therefore, the drug fails the hypothesis and a clinical failure does not necessarily mean that the basic hypothesis was incorrect.
As an example of the drug metabolism and different neurotransmitter physiology issue, using computer-based mechanistic disease simulation, we identified a possible explanation as to why two similar D2R partial agonists, aripiprazole and bifeprunox, had such a different clinical outcome (Spiros et al., 2010). Both the presence of a unique human metabolite and the specific primate striatal dopamine dynamics that were different from the rodent physiology favored the pharmacology of aripiprazole. It is of interest to note that the second aspect has subsequently been verified experimentally in preclinical models (Natesan et al., 2010).
While the last three problems could be addressed relatively easily in preclinical drug discovery, the first three problems are probably beyond reach of preclinical animal models. Therefore—in line with other successful industries such as aeronautics and microelectronics—we have proposed to supplement the “traditional” preclinical models in drug discovery with a computer-based mechanistic disease simulation (Geerts, 2010) that is implemented using preclinical neurophysiology but is parameterized with human neuroanatomy, neuropharmacology, pathology, and genotypes, and calibrated using retrospective clinical data. Besides being helpful in progressing clinical candidate compounds, it also provides additional insights into the biology and can serve as a “quantitative” knowledge repository that can be tested and improved with each new trial outcome.
Geerts H (2009) Of mice and men: bridging the translational disconnect in CNS drug discovery. CNS Drugs 23:915-926. Abstract
Geerts H. (2010). Mechanistic Disease Modeling as a Useful Tool for Improving CNS Drug Research and Development, Drug Discovery Development, in press.
Natesan S, Reckless GE, Barlow KB, Nobrega JN, Kapur S (2010) Partial agonists in schizophrenia—why some work and others do not: insights from preclinical animal models. Int J Neuropsychopharmacol. 1-14. Abstract
Spiros A, Carr R, Geerts H (2010) Not all partial dopamine D(2) receptor agonists are the same in treating schizophrenia. Exploring the effects of bifeprunox and aripiprazole using a computer model of a primate striatal dopaminergic synapse. Neuropsychiatr Dis Treat 6:589-603. Abstract
View all comments by Hugo Geerts