The first day of the workshop included concurrent breakout groups intended to facilitate in-depth analysis of the translational success of animal models in six areas of neuroscience research. Those areas are Alzheimer’s disease, neurodegeneration, stroke, addiction, schizophrenia, and pain. In preparation for the subsequent workshop sessions, breakout groups focused their discussions on three key questions:
1. Would this research area benefit from a new or improved standardized animal model?
2. How well do animal model and human clinical endpoints correlate in this area of research?
3. What is needed to bridge the translational gap between animal models and clinical science in this area?
These smaller breakout groups enabled discussions about the role and effectiveness of animal models in the development of therapies for nervous system disorders. Following the breakout discussions, each group moderator summarized the main points of discussion for all attendees.
ANIMAL MODELS FOR ALZHEIMER’S DISEASE
In discussing current animal models for Alzheimer’s disease it is important to think about the human phenotype and what is being modeled in terms of the animal phenotype. The moderator, Bradley Hyman, professor of neurology at Harvard Medical School, said that animal models of
Alzheimer’s disease, based on the genetics of the disease and the closely related frontotemporal dementia, replicate at least some of the pathology. Researchers have been successful at modeling very specific aspects of Alzheimer’s disease in the mouse (e.g., plaques, tangles). Although these are incomplete models of the human disease, they have been well received in the field as potentially relevant models for use in drug discovery.
Patients with Alzheimer’s disease will display both amyloidopathy and tauopathy; however, scientists often focus, in a reductionist way, on one or the other in an animal model. A participant added that even though the anatomy in the mouse is different than the human, mutant tau mice are relatively good models in that they recapitulate tau-dependent neurodegeneration. This has led a number of companies to focus on antibodies that block tau-dependent neurodegeneration in these mouse models.
Hyman reiterated that mouse models are partial, or incomplete, models of the overall human phenotype. In an animal model, pathological changes are studied in the context of a unique and isolated event (i.e., lesion) over a relatively short period of time. Alzheimer’s disease, as it occurs in humans, is the sum of how lesions occur and evolve over the course of many years or decades. Mapping where in the evolution of the human disease an individual mouse phenotype model fits is an important and often uncertain piece of information. Hyman questioned the hypotheses tested in humans that do not have exact correlates in animal models (e.g., differences in when amyloid deposition occurs between animal models and in human disease).
Several participants also discussed the use of imaging and fluid biomarkers in both animal models and clinical research (e.g., positron emission tomography or PET); ligands that can identify beta-amyloid load in the brains of humans; analysis of beta-amyloid tau and phospho-tau as biomarkers in cerebrospinal fluid. These types of biomarkers are now used for early diagnosis and to monitor disease progression in humans and many participants discussed the need to translate them back into animal models. As new therapeutics are examined, using similar biomarkers in both animal models and humans may allow for better translation of animal findings into humans. It was noted that the Alzheimer’s Disease Neuroimaging Initiative (ADNI) is working toward Standardization of Alzheimer’s disease biomarkers across the 57 ADNI sites, enhancing quality assurance, quality control, and better analysis of the clinical and imaging data in the ADNI public database.
In summary, Hyman emphasized the following about Alzheimer’s disease animal models:
• Some mouse models exist that are close genocopies of inherited early-onset disease and have been well received as potentially relevant models.
• Mouse models are incomplete models of the human phenotype.
• Behavioral results in mouse models are not as robust as biochemical and neuropathological readouts.
• There is a need to better match animal models to the appropriate stage of human clinical disease.
Animal models are limited in terms of how far they can be extrapolated toward the human condition. However, compared to many other types of neurologic diseases, Alzheimer’s disease research has some very promising successes that can be built upon.
ANIMAL MODELS FOR NEURODEGENERATION
Neurodegeneration research spans Parkinson’s disease, Huntington’s disease, ammyotrophic lateral sclerosis (ALS), and multiple sclerosis, for example. Robert Ferrante, professor in the departments of neurological surgery, neurology, and neurobiology at the University of Pittsburgh, noted that much of this breakout group’s discussion centered on standardization of models and whether they accurately reflect neurodegenerative diseases. Ferrante suggested that current animal models for Huntington’s disease and ALS may accurately reflect not only patho-physiological mechanisms of human disease, but also neuropathology and behavioral phenomena. For other disorders, however, it is much more difficult.
In addition to emphasizing the need for standardization of animal models of neurodegenerative diseases, participants in this breakout also discussed enforcing standards for preclinical studies in animals and raised concerns about the publication of research that cannot be replicated. It was noted by some participants that although the NIH has set standards for conducting animal research1 and papers have described these standards, widespread adoption of these recommendations has been slow (Kilkenny et al., 2010).
Breakout session participants also discussed reevaluating the scientific approach to drug discovery for neurodegenerative diseases. In general, the target-centric approach to neurodegenerative diseases has failed during the past 50 years and there was discussion of a systems biology approach to disease research, as well as in silico models of disease.
In discussion whether animal models accurately reflect human neurodegenerative disease, the issue was raised as to whether animal model studies might be replaced with more Phase 0 clinical trials in humans. In this regard, there was a call for the identification of pharmacodynamic markers and biomarkers that are clearly reflective of the disease. For example, some noninvasive mechanisms, such as high-definition fiber tracking in traumatic brain injury and other disorders, reflect what is occurring in the brain and could be developed as a biomarker for many neurodegenerative disorders. Several participants in this group also noted the need for correlation between the biomarkers used in patients and in animal models.
Ferrante summarized the main points of discussion in this breakout session as
• There is a need for more standardization of models reflecting neurodegenerative disease in patients.
• The scientific approach to neurodegeneration research (e.g., target-centric versus systems-based, in silico, etc.) may need to be updated.
• Research could benefit from increased focus on Phase 0 human clinical trials.
• There is a need for pharmacodynamic markers and biomarkers that clearly reflect the disease.
ANIMAL MODELS FOR STROKE
The biology of ischemia is different from that of neurodegeneration, explained group moderator Constantino Iadecola, professor of neurology and neuroscience at Weill Cornell Medical College. The stroke process starts with an arterial occlusion, which can be reproduced effectively in animals. Several participants noted that animal models of stroke are generally predictive and adequate. However, the models are not perfect, and Iadecola said it could be argued that the mechanisms whereby a thrombosis or embolus forms in humans may not be mimicked exactly by the
surgical occlusion of an artery in an animal model. Nevertheless, the basic reaction of the tissue to the occlusion is fairly standard between different species.
Despite the relative suitability of the animal model for the study of stroke, clinical trials have not produced effective treatments and many pharmaceutical companies have scaled back or abandoned stroke research programs. A few participants suggested that failure in the clinic is partly because endpoints used in preclinical animal studies are different from those used in clinical trials. For example, animal studies often assess stroke volumes histologically using a chemical marker, which does not really reflect cell death, while clinical trials measure functional outcomes. Fortunately, Iadecola said, advanced technologies, such as diffusion-weighted imaging, can be done in animals and humans, allowing for better correlation between animal and human studies.
One participant noted that there is an ongoing disconnect between animal studies and clinical trials with regard to what happens in human stroke and how animal stroke data are obtained. For example, in some animal studies, the investigational drug is given before the stroke is induced and protection conferred. However, this is not possible in patients as drugs are administered 6, 12, or 24 hours after the stroke, with no effect.
As a result of discussions at a number of symposiums, organizations have been formed to address this issue. Researchers are working toward studying stroke in the animal models such that the studies mirror more closely what happens in the clinic. For example, Iadecola said, researchers are using more animals with risk factors for stroke, such as diabetic animals and aged animals as opposed to the younger “teenage” animals on which classic stroke work has been based. At the same time, clinical trials are taking into account the conditions under which protection has been observed in animal studies (e.g., sex of animals, time window of drug administration).
In summary, Iadecola highlighted the main points from this breakout session:
• Although neuroprotection has been demonstrated in numerous animal studies, treatment of humans has not been effective.
• Adequate animal models of stroke exist, but successful translation of the science from animal models to humans has been limited.
• The discordance between animal and human studies may be due to bias in study design (e.g., different endpoints) or to the failure of animal models to mimic clinical disease adequately.
ANIMAL MODELS FOR ADDICTION
Athina Markou, professor in the department of psychiatry at the University of California, San Diego, described the development of the smoking cessation drug varenicline as an example of successful development of a therapy for addiction. Markou speculated, however, that the success of varenicline in clinical trials is attributable more to the fact that there was a very strong theoretical rationale and less to the translation of preclinical animal studies. The animal models used were valid models, she said, but they were simple.
In contrast, despite the significant amount of research that has focused on dopamine, there are almost no drugs that have made it to the market for the treatment of addiction. Markou noted that there is disagreement as to why this is the case. Some argue that the hypothesis that dopamine mediates dependence and addiction for all drugs of abuse is incorrect. A counterpoint is that the clinical trials of dopaminergic drugs have not been done properly or perhaps have not been done at all. There are potential targets, such as the dopamine D3 receptor, but Markou noted there is currently little interest on dopaminergic targets by pharmaceutical companies. Other potentially good targets have not been explored sufficiently and breakout group participants discussed the need to incentivize industry and to educate pharmaceutical manufacturers that there is a significant market for addiction treatments.
Unlike some other nervous system disorders (e.g., schizophrenia where the etiology is not definitively known), the etiology of drug dependence is known to be excessive exposure to the drug. As such, animals can be similarly exposed to a drug and studied. Several breakout participants noted that models of addiction exist, although there is always room for improvement. For example, it was noted that additional emphasis is needed on more sophisticated models, such as models that examine the switch from drug experimentation to addiction or of relapse. A few participants in this breakout session raised concerns that a standardized animal model of addiction might not be the best approach. Rather, it was suggested that studying a variety of models that employ different approaches could provide converging evidence.
One issue for animal models of addiction is the heterogeneity of the human population with regard to addiction. The vast majority of people who experiment with drugs do not become dependent, suggesting a genetic component to addiction. In fact, genetic studies have provided some potential targets and it was suggested that one way to move forward is to try to over- or under-express these genes in mouse models and to study the heterogeneity of addiction development.
Finally, participants in the breakout session discussed concerns with clinical trials. As in other breakout groups, the need for cross-validation of animal model endpoints with clinical measures was noted. Also, current clinical measures for addiction studies were said to be inadequate in that the primary outcome measure is drug consumption—did the patient stop taking the drug or not? There are many processes that lead to addiction or to lack of abstinence that are not assessed in clinical trials, Markou noted. For example, is drug consumption the result of physical withdrawal or perhaps due to some cue that reminded the patient of the drug? For alcohol dependence, is controlled use of alcohol an acceptable endpoint? Patient compliance is also an issue. In many clinical trials for addiction, it is not clear whether the patients have actually taken the therapeutic drug and failure of the trial may be because the patients do not achieve adequate levels of the therapeutic drug in their system.
In summary, Markou highlighted the following points made by various participants in this breakout group:
• There are good animal models of addiction. Rather than standardization of models, many participants noted that the use of multiple models that employ different approaches could provide converging evidence.
• Genetic animal models may be helpful in understanding the heterogeneity of human addiction.
• The many potential therapeutic targets for addiction that have not been adequately researched and incentives for research in this area may be needed.
• Cross validation among animal models, clinical endpoints, and processes that are assessed in human trials is lacking.
ANIMAL MODELS FOR SCHIZOPHRENIA
In the breakout session on animal models for schizophrenia, much of the discussion focused on processes, explained breakout moderator Holly
Moore, associate professor of clinical neurobiology in psychiatry at Columbia University. Topics included the neurobiological processes that might underlie psychological processes disrupted in schizophrenia; the process of doing research; and the process of dialogue between clinicians and researchers using animal models.
Many participants in this breakout session believed that current animal models for schizophrenia, while informative, are not adequate. It was noted that there are useful assays of behavior and cognition and of the neurocircuitry mediating the cognitive process affected in schizophrenia. However, divergent opinions were expressed on how useful those assays are and whether it is necessary to assay neurocircuitry or whether looking for direct impacts of therapeutics on behavior is sufficient. Animal models are being developed to probe the neurocircuitry underlying cognitive deficits, as well as the basic processes underlying psychosis and negative symptoms in schizophrenia.
Moore noted that some breakout group participants thought that the path forward is to go back to clinical and epidemiological research and ask “what is wrong” in schizophrenia. One simple approach to answer this question would be to examine patient behaviors while imaging their brains. This would allow researchers to determine what is behaviorally and cognitively aberrant and what neurocircuits are activated in correlation with the observed deficits.
Armed with that information, researchers could develop assays in animals that have homology with assays used in the clinic. First, however, there needs to be reliable and objective assays for humans that can predict a clinically significant change such as worsening or improvement in the patient’s clinical profile. For animal modelers, clinical outcomes such as reduction in symptoms based on subjective scales are not useful. On the other hand, an objective assay of cognition and behavior in humans without data on the clinical significance of these outcomes is also not helpful.
In some cases, objective assays in humans and the homologous assays in animals may be very similar. For example, prepulse inhibition measurement of sensorimotor gating is similar across animals and humans and is mediated by the same circuits in the brain. In other cases, assays that are guided by similar circuits do not look the same in an animal as they do in a human from a phenomenological point of view.
Moore noted that many breakout session participants thought that homology at the level of neurocircuits might be a useful starting point for dialogue between clinicians and researchers that use animal models about
what mediates symptoms or behavioral pathology. Others suggested that it is not necessary to understand how neurocircuits mediate an aberrant behavior; that it would be possible to have a reliable and validated assay of a problematic behavior from a psychological point of view (e.g., an assay of a sensorimotor deficit). Another concern raised by one participant was that many animal models of schizophrenia focus on primary pathology and not how drugs might act on or become a compensatory mechanism.
Finally, once there are assays in animals that have some homology with the assays used for humans and which have been shown to predict clinically significant outcomes or functional outcomes, are those animal models being fully used? Studying systems in control, or intact, animals is relevant for target validation and pharmacodynamics. Once studies in an intact animal establish that the drug is binding to circuits of interest and modulating both circuit activity and behaviors known to be mediated by that circuit, the question remains whether the drug will work on that same circuit and to the same extent in humans. Several group participants emphasizes that this is where an animal model of disease guided by epidemiology and symptomology is important.
Moore summarized the main points of this breakout session as
• Animal models and human clinical research inform each other.
• Control, or intact, model systems are useful for target validation and pharmacodynamics.
• Animal models of disease would benefit if they were guided by epidemiology and symptomology.
• There is a need for animal assays that are translatable and predict clinical outcomes and for assays to have some homology across species and determinants.
Ideally, once an appropriate animal model is in place, the clinical trials would be designed to ask the same questions that the animal models asked, using the same objective assays in the clinical trials that were chosen for use in the animal studies because, at the very beginning of the process, they were objective assays that had some clinical relevance.
ANIMAL MODELS FOR PAIN
A large number of animal models are used by pain researchers. Participants in this breakout session discussed the adequacy of these models and the appropriateness of the assays used relative to clinical outcome measures.
In some ways, the field of pain research is unique in that it is possible to mimic the initial inciting events, explained A. Vania Apkarian, professor in the Neuroscience Institute at Northwestern University and group moderator. Researchers can cause peripheral neuropathy, for example, and study diabetic neuropathic pain-related behavior in animals. Classically, the outcome measure in pain studies has been nociception, on the assumption that reflexive outcomes (e.g., sensitivity to touch or heat) correlate to some extent with the human condition.
Apkarian relayed that many group participants felt there were many useful animal models of pain and that a standardized model was not needed. Rather, to make the most of existing models, it is important to ask the right questions. As in other sessions, participants also discussed the need for biomarkers that can be assayed in humans and animals alike.
Human brain imaging studies are changing the field of pain research through investigation of chronic pain conditions in humans, Apkarian said. There was discussion about the need to start looking at correlates of chronic pain in animal models. As current models are essentially models of inciting a painful condition, the question has not been asked as to what is the causal or a critical parameter that induces the maintenance of pain. Chronic pain is not just nociception. Pain interacts with and reorganizes the brain. Injuries in humans may or may not lead to chronic pain, suggesting that something genetic in the brain needs to be considered in addition to the injury. Many participants indicated that much can be learned from genetic models in mice that might inform research on the human condition.
In summary, participants in this breakout session raised the following issues with regard to animal models of pain:
• Many existing animal models of pain might be more useful if researchers ask the right questions.
• Pain is more than just a sensation and appropriate measures are needed in existing animal models to address this complex issue.
• In particular, several participants were interested in identifying mechanisms for inciting pain versus maintenance of pain and
understanding the mechanisms of chronic pain in humans.
• The usefulness of mouse genetic models and corresponding animal and clinical neuroimaging biomarkers was also discussed.
ANIMAL MODELS ADDRESSING NEURODEVELOPMENT
In the open discussion following the breakout group summaries, a participant raised another subarea of neuroscience research as an offshoot of the discussions of models for schizophrenia and addiction—animal models of what may essentially be developmental disorders.
Moore pointed out that although models of schizophrenia in adult animals are used for pharmacologic studies, knowledge of the epidemiology of schizophrenia has led to the development of models where the perturbation is made quite early in development, when risk factors for the disease come into play. Although the perturbations are made early in development, behavioral and neurological outcomes traditionally have not been studied until those animals were adults, presumably because that is when the disease emerges in humans. Only recently are researchers starting to think about looking at different stages in disease development and potential strategies for prevention.
Moore noted that people who are at high risk for the psychopathology associated with schizophrenia are not asymptomatic before they become psychotic. Rather, they have phenotypes that could be identified, characterized, and targeted for treatment (Kaur and Cadenhead, 2010). That treatment may delay or prevent psychosis might significantly impact functional outcome even though the person is still undergoing a psychotic episode. Researchers can start looking for signs earlier in people who have a first degree relative with schizophrenia and can look at prodromal patients who have been clearly identified as at risk using well-characterized and accepted scales, and ask what treatment is needed prior to the onset of symptoms.
Moore suggested that there should be less focus on predicting who may become psychotic and trying to prevent that and focusing more on treating the pathology affecting them at any particular time in their lives. There is a real need for a developmental perspective to schizophrenia, she said, and animal models can help elucidate this.
Many changes in the brain occur during adolescence, but that does not mean that adolescence is a pathology. Perhaps the parts of the brain
that are changing at the fastest rate during adolescence may be the most vulnerable. If those areas overlap with the circuits that are implicated in anxiety, drug abuse, or depression, for example, it may provide clues to the points of vulnerability in that circuit at that time. These are still basic research questions.
Markou noted that people who start tobacco smoking in adolescence have the hardest time quitting. Therefore, it would be important to extend animal models of addiction to this developmental stage as well.