Opinion on the Use of Animal Models in Nonclinical Safety Assessment: Pros and Cons

Abstract

Nonclinical evaluation of human safety risks for new chemical entities (NCEs) is primarily conducted in conventional healthy animals (CHAs); however, in certain instances, animal models of diseases (AMDs) can play a critical role in the understanding of human health risks. Animal models of diseases may be especially important when there is a need to understand how disease conditions associated with the intended indication might impact risk assessment of NCEs or when CHAs lack the human-specific target of interest (receptor, etc). Although AMDs have potential benefits over CHAs, they also have limitations. Understanding these limitations and optimizing the AMDs of interest should be done prior to proceeding with studies that will guide development of NCE. The purpose of this manuscript is to provide an overview of the major pros and cons of utilization of AMDs in nonclinical safety assessment.

Introduction

Laboratory animals serve as an initial test platform to simulate human physiology by eliciting a biological response to the drug under development. The data from these nonclinical studies are used to initiate early phase human clinical trials and support subsequent nonclinical studies to characterize new clinical entity (NCE)-related effects on toxicity end points to further guide product development. Safety and efficacy data from nonclinical studies utilizing conventional (“not genetically, surgically, or chemically altered to produce disease”) healthy animals (CHAs) are an integral component of drug development. Conventional healthy animals are intended to support product development by helping to address regulatory safety concerns while evaluating the proof-of-concept/efficacy of the product in humans. Studies with CHAs are intended to provide information on the biological plausibility, feasibility and safety of the intended clinical route of administration, and identification of physiologic and biochemical parameters to guide clinical monitoring, as well as to identifying potential human risks. ¹

Animal models of diseases (AMDs) are often used as tools to understand disease progression and proof-of-concept efficacy testing in the early discovery phase for many pharmaceuticals under development and may also be helpful adjuncts in nonclinical safety assessment in the pharmaceutical industry. ² A survey of pharmaceutical studies investigating the use and utility of AMDs for nonclinical safety assessments was conducted in 2017. ³ The survey revealed that the most frequent use of AMDs for safety assessment was in the discovery phase and included evaluation of potential safety concerns prior to the conduct of toxicology studies in CHAs, investigations to better understand toxicities associated with exaggerated pharmacology in CHAs or to better understand issues when the target is expressed only in the disease state. Less frequently, AMDs were utilized in the development phase, being used primarily to investigate nonclinical safety issues associated with targets expressed only in the AMD and/or in response to requests from regulatory agencies. ³

Animal models of diseases are also used for efficacy testing during the development of vaccines and therapeutics that are intended to treat or prevent life-threatening conditions caused by exposure to chemical, biological, radiological, or nuclear substances, where human trials are not ethical and field trials are not feasible. The US Food and Drug Administration has put forth guidance documents for the qualification of AMD, which can be used for efficacy testing in drug development programs for NCEs for some targeted diseases or conditions. This review will focus on the use of AMDs during nonclinical safety assessment and is not intended to provide an overview of use of AMDs for efficacy testing.

Animal models of disease may be naturally occurring models of the human disease, may be developed using targeted genetic manipulation, or induced by chemical or surgical intervention. The ideal AMD should mimic complex human disease with multiple contributory factors such as genetics, nutritional status, and environmental influences. Although potentially providing benefit over CHAs, no AMD is expected to be a substitute for the complexities of the human body but it is expected at a minimum to replicate certain aspects of the disease under investigation. ⁴ Fortunately, for some conditions there are multiple different types of AMD available to aid in the modelling of complex human diseases. For example, in the case of osteoarthritis, models including naturally occurring, genetically modified, surgically induced, chemically induced, and noninvasive induction are available to aid research. However, while choosing an appropriate model, consideration needs to be given to the underlying intent/rationale for the investigation (eg, determination of osteoarthritis pathophysiology, progression or therapy; pain behavior; or post-traumatic osteoarthritis studies) ⁴ as well as the timing of NCE administration relative to the onset of disease.

Nonclinical evaluation in an AMD may provide information that supports the identification of appropriate doses to use in initial human clinical studies, the identification of target organs, the determination of reversibility of effects, and the identification of safety parameters for clinical monitoring while also providing insight into potential mechanisms of nonclinical and/or clinical toxicity. ¹ Animal models of disease can also help define the risk–benefit ratio associated with NCEs and provide the opportunity for identification of biomarkers, which may be useful during clinical trial monitoring. When AMDs are utilized for safety assessment, evaluations/studies are hypothesis driven and typically supplement the evaluation performed in CHA by providing insight into the relationships of dose to biological response and toxicity.

When selecting an AMD, it is important to determine whether it is fit for the purpose and an appropriate model to address the question. High priority should be given to selecting an AMD that represents human anatomy and physiology for the disease paradigm as closely as possible. Each AMD will have inherent strengths and weaknesses, which should be well-characterized and will need to be described in the regulatory narrative. This review provides additional detail beyond a previous manuscript which provided an overview of recommendations/best practices for the use of AMDs. ³ Additionally, some of the in vivo models currently in use in nonclinical safety assessment, their relevance in the context of generated results and applicability, and the pros and cons of using various AMDs in different nonclinical settings are outlined. A detailed description on the pros and cons of the use of AMDs for the wide range of therapeutic indications is beyond the scope of the current manuscript, however, general considerations for pros and cons for the uses of AMDs will be outlined.

General Basis of Use and Advantages (Pros) of Using AMDs

Animal models of diseases are frequently used to facilitate understanding of disease pathogenesis. They help to establish nonclinical proof-of-concept with respect to target validation and efficacy in drug discovery programs. Traditionally, the use of these models in nonclinical safety assessment is limited to specific hypothesis-driven studies. In recent years, regulatory agencies have accepted and, in some cases, encouraged the use of AMDs in support of drug development programs. ^2,5 For example, when the target or substrate is only present in the disease state, toxicity studies may be performed in AMD(s) as an alternative to CHA(s). ^{1,6 –10} Therapeutics that target inborn errors of metabolism, specific tumors or viral antigens that are absent in the CHAs are good examples of such applications in safety assessment. Human diseases are characterized by complex and multifactorial pathogeneses; hence the most appropriate AMD will be one which is well characterized with a detailed mechanistic understanding of the concordant and discordant features of the model compared to the human disease.

In instances in which the target is only present in the disease state, AMDs may be more predictive of potential adverse events in humans than CHAs. This scenario is best exemplified by the use of acid sphingomyelinase (ASM) knock-out mice in the nonclinical safety assessment of recombinant human acid sphingomyelinase (rhASM). ¹¹ Hereditary deficiency of ASM, a lysosomal enzyme required for catabolism of sphingomyelin to ceramide and phosphorylcholine, results in types A and B Niemann-Pick (NP) disease in humans, in which sphingomyelin accumulates in lysosomes in the spleen, liver, lungs, bone marrow, and brain. ^{12 –14} The ASM knockout mouse lacks ASM activity from birth and exhibits tissue accumulation of sphingomyelin and progressive neurologic disease, similar to type A NP disease, with death in mice from neurologic disease between 6 and 8 months of age. ¹⁵ In nonclinical toxicity and safety pharmacology studies in conventional healthy Sprague-Dawley rats, beagle dogs, and/or cynomolgus monkeys to assess rhASM as a therapeutic for NP disease, rhASM was well-tolerated at intravenous (IV) rhASM doses up to 30 mg/kg with no clinical signs or functional changes. In contrast, ASM knockout mice exhibited dose-dependent toxicity and moribundity within 6 hours following a single IV dose of rhASM at 30 mg/kg and within 12 to 24 hours at 20 mg/kg. The ASM knockout mice also exhibited elevations in plasma ceramide levels and dose-dependent increases in serum proinflammatory cytokines after dosing with rhASM. Additional studies directly comparing rhASM toxicity between healthy wild-type C57Bl/6 and ASM knockout mice after a single IV dose of rhASM at 0.03 to 10 mg/kg, resulted in death or euthanasia due to a shock-like syndrome in ASM knockout mice given 10 mg/kg, while their wild-type counterparts exhibited no clinical signs. Toxicity of rhASM in ASM knockout mice was prevented by administering low doses of rhASM (3 mg/kg) over several days prior to repeated administration of a higher, previously toxic dose (20 mg/kg) of rhASM. The toxicity observed in ASM knockout mice was attributed to the rapid breakdown of sphingomyelin into ceramide and/or other metabolites with cardiovascular and proinflammatory effects. ¹¹ In this example, the AMD was more sensitive than CHAs to detection of toxicity relevant to the proposed human patient population and elucidated a gradual dose escalation strategy to mitigate the toxicity. ¹⁶

Utilizing CHAs to investigate the potential toxicity (or therapeutic benefit) of a compound that is intended to alter or ameliorate abnormal physiology may also be particularly problematic. For example, the use of antihypertensive agents in normotensive animals may result in significant adverse effects at therapeutic doses of the test article. Although beyond the scope of this manuscript, a recent review article provides an excellent overview of the potential benefit of the use of animal models of hypertension in the elucidation of the understanding of the pathophysiology, genetic mechanisms, identification of new disease markers, and potential therapeutic targets as it pertains to hypertension. A variety of animal models of hypertension are available, including genetic models and induced models, as well as models that allow the exploration of the role of obesity in the development and progression of hypertension. Studies in multiple AMDs are recommended by the authors based on the often multifactorial and incompletely understood pathogenesis of hypertension in humans as well as potentially general (eg, drug pharmacokinetics) and specific (eg, amino acid sequence of renin, especially in rodents) differences between animal models and humans. ¹⁷ As is generally the case with animal models, the potential benefit of the use of animal model(s) of hypertension will be highly dependent upon optimal model selection as well as study design, implementation, and interpretation of results.

Due to the wide variability of phenotypic manifestations of human disease and the frequent presence of comorbidities (eg, adverse finding that may be seen as an expected coexisting condition), it may be challenging to determine whether a therapeutic agent plays a role in the development of adverse effects that can be seen as an expected comorbidity of a disease entity. Animal models of disease may help to determine whether or not adverse findings noted in a clinical setting are related to administration of a therapeutic agent or are instead more likely related to the pre-existing or expected disease state/comorbidity in the patient population. An example is the identification of pancreatitis in diabetic patients treated with glucagon-like peptide-1 (GLP-1) receptor agonists. ^{18 –20} Since type 2 diabetes and/or obesity are risk factors for the development of pancreatitis, the relationship of the observed pancreatitis to administration of the GLP-1 receptor agonist(s) was considered to be uncertain. Initial testing in CHAs resulted in contradictory data, making it unclear what role GLP-1 agonists played in the development of pancreatitis. In order to understand this relationship better, a robust study was conducted in normal and diabetic rodents with a specific GLP-1 receptor agonist. Both conventional and diabetic rodents received the GLP-1 receptor agonist in the absence and presence of experimentally induced pancreatitis (2 models of chemically induced pancreatitis). End points included evaluation of plasma lipase, amylase, and inflammatory mediators as well as histopathologic evaluation of the pancreas. Analysis revealed a lack of relationship between the administration of GLP-1 receptor agonists and an increased risk of pancreatitis and instead it actually revealed a partial amelioration of chemically induced pancreatitis in both normal and diabetic rodents. ²¹

Another example of an AMD utilized for retrospective assessment of the potential risk or significance of an NCE to a clinical finding includes evaluation of the relationship of a transporter (uridine 5′-diphosphoglucuronosyltransferse [UGT1A1]) to the development of indirect hyperbilirubinemia. The Gunn rat, being deficient in all members of the UGT1A family of isozymes, may serve as a useful model to determine whether or not the mechanism of clinically noted indirect hyperbilirubinemia may be related to an effect on UGT1A1. ²²

In some instances, AMDs may enable better characterization of underlying risk factors that could potentially result in particular human populations being at increased risk of developing adverse effects associated with the administration of NCEs. For example, spontaneously hypertensive rats (SHR) have been shown to be more sensitive to doxorubicin-induced cardiac toxicity than normotensive Wistar-Kyoto rats (WKY), suggesting that hypertension plays a critical role in the toxicity and that hypertensive populations may have an increased risk for cardiac toxicity. ²³ Similarly, the spontaneous hypertension-heart failure (SHHF) rat provided additional insight into the potential pathogenesis of differential sensitivity to doxorubicin toxicity. As expected, the SHR model exhibited greater cardiac and renal histologic lesions as well as mortality as compared to the WKY model. Surprisingly, while the SHHF rats had greater histologic renal and cardiac microscopic lesions as compared to the WKY rats, the microscopic cardiac lesions were less pronounced than those noted in the SHR model. The finding of less severe cardiac toxicity in a model that was both hypertensive and genetically predisposed to heart failure was unexpected and was attributed to potential strain differences in arachidonic acid metabolism (increased leukotriene D4 in SHR rats but not SHHF rats) after treatment with doxorubicin. ²⁴ Thus, use of AMDs may assist in determining the effect of a compound in human populations with pre-existing disease (eg, hypertension) and may also help to understand some of the reasons behind differences in sensitivity within populations.

Certain AMDs can also help distinguish on-target and off-target effects. Diabetic patients undergoing intensive insulin treatment often develop hypoglycemic neuropathy, which is a leading complication of diabetes mellitus. In 2010, Ozaki et al demonstrated severe hypoglycemic peripheral neuropathy in spontaneously diabetic Wistar Bonn Kobori rats that were treated with intermittent insulin via insulin implants. ²⁵ In this rat strain, the insulin-associated hypoglycemic lesions were characterized by axonopathy followed by severe breakdown of the myelin sheath, whereas the hyperglycemic lesions were characterized by segmental demyelination, making this model suitable for morphologic distinction of hypoglycemic peripheral neuropathy from hyperglycemic peripheral neuropathy. Additionally, for animals with the same plasma insulin concentration, hypoglycemic lesions were identified in peripheral nerves of animals that were hypoglycemic to slightly hyperglycemic, but not in animals that were normoglycemic to slightly hyperglycemic throughout the course of the study. These data demonstrate the utility of the AMD in distinguishing peripheral nerve pathology induced by variations in blood glucose level from a direct effect of insulin on peripheral nerves, as normoglycemia would be difficult to achieve in an insulin-treated nondiabetic animal.

Disadvantages (Cons) of Using AMDs

Although AMDs often provide advantages during safety assessment, there are disadvantages associated with their use in some cases. Some of the potential limitations of AMDs include limited historical control data, technical limitations with respect to the anatomy and physiology of the animal model, animal breeding and care issues, especially for genetically altered animals, and limitations in replicating human disease conditions. ²⁶

Additionally, in some instances AMDs may mask the development of toxicities that are apparent in CHAs such as that which occurs with the use of adenoassociated virus (AAV) gene therapies to treat Tay-Sachs disease (TSD) and Sandhoff disease (SD), which are lysosomal storage diseases resulting from hexosaminidase deficiencies. In TSD, there is a deficiency of the α subunit hexosaminidase and in SD, there is a combined deficiency of the α- and β-hexosaminidase subunits. The resulting accumulation of gangliosides in the brain leads to neurologic signs and early death. Intracranial delivery of AAV gene therapy vectors encoding species-specific hexosaminidase subunits to transgenic mice or to naturally occurring feline models of SD or an ovine model of TSD all resulted in widespread distribution of hexosaminidase enzyme activity and therapeutic efficacy, without evidence of toxicity. ^6,27,28 In contrast, intracranial injection of AAV gene therapy vectors encoding species-specific hexosaminidase subunits in conventional healthy cynomolgus macaques resulted in the development of dose-dependent neurologic signs with microscopic evidence of neurotoxicity. The neurotoxicity in cynomolgus monkeys was attributed to overexpression of the hexosaminidase subunits in a healthy animal.

Other potential disadvantages of AMDs include confounding effects resulting from heterogeneous expression of disease. Naturally occurring hereditary diseases may have variable degrees of penetrance or disease expression may be influenced by environmental factors and chemically or surgically induced diseases may be induced in a nonuniform manner. Mitigation of these confounding factors often requires using a larger number of animals and careful attention to randomization prior to the initiation of therapeutic administration. The generation of sufficient numbers of animals for a given study may require a substantial investment of time and money and staggering of the study initiation may be required. Another potential disadvantage of using AMDs may be related to limitations in the timing and duration of treatment that can sometimes prevent recapitulation of the paradigm intended to be used in humans. The pharmacologic response to the therapeutic agent in the AMD may be overestimated if the treatment is initiated prior to the onset of overt clinical manifestations of disease.

Genetic background of the AMD is an additional important consideration when selecting models for nonclinical safety assessment. Currently, most AMDs are developed from inbred strains that have relatively small gene pools with unique phenotypic characteristics and genetic backgrounds. ¹ Knowledge of the genetic background is valuable in interpretation of the data. During evolution, homologous proteins do not always remain functionally equivalent at the species level. This leads to functional differences, which might emerge during the conduct of studies. A vast majority of human phenotypic conditions such as diabetes, obesity, or cardiovascular abnormalities are a manifestation of interaction between genes and environmental factors. ⁷ Incomplete backcrossing to a single strain in animals can lead to the development of a variable phenotype and in some instances may lead to loss of a phenotype or possibly strengthening of another. ²⁹ Simplification of these conditions in AMDs by introducing a single intervention against a background of complex variables creates bias. Nevertheless, it is beneficial to have a consistent phenotype regardless of the background strain to allow proper interpretation of the generated data/findings. Environmental factors can also have a profound impact on phenotypes. For example, quality of life and physical and mental health can have a direct impact on the outcome of a disease, which is hard to replicate in AMDs. Likewise in animal models of different genetic backgrounds, diet can have varying effects thus stringent measures need to be implemented to avoid variability. ⁷

Although an AMD may provide valuable insight into the pathogenesis of a disease, the information they provide on potential therapeutic benefit and/or toxicity as well as the reliability of this information is, of course, highly dependent upon the suitability of the model. For optimal interpretability and predictability, the AMD should reliably exhibit all critical manifestations of the disease relevant to the investigation. ^2,3,26 Selection of the animal model with respect to suitability is essential and investigators should ensure that the model and the design of the study are carefully planned. For example, while there are numerous mouse models of Alzheimer disease that have been utilized in the identification of therapeutic agents, the various models have some but not all the histologic features observed in humans. ^30,31 It is imperative that the specific model is carefully selected with the desired end points kept in mind as some models develop β-amyloid peptide deposits while others do not, neural loss is not typically encountered, and none develop neurofibrillary tangles. ^32,33 Thus, not all aspects of the human disease can be recapitulated, which may confound prediction of potentially beneficial therapeutic agents.

Conclusion

Animal models of disease continue to play a crucial role in characterizing the potential efficacy of NCEs. Increasingly, AMDs are being used in early discovery and in later development studies to elucidate risks in humans to characterize and ultimately support regulatory acceptance of AMDs. Although CHA continues to be a critical component required for many regulatory submissions, in some instance the AMD alone may be suitable. The greatest utility of AMDs in drug development is in targeted or hypothesis-driven studies that help to identify the mechanism of toxicity or to evaluate the potential for target-related toxicities in instances where the target of interest is not present in CHAs.

Although utilization of an AMD in research may provide distinct advantages over CHAs in some instances, it is clear that no one model is “perfect,” and careful consideration is required concerning selection of the model to obtain the desired result. Other critical factors to consider prior to selection of an AMD include rigorous characterization of the AMD and assurance that adequate similarity exists within the model with respect to the human disease of interest.