Creating a Roadmap to Quantitative Systems Pharmacology–Informed Rare Disease Drug Development: A Workshop Report

Abstract

One of the goals of the Accelerating Rare Disease Cures (ARC) program in the Center for Drug Evaluation and Research (CDER) at the US Food and Drug Administration (FDA) is the development and use of regulatory and scientific tools, including drug/disease modeling, dose selection, and translational medicine tools. To facilitate achieving this goal, the FDA in collaboration with the University of Maryland Center of Excellence in Regulatory Science and Innovation (M-CERSI) hosted a virtual public workshop on May 11, 2023, entitled “Creating a Roadmap to Quantitative Systems Pharmacology–Informed Rare Disease Drug Development.” This workshop engaged scientists from pharmaceutical companies, academic institutes, and the FDA to discuss the potential utility of quantitative systems pharmacology (QSP) in rare disease drug development and identify potential challenges and solutions to facilitate its use. Here, we report the main findings from this workshop, highlight the key takeaways, and propose a roadmap to facilitate the use of QSP in rare disease drug development.

The Orphan Drug Act defines a rare disease as a disease or condition that affects less than 200,000 people in the United States. There are ~7,000 rare diseases impacting an estimated 25–30 million people in the United States,¹ collectively representing a major unmet medical need. However, developing drugs for rare diseases is challenging due to the small number of patients available for clinical trials, which is often further complicated by disease heterogeneity, poorly characterized natural history of the disease, and other factors. Thus, innovative tools and approaches are needed to overcome these challenges and foster development of safe and effective therapeutics for rare diseases. Quantitative systems pharmacology (QSP) is a modeling approach based on disease pathophysiology and mechanism of action of the drug to computationally characterize pharmacological and clinical responses across molecular, cellular, and tissue/organ levels.^2,3 One unique aspect of QSP is the ability to mechanistically integrate the key molecules within the biological network of specific target(s) of a drug candidate, and to connect them, in a quantitative and multi-scale manner, to the pharmacodynamic biomarkers and clinical end points of the target disease. As such, QSP may provide advantages over population pharmacokinetic or pharmacokinetic/pharmacodynamic modeling; the former is statistically oriented while the later lacks the mechanistic details of the pharmacological effects of the drug within the context of a disease. Therefore, QSP may be applied to address critical gaps in the rare disease drug development setting, by aiding in dose selection, guiding clinical trial design, and predicting clinical response to an investigational drug.

On May 11, 2023, the US Food and Drug Administration (FDA) in collaboration with the University of Maryland Center of Excellence in Regulatory Science and Innovation (M-CERSI) hosted a virtual public workshop entitled “Creating a Roadmap to Quantitative Systems Pharmacology–Informed Rare Disease Drug Development.” Representatives from the FDA, academic institutions, and pharmaceutical industries convened for a full-day scientific workshop about challenges and potential utility of QSP in rare disease drug development. The workshop included three sessions with presentations followed by panel discussions on the following topics: (i) identifying rare disease drug development needs that may be addressed via QSP approaches; (ii) overview of QSP and successful uses of QSP in drug development; and (iii) how can we develop useful QSP models in rare diseases. Herein, we provide a summary of the presentations and discussions that took place at the workshop, including an overview of the current state of quantitative analyses and natural history analyses in rare diseases, details of FDA initiatives to address drug development needs for rare diseases, and discussion of case studies of QSP application in rare diseases, thus expanding on recent discussions of QSP models used in rare disease drug development and submitted to the FDA.⁴

RARE DISEASE DRUG DEVELOPMENT NEEDS THAT MAY BE ADDRESSED VIA QSP APPROACHES

The objective of Session 1 was to identify challenges in the development and approval of drugs for rare diseases that may be aided with QSP and other quantitative modeling approaches. Throughout the presentations and discussion, multiple rare disease drug development challenges were identified, current efforts to address these challenges were outlined, and opportunities for modeling approaches were discussed. Among the myriad challenges associated with rare disease drug development, some of the key factors include that the natural history of the disease is often poorly understood; genotypic and phenotypic diversity often exist within the patient population of a rare disease; many rare diseases have pediatric onset; small patient populations often restrict study design options; and drug development tools such as clinically meaningful end points, outcome measures, and biomarkers are often lacking. Currently, numerous efforts are underway to facilitate rare disease drug development. For instance, one key program highlighted by speakers from the FDA was CDER’s Accelerating Rare Disease Cures (ARC) program.⁵ The goals of the ARC program include driving the development and use of regulatory and scientific tools. This initiative may help to create platforms to facilitate natural history studies, methodologies to construct novel end points, drug/disease modeling, dose selection, and translational medicine.

Modeling & simulation has been recognized as a pathway to streamline development and assist regulatory decisions.⁶ As such, modeling & simulation has been used to address drug development challenges encountered in rare diseases, and there is potential for the application of QSP modeling to rare disease drug development and regulatory decision making. Opportunities for QSP in rare diseases include aiding in clinical trial design (patient selection/enrichment, trial duration, etc.), dose selection/optimization, biomarker selection (informing dose dependency / time dependency, biomarker dynamics in target tissues, relationship of the biomarker to the clinical outcome), informing the strength of a hypothesis, mechanistically informed safety assessments, extrapolation of efficacy/safety findings, and reverse translation (understanding response variability). However, the lack of data available for model development is a major constraint. As such, properly conducted natural history studies and patient registries that facilitate a better understanding of the disease may help strengthen model development efforts. In addition, careful attention should be made in the design and conduct of both natural history studies and clinical trials to collect biomarker and relevant phenotype data (as well as pharmacokinetic (PK) data when relevant). Importantly, working collaboratively in a precompetitive space may facilitate the availability of additional data needed for QSP model development.

Aside from QSP, additional modeling techniques should be considered in the rare disease setting. One modeling approach discussed during Session 1 of the workshop was “Quantitative Retrospective Natural History Modeling,” which can inform survival time, disease stratification (by severity), and biomarker–phenotype associations for rare diseases.⁷ This approach has been utilized for multiple rare diseases, including models developed for molybdenum cofactor deficiency, galactosialidosis, Krabbe disease, and sialic acid storage disease. Moreover, this technique could potentially be coupled with a QSP approach, by providing critical information that may be leveraged to develop and validate the QSP model. While the challenges inherent in rare disease drug development make mechanistic modeling approaches desirable and many potential applications of QSP in rare diseases exist, a number of current challenges must continue to be addressed for this potential to be realized, such as alignment on the utility of QSP in drug development and the application of risk-informed credibility assessment.

SUCCESSFUL USES OF QSP IN DRUG DEVELOPMENT

The objective of Session 2 was to provide an overview of QSP and discuss successful cases of QSP-informed drug development. QSP is valuable for informing dosing regimens since this approach utilizes the mechanistic pharmacology of a drug and physiologically-based pharmacokinetic (PBPK) modeling. Nonetheless, the challenges facing QSP modelers include complex model structure, requirement for a large amount of high-quality data, scaling to predict long-term outcome, and lack of standards and consistent terminology. To address these challenges, it is important to ask three questions: (i) What do we want to know? (ii) How certain do we need to be? (iii) What are we willing to assume? The right tools are needed to answer specific questions that arise in different drug development phases.⁸ Quantifying exposure–response variability is critical for drug development, especially for drugs with a narrow therapeutic window. QSP can potentially address this challenge by enabling preclinical–clinical translation and optimization of combination therapy. Importantly, building on consensus approaches and model reusability, as well as fostering transdisciplinary approaches will help overcome the barriers and facilitate QSP-informed drug development.

One successful example of QSP modeling to support rare disease drug development is shown in the development of olipudase alfa for treatment of non–central nervous system manifestations of acid sphingomyelinase deficiency in pediatric and adult patients. Acid sphingomyelinase deficiency is an inborn error of metabolism caused by pathogenic variants in the sphingomyelin phosphodiesterase 1 gene, resulting in lysosomal hydrolase acid sphingomyelinase (ASM) deficiency. A QSP model was developed which included a minimal PBPK model of olipudase alfa, effects of olipudase alfa on lysosomal dynamics of sphingomyelin and on sphingomyelin accumulation in splenic and alveolar macrophages across molecular, cellular, and organ levels.³ Model outputs included plasma concentrations of olipudase alfa and biomarkers (ceramide and lysosphingomyelin) as well as spleen volume and diffusing capacity of carbon monoxide. Model calibration and validation was performed with clinical data. Briefly, the strategy to quantify the mechanistic similarity of disease and olipudase response between adult and pediatric patients utilized the virtual patient twin analysis and illustrations of similar model structure, similar biology between pediatric and adult patients, similar parameter values, and similar parameter sensitivity between pediatric and adult virtual twins.

Another case of QSP-informed drug development is QSP modeling of receptor occupancy,⁹ which was applied to support the postapproval addition to a dosing regimen with less frequent administration. This mechanistic receptor binding model leveraged accumulated nivolumab clinical pharmacokinetic data, a previously developed pharmacokinetic model, and exposure–response analysis for simulations of receptor occupancy across melanoma, non-small cell lung cancer, and renal cell carcinoma. The model consisted of a population PK model, tumor transport model, and cell binding model (affinity, avidity, membrane transport, synapses, ligand competition in the PD-1/ligand pathway). Nivolumab internalization rates were informed with in vitro continuous and pulse phase fluorescence quench assay. Simulations compared differences of receptor occupancy over the first 28 days among various dosing regimens under several scenarios of varying sensitive tumor binding parameters.¹⁰ The receptor occupancy QSP model supported approval of nivolumab 480 mg every 4 weeks for treating various cancers including several tumors with orphan drug designations.¹¹

These examples demonstrate how QSP modeling can successfully inform drug development. The credibility of a model is associated not only with the quality of the model structure, but also with the amount of high-quality data for model calibration and validation. In addition, application of model simulations entails a level of risk that is influenced by the question of interest (the model is developed to answer), model influence, and decision consequence. To ensure the successful use of QSP, establishing risk-based model credibility is important, regardless of the complexity and scope of the QSP model.

HOW CAN WE DEVELOP USEFUL QSP MODELS IN RARE DISEASES?

The third session focused on what is needed to develop QSP models for rare disease drug development. One important question is whether emerging technologies, such as digital twins (DT) can aid QSP-informed drug development (please see the note below for a brief description of DT and virtual patient twins). DT utilizes real-world information to represent historical and present time data for prediction of treatment effectiveness. For example, the DT for patients with cancer (DTCP) utilizes multi-scale and multimodal data (clinical characteristics, omics, environmental, and social data) from individual patients to train the DTCP for predicted patient care; the predictive power of DTCP can be enhanced from the process of continuously informing clinical trials and basic research as well as learning from new knowledge and data.¹² DT can potentially enable realistic disease models, virtual patient trials to predict the effects of a drug candidate, optimization of clinical trials, and monitoring of real-time treatment outcomes. The challenges to the credibility of DT include data, modeling, clinical integration, and ethics and equity. Integration of mechanistic QSP and data-driven models will increase the predictive power of DT. Informing future rare disease clinical trial decision making and other drug development goals may be enhanced by integrating the aggregate data from individual patients of diverse backgrounds, clinical trials, and population studies into a QSP-informed DT framework. Consequently, DTs may ultimately be informative for rare disease drug development.

One QSP informing rare disease drug development highlighted in Session 3 was the case of developing a drug product to treat Hunter Syndrome. Hunter Syndrome is a rare genetic lysosomal storage disease caused by iduronate-2-sulfatase (IDS) deficiency, a lysosomal enzyme metabolizing glycosaminoglycans (heparan sulfate and dermatan sulfate). Two-thirds of patients have progressive cognitive impairment and behavioral symptoms. To overcome lack of effectiveness in treating central nervous system symptoms of intravenous IDS, an enzyme transport vehicle (ETV) was designed to target transferrin receptors to deliver IDS across the blood-brain barrier to the brain.¹³ To model and simulate the effectiveness of delivering IDS to the brain via intravenous ETV:IDS and predict an effective dose, QSP modeling and simulations showed delivery of IDS to the brain and subsequent effect on substrate turnover.¹⁴ The model considered transport between blood circulation, peripheral tissues, central nervous system endothelial space, parenchymal (including superficial and deep regions), and cerebrospinal fluid. The model also incorporated the effect of antidrug antibodies and was used to predict the effective dose in humans, demonstrating the utility of QSP approaches to inform critical drug development decisions for rare diseases.

To translate preclinical findings of an investigational drug product and demonstrate its efficacy in the target patient population, we must overcome the challenges resulting from biological heterogeneity and epistemic uncertainty. A rational approach would consist of hypothesizing the mechanisms, utilizing a minimal model with desired features of the system, encompassing biological heterogenicity and uncertainty, and iteratively improving the model with more diverse data. Therefore, these challenges can potentially be addressed by applying the “maximum entropy principle” to the least biased statistical model to reproduce a disease or response phenotype, which encompasses biological heterogeneity and contains specific interaction rules between molecules, meaning that within the constraints of the model (including model parameters and structure), biological variation and heterogeneity are allowed to enable robust prediction. One example of extending this principle to mechanism-based dynamic models is agent-based modeling of the time series profiles of cytokines and organ failure scores in sepsis.¹⁵ Clinical heterogeneity can inform refinement of a high-dimensional agent-based modeling and discovery of interaction rules by applying genetic algorithms.¹⁶ Furthermore, synthetic populations that encompass biological heterogeneity and utilize artificial intelligence–enabled control discovery could be employed to conduct in silico clinical trials, thus providing valuable information to aid in the development of candidate products to bring more effective therapies for precision medicine.

This session conveyed how other modeling approaches enhance QSP applications to inform critical aspects of rare disease drug development. Some important lessons include that data-driven statistical analysis can produce good data fit but the results may not be credibly extrapolated outside the data set. To be able to extrapolate for predictions in the rare disease population, a DT of a rare disease population generated with a hybrid model would be useful. The hybrid model consists of a mechanistic model (QSP) that represents the underlying human biology and can be refined by new emergent data, as well as a data-driven artificial intelligence/machine learning model that utilizes real-world data to produce synthetic patient data and relate key biomarkers of a disease to response to a drug. As such, the combination of QSP virtual patient population/twins and data-driven synthetic patient data could be useful in drug development.

KEY TAKEAWAYS AND A POTENTIAL ROADMAP TOWARD QSP-INFORMED RARE DISEASE DRUG DEVELOPMENT

Potential applications of QSP in rare disease drug development

• Key opportunities for QSP in rare diseases include preclinical to clinical translation, aiding in clinical trial design (biomarker selection and dynamics), dose selection/optimization, mechanistically informed safety assessments, and extrapolation of efficacy/safety findings.

• Critical needs identified include integration of quantitative modeling of the natural history of a rare disease with QSP modeling, such as through “Quantitative Retrospective Natural History Modeling,” to facilitate exploration and identification of biomarkers and clinical end points to quantitatively benchmark the progression of a rare disease.

• Collaboration in the precompetitive space, broad efforts to advance drug development for rare diseases, and thoughtful collection of biomarker data in clinical trials for rare diseases may facilitate better characterization of rare diseases and create a solid translational science foundation to build on, which could in turn identify gaps in knowledge where modeling approaches may be useful and provide valuable data to facilitate model development.

Developing successful QSP models in rare disease

• Applying lessons learned in establishing PBPK as a useful drug development tool and gaining regulatory acceptance may accelerate use of QSP modeling for rare diseases. However, the uniqueness of the QSP model to each individual disease presents additional challenges to QSP-informed rare disease drug development. Therefore, utilizing some sub-models or components of existing credible models to build a new model to inform development of a rare disease drug candidate may facilitate applications of QSP models in additional settings.

• Lack of data is a challenge in calibrating and validating models in rare diseases; however, for diseases with a well-understood pathophysiology (such as some monogenic diseases) it may be possible to leverage data from healthy individuals (e.g., for enzyme kinetics) to facilitate model development.

• Regulatory support of QSP as a model-informed drug development tool may aid rare disease drug development programs by providing valuable information on appropriate dosing and clinical trial design.

Advancing the use of QSP in rare diseases

• For rare diseases we will need to use knowledge from all relevant sources to inform model building. Using knowledge of biological processes and disease mechanisms, deep phenotyping approaches when available, supplementing knowledge with sparse phenotyping data, and integrating other innovative data sources may fill gaps in data to aid model development.

• The approaches of virtual patient population could be useful to simulate many scenarios. When validated, the virtual patient population can be used to simulate the quantitative changes of response biomarkers to inform design of clinical studies.

• The concept of DT is appealing but the potential utility has not yet been defined for drug development. Since virtual patient twins have been used in QSP modeling, a framework for use of DT should be clearly articulated, which may facilitate its application to drug development.

• Though there are successes in QSP-informed drug development for rare diseases, broad, consistent collaboration among the stakeholders, instead of siloed approaches, would help facilitate more successful applications of QSP-informed rare disease drug development.

(Note: Digital twin is an in silico representation of an individual or a disease population, which uses real-time data and computationally reflects in real time the dynamic molecular and physiological changes in response to a treatment or lifestyle changes, for the goal of reflecting real-time responses and health status at the individual or population level. Virtual patient twins are generated with a QSP model by sampling throughout the parameter space of sensitive parameters, which is then computationally optimized to reflect the responses of the treatment population).

A roadmap for QSP-informed rare disease drug development is shown in Figure 1. This roadmap is based on discussions at the workshop and emphasizes facilitating collaborations among the stakeholders and establishing a consensus framework for developing QSP models in rare diseases. Potential considerations for a consensus framework include (i) how to utilize the data in electronic health records and a rare disease registry and analyze the quantitative characteristics of the natural history of the disease; (ii) applying data-driven computational methods to identify key disease biomarkers; (iii) incorporating the quantitative, heterogenous characteristics of the natural history of a disease along with the quantitative changes of key disease biomarkers into the QSP-based virtual patient population twin (mechanistic and data driven); and (iv) continuously building the credibility of the virtual patient population twins with emerging global patient registry data and clinical data of a drug candidate to conduct in silico trials to inform the designs of clinical trials and conduct risk-based applications.

Figure 1.

Roadmap to QSP-informed drug development. First, collaboration among stakeholders to better characterize the natural history of rare diseases, including quantitative changes of disease biomarkers (genomics, proteomics, and phenotypes), will lay the translational foundation for development of QSP models. Second, a consensus framework is developed for the credible development of QSP-based virtual patient/digital twins of a rare disease patient population that can be used to address challenges of rare disease drug development. Lastly, maximizing the credibility of the virtual patient twins with emerging data of a drug candidate would inform the designs of clinical trials for a drug development program.