To define the principles and illustrate the complexity of de-risking targets in the absence of predictive animal models of disease, workshop participants explored two case studies—Parkinson’s disease (PD) and schizophrenia. In the first of these cases, strong genetic evidence exists linking a target to PD, said Stevin Zorn, yet the tools for translating this evidence into treatments has been limited. In schizophrenia, the genetic evidence is less developed, said Steven McCarroll, increasing the difficulty of creating animal models.
PARKINSON’S DISEASE
Todd Sherer, chief executive officer of The Michael J. Fox Foundation for Parkinson’s Research (MJFF), and Jan Egebjerg described recent efforts to develop effective therapeutics for PD by targeting the leucine-rich repeat kinase 2 (LRRK2) pathway. LRRK2, a protein kinase, was first linked to PD in 2004, when mutations in the LRRK2 gene were identified in a number of families with autosomal dominant PD (Paisan-Ruiz et al., 2004; Zimprich et al., 2004). According to Sherer, LRRK2 kinase inhibitors have since emerged as a leading strategy to treat PD, based largely on a wealth of genetic data linking LRRK2 mutations to familial forms of the disease, as well as broader genetic studies suggesting that genetic variants in LRRK2 increase the risk of sporadic PD. The fact that the genetic and sporadic forms of the disease appear indistinguishable by clinical and pathological phenotype lends credence to the idea that both forms of the disease may arise from common biological pathways, he said. Importantly, LRRK2 is a kinase, making it an attractive, druggable target for pharmaceutical companies, who are experts in making selective kinase inhibitors, said Sherer.
Because LRRK2 functions throughout the body, its inhibition has the potential to have undesirable effects (Herzig et al., 2011). Sherer said animal models have been very useful in examining potential safety issues. For example, a study in nonhuman primates uncovered a potential lung abnormality in animals treated with LRRK2 inhibitors (Fuji et al., 2015). Interestingly, said Sherer, rodent knock-out models showed similar findings, suggesting that this may be an LRRK2 target-related effect. These observations led to a collaboration among MJFF, Genentech, Pfizer, and Merck to explore this safety issue in primates using three different LRRK2 kinase inhibitors with different chemical scaffolds. At high doses, these compounds all produced the same lung problems, but some
were able to achieve full inhibition of LRRK2 at lower doses without affecting lung function, suggesting there may be a safe therapeutic window, said Sherer.
However, he noted there are no good preclinical LRRK2 models to demonstrate efficacy or to establish a therapeutic index or window. Egebjerg added that models are also needed to determine whether drugs can reach a sufficiently high level in the brain to achieve the desired level of inhibition. Multiple mouse models have been developed, including LRRK2 knock-outs, knock-ins, overexpressers, and other technologies, yet none of them show the classic signs of PD—dopamine degeneration and motor dysfunction. Despite the development of multiple transgenic models of PD, Egebjerg noted that there is no animal model that displays Parkinson’s symptomatology or the core pathology of PD, accumulation of the protein α-synuclein. Some of the models show dysfunction in locomotor activity or in striatal dopaminergic circuitry, he said, but these findings are not sufficiently robust for screening or dose finding in pharmaceutical development. Information from different models can be garnered to provide some valuable information, said Egebjerg.
Egebjerg and Sherer described several efforts under way to develop novel technologies for preclinical drug development. For example, MJFF has developed collaborations with investigators in academia and industry to try to develop pharmacodynamic markers of LRRK2 activity in human cells, which could enable experimental medicine approaches for early clinical studies. To determine whether the drug is getting into the brain, positron emission tomography ligands could be useful, said Egebjerg, although LRRK2 is a challenging target because it is a low-abundance protein that is affected by high levels of fat in the brain. Another new approach mentioned by Egebjerg as a potential measure of LRRK2 inhibition is the measurement of LRRK2 phosphorylation in cerebrospinal fluid exosomes (Williams et al., 2015).
According to Egebjerg, two additional questions need to be answered to advance LRRK2 inhibition as a therapy for PD: (1) at what stage in the disease is LRRK2 dysfunction critical, meaning, when is the appropriate time for intervention, and (2) what clinical outcome measure would be most sensitive to LRRK2 inhibition? These questions will require studies in human patients, including natural history studies, said Egebjerg. MJFF has established patient registries and begun cohort studies of patients carrying the LRRK2 mutation, as well as individuals at risk of developing PD, in preparation for future clinical trials and to identify biomarkers that may be relevant to LRRK2 function.
Given the tremendous heterogeneity among PD patients, there is also an increased focus on disease stratification and segmentation, which would enable better patient selection for clinical trials and treatment, said Egebjerg. He suggested the use of iPS cells as a translational tool to model disease heterogeneity. Sherer agreed and advocated for working back from the human condition—whether through genetics or the phenomenology of the clinical syndrome—by linking genes or clinical phenotypes to the biology, and then using model systems to make better sense of that information to move therapies forward.
SCHIZOPHRENIA
In PD, increased genetic understanding has been enabled by relatively high-penetrance mutations (e.g., LRRK2 mutations) that segregate for generations. However, other complex, heterogeneous, and highly heritable diseases that lack these segregating Mendelian forms, such as schizophrenia, have been more challenging to understand genetically, according to McCarroll. Nonetheless, he said, there have been some early victories toward revealing the genetic architecture of schizophrenia in humans through large association studies of tens of thousands of people. For example, a meta-analysis by the Psychiatric Genomics Consortium identified 108 loci associated with human schizophrenia (Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014).
While substantial progress has been achieved through genetic studies, they have yet to result in development of useful animal models of genetically complex disorders. McCarroll said that using these association findings to create animal models has been challenging for several reasons: (1) each of these loci individually contributes only small effects; (2) mutations found tend to be regulatory rather than protein coding; (3) none of the rare variants found thus far are fully penetrant, indicating that there are environmental and genetic modifiers; (4) in human patients, most of the large-effect variants found so far cause constellations of symptoms, making the interpretation in animals more challenging; and (5) some of the strongest genetic risk factors identified appear to reflect relatively recent evolutionary events and thus are not shared with species other than primates. Another challenge with animal models, raised by Roger Little, acting director of the Division of Neuroscience and Behavior at the National Institute on Drug Abuse, is that the genetic background of the animal can modify penetrance of the phenotype to such a
degree that it may be unclear whether the observed phenotype is the result of the mutation or the background. According to Niels Plath, one further complication that may limit the usefulness of genetic models is that common variants usually have a modest impact on disease status, whereas rare alleles often have a very high effect size. Research in polygenic disorders like schizophrenia therefore turns toward understanding the interaction of individual genetic risk factors on a network level rather than the impact of rare alleles alone.
McCarroll said that lower costs for whole exome and genome sequencing have begun to allow scaling toward large cohorts, which may enable identification of variants in specific genes more strongly associated with schizophrenia (Singh et al., 2016). However, so far the substantial genetic progress that has been made has not lent itself to new animal models. He said the most useful levels of analogy to mice and other animals may be for phenotypes at the molecular, cellular, and circuit levels rather than at the level of behavior or an assumed model for the disease itself.
Another roadblock to modeling schizophrenia is the substantial degree of heterogeneity, both etiological and phenotypical, said Plath. There will never be one animal model of schizophrenia, but there may be models for specific subtypes of schizophrenia, he added. These models may enable investigators to address phenotypic heterogeneity.
Copy number variants (CNVs)—stretches of DNA that are either deleted or duplicated—are a common type of genetic variation often found in schizophrenia, said Plath. Some are relatively rare, but have a high biological impact, and in some cases, they are evolutionarily well preserved with orthologues found in rodents. For example, one CNV associated with schizophrenia is found in the genetic locus 15q13.3. A knockout mouse model has been created, which recapitulates some phenotypic characteristics of schizophrenia (Fejgin et al., 2014). Plath emphasized that this represents a model of the 15q13.3 deletion, not a model of schizophrenia, and that other deletion mutants have also been identified. By identifying endophenotypes in mice that are also seen in humans and that can be linked to functional outcomes (e.g., impairments in electrical activity in the brain), investigators can begin to tease out mechanisms involved in schizophrenia and potentially identify therapeutic targets and biomarkers. If the endophenotypes are preserved across different CNV backgrounds, said Plath, this would increase confidence in the relevance of the mechanism.
There is additional need to tightly integrate research in animals with exploratory research in humans, noted Plath (see Figure 3-1). Models based on high-penetrance genetic variations (e.g., the 15q13.3 model), by identifying in vitro and in vivo phenotypes, could allow the identification of a laboratory assay for target identification, and potentially for higher throughput screening of compounds. These assays might also be used for validation studies, and eventually for clinical segmentation. Such assays, of course, require identifying intermediate phenotypes in the model system that are relevant to human disease.
TRANSLATING DISCOVERIES TO TREATMENTS
Frances Jensen applauded the use of animal models to deconstruct complex diseases into component parts, but noted that it may be overreaching to expect an animal model to provide a systems-level phenotypic mimic of human disease. Given that these component mechanisms
may play a role in multiple diseases and may have characteristic imaging or biochemical signals, she suggested that the field may want to start cataloging these component mechanisms to accelerate translation of discoveries into novel treatments across a range of nervous system disorders. McCarroll agreed, noting that although mice may not be useful to establish efficacy at a global level, they can be used to demonstrate how something acts on cells and circuits, or show basic interactions among different cell types. Egebjerg added that a similar type of deconstruction at the clinical level could also prove beneficial as a step toward identifying which patients might benefit from therapeutic approaches that target particular mechanisms. Frank Yocca concurred, noting that from a pharmaceutical industry perspective, there may be no choice but to find common components among very heterogeneous diseases.
A complication in the translation of genetic risks to new treatments for both schizophrenia and PD is that both disorders develop over many years, commented George Koob, director of the National Institute on Alcohol Abuse and Alcoholism. Therefore, clinically relevant targets may be present from the outset (i.e., a risk variant) or represent a downstream effect. McCarroll commented that whether a disease such as schizophrenia is truly neurodevelopmental (i.e., resulting from an early insult) remains unclear. For a genetic risk factor, such as LRRK2 in PD, acquiring the confidence needed to move forward in drug development will require a more refined understanding of the pathways from genetic variant to outcome, as well as read-outs that build a link to the biology, said Egebjerg.
Plath added that the possibility that a disease such as schizophrenia is neurodevelopmental also adds challenges with regard to testing an intervention in the clinic: If the disease has built up gradually over many years, is it likely that reversal could be seen in 12 weeks? If a treatment response is seen, is this merely symptomatic rather than finding the root cause of the disease? The question of whether these diseases are neurodevelopmental also heightens the need for more natural history studies to clarify disease trajectories and their correlation with multiple biomarkers, said Steven Hyman.
This page intentionally left blank.