Sharon Rosenzweig-Lipson of IVS Pharma Consulting, and session chair, began by posing several questions for consideration: Will corresponding endpoints be useful for predicting clinical efficacy? What is an animal model meant to predict and what is the corresponding endpoint intended to predict? For example, decreases in amyloid beta can be measured both in animals and in patients with Alzheimer’s disease. However, having this corresponding animal and clinical endpoint is not necessarily sufficient to make a prediction about the clinical efficacy of a potential therapeutic. Similarly, specific behavioral changes in animals may correspond to changes in humans, but these may, or may not, translate into a prediction of disease course. There is value to translation of clinical endpoints, she said, but it is important to understand what that value is.
ROLE OF MATCHING ENDPOINTS
Invited panelists used specific case examples to discuss the role of corresponding endpoints, and the impact of experimental parameters on corresponding endpoints and bidirectional translation. Neal Swerdlow described prepulse inhibition as an example of the ability to study the same endpoint in both an animal model and humans. Larry Steinman discussed experimental autoimmune encephalomyelitis (EAE) as an example of how differences in the way an animal model is tested can have profound differences on the findings. Michela Gallagher described how
neuroimaging tools have demonstrated that functional components of hippocampal circuits are very similar between animal models and humans.
Neal Swerdlow, professor in the department of psychiatry at the University of California, San Diego, used the prepulse inhibition assay as a discussion case for the role of corresponding endpoints. Prepulse inhibition is defined as the automatic inhibition of the startle reflex, the contraction of the facial and skeletal musculature in response to an intense, abrupt stimulus, when the startling stimulus is preceded by a weak lead stimulus or prepulse. A primary measure of the startle reflex is movement of the orbicularis oculi muscle, or eye blink, often determined using surface electrodes attached to the muscles around the eye. In the laboratory, prepulse inhibition is an operational measure of sensorimotor inhibition, the inhibition of a motor response by a weak sensory event.
Prepulse inhibition is markedly diminished in a number of different neuropsychiatric disorders, including schizophrenia, Huntington’s disease, Tourette’s syndrome, Asperger’s syndrome, fragile X syndrome, and obsessive-compulsive disorder (reviewed in Braff et al., 2001). In schizophrenia, for example, patients show deficits in prepulse inhibition regardless of whether the startling stimulus is a tactile (e.g., an air puff) or acoustic stimulus. Although this phenotype is not specific to schizophrenia, it is robust and replicable.
Commonalities and Differences
From an experimental perspective, the stimulus delivery and response acquisition hardware and software that are used for prepulse inhibition testing are very similar across species. The most obvious difference in testing is physical restraint; human subjects voluntarily sit in a chair during testing while mice are enclosed in a tube.
The response characteristics are strikingly similar across species, including sensitivity to stimulus parameters (e.g., prepulse, intensity, and interval), cross-modal inhibition, habituation, and latency facilitation. That is primarily because the startle reflex involves neural circuitry that is common across all mammalian species, Swerdlow explained (Swerdlow et al., 1999, 2008). There are some obvious, although rela-
tively subtle, differences in response characteristics; for example, axons are longer in humans and therefore reflex latencies tend to be longer. Waveform morphology can differ depending on whether the electrodes are collecting a whole-body response (as in the mouse model) versus a single muscle group (eye blink in humans).
There is also evidence that similar biological substrates are involved in regulating prepulse inhibition across species. There is interesting sexual dimorphism, he noted, with males tending to be more inhibited than females across species (mice, rats, and humans). Prepulse inhibition is also a highly heritable phenotype.
Some of the practical uses of prepulse inhibition testing relate to its predictive validity of antipsychotic drug effects. Prepulse inhibition is disrupted by dopamine agonists such as apomorphine. Swerdlow described early work that he and others conducted which showed that administration of the typical antipsychotic, haloperidol, or the atypical antipsychotic, clozapine, produced dose-dependent normalization of prepulse inhibition in apomorphine-treated animals (Swerdlow and Geyer, 1993). This ability of a compound to reverse the disruptive effects of a dopamine agonist on prepulse inhibition in an animal has been used as a predictive model of antipsychotic efficacy in humans.
While establishment of these cross-species comparable endpoints has created robust systems for predicting clinical efficacy of antipsychotic therapies for schizophrenia, the larger picture is that the system facilitates the identification of “me-too” drugs, Swerdlow said. New compounds may have a different profile but drugs identified in using this system are all basically antipsychotics that affect positive symptoms.
Whether drug effects on prepulse inhibition have corresponding endpoints across species is dependent on many variables (e.g., species and strain, stimulus parameters, drug dose and route of administration, concomitant drug effects on startle magnitude, subpopulations). Most of these, Swerdlow explained, can be controlled or addressed post hoc to ensure matching endpoints.
For example, Long-Evans and Sprague-Dawley outbred rats have very different prepulse inhibition profiles. While Long-Evans rats show only slightly less prepulse inhibition than Sprague-Dawley rats, their response to a dopamine agonist is dramatically different. At shorter
prepulse intervals, Long-Evans animals show a robust potentiation of prepulse inhibition, while Sprague-Dawley animals show a profound disruption (Swerdlow et al., 2004). This is a highly reliable strain difference that was first viewed as noise, but is now known to be related to differences in biology. Importantly, studying the biology of this difference has led to much useful information that Swerdlow pointed out would not be known if only one standardized strain of animal was used.
There are corresponding endpoints across species in terms of the neural circuitry. The primary startle circuit is fairly constant, with more variability and interesting differences in the downstream circuitry, Swerdlow explained. Researchers have shown that activation of basal ganglia and cortical regions are relevant for regulating prepulse inhibition in rodents. Correspondingly, a number of human disorders that display deficits in prepulse inhibition have identifiable abnormalities within portions of the basal ganglia or limbic cortical circuitry. In other words, the human anatomy maps well to the rat anatomy of circuit regulation.
A number of developmental animal models will produce a deficit in prepulse inhibition, such as the neonatal ventral hippocampal lesion model. As one example of construct validity of these models, these deficits can be corrected in a dose-dependent manner with clozapine. This endpoint correlates well to the human condition, where control subjects display about half as much prepulse inhibition compared to patients with schizophrenia, a deficit that can be normalized substantially by clozapine (Kumari et al., 1999). Swerdlow also described corresponding endpoints relative to disease gene effects on prepulse inhibition across species. Patients with Huntington’s disease, for example, show profound deficits in prepulse inhibition (Swerdlow et al., 1995), a phenomenon that has been reproduced in a transgenic mouse model of the disease (Carter et al., 1999).
The Role of Corresponding Endpoints
In summary, the conditions (e.g., eliciting stimuli, response acquisition) for studying startle and prepulse inhibition across species are nearly identical (with the obvious difference of physical restraint of animals);
response characteristics are comparable; and there is evidence for similar biological substrates across species. The prepulse inhibition assay has predictive validity in developing and testing the activity of antipsychotics and for developing typical and atypical antipsychotics.
That said, the anatomy, neural circuitry, and neural substrates of schizophrenia are very complex. While drugs can be developed to control some of the simpler symptoms of this disorder it is less clear that these predictive models will be helpful in developing interventions that offer long-term benefit in terms of function, interventions that may act through a completely different mechanism. The diffuse neuropathology in schizophrenia may reflect events very early in development, years or decades before patients seek medical intervention (Halliday, 2001). Swerdlow suggested that even a “perfectly corresponding” animal model cannot generate a therapeutic (drug, gene, protein, etc.) that will substantially restore healthy neural function to patients with schizophrenia, addressing the variable web of absent and misguided neural connections that have developed over the course of a lifetime. In developing therapeutics for schizophrenia, our “endpoint” should reflect these limits, he concluded.
EXPERIMENTAL AUTOIMMUNE ENCEPHALOMYELITIS
The EAE animal model was first described in 1933 by Rivers and colleagues while they were seeking to understand how some viral infections lead to neurologic reactions. Nearly 80 years later, EAE remains “one of the most enduring models of human disease,” said Larry Steinman, professor in the departments of Neurological Sciences and Pediatrics at Stanford University (see Steinman, 2003).
EAE is not a model of multiple sclerosis, Steinman stressed, but there are similarities, and EAE is often used for the study of demyelination and examination of potential therapeutics for multiple sclerosis. Steinman shared two case examples of how experimental parameters can impact outcome; one is highlighted here.
Steinman shared the study of tumor necrosis factor (TNF) blockers for treatment of autoimmune diseases. One approach does not suit all autoimmiune diseases, he noted. It is known that blockade of TNF is effective in treating about 70 percent of patients with rheumatologic diseases (e.g., rheumatoid arthritis, psoriasis, inflammatory bowel disease), while it is not effective for treatment of multiple sclerosis and can even
exacerbate the disease in some patients (Steinman et al., 2012). Early research in this area, however, illustrates the impact of experimental design on outcomes.
Based on data showing an association between elevated levels of TNF in spinal fluid and disease progression in multiple sclerosis (e.g., Sharief and Hentges, 1991), it was thought that TNF blockade might have therapeutic value. In pursuit of this hypothesis, two groups published studies with very different conclusions. Feldmann and colleagues demonstrated control of established EAE in mice by inhibition of TNF using a monoclonal antibody (Baker et al., 1994). Another group, however, induced EAE in animals with a disruption in the TNF gene, suggesting that TNF was not essential for development of demyelinating lesions. They also found that TNF treatment reduced the severity of EAE in the animals (Liu et al., 1998).
Steinman explained that a potentially critical difference between the experiments was that the Feldman group used complete Freund’s adjuvant in induction of EAE while the Bernard group used adoptive transfer. As it turns out, Freund’s adjuvant induces production of TNF-alpha and causes leakage of the blood-brain barrier (Müssener et al., 1995; Rabchevsky et al., 1999). Steinman suggested that what Feldmann and colleagues may have seen was an amelioration of EAE due to the TNF-enhancing effect of the complete Freund’s adjuvant. Both of these studies were “EAE experiments,” Steinman pointed out; however, the results were completely opposite.
Around the same time, clinical studies of TNF blockade in humans with multiple sclerosis also showed exacerbation of disease instead of reduction of lesions (Lenercept, 1999; van Oosten et al., 1996). Steinman added that the TNF antagonist, Enbrel, now carries a label warning about the increased risk of demyelination.
Steinman concluded that an appropriate animal model should have a strong link to human disease. It is also important to avoid inclusion of any unnecessary steps or components, such as Freund’s adjuvant. EAE can be a good model system. However, there are many different EAE models and Steinman noted that more than 10,000 publications on EAE are listed in PubMed.
The bigger issue, Steinman suggested, is not animal models, but how to facilitate faster, less expensive trials in humans. Animal models will
always be imperfect, Steinman opined. Even successful use of a model cannot predict all possible outcomes, such as the risk of PML with natalizumab treatment. Steinman went on to state that human clinical trials will ultimately guide therapy. He added that many approved drugs could be repurposed, but finding someone to support and conduct the trials can be challenging. Koroshetz and another participant noted that the National Institute of Neurological Disorders and Stroke has a Network of Excellence in Neuroscience Clinical Trials that was set up to facilitate testing of new therapies in patients with neurological disorders. Academic investigators, industry, advocacy groups, and others with a novel therapeutic can apply to conduct a study within the network.1
IMPROVING BIDIRECTIONAL TRANSLATION FOR NERVOUS SYSTEM DISORDERS
Expanding on the previous discussion by Swerdlow, Michela Gallagher, professor of psychology and neuroscience at Johns Hopkins University, agreed that prepulse inhibition is a good example of corresponding endpoints across species and can facilitate the study of neuronal circuitry. There are new in vivo imaging tools that provide information about the functional position of networks in the normal state and in disease states.
Gallagher described her work with an aged rat model of memory loss. Gallagher noted that although this is referred to as an animal model of aging, it is important to understand that it is not a surrogate of aging like some other models; it actually is aging. Studying the hippocampal circuitry using this model predicts circuit overactivity and its localization. Gallagher noted that in this model, there is no neuronal loss and the numbers of synapses are maintained in old animals with memory loss. The hippocampal circuit most affected in terms of integrity of synaptic connections is the entorhinal cortex layer 2, which sends input to the dentate gyrus and CA3. Electrophysiological recording experiments showed that CA3 pyramidal neurons have elevated firing rates in this animal model. In animals with memory loss, those neurons fail to encode new information (Wilson et al., 2003). In this condition, overactivity is a sign of dysfunction.
In the hippocampus of an aged brain, it is predicted that these neurons encode less distinctive representations when animals experience
overlapping elements, a property referred to as pattern separation or the encoding of a new environment so it has little overlapping property. Because the majority of the synaptic inputs to CA3 neurons come from auto-associative networks,2 this elevated activity drives a complementary process referred to as pattern completion. Cognitively, Gallagher explained, there is a shift from pattern separation to pattern completion, which does not encode something new, but rather retrieves something old.
This was observed empirically in aged animals with memory loss by recording an ensemble of neurons in the CA3 circuit. The recordings show spatially localized neuronal firing in a familiar environment. When young animals are moved to a new environment, they exhibit the phenomenon of pattern separation, either in terms of firing rates or neurons involved in that representation. This distinguishes one episode and environment distinctively from another. In aged rats with memory loss that are presented with a new environment, there is a failure of CA3 neurons to rapidly encode a new representation. Aged rats with normal spatial memory have CA3 neurons that encode new information comparable to young animals.
Gallagher explained that there are modalities for testing people that can also capture this kind of pattern-separation/pattern-completion process. One example is running a recognition task where patients identify a visual stimulus as old, new, or similar but not identical to something already seen, while undergoing functional MRI (Bakker et al., 2008). A correct response of “similar” reflects pattern-separation ability, while an incorrect response that the stimulus is “old” indicates pattern completion. Using this approach, Bakker et al. (2012) provide evidence for the role for the human dentate gyrus/CA3 region in pattern separation. High-resolution neuroimaging tools have shown that hippocampal overactivity in patients with mild cognitive impairment is isolated to the human dentate gyrus/CA3 region (Yassa et al., 2010).
2Autoassociative neural networks are feedforward nets trained to produce an approximation of the identity mapping between network inputs and outputs using backpropagation or similar learning procedures (Kramer, 1992).
In the aged animal model, the functional contribution of hippocampal overactivity is studied further by using a variety of treatments to try to lower activity (e.g., viral transfection of an inhibitory peptide, drugs). When overactivity in the CA3 region was reduced, the performance of the network improved, Gallagher explained. In the process, it was found that an atypical antiepileptic, levetiracetam, could restore behavioral performance and network function in aged animals with memory loss. Levetiracetam preferentially reduces the activity of neurons that are in burst-firing mode, and when old animals have increased firing rates, they generate more spikes per burst.
Gallagher and colleagues took this finding forward into a human study and found that a subclinical dose of levetiracetam reduced hippocampal overactivation and improved task-dependent memory performance in patients with mild cognitive impairment (Bakker et al., 2012). Therefore, it would appear that hippocampal overactivity is a condition of network dysfunction, not a compensatory beneficial recruitment of resources in the hippocampus.
Further experiments are needed, Gallagher said, and they will be complex. Analysis of data sets from the Alzherimer’s Disease Neuroimaging Initiative (ADNI) and other studies indicate that the degree of increased activation of the hippocampus predicts subsequent cognitive decline, and can predict conversion to Alzheimer’s disease (Putcha et al., 2011). An unanswered question is whether there is any correlation or causal relationship between the loss of the layer 2 entorhinal cortex neurons that occurs in prodromal Alzheimer’s disease and hippocampal overactivation, as these neurons form the input pathway to the dentate gyrus and the CA3 region. Analysis of ADNI data on cortical thickness of the entorhinal cortex shows that during mild cognitive impairment there is ongoing thinning of the entorhinal cortex that might represent the neurodegeneration that has been seen in autopsy.
In conclusion, Gallagher noted that this is just one example of how neuroimaging tools have allowed us to understand that the functional components of these circuits are very similar in their core functions across animal models and humans.