Can quantitative pharmacology improve productivity in pharmaceutical research and development?

KEYWORDS:

• Translational biomarkers
• R&D productivity
• model-informed drug discovery and development (MID3)
• quantitative pharmacology
• modeling and simulation

1. Introduction

Average R&D productivity in the pharmaceutical industry is less than 5% from the conception of idea to registration. Despite an increase in FDA-approved new drugs in the last 5 years (247 between 2015 and 2019 versus 184 between 2010 and 2014), overall productivity remains low, driven by growing complexity in drug R&D and continued high attrition rates [1]. Lack of efficacy and unclear differentiation of novel therapies at doses that can safely be administered to patients are prominent reasons for clinical failure and contributors to low R&D productivity. Much expenditure can be avoided and resources deployed elsewhere if unworthy molecules or targets are terminated at an earlier point or directed toward patients that have the greatest likelihood to respond (i.e. personalized medicine).

Historically, pharmaceutical companies selected clinical candidates based on established proof of efficacy in animal models that recapitulate behaviors or pathophysiological features associated with human disease. However, in therapeutic areas like neuroscience and oncology, concerns are mounting over the inability of such models to predict clinical efficacy. Moreover, with the rapidly expanding immunomodulatory, gene, and cell therapies, targets are studied that are not expressed or have different functions in preclinical species. Also, in a recent review, Jardim et al. identified lack of a biomarker-driven clinical development strategy as one of the key risk factors leading to late-stage failure in oncology [2]. In response, companies have started to emphasize human disease biology and developed translational sciences in search for biomarkers that may offer an improved line of sight to the clinic. Thus, the research paradigm is shifting from establishing proof of efficacy in animal models to proof of pharmacological mechanism using biomarker(s). Translational biomarkers are measured preclinically and clinically. They also display a quantitative relationship in response to a drug which enables cross-system extrapolation (i.e. human dose prediction, patient populations, etc.) [3]. Translational biomarkers enable the application of quantitative mathematical modeling to characterize the kinetics of drug, target pharmacology, and impact on disease biology [3]. For DPP4 inhibitor sitagliptin, active GLP-1 and DPP4 activity were used as proximal biomarkers based on pharmacology of the DPP4 pathway. Preclinical data suggested that 80% target engagement would result in maximum efficacy. Strong biomarker results in Phase 1 studies combined with a favorable tolerability profile allowed confident, immediate progression of sitagliptin to Phase IIb, thereby saving significant time [4]. Biomarkers also enable root cause analyses for a failed trial. They help to distinguish failure due to the target not playing a pivotal role in the human disease biology versus insufficient quality of the molecule to test the full potential of pharmaceutical intervention directed toward the target.

Recent developments in genomics, transcriptomics, proteomics, and metabolomics have revolutionized biomarker research. It has unveiled complex biological pathways and identified new ways to study physiochemical and pharmacological changes associated with human disease progression and intervention. These platforms are also used to discover diagnostic biomarkers that identify patients that respond favorably to a therapy. This can be used to enrich clinical trials to better detect a therapeutic effect. For example, PD-1 checkpoint inhibitor pembrolizumab blocks the interaction of programmed death-ligand 1 (PD-L1) with its receptor PD-1 on activated immune cells. In clinical studies, PD-L1 expression was positively associated to pembrolizumab efficacy [5]. Moreover, it can be used to detect mutations in proteins that can be developed as novel targets such as is done for Kras [5]. Like translational biomarkers, this information can be integrated into quantitative models.

Over the last decades, the use of quantitative modeling in drug development has matured. Major regulatory agencies have published guidelines for submitting quantitative analyses as part of New Drug Application dossiers [6]. Quantitative modeling has become an integral part of modern drug development and is used for dose and schedule optimization, clinical trial design, to provide risk assessment for intrinsic (e.g. special populations) and extrinsic factors (e.g. DDI) and to support label claims for efficacy or safety. Moreover, as quantitative pharmacology models integrate understanding of pharmacokinetics, target pharmacology, disease biology, and physiology, they are used to interpret experimental results, generate hypotheses, and inter- or extrapolate to new scenarios. Nayak et al. recently highlighted impactful examples of using quantitative pharmacology in drug development [7]. One example worth highlighting is the application of quantitative modeling to support the approval of the 2 mg/kg dose of pembrolizumab for treatment of melanoma while this dose was not studied in the pivotal trial (Keynote-006). The selection of the dose was justified by an integrated analysis that included data from another clinical trial. This saved 6 months to 1 year on the overall timeline and made the drug available to patients earlier [7]. Traditionally, oncology drugs often have been tested at the maximum tolerated dose (MTD), but novel biological therapeutics show high specificity and maximum efficacy may be obtained at lower doses than maximum tolerated. Moreover, the MTD approach can be intractable for immunostimulatory drugs and several other novel therapies. Integration of translational biomarkers and quantitative modeling enabled the research paradigm to shift from the MTD to the pharmacologically active dose that achieved the desired target engagement and downstream pharmacology to inform dose selection for immuno-oncology drugs [8].

Physiologically based pharmacokinetic (PBPK) models are grounded in human physiology and capture physiological attributes that govern drug absorption, distribution, and elimination, including organ blood flows, body size, organ weights, expression level of drug-metabolizing enzymes and transporters. Observed clinical pharmacokinetics may be simulated, when combined with physicochemical properties and in-vitro disposition characteristics of a specific drug. These mechanistic models can be used to prospectively predict changes in drug pharmacokinetics and pharmacodynamics related to the impact of genetic polymorphisms, concomitant medication that interferes with distribution or elimination, ethnicity, ontogeny, organ impairment, and more. Such information is commonly used to support decisions concerning dose selection and adjustments, risk assessment and trial deferral, study design, and more recently to support label claims [9]. To illustrate, letermovir, used to prevent cytomegalovirus infection in bone marrow transplant recipients, has a complex DDI profile and inhibits, among others, cytochrome P450 CYP2C8. Using a qualified PBPK model to simulate the interaction between letermovir and repaglinide, it was shown that letermovir modestly increased the exposure of sensitive CYP2C8 substrates [10]. These results were used to inform the US prescribing information, obviating a dedicated DDI study.

Recently, there is notable interest in applying quantitative model-informed approaches earlier and more consistently in the discovery process. Applications include selection of the most appropriate clinical candidate, prediction of active human dose, clinical utility and differentiation, preclinical study design, target properties and impact of formulation, route of administration, and more. Early quantitative integration of data offers scientists insight in specific molecules of interest and an opportunity to avoid incurring cost if the attributes of the asset are projected to be insufficient to advance to clinical proof of concept. By applying quantitative reverse translation, prior clinical experience can be leveraged to determine the key distinguishing feature that is required for a new drug to be successful. Equally important, quantitative integration also offers learnings on the biological system that is being studied. As different molecules are being studied, their pharmacological signatures may provide refined or new insight into underlying biology. The rapid emergence of sophisticated, multiscale quantitative systems pharmacology (QSP) models that maximize integration of knowledge within or across biological systems, is a testament to this strategy. With advancements in chemistry, protein sciences, and molecular engineering, we are now better equipped than ever before to design any specific, or a combination of, property(ies) in a molecule to interact in a specific fashion with its target(s). Quantitative pharmacology frameworks ought to be utilized to help define which key design features are required to create novel molecules with the right intrinsic pharmacological and pharmacokinetic profile for a low-dose and convenient dosing schedule [11]. This includes consideration of different drug modalities, such as small molecules, peptides, antibodies, proteins, oligonucleotides, etc. Application of quantitative modeling is not limited to characterize and translate therapeutic benefit but has also been successfully applied to drug-related toxicities. Venkatakrishnan et al. describe various quantitative approaches to characterize exposure–toxicity relationship for common drug-related toxicities in oncology like myelosuppression and QT prolongation, with successful translation from the preclinical to clinical setting [12]. Clinical success of any drug is tied to its benefit over risk balance and application of quantitative modeling to characterize it can help guide clinical development. Bottino et al. integrated the preclinical exposure–response relationship for combination therapy of a Pi3Kα and a TORC1/2 inhibitor with clinical exposure–toxicity relationship and predicted superior therapeutic benefit of monotherapy over combination therapy at tolerated doses [13]. This was later confirmed in a clinical trial [13].

Implementation of model-based strategies early in discovery also uncovers gaps in translational approaches and lack in understanding of biology. When recognized early, organizations have an opportunity to make directed investments to mitigate perceived gaps and associated higher risk to downstream failure [3]. Even when unaddressed, it allows executive leadership to incorporate this elevated risk in its decision whether to proceed with a program.

In summary, biomarkers and quantitative modeling approaches are complementary in nature and inform optimal molecule design that enables effective clinical testing of the biological hypothesis and development of life-changing therapies for patients.

2. Expert opinion

Application of quantitative pharmacology and translational biomarker research has made significant impact on drug development. It will take additional years before the full effect of these strategies can be evaluated, but there are reasons to be optimistic. AstraZeneca ’s 5 R framework is beginning to have an impact, with success rates from candidate drug nomination to phase III completion improving from 4% in 2005–2010 to 19% in 2012–2016, and with fewer programs failing in discovery [14]. Several biotech companies have reported efficiency gains and cost savings, as well as acceleration of patient access to innovative therapies, as a result of quantitative pharmacology applications [7].

One of the most difficult and critical aspects of drug discovery is the selection of the right target. Thus far, a primary strategy has been to identify and stop clinical candidates with low probability of success early to avoid incurring large costs. However, to become truly transformative, we need to become more efficient in identifying the most promising targets or combination of targets to pharmacologically intervene in a disease. As we continue to unravel human biology, the application of various mathematical models to assess therapeutic potential will accelerate learning. This is achieved by relatively uncomplicated models to compare different mechanisms of action and identify design properties, and translatable QSP models that allow connecting targets or pathways to human outcome by integrating pharmacology, human disease biology, and variability in patient response. It may not be possible to fully appreciate the therapeutic potential of a novel target without clinical data, but a robust translational approach will increase the probability of success. In such cases, one should focus on mitigating risks one can control, like assuring appropriate pharmacokinetic and intrinsic potency properties to achieve desired exposure for target engagement and progress as fast and nimble as possible to the clinic to conduct the ultimate experiment.

With technological advances, large datasets are being generated during a drug‘s discovery and development stage, which is not being fully leveraged today. This can be supplemented with external health- and disease-related data. Next-generation technology like machine learning and artificial intelligence will facilitate the use of these untapped riches of data to augment drug discovery and development processes. Machine learning and artificial intelligence approaches will streamline certain drug discovery processes, accelerate novel biomarker and target identification in the clinic, and lead to an era of personalized medicine by identifying personalized therapy using omics data. With technology like smart watch and smart tracker, future clinical trials may look different and a continuous stream of physical and mental health data will become readily available as digital biomarkers. Pharmaceutical companies are often conservative in adopting new technology, in part due to regulatory- and compliance-related challenges. For a successful future, it is important to invest in the development of technology and talent to address these future needs. Many pharmaceutical companies have started exploring next-generation technology to integrate diverse data streams and adopting more evidence-based decision-making models, which is expected to further increase R&D productivity in the future.

Declaration of interest

The authors are both employees of Merck & Co. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer disclosures

One referee is an employee of Bayer Pharma AG. Peer reviewers on this manuscript have no other relevant financial relationships or otherwise to disclose.

Additional information

Funding

The authors are supported by their employers, Merck & Co.