Despite the diversity of the animal models discussed thus far, session chair Mark Geyer, professor in the department of psychiatry at the University of California, San Diego, pointed out they all share some fundamental problems. In this session, panelists contemplated approaches that might help to bridge the gap between preclinical animal studies and human clinical trials. Geyer noted that new strategies are needed to better enable bidirectional translation of knowledge between preclinical and clinical researchers and development of common terminology and sharing of resources across disciplines and stakeholders.
Richard Ransohoff opened the session by discussing the experimental autoimmune encephalomyelitis (EAE) model a case example of some of the challenges of translating findings from bedside to bench and back again.
This was followed by presentations of current efforts to promote discussion about the translational disconnect.1 Deanna Barch discussed the need to provide concrete opportunities for basic and clinical scientists to come together and learn from one another. Gerry Dawson described P1vital, a clinical research organization focused on standardizing, improving, validating, and sharing of human experimental medicine tools in the precompetitive space. Consortium members can then use these tools in their intellectual property–protected drug development programs. Hugo Geerts explained how quantitative systems pharmacology can be an effective translation tool, applying mathematical model-based decision support to drug development, and Thomas Steckler provided further
1This selection of presentations are not intended to be a complete listing of all programs addressing translational issues for nervous system disorders.
information on New Medications in Depression and Schizophrenia (NEWMEDS) as an example of a government-facilitated, precompetitive public–private partnership focused on developing new models and methods for drug development.
DEVELOPING BETTER ANIMAL MODELS OF ETIOLOGY AND PATHOPHYSIOLOGY
Richard Ransohoff, director of the Neuroinflammation Research Center at the Cleveland Clinic, discussed EAE and multiple sclerosis as a case example to illustrate some of the challenges in translating research from bedside to bench and back again.
Multiple sclerosis is a chronic, inflammatory, demyelinating disease that is specific to humans. There are no spontaneous conditions that occur in any other animal that recapitulate multiple sclerosis to the extent that it could serve as a model for study. As discussed by Steinman (Chapter 5), EAE is the most common animal model used for the study of multiple sclerosis. EAE as a model has become very highly refined, Ransohoff explained, evolving from inducing the disease within whole brain, to white matter, to myelin, to myelin proteins, and now, myelin peptides; and from studying monkeys, to rabbits and guinea pigs, to rats, to mice, and now most often, C57BL/6 mice. Studying EAE has helped to expand knowledge of autoimmunity and tolerance, immunity and inflammation, cytokines, chemokines, the blood-brain barrier, microglia in the brain, oligodendrocyte cell injury, and the capacity and mechanisms for remyelination. EAE is poorly predictive for therapeutics, however, as illustrated earlier by Steinman.
At the phenomenological level, mice with EAE lose weight and develop a range of neurobehavioral impairments. Importantly, EAE in C57BL/6 mice is a monophasic illness and therefore, the model lacks ability to study progressive or relapsing multiple sclerosis, making it a useful, but only partial, model. Against the spectrum of a lifetime with MS, the EAE model only captures a few weeks, recapitulating the initiation phase in which something breaks the tolerance and autoimmunity is expressed in the myelin, leading to a single demyelinating event.
Ransohoff mentioned other animal models of multiple sclerosis, each with varied strengths and limitations. There are viral models (e.g., Theiler’s murine encephalomyelitis virus, mouse hepatitis virus, herpesvirus), but the focus of these studies is often on the virology, not
demyelination/remyelination mechanisms. Another model uses the mitochondrial toxin, cuprizone, which causes very predictable oligodendrocyte apoptosis followed by remyelination, and offers a good model to study demyelination/remyelination, he said. There are also emerging models with unknown potential, such as transgenic mice with inducible oligodendrocyte cell death. Finally, a new mouse model of spontaneously developing EAE does recapitulate many of features of relapsing-remitting multiple sclerosis (Berer et al., 2011). The model is difficult and slow, involving multiple transgenic mouse strains and manipulating the gut microbiome. Ransohoff suggested that while it may be the “EAE gold standard,” it is unclear whether the platform will be useful for studying much other than the science of disease.
Ransohoff offered his thoughts on research practices found in some EAE studies in order to introduce concepts for discussion:
• If a mouse with EAE dies in the course of an experiment, some researchers include the dead mouse every day until the experiment is over. Ransohoff suggested it should be removed from the experiment. It is also incorrect to count a mouse that does not get sick every day of the experiment.
• Extensive reporting of downstream events in mice that do not show inflammation of the CNS is not informative. Animals that simply fail to become sick are not necessarily demonstrating neuroprotection.
• How limp the tail appears is not a useful parameter.
• Studies of prophylactic treatment are irrelevant for multiple sclerosis.
• There is often poor-quality tissue analysis of demyelination. For example, loss of Luxol Fast Blue staining for myelin does not necessarily indicate demyelination. Other factors can impact uptake of stains, including edema, axonal destruction, poor tissue preparation, infiltration of cells, etc.
• Lack of blinding during rating of disease severity and lack of correction for cage effects during interventions.
• Using phosphate buffered saline as the control for intraperitoneal injection of protein therapeutics instead of irrelevant antibodies or proteins.
• Comparing C57BL/6 mice from a commercial vendor with germline knockout mice of mixed background rather than littermate controls.
• Underuse of chemical demyelination (e.g., cuprizone) to study mechanisms of demyelination and remyelination.
• Failure to follow longitudinally the fate of authentically demyelinated axons.
• Underapplication of adoptive transfer.
• Minimal exploitation of spontaneous EAE models.
In closing, Ransohoff said that animal models have been helpful and have materially contributed to the science of disease. EAE is difficult to do well, and its application for direct drug development has been weak. The potential for remyelination or neuroprotective therapy is unknown. Although much has been learned about myelin production, maintenance, and repair, researchers do not necessarily know how to apply that knowledge.
To optimize the chances of translational success for this model, he proposed that there be a central or regional facility where EAE is done properly and in a standardized fashion, particularly for drug development. Discussion of such facilities should consider whether spontaneous EAE models would also be housed there and whether there should also be contract facilities for demyelination/remyelination models, where accurate quantitative tissue analysis is critical.
EFFORTS TO ADDRESS THE TRANSLATIONAL DISCONNECT
Opportunities to Bring Researchers Together for Consensus Building
Deanna Barch, professor of psychology, psychiatry, and radiology at Washington University, offered an illustration of what she called “the three-way problem” of linking basic and clinical science: The “basic animal researcher” studies spatial learning and memory using the Morris water maze. The “basic human researcher” may study the same mechanisms or constructs, but uses very different terminology and paradigms, and potentially different tools or methods. The clinical researcher may study thinking, memory, and functional abilities in patients with psychosis. This researcher is distinct from the other two and is not necessarily developing or using tools in humans that translate to the animal models, or vice versa.
Getting this trio of researchers to work together can be a challenge because there are many silos and very different languages. Barch offered several suggestions for enhancing interactions among basic and clinical researchers studying animals and human subjects:
• Provide concrete opportunities for basic and clinical scientists to exchange and develop ideas and to learn each others’ “language.” As examples, Barch mentioned the Measurement and Treatment Research to Improve Cognition in Schizophrenia (MATRICS) initiative,2 the Cognitive Neuroscience Treatment to Improve Cognitionin Schizophrenia (CNTRICS) initiative,3 and this Institute of Medicine workshop.
• Identify and clearly define the constructs of interest in terms that both the basic and clinical researchers can understand. Foster an understanding of how and why constructs have been operationalized differently across species.
• Provide concrete funding mechanisms to encourage basic and clinical research collaboration and funding support for tool development.
• Develop a consensus on what it means to measure a homologous process or construct and identify ways of improving that homology, either in existing paradigms or in new paradigms, with the goal of improving predictive utility.
• Raise awareness among the human researchers about what is needed to make their paradigms more amenable and attractive to animal researchers.
• Be willing to challenge existing ideas about animal models.
Bringing Validated CNS Experimental Medicine Methods from Academia to Industry
Seeing an opportunity to help bridge the gap between his clinical and preclinical colleagues, Gerry Dawson co-founded P1vital,4 a clinical research organization focused on experimental medicine for central nervous system (CNS) disorders. Dawson is chief scientific officer of the
company, which is headquartered in the department of psychiatry at the University of Oxford (in the United Kingdom [UK]). The specific idea, he explained, was to take the existing CNS experimental medicine methods for schizophrenia, depression, cognition, and anxiety that are used in academia and make them accessible for use by drug companies through management and coordination by P1vital, including the validation of models.
P1vital assembled an academic network of five UK academic institutions with expertise in CNS research (University of Oxford, Institute of Psychiatry London, University of Manchester, Cardiff University, and Imperial College London), and a precompetitive consortium of five pharmaceutical companies that provided financial and practical support for the validation studies (AstraZeneca, GSK, Lundbeck, Organon [a Merck subsidiary], and Pfizer Inc.).
Dawson described a consortium study on schizophrenia to illustrate the P1vital process. Schizotype is a psychological concept that describes a continuum of personality characteristics and experiences related to psychosis and schizophrenia. The basic hypothesis of the study is that high schizotypes can be identified from those who are low or mean scoring using a simple signal-detection test that exposes individuals to white noise and voices. High schizotype participants will have more false alarm responses (e.g., perceiving spoken human voices embedded in white noise when no voice is present) and faster-to-respond, or “jumping-to-conclusions” decision-making processes. Upon hearing voices, high schizotypes will display the same areas of activation on functional magnetic resonance imaging (fMRI) as patients having hallucinations.
College-age participants were given a nicotine patch as a putative cognition enhancer and a placebo capsule, or a placebo patch and either a placebo capsule, an amisulpride capsule, or a risperidone capsule. Combined data from three test sites showed very significant drug-by-group differential effects. In high schizotypes, amisulpride improved performance, and in low schizotypes it made it worse.
To try to back translate to animal models, Dawson and colleagues first brought forward an animal model system, the Morris water maze. Human participants were observed with fMRI while navigating a virtual reality water maze, or arena task, via 3-D goggles and a joystick. In a preliminary experiment, with only high and mean schiotypes, testing encoding and retrieval, hippocampal activation was observed in young males, but not in elderly males. In assessing schizotypes, it was found
that mean and high schizotypes performed the task comparably. High schizotypes showed decreased hippocampal activation, via fMRI, during encoding (i.e., fewer resources were recruited), and increased activation during retrieval compared with controls. This indicates that while there were no behavioral differences between the groups there were differences evident by fMRI. One participant noted that this would suggest a potential biomarker for schizophrenia that would be present prior to onset of symptoms.
Similarly, young people who have a family history of depression complete simple tasks such as N-back successfully, but fMRI shows they need to recruit more resources to do so. As these individuals get older, Dawson said, this will become more difficult and they will become more vulnerable. The question, then, is when to intervene with drugs and/or therapy.
Dawson noted that P1vital has a variety of methods validated thus far and the pharmaceutical consortium members are using them in their drug-discovery processes. He added that the pharmaceutical industry has many compounds with specific and well-characterized mechanisms of action that were assessed in primary efficacy clinical trials and subsequently shelved for lack of development resources. The next step for P1vital will be to select four to five unique compounds from these collections and assess them in a range of these validated experimental models for signals of efficacy. This effort will be sponsored by a European College of Neuropsychopharmacology (ECNP) led experimental medicine network of companies and academic groups. In this regard, Geyer added that AstraZeneca, which has decreased its focus on psychiatry research, has prepared a list of compounds that it will make available for research under an agreement with the company.
Quantitative Systems Pharmacology as a Translational Tool
Hugo Geerts, chief scientific officer for In Silico Biosciences, discussed quantitative systems pharmacology as a translational tool. To begin, he took stock of what the pharmaceutical industry could potentially learn from other successful industries, such as microelectronics and aeronautics. First, these other industries, Geerts explained, have formalized their “collective knowledge,” which helps them to move from information to knowledge to building prototypes. These groups use an
advanced modeling and simulation approach to capture community-wide expertise and knowledge.
Next, these companies embrace complexity. The focus is on circuit analysis because the networks give rise to emergent properties that are not explained by single targets. In many cases, Geerts said, multitarget pharmacology is needed to really affect the complete outcome of an emergent property in the brain. However most pharmaceutical companies prefer to work on very selective molecules. Geerts added that nonlinear processes, including those going on in the brain at any given time, are very hard to capture without mathematical modeling.
In addition, these other industries focus more on failure analysis, learning from negative outcomes. In many cases with failed clinical trials, products are simply dropped and the results never published. But those failed trials offer a lot of information (e.g., Why did that trial fail? Was it the drug? The target? The patient population?).
How can the successes in other industries be applied to CNS research and drug development? As observed throughout the workshop, animal models are helpful in unraveling individual biological processes in great detail, Geerts said. They are less helpful for translating this knowledge into the clinic. There are a number of different reasons for this, he said. Human neural networks and clinical outcomes are much more complex than animal networks and experimental readouts. Drugs may behave differently in animal models versus humans, due to factors such as pharmacology, genotypes, comedication, or comorbidities. There is growing realization that complex CNS diseases need multitarget pharmaceutical approaches (Swinney and Anthony, 2011). Thus while the reductionist approach is good for understanding the biology of disease, it may not be the best approach for bringing drugs to the clinic.
A large amount of untapped clinical data exists for CNS drug development and a possible solution is an approach Geerts termed “computer-aided research and development.” This approach combines the best of the animal world with the clinical world, in an animal-independent implementation, he explained.
One of the major issues in bridging the gap between biomarkers and clinical functioning is that in many cases, it is a very large jump from the individual biochemical change to the cognitive performance. In Alzheimer’s disease, for example, the biochemical marker may be amyloid pathology, which affects the synaptic interactions. However, those synapses are part of pyramidal cells, and those pyramidal cells are part of
networks. Changes in emergent network properties are then assessed with clinical scales, such as cognitive performance.
Geerts listed several possible reasons why animal models might not always translate to the clinic:
• Differences in drug metabolism, both in clearance and absorption, and in the formation of metabolites.
• Different drug affinities for human versus animal targets, which Geerts noted is often underestimated.
• Absence of functional genotypes in animals, which in humans can affect pharmacology.
• Incomplete pathology in transgenic mice.
• Different neurotransmitter wiring. Human regions of the brain have many different receptor densities as compared to animals.
• Placebo responses in humans.
To address these issues, Geerts suggested a systems pharmacology approach. Systems pharmacology considers the interaction of a drug with its target not only at increasing structural levels (e.g., cells, organs, animals, humans, populations), but also at the cellular and multicellular networks levels, trying to simulate in a quantitative fashion the effect of an intervention on emergent properties, using mathematical models and simulations (see NIH, 2011, for further discussion).
To make this approach usable and actionable for pharmaceutical development, In Silico Biosciences leverages the preclinical data from academia over the past 60 years and integrates information on receptor physiology, CNS drug pharmacology, target exposure, human pathology, and imaging to create a computational neuropharmacology translational model bridging preclinical and clinical research. The goal is to have calibrated and validated platform applications to take compounds from pharmacology to clinical readout for different CNS diseases.
Precompetitive Public–Private Consortium for Methods Development and Validation
Thomas Steckler, head of Translational Research CNS at Johnson & Johnson, expanded on the discussion by describing NEWMEDS (Novel Methods Leading to New Medications in Depression and Schizophrenia), one project of the Innovative Medicines Initiative (IMI).5 IMI is Eu-
rope’s largest public–private initiative aiming to speed up the development of better and safer medicines for patients. Steckler noted that the lack of predictive validity remains the most frequent reason for attrition in drug development (Kola and Landis, 2004) and animal models are often blamed for failure of a compound to show efficacy in the clinic (Wang et al., 2011). In working to address this relative to the development of psychiatric drugs, the NEWMEDS consortium is focused on three key challenges:
1. How can we better understand the disorders (schizophrenia and depression) and make sense of the genetic and molecular findings, the pathways of the neuronal circuits, and symptom clusters?
2. Are there novel approaches that can improve success rates for new compounds taken into humans? Are there new animal models that would be more predictive of the efficacy of compounds in the clinic? Can imaging tools (e.g., PET, fMRI) and experimental medicine approaches be used to increase success?
3. Are there ways of doing clinical trials more efficiently (e.g., clinical databases, novel endpoints)?
Composed of work packages (WPs) that address issues across the drug development process, NEWMEDS creates a translational network. For example, WP07, “Identifying risk pathways via CNV (copy-number variation) genetics,” is linked to WP04, “Cross-species and functional imaging models for drug discovery” to determine whether there are different activation patterns in patients depending on the CNVs they carry. Both are also linked to WP03, “Human cognitive testing,” to the component of WP04 focused on rodent fMRI phenotyping, to WP01, “Linking animal and clinical models via electrophysiology,” and to WP02, “Animal models of cognitive dysfunction that relate to clinical endpoints,” as there are mouse models that carry these CNVs to be tested not only for behavior, but also electrophysiologically and via imaging. Proteomics and metabolomics (WP09) are used to help in characterizing the rodent model.
NEWMEDS activities also interact with other IMI projects, such as PharmaCog (focused on Alzheimer’s disease) and the EU-AIMS (focused on autism). All three projects, for example, use touchscreen technology for development of model paradigms.
Activities within the consortium involve multimodal approaches; sharing of expertise, knowledge, methodology, and resources; and translation and back-translation. Across the consortiums there is overlapping molecular biology, neuronal circuits, symptom domains, and technologies.
Important aspects of NEWMEDS are replicability, reliability, and data sharing. Steckler noted that the consortium has lead to 15 published papers and 60 conference publications, as well as the development of Web-based tools, such as a clinical significance calculator for biomarkers in depression (Uher et al., 2012).
Steckler shared his personal assessment of the strengths, weaknesses, opportunities, and threats for the NEWMEDS initiative (see Figure 6-1). The strong scientific focus, multimodal approach, complementary capabilities, cross-species and back-translation activities, and high level of scientific and technical expertise are strengths of the NEWMEDS approach. The commitment to the initiative by the different partners and the openness to sharing data are the most valuable aspects of the consortium, he said. There are many opportunities to gain knowledge, communicate across partners, and share the workload.
A primary limitation, however, is that the initiative is limited to the European Union. Steckler also noted that there is room for improvement in the interactions between the different work packages. A participant added that although there is much U.S. interest in NEWMEDS, companies have raised concerns about the level of bureaucracy.
The most significant threat, according to Steckler, is the instability of the pharmaceutical industry, with three pharmaceutical partners already opting out of the consortium because they decided to drop some or all of their psychiatry research programs. They may still provide expert input, but they are no longer actively generating data.