Improving productivity of modern-day drug discovery

Abstract

The pharmaceutical industry is confronted by increasing costs of clinical development and diminishing productivity. The most challenging aspect of drug development has been the failure of therapeutics in expensive Phase II or III trials, and this is most commonly due to lack of efficacy. More can be done during the drug discovery phase to optimize efficacy-testing in animal models by expending resources to explore the congruence of the animal model with the human disease. Historically, relatively little attention has been paid to validation of these models, but access to molecular mRNA and genetic profiling offers a new lens through which the similarity of these disease models to human diseases can be examined and their utility for exploring therapeutic efficacy can be optimized. Exploring congruent experimental end points in clinical and preclinical experiments will also increase confidence of success in late phase clinical development. The expense of this investment is trivial compared to the costs of a failed clinical trial, more than justifying this endeavor.

Keywords::

• animal model
• clinical trial
• experimental design
• human disease

At a time when our mechanistic understanding of disease pathophysiology has never seen more rapid advances, progress in drug discovery and development has slowed, and launching a new drug has become progressively more difficult and expensive [1]. While the number of new molecular entities launched per year has gradually decreased [2], over the past six decades, the cost of launching a drug has doubled every 9 years in inflation adjusted dollars [3]. While many drugs fail preclinically (usually for toxicity), the most costly stage for failure is Phase II or III, due to lack of efficacy which approaches 50% [4,5]. Since demonstration of efficacy in an animal disease model is a typical prerequisite for advancing a molecule into clinical testing, these clinical failures have been attributed to incongruities between animal and human biology and disease pathogenesis [6,7]. However, animal models have proved predictive in many cases. An analysis of 76 targets published in highly cited journals, showed that 28 studies (37%) yielded concordant clinical results, while 14 failed and 34 were not clinically tested [8]. While much can be done to improve the clinical phase of drug development, this commentary will focus on the improvements that can be implemented in the drug discovery phase which could mitigate costly efficacy failures in the clinic.

Inattention to animal disease model characterization represents a significant gap in many preclinical drug development programs. In part, this gap may be viewed as a predictable consequence of new emerging diseases and shifting spectrum of disease afflicting humanity requiring establishment of new models. Over the past half century, the spectrum of disease has shifted away from bacterial infections and acute illnesses, and has shifted toward chronic diseases, including cancer, Alzheimer's disease, chronic obstructive pulmonary disease (COPD), AIDS, ischemic heart disease and kidney failure [9]. This shift has required de-emphasis of more expeditious pharmacologic assays used to assess bactericidal activity, smooth muscle relaxation and blood pressure reduction, toward longer efficacy models, including cancer xenografts, models of neurodegeneration, kidney disease and heart failure. While animal models for chronic diseases have been developed, there has been scant information comparing disease models head-to-head, so as to inform which model is best used for a given human disease.

A critical step in animal model development is establishing a deep pathophysiological understanding of human disease to be modeled. For example, today, breast cancer is no longer thought of as a monolithic entity but is separated into distinct treatment-sensitive subtypes, including HER2⁺, ER⁺, PR⁺ and triple negative. It is axiomatic that drug testing in one breast cancer model would not necessarily be applicable to all others [10]. It is likely that many drug failures not only in oncology but also in diabetes, psychiatric disease or inflammation derive from an inadequate understanding of the diverse pathophysiology of human disease. Significant resources should be devoted to characterizing unique environmental and genetic factors contributing to disease complexity. Key to characterization of human disease is the ability to access diseased human tissue and the ability to link early molecular and histopathological features of these samples to the subsequent clinical course. Identification of disease endotypes among patients needs to be expanded to inform pathophysiological heterogeneity as well as heterogeneity with an individual patient as suggested for COPD and renal cell carcinoma [11,12]. While there are many sources of human biological samples and tissues, most of these are obtained at death, providing only cross-sectional clinical information [13,14]. Still, even available cross-sectional information is generally poorly annotated with only the most superficial clinical information provided (e.g., age, gender and major diagnoses). Invaluable information could be obtained through biological samples linked to subsequent disease course. Similarly, genetic information from tissue samples obtained at biopsy or surgery (e.g., unaffected tissues at tumor margins) is typically annotated only with clinical information obtained at the time of sampling [15,16]. Information obtained by linking this sample information to subsequent clinical course by utilizing electronic health records would provide invaluable information for biomarker discovery and disease heterogeneity regarding rates of disease progression and outcomes. Only through optimizing our understanding of clinical disease to be modeled can an appropriate animal model of the disease be identified.

It would be naïve to think that the complexities of a chronic disease in the diverse human population could be accurately captured by a single animal model of disease. Given our current limited understanding of human disease, a deeper understanding of similarities and differences between animal model systems and human disease would substantially facilitate drug development. While the vast majority of genes are shared between human and model mammalian species, it is also clear that there are subsets of genes, especially those with roles in reproduction and immunity that are significantly different [17,18]. For example, there are at least 3767 murine genes for which no human ortholog exists, mainly involved in reproduction [17,19]. Similarly, natural killer (NK) cell function is regulated by non-orthologous NK receptor families in mouse versus humans [20], contributing to differences between murine and human immune functions. Thus, modeling autoimmune diseases or diseases of the reproductive system in mouse is particularly problematic since the genes comprising these systems exhibit low orthology [6,7]. Polymorphisms in the ApoL1 gene have been strongly associated with kidney disease and trypanosome immunity in humans, but the ApoL family is completely absent in rodents or other mammalian orders [21,22], making the mouse an inappropriate organism for studying these diseases. Nevertheless, recent studies comparing the molecular signatures of tissues in mouse models reveal areas of similarity between human and murine inflammation, Alzheimer's disease and diabetic kidney disease [7,15,23]. Messenger RNA profiling provides a valuable approach to elucidate overlapping molecular pathways in human disease and mouse models, thereby informing the translational validity of therapeutic efficacy in a disease model.

In this era of genomic medicine, identification of human diseases caused by a single major gene defects has prompted the development of new animal models carrying homologous gene defects and has offered a new method of engineering animal models relevant to human disease. Even in this case, the complexities of human disease pathogenesis have stymied the development of relevant therapies for the human disease, as illustrated by the cases of sickle cell anemia and cystic fibrosis [24,25]. Nevertheless, humanized animal models of disease offer an exciting new means to develop disease models with direct relevance to human disease.

Although current animal disease models exhibit limited molecular similarity to human diseases, it is notable that different mouse models of diabetic kidney disease capture different aspects of human disease and only incompletely overlap with each other [15]. Identification of those pathways which overlap between humans and mice may inform which models are best used to test the benefit of a particular therapeutic intervention. Nevertheless, until the diversity in human disease is fully identified, and captured in disease models, it behooves one to establish efficacy of a therapeutic in multiple animal disease models (and species) before placing bet on testing for clinical efficacy.

Another hurdle to drug discovery is the common disconnect between animal model efficacy testing and human clinical trial outcomes due to incongruities in the experimental end point tested. Gaps are especially evident in studies of dementia, cardiovascular and renal diseases. Many preclinical studies of these models rely inordinately on histopathological changes or changes in biomarkers of the disease that are not accepted as end points for efficacy in clinical trials. For example, numerous drugs have shown efficacy in reducing cardiovascular or renal fibrosis; yet, fibrosis is not an accepted registration efficacy end point in Phase III trials and cannot be assessed without cardiac or renal biopsy. It is, therefore, disappointing to see anti-fibrotic therapies advance into the clinic before establishing a preclinical biomarker of efficacy that can be translated into the clinic [26,27]. Although there are evolving imaging techniques for tissue fibrosis that may be developed in the future, at present, improvement in fibrosis per se is not an acceptable trial end point for regulatory agencies. In the absence of a clinically translatable biomarker reflective of fibrosis, there is no clear path for testing efficacy of anti-fibrotic therapies, short of taking a ‘leap-of-faith’ that their anti-fibrotic effects will provide a previously unestablished benefit in human disease. Identification of biomarkers of fibrosis in animal models that translate into the clinic will be critical to develop anti-fibrotic therapies.

Similarly, gaps exist for development of anticancer therapies in animal cancer models. Recent cancer models are typically developed to study cells specific for commonly occurring clinical cancers by using subcutaneous patient-derived xenografts. While an advance over other models, they neglect the organ tissue microenvironment that might be critical tumor growth including the host microvasculature. This might be tested using the more laborious, but likely more relevant models examining tumor growth in situ or implantation in orthotopic sites known to be predisposed to metastatic spread in the clinic [28-30].

At no time has the opportunity for discovering treatments for human afflictions been greater. While there is much to be learned, animal models of disease remain a cornerstone of drug discovery. Through improved characterization, validation and experimental design, animal models should better inform the efficacy of therapeutics in the clinic and thereby substantially decrease costly efficacy failure during clinical development.

Declaration of interest

The authors are employees of Eli Lilly and this paper has been funded by Eli Lilly and Co.