“Leveraging Modeling and Simulation to Optimize the Therapeutic Window for Epigenetic Modifier Drugs”

Abstract

Dysregulated epigenetic processes can lead to altered gene expression and give rise to malignant transformation and tumorigenesis. Epigenetic drugs aim to revert the phenotype of cancer cells to normally functioning cells, and are developed and applied to treat both hematological and solid cancers. Despite this promising therapeutic avenue, the successful development of epigenetic modulators has been challenging. We argue that besides identifying the right responder patient population, the selection of an optimized dosing regimen is equally important. For the majority of epigenetic modulators, hematological adverse effects such as thrombocytopenia, anemia or neutropenia are frequently observed and may limit their therapeutic potential. Therefore, one of the key challenges is to identify a dosing regimen that maximizes drug efficacy and minimizes toxicity. This requires a good understanding of the quantitative relationship between the administered dose, the drug exposure and the magnitude and duration of drug response related to safety and efficacy. With case examples, we highlight how modeling and simulation has been successfully applied to address those questions. As an outlook, we suggest the combination of efficacy and safety prediction models that capture the quantitative, mechanistic relationships governing the balance between their safety and efficacy dynamics. A stepwise approach for its implementation is presented. Utilizing in silico explorations, the impact of dosing regimen on the therapeutic window can be explored. This will serve as a basis to select the most promising dosing regimen that maximizes efficacy while minimizing adverse effects and to increase the probability of success for the given epigenetic drug.

1. Introduction

Epigenetic modulators are promising therapeutic targets for the treatment of hematological and solid tumors. Components of the epigenetic machinery include DNA and histone methylation, histone modification and chromatin remodeling (Cheng et al., 2019; Jaenisch & Bird, 2003) and are required for an organism’s normal development and response to environmental changes (Baylin & Jones, 2011; Cheng et al., 2019; Jaenisch & Bird, 2003). Epigenetic alterations have been associated with tumor growth and tumor progression in a variety of cancers leading to the dysregulation of key biological pathways involved in proliferation, differentiation and tumor suppression (Di Costanzo et al., 2014). Accumulation of epigenetic alteration in cancer cells has been attributed to the emergence of resistance to drug therapy (Quagliano et al., 2020). Those pathophysiological epigenetic modifications offer suitable drug targets for small molecules inhibitors and can be categorized into three epigenetic protein families - readers (eg., bromodomains), writers (e.g., DNA methyltransferases (DNMTs) and Lysine methyltransferases (KMTs) that methylate DNA and histones, respectively) and erasers (e.g., Histone Deacetylases (HDACs) and Lysine Demethylases (KDMs) that deacetylate and demethylate histones, respectively) (Ganesan et al. 2019). Epigenetic treatment strategies targeting those proteins aim to restore the normal cellular functionality by reverting the epigenetic alterations of the cancer cell (Baylin & Jones, 2016; Biswas & Rao, 2018; Stahl et al., 2016). To date, the majority of approved epigenetic modulators have achieved clinical success for the treatment of hematological malignancies but have depicted only limited activity in solid tumors (Jin et al., 2021). Despite the promising therapeutic avenue, the development of epigenetic drugs seems less straightforward as compared to classical chemotherapies. Here, we highlight some of the specific hurdles and show how those can be resolved by utilizing modeling and simulation approaches.

2. Challenges in drug development of epigenetic modulators

A widespread dose-finding strategy during clinical trials is the identification of a maximum tolerated dose (MTD). This is based on the assumption that the higher the dose the greater the efficacy. For chemotherapy, this has been demonstrated as a useful concept, however it is not generally applicable to epigenetic modulators (Sachs et al., 2016). For example, the DNMT inhibitor decitabine, which is approved in acute myeloid leukemia (AML) by the EMEA and myelodysplastic syndromes (MDS) by the FDA, was originally tested at very high doses close to the maximal tolerated dose and showed disappointing efficacy. As a consequence, the development was put on hold. A few years later, the drug was resurrected and tested in AML at a 10-fold lower dose but with optimal biological activity based on the re-expression of genes (i.e., ER and p15) that are frequently methylated in AML (Blum et al., 2007). In general, the selection of optimal biological dose is typically done using an effect marker of target engagement, target modulation or disease progression marker (Sachs et al., 2016), which is typically much lower than the MTD (Corbaux et al., 2019; George et al., 2016). In this case, target modulation monitored as gene re-expression was used as a marker for biological activity. In conclusion, the development of decitabine finally succeeded with a paradigm shift in the dose selection strategy. Instead of selecting the maximal tolerated dose, the optimal dose was based on re-expression of silenced genes considering the expected pharmacodynamic effects related to the mode of action (Blum et al., 2007).

For many epigenetic drugs, the safety margin is often narrow (Mohammad et al., 2019). Therefore, their successful clinical drug development not only requires identifying the tumor type that is likely to respond to the given epigenetic drugs but also the identification of a dosing regimen that maximizes the therapeutic benefit while minimizing the adverse effects (Shen & Laird, 2012). This requires a thorough understanding of how the timing and frequency of dosing is related to the therapeutic and adverse drug reaction. Myelosuppression such as neutropenia and thrombocytopenia are the most frequently reported adverse effects in this drug class (Sermer et al., 2019) and has been observed in the majority (8/9) of the approved epigenetic modulators (for overview, see table 1). Whereas the exact cause is unknown, it is assumed to be an on-target driven effect through interference with the epigenetic machinery of cells with a high turn-over, such as the stem cells implicated in thrombopoiesis (Ali et al., 2013). Sun and colleagues systematically analyzed the reported safety and efficacy profiles of 12 bromodomain and extra-terminal motif (BET) protein inhibitors in Phase I/II clinical trials and reported hematological adverse events (AEs) such as thrombocytopenia, anemia, and neutropenia as the most common and severe (grade ≥3) adverse effects and all 12 BET inhibitors exhibited exposure-dependent thrombocytopenia (Sun et al., 2020). These AEs are attributed to on-target effects due to the interference of BET inhibitors with the transcription factor GATA1 (Spriano et al., 2020) which plays a critical role in the lineage commitment of hematopoietic stem cells toward erythroid/megakaryocytic lineage and subsequent maturation (Matsumura & Kanakura, 2002). While those hematological toxicities are reversible and can be clinically managed by dose interruption or dose reduction (Mohammad et al., 2019), this may limit their clinical application. Therefore, a more promising way to increase the chances of success for those molecules is to select the optimal dosing schedule that permits the recovery of systemic blood cell counts without adversely affecting antitumor effects. This can only be achieved with a systematic approach based on a thorough understanding of the dose-exposure-response relationship between the drug pharmacokinetics and pharmacodynamics including target engagement and biomarker response related to safety and efficacy for a given drug, which may vary considerably from one drug candidate to another. This may facilitate the development of epigenetic drugs as it helps to better understand how the drug exposure and the drug potency is related to the onset, magnitude and duration of the downstream effects and what this means in terms of an optimized dosing schedule. Ideally, this is a continuous learning exercise that starts already during the drug discovery stage with the selection of the most promising drug candidates and continues throughout the drug development and life cycle of the drug (Chien et al., 2005; Kim et al., 2018). With inappropriate exposure-response quantification, there is a higher risk to select a less favorable molecule for clinical testing or to prematurely stop the clinical development of a promising molecule because of a suboptimal dosing regimen resulting in a less favorable risk-benefit ratio, which is assumed as one of the key obstacles of epigenetic drugs (Fardi et al., 2018; Mohammad et al., 2019; Sun et al., 2020).

Table 1:

Hematologic abnormalities observed with epigenetic therapeutics approved or in clinical development

Epigenetic therapeutics	Target/ Indication*	Year of FDA approval or Clinical Phase	Hematologic toxicity reported	Reference
Azacitidine (Vidaza)	DNMT-1 inhibition/ MDS	2004	Yes (neutropenia, thrombocytopenia, anemia, leukopenia)	(García-Delgado et al., 2014) See the Label from Drug@FDA https://www.accessdata.fda.gov/drugsatfda_docs/label/2020/050794s032lbl.pdf
Decitabine (Dacogen)	DNMT-1 inhibition/ MDS	2006	Yes (neutropenia, thrombocytopenia, anemia)	(B.-H. Lee et al., 2020) See the Label from Drug@FDA https://www.accessdata.fda.gov/drugsatfda_docs/label/2020/021790s025lbl.pdf
Vorinostat (Zolinza)	Class I and II HDACs/ CTCL	2006	Yes (thrombocytopenia and anemia)	(Shah, 2019) See the Label from Drug@FDA FULL PRESCRIBING INFORMATION 1 INDICATIONS AND USAGE ZOLINZA1
Romidepsin (Istodax)	Class I HDACs primarily/ CTCL and PTCL	2009	Yes (thrombocytopenia, neutropenia, lymphopenia and anemia)	(Shah, 2019) See the Label from Drug@FDA https://www.accessdata.fda.gov/drugsatfda_docs/label/2021/022393s017lbl.pdf
Belinostat (Beleodaq)	Class I, II and IV HDACs/ Relapsed or refractory PTCL	2014	Yes (thrombocytopenia, leukopenia (neutropenia and lymphopenia) and anemia)	(Shah, 2019) See the Label from Drug@FDA https://www.accessdata.fda.gov/drugsatfda_docs/label/2020/206256s003lbl.pdf
Panobinostat (Farydak)	Class I, II and IV HDACs/ MM	2015	Yes (thrombocytopenia, leukopenia, neutropenia and lymphopenia) and anemia)	(Shah, 2019) See the Label from Drug@FDA https://www.accessdata.fda.gov/drugsatfda_docs/label/2015/205353s000lbl.pdf
Enasidenib (Idhifa)	IDH2 mutant enzyme/relapsed or refractory AML (IDH2 mutation)	2017	Yes (thrombocytopeniaa, leukocytosisb)	(Cerchione et al., 2021) bSee the Label from Drug@FDA https://www.accessdata.fda.gov/drugsatfda_docs/label/2017/209606s000lbl.pdf
Ivosidenib (Tibsovo)	IDH1 mutant enzyme/ relapsed or refractory AML (IDH1 mutation)	2018	Yes (thrombocytopeniaa, leukocytosisb)	(Cerchione et al., 2021) bSee the Label from Drug@FDA https://www.accessdata.fda.gov/drugsatfda_docs/label/2018/211192s000lbl.pdf
Tazemetostat (Tazverik)	EZH2 inhibition/ (R/R) FL, Metastatic or locally advanced epithelioid sarcoma	2020	No	(Hoy, 2020) See the Label from Drug@FDA https://www.accessdata.fda.gov/drugsatfda_docs/label/2020/211723s000lbl.pdf
Guadecitabine (SGI-110)	DNMT inhibition/advanced HCC	Phase II	Yes (neutropenia, leukopenia	(V. Lee et al., 2018)
Entinostat (SNDX-275, MS-275)	HDAC inhibition/R&R Hodgkin Lymphoma	Phase II	Yes (thrombocytopenia, anemia, neutropenia, leukopenia)	(Batlevi et al., 2016)
Mocetinostat (MGCD0103)	HDAC inhibition/MDS	Phase II	Yes (thrombocytopenia, anemia, neutropenia)	(Luger et al., 2013)
Iadademstat (ORY-1001)	LSD1 inhibition/AML and solid tumor	Phase II	Yes (thrombocytopenia, neutropenia)	(Fang et al., 2019) (Genomics & A., 2020)
Molibresib (GSK525762)	BET Inhibition/NMC, other solid tumors	Phase I/II	Yes (thrombocytopenia, anemia)	(Jin et al., 2021) (Cousin et al., 2020)

DNMT-1 DNA demethyltransferase-1, DNMTi DNA methyltransferase inhibitor, FL follicular lymphoma, HDACi histone deacetylase inhibitor, IDH isocitrate dehydrogenase, MDS myelodysplastic syndrome, MM multiple myeloma, PTCL peripheral T-cell lymphoma, CTCL cutaneous T-cell lymphoma

3. Facilitating drug discovery and development with modeling and simulation

Modeling and simulation is suited to address those key questions, namely the optimal dose finding strategy in early clinical trials as well as the identification of an optimized dosing regimen that balances therapeutic and adverse effects. This is typically done with mathematical models that are developed to recapitulate the current knowledge of the pharmacological processes in a formal way and, based on this, predict the expected future outcome (Kim et al., 2018; Mould et al., 2015). This process starts with model building to describe the observed data, often by fitting the model to PKPD data from dedicated in vitro and in vivo pharmacology studies or human clinical trials. The predictive power of the model can be tested by comparing a prior prediction to the observed data. A model is considered to be validated if the model predictions are in line with the observations. In case there is a mismatch, the model structure will be adjusted and re-tested. This so-called learning-confirming cycle is an important process in drug discovery and development because it reveals potential knowledge gaps that can trigger follow up experiments or studies to close those (Chien et al., 2005). The model can then be further used to predict the untested scenarios and enable in silico exploration of “what if'' scenarios, which can enhance the decision making process at all stages of drug discovery and development, from suggesting the need for follow-up in vitro or in vivo experiments to directly informing dosing regimens for human clinical trials (Lavé et al., 2016; Miller et al., 2005).

For epigenetic modulators, the anti-tumor effect is the result of the drug’s ability to modulate the epigenetic machinery in such a way that epigenetic alterations are reverted and the normal cellular function can be restored (Figure 1A) , which can be monitored by the increase or decrease of a relevant biomarker and eventually by the anti-proliferative effects. All those drug-induced effects occur at a different pace and may vary with regards to magnitude and duration in response to drug exposure. With PKPD or systems pharmacology models, those pharmacodynamic processes happening on a molecular, cellular and phenotypical level can be predicted as a response to the drug treatment and the associated pharmacokinetic processes (Figure 1B). Another aspect is to apply mechanistic modeling to both efficacy and safety. This allows to predict the time course of the expected biomarker, safety and efficacy response (Figure 1C) and can be utilized to identify the most favorable dosing regimen maximizing therapeutic and minimizing safety effects (Yates & Fairman, 2021). In general, it has been broadly acknowledged that a good understanding of those PKPD aspects is closely linked with a higher probability of success in clinical development (Cook et al., 2014; Morgan et al., 2012; Visser et al., 2013). In the following chapters we demonstrate how modeling and simulation can guide and facilitate the development of epigenetic drugs.

Figure 1:

Schematic overview of the mode of action of epigenetic drugs (A) and its formal representation in a mechanism-based model (B) to enable simulations of the drug response (C A: Alterations in the epigenome are assumed to affect the gene expression which may result in uncontrolled cell proliferation. Epigenetic drugs target proteins that modulate the epigenome and contribute to the reprogramming of normal cellular function. Figure 1B: Mathematical model structure to recapitulate epigenetic drug response. After dosing, the drug concentration time profile (PK) in plasma or serum (central) is described with a compartmental model and linked to the PD and safety model. The disease response starts with binding of the free drug to the epigenetic target that is captured as target engagement. This will affect the expression of down-stream proteins that can be monitored as the biomarker response and which will eventually restore the normal cellular function. Figure 1 C Utilizing the mathematical model, simulations of the PK, biomarker, efficacy and safety profile over time can be conducted.

4 . Case examples applying M&S to guide epigenetic drug development

Here, we present three case examples that utilize different modeling strategies to facilitate the development of epigenetic drugs. We illustrate which data are suited to build predictive PKPD and systems pharmacology models supporting informed decision making during drug discovery and development. Table 2 provides an overview of the presented use cases, its application and limitations. The first example (Yamazaki et al., 2020) is an illustration of optimal biological dose prediction in cancer patients based on nonclinical data. The goal is to predict the efficacious exposure in cancer patients based on nonclinical in vitro and in vivo data. The authors build a translational PKPD model that predicts how much target inhibition is required to achieve a desired disease response suited to predict the optimal biological dose in cancer patients. The second example (Moji et al, 2017) illustrates how to bridge from adult to pediatric cancer patients accounting for population specific differences and how confidence in model prediction was increased with a systematic learn-and-confirm cycle. With the third example (Bouhaddou et al., 2020), the authors present a mechanistic model which is purely built on vitro data that reliably predicts the in vivo anticancer drug response in xenograft mice of different dosing schedules. Applying such an approach will greatly reduce animal usage and speed up the early development due to decreased cycle times during the nonclinical development.

Table 2:

Case examples illustrating M&S applied strategies for epigenetic drugs

	Case example 1	Case example 2	Case example 3
Literature reference	(Yamazaki et al., 2020)	(Moj et al., 2017)	(Bouhaddou et al., 2020)
Applications	Predict OBD and efficacious concentration in DLBCL patients based on nonclinical data	Predict pediatric dose of vorinostat approved in adult cancer patients and explore alternative dosing schedules with improved therapeutic window	Support in silico exploration of optimal dosing with regards to efficacy based on mechanistic PD
Modeling approach	Semi-mechanistic PK/biomarker/tumor growth inhibition model to predict the required target inhibition in tumor to achieve tumor stasis	Physiologically based (PB)PK/PD model with verification steps to check for consistency of model predictions with observed data	Mechanistic PD model predicting kinetics of target inhibition and how this translates to biomarker response and tumor cell killing
Data requirements	Nonclinical in vitro and in vivo data, time course on TGI: Dose response in xenograft mice Biomarker: collecting tumor lysate for target inhibition PK in mice and plasma protein binding	Nonclinical (in vitro) and clinical data of PK, biomarker response and thrombocytopenia	Nonclinical in vitro data for model building, varying conditions related dose strengths and incubation time, confirmatory in vivo data
Key insights	Tumor stasis associated with 70% of target inhibition in tumor	Bridging to pediatric patients requires to account for age-related physiological differences Therapeutic window depends on dosing regimen	In vitro in vivo prediction: PD response kinetics can be scaled from in vitro to in vivo by correcting for the tumor proliferation rate
Limitations	Does not account for resistance to drug treatment	Biomarker response in PBMCs not mechanistically linked to disease response	Restricted to efficacy prediction, no information on therapeutic window

4.1. Guiding optimal dose selection based on translational PK/PD modeling: (Yamazaki et al., 2020)

The objective of Yamazaki and colleagues was to support the selection of an optimal biological dose in cancer patients based on nonclinical data. Therefore, the authors developed a translational PKPD model and predicted the required target modulation related to a desired disease response of the orally administered PF06821497 and the corresponding free drug concentration. PF06821497 is an inhibitor for Enhancer of zeste homolog 2 (EZH2). EZH2 plays a role in the methylation of histone H3 and mutants of EZH2 have been reported in various malignancies such as diffuse large B-cell lymphoma (DLBCL), follicular lymphoma, and multiple solid tumors (Yamagishi & Uchimaru, 2017). Due to its repression of tumor suppressor and cellular differentiation genes, mutated EZH2 is an attractive target for drug therapy with promising therapeutic benefits as observed with the EZH2 inhibitor tazemetostat which is approved for the treatment of relapsed or refractory follicular lymphoma and metastatic or locally advanced epithelioid sarcoma (Hoy, 2020).

The model was built on in vivo PKPD studies conducted in xenograft mice implanted with a DLBCL cell line (Kung et al., 2018). The drug was tested with a broad range of doses administered orally or subcutaneously. The model is composed of three components - PK, target inhibition, and tumor growth inhibition model - and allows for a mechanistic interpretation of drug action. It predicts the quantitative relationship between the desired disease response expressed as tumor growth inhibition, the required level of target inhibition (i.e., reduction in H3 methylation) in the tumor and the corresponding free plasma exposure of PF06821497. For target inhibition, terminal sampling was performed to collect tumor lysates for the measurement of target inhibition in the tumor. For PK analysis, serial blood sampling was conducted. The concentration-dependent inhibition of the drug target, namly H3 methylation is used as a biomarker and linked to the further downstream effects on tumor growth inhibition. The PK model predicts the time course of plasma concentration as a function of the dose using an empirical PK modeling approach. The PK model is then linked to the target inhibition model using an indirect response model, which predicts a constant target baseline of H3 methylation in the control group and a concentration-dependent decrease of H3 methylation in the treated group. Since the PD effect on tumor growth is typically delayed relative to dosing, a signal transduction model was applied to predict the gradual changes in tumor growth as a result of the concentration-dependent inhibition of H3 methylation in the tumor. For translation of these findings to cancer patients, the authors corrected for species differences in the plasma protein binding of the drug. As a result, a minimal efficacious concentration was predicted for DLBCL cancer patients with a required target inhibition in the tumor of 70% in the tumor. In conclusion, this approach demonstrates how the optimal biological dose prediction can be derived based on the quantitative PKPD relationships of target engagement, disease response and unbound drug concentration assessed in nonclinical studies.

4.2. Bridging from adult to pediatric population based on physiologically-based PK/PD (Moj et al., 2017).

The prediction of the anticipated exposure-response relationship in pediatric patients based on adult data relies on a good quantitative understanding of drug- and system-related processes. Modeling and simulation techniques allow us to test our current understanding of the exposure-response relationship related to efficacy and/or safety by comparing model predictions with the observed data. Such a strategy was applied by Moj and colleagues who developed a physiologically-based PK/PD (PBPK/PD) modeling framework incorporating several validation steps for the recommendation of a pediatric dosage of the histone deacetylase (HDAC) inhibitor vorinostat. In addition, they explored in silico a more effective regimen as compared to the approved standard dosing regimen in adults. Vorinostat is FDA-approved for the treatment of cutaneous T-cell lymphoma (CTL) in adult cancer patients and acts by inhibiting histone deacetylase (HDAC) activity. The challenge for pediatric dose selection is that it requires a good understanding of the pharmacokinetic processes, such as drug absorption and drug metabolism, and how those change as a function of maturation and age (Templeton et al., 2018). Physiologically-based pharmacokinetic (PBPK) models are mathematical models based on physiology, biological processes, organ function, enzyme/transporter abundance and function, as well as blood flow to recapitulate the pharmacokinetic processes of a given drug. PBPK models are suited to predict from in vitro and animal species to humans or from one patient population to another such as from adult to pediatric populations accounting for age-dependent physiological differences on drug disposition (Verscheijden et al., 2020). The authors considered the ontogeny of various metabolising enzymes relevant for the metabolism of vorinostat, which has implications on the expected drug exposure in children with yet to complete enzyme maturations. Aiming to identify relevant pharmacokinetic processes, a PBPK model was developed for adults, based on clinical PK data with intravenous and oral dosing of vorinostat. With the successful validation of the adult PBPK model, the authors concluded that the model captured all relevant pharmacokinetic processes and were confident that it is suited to predict the PK in children (0-17 years) by accounting for age-related physiological differences, including enzyme ontogeny. Model evaluation of the pediatric PBPK model was carried out comparing observed and predicted PK. This series of learn-confirm cycles is a good way to identify potential knowledge gaps and to increase confidence in the model predictions.

The second part of the proposed PBPK/PD model focuses on the optimization of the dosing schedule by integrating pharmacodynamic processes related to safety and efficacy. As an indirect readout for efficacy, the authors proposed to predict the decrease in HDAC activity in peripheral blood mononuclear cells (PBMCs) as a biomarker for target inhibition. The safety profile of vorinostat has been extensively characterized in mono- and combination therapies (Siegel et al., 2009). In addition to fatigue, thrombocytopenia has been identified as the most common grade 3 and 4 adverse event with vorinostat. For drugs exhibiting hematological toxicities such as thrombocytopenia and neutropenia, the drug typically acts on the progenitor cells of platelets in the bone marrow resulting in a delayed decrease of circulating platelets in blood relative to the time of dosing. Of particular value is the application of semi-mechanistic PKPD models recapitulating these processes for the prediction of both the degree and duration of hematological toxicity, after different schedules of administration (Chalret du Rieu et al., 2014; Friberg et al., 2002). For safety predictions, Moji and colleagues linked the validated PBPK model to a refined thrombocytopenia model (Chalret du Rieu et al., 2014) and verified the model-based predictions against available observed clinical data from adult and pediatric populations. With the verified model, the authors performed in silico simulations testing various dosing regimens and compared this to the approved regimen of 400 mg/day for adult patients. As a result, the authors concluded that some of the simulated dosing regimen showed superiority versus the standard dosing regimen with higher and more sustained target inhibition with a comparable safety profile.

4.3. Predicting in vivo response based on in vitro data with mechanistic PD model (Bouhaddou et al. 2020).

An important goal for mechanistic PD models is their ability to predict in vivo behavior based primarily on in vitro data that is available at early drug development stages. Empirical pharmacodynamic models struggle in that regard, but mechanistic pharmacodynamic models may do better if they capture relevant processes. Bouhaddou and colleagues proposed such a mechanistic PD model leveraging in vitro data that is able to accurately predict in vivo response of intermittent versus continuous dosing regimen in xenograft mice. In this work, it was assumed that ORY-1001, a potent and selective inhibitor of LSD1, will cause epigenetic reprogramming of cancer cells that can be monitored by an increase of gastrin releasing peptide (GRP) expression which in turn is expected to suppress tumor proliferation (Figure 1B). LSD1 is a lysine-specific histone demethylase enzyme and is overexpressed in many cancers and associated with poor prognosis (Maes et al., 2018; Wu et al., 2015). The model suggests that the anticancer activity of the ORY-1001 starts by inhibiting the epigenetic target protein LSD1, which in turn will activate the expression of downstream proteins such as GRP. As a consequence, this will convert uncontrolled proliferating cells into quiescent cells and eventually inhibit tumor growth. Figure 1B shows how the model recapitulates key pharmacological processes triggered by ORY-1001. The model was trained on in vitro data using a small cell lung cancer cell line (NCI-H510A). Multiscale PD responses were collected such as target engagement reflected by relative changes in free LSD1 protein levels, the resulting biomarker response captured as mRNA expression of GRP and its downstream effect on tumor viability. The strength of the mechanistic model is that it correctly predicts the magnitude and duration of these processes with continuous and intermittent drug exposure by dissecting the different kinetic processes and their interplay controlling the drug response on a molecular (i.e., target engagement and mRNA expression), cellular (i.e., tumor cell viability) and phenotypical (i.e., tumor growth control) level. The model was subsequently scaled to the in vivo xenograft model engrafted with the same tumor cell line and correctly predicted tumor growth inhibition as a function of different dosing regimens. This was done by assuming the same quantitative relationships of target engagement, biomarker response and tumor growth inhibition in the in vitro and in vivo system and by accounting for system-related differences between in vitro and in vivo. These were differences of tumor cell growth kinetics observed in vitro and in vivo and differences related to the pharmacologically active drug concentration in vivo by accounting for the fraction of free drug in plasma that is not bound to plasma proteins. In summary, the proposed model accurately predicts the time course of target engagement, PD biomarker response and tumor growth inhibition to the administered dosing regimen in xenograft mice and is a good quantitative foundation to identify dosing regimen that will maximize the therapeutic effect.

5. Future perspectives: defining optimal dosing schedules in silico

The successful development of epigenetic modifiers relies on finding the sweet spot with maximized therapeutic effect and minimal toxicity. Since, for epigenetic drugs, the kinetics of the drug responses related to efficacy and safety are expected to differ, it is assumed that the therapeutic window will highly depend on the dosing regimen (Sun et al., 2020). Optimizing the dosing regimen to achieve maximal tumor reduction with acceptable safety can occur empirically in the clinic by comparing regimens in a randomized trial or using an early responding biomarker if available. Yates and Fairman suggest utilizing a translational modeling framework leveraging nonclinical data to predict the impact of both dose and regimen sensitivities in cancer patients (Yates & Fairman, 2021). Here, we provide an outlook on how this can be applied for epigenetic modulators exhibiting myelotoxicity. We suggest the combination of efficacy and safety prediction to facilitate the selection of an optimal dosing regimen with a broader therapeutic window based on in silico exploration. This concept is summarized in figure 2. Backbone of this optimization strategy is a thorough understanding of the exposure response relationship related to efficacy and safety of the given drug. Efficacy models enable us to predict the level of target engagement or target inhibition as a function of the PK and how this in turn results in downstream effects on target modulation measured as biomarker response and the effects on disease progression captured by tumor growth inhibition (Figure 2B). Complementary to this, semi-mechanistic PKPD models (Chalret du Rieu et al., 2014; Friberg et al., 2002) are suited to predict the nadir and duration of myelotoxicity as a result of drug potency and the administered dosing regimen (Figure 2C). Combining the two modeling efforts will allow us to compare the clinical utility of different dosing regimens explored in silico. As a metric for the clinical utility, we propose quantifying its therapeutic effect expressed as percent tumor growth inhibition and its adverse effect captured as the relative change from baseline of the blood cell count (e.g., platelets or neutrophils) of interest over time and to plot those values against each other as illustrated in figure 2. For illustration, two different dosing regimens with increasing dose levels were simulated and graphically represented in Figure 2A. It shows that the therapeutic window is highly dynamic and changes with the dosing regimen. As such, suitable dosing regimens can be selected that offer a good balance of achieving no or low-grade thrombocytopenia while maintaining adequate anti-tumor activity. For example, dosing regimens with low toxicity and low efficacy are shown in the top left corner, whereas the desirable dosing regimens appear in the upper right corner with maximal antitumor effect and minimal myelotoxicity (figure 2A, blue shaded area). Unacceptable dosing regimens are those that result in high toxicity regardless of the predicted antitumor effect (figure 2A, red shaded area). The graphical representation in figure 2A is designed to permit generalizability to other drugs and indications. This framework can be adjusted with the appropriate scales of toxicity and efficacy tailored to the given indication and project needs and based on the corresponding thresholds that are clinically relevant for their specific project.

Figure 2:

Concept of the proposed in silico optimization strategy to identify a dose and dosing regimen with maximal efficacy and minimal toxicity. (A) Clinical utility plot of the predicted toxicity and efficacy of a given drug. Blue shaded area is the desired therapeutic window, red shaded area is area of unacceptable toxicity. Two dosing regimens (dotted and broken lines) at increasing dose levels (arrows indicate increase in total dose) with the predicted efficacy and toxicity metric. (B) Tumor growth kinetic predictions as a pillar to predict the relative tumor growth inhibition. ( C) Thrombocytopenia model to predict changes in platelet count over time and to assess the relative changes of platelet decrease.

We have tested this concept and simulated the drug response of a hypothetical epigenetic modifier drug that elicits anti-proliferative effects as proposed by Bouhaddou et al (Bouhaddou et al., 2020). and that is expected to exhibit thrombocytopenia as described by Chalret du Rieu (Chalret du Rieu et al., 2014). We compared the predicted efficacy (Figure 3A) and safety (Figure 3B) profiles of the hypothetical drug that are achieved via different dosing regimens. Our comparison includes continuous (i.e., daily dosing; QD) and intermittent (e.g., once every second week (Q2W) or once every fourth week (Q4W)) dosing through administration of the same total dose (1000 mg/kg) over the entire treatment period, resulting in the same total exposure assuming linear PK properties. Despite the same total exposure, the PD response profiles differ significantly. The most efficacious dosing regimen here is the daily dosing, however this will not be clinically acceptable due to high grade thrombocytopenia. In contrast, the intermittent dosing regimens appear to be less efficacious but can be safely administered. Next, we compared varying dosing regimens with regards to their therapeutic window and their clinical utility as introduced in figure 2. The results are displayed in figure 4. For all tested dosing regimens, the increased efficacy is associated with an increase in toxicity as defined by the platelet count decrease from baseline, which can be graded according to the National Cancer Institute (NCI) Common Terminology Criteria for Adverse Events (CTCAE v5.0) hematologic toxicity criteria (NCI CTCAE 5.0). However, there are subtle differences between the different dosing regimens, which become apparent when comparing the slopes. This is most obvious when we compare two extreme cases in terms of dosing frequency. With a daily dosing regimen, when connecting the dose levels, there is a steep slope indicating that there is only little gain in efficacy predicted as a level of tumor growth inhibition at the lower dose levels, while there is a significant increase of thrombocytopenia. In contrast to the continuous dosing regimen, the pulsed dosing regimen (i.e., Q4W) shows a shallow slope indicating a greater increase of anti-tumor effects with lower toxicity at increasing dose levels. In order to identify the threshold of unacceptable toxicity, we defined a critical level of platelets over time that will result in sufficient recovery of the platelet levels prior to re-dosing that have to be maintained. We compared the predicted area under the platelet-time course relative to the baseline (figure 5, supplementary figure S1). Based on the simulations for the given thrombocytopenia response, we have identified a threshold level of 40% to allow sufficient recovery (Figure 5A) reaching a minimal platelet count of 100 × 10⁹ cells/liter (Piette & Broussard-Steinberg, 2021) prior to re-dosing. It should however be noted that with this threshold, grade 3 hematologic toxicity (according to NCI CTCAE v5.0 hematologic toxicity criteria) may be observed for some of the dosing regimen (Figure 5B). Furthermore, it should be noted that these predictions are only valid for the hypothetical drug and cannot be generalized. The peculiar efficacy and safety profile of the respective drug will be driven by the underlying mode of action and the PK properties. Therefore, we suggest building such a quantitative modeling framework for the given drug of interest and use the in silico explorations to predict safety and efficacy in order identify the most promising dosing regimen. Such information may be particularly useful prior to designing the first-in-human study. The model code is provided in a GitHub repository (https://github.com/PKPD-coder/PKPD_modeling_epigenetic_modifiers.git) to allow readers to apply it accordingly.

Figure 3:

Simulated profiles of therapeutic response captured as tumor growth profiles (3A) and the safety profile expressed as platelet cell count (3B) over time as a result of the administration of 1000 mg/kg total dose administered once daily (QD, blue line), once every 2 weeks (Q2W, orange line) or once every 4 weeks (Q4W, green line). Control group is shown with a dotted red line. Shaded areas in Figure 3B indicate the grade of thrombocytopenia, grade 1 (green shaded area), grade 2 (yellow shaded area), grade 3 (orange shaded area), grade 4 (red shaded area).

Figure 4:

Predicted clinical utility of each dose levels tested with different dosing regimen of a hypothetical drug. Total dose levels are shown with the same colour code from 10 mg/kg (blue) to 10000 mg/kg (red). 1. Symbols represent the dosing regimen with Q4W (triangle), Q3W (diamant), Q2W (rhombus), 3 on/4 off (cross), 5 on/2 off (square), 1 on/6 off (rotated cross).

Figure 5.

Simulated endpoints utilizing the myelosuppression model to identify threshold for acceptable toxicity (A) and to predict severity of myelotoxicity (B). A: Simulations of different dosing regimens were tested to allow recovery to a critical platelet level (100×10^9 cells per liter) prior to re-dosing versus the predicted changes in platelet response. Assessment of a threshold level of 40% reduction of platelet response over time measured as the relative changes of the area under the platelet cell count over time versus baseline. B. Relationship of the predicted severity of thrombocytopenia defined as the minimal platelet count level (nadir) versus the relative changes of platelets over time captured by the area under the effect curve.

6. Conclusions

There is increasing clinical evidence that epigenetic drugs offer a promising therapeutic strategy for cancer treatment. The development of epigenetic drugs can succeed if the right tumor type, the right combination partner at the right dosing regimen has been identified. We suggest conducting mechanistic PK/PD modeling studies to better understand the kinetic processes related to the drug response with respect to safety and efficacy and its implication on the optimal dosing.

Defining the right dosing regimen seems often an empirical approach and is rarely guided by the rich information collected during the nonclinical development. The proposed PKPD framework may offer a more rational approach in selecting the right dosing regimen and may increase the probability of a patient benefiting from it.

Abbreviations:

AE

Adverse events

AML

Acute myeloid leukemia

CTCAE v5.0

Common Terminology Criteria for Adverse Events; version 5.0

CTLC

Cutaneous T-cell lymphoma

DLBCL

diffuse large B-cell lymphoma

DNMT

DNA methyltransferase

EMEA

European Medicines Agency

EZH2

Enhancer of zeste homolog 2

FDA

U.S. Food and Drug Administration

FL

Follicular lymphoma

GRP

Gastrin releasing peptide

HDAC

Histone deacetylase

IDH

Isocitrate dehydrogenase

KDM

Lysine demethylases

KMT

Lysine methyltransferase

LSD1

Lysine-specific demethylase 1

MDS

Myelodysplastic syndromes

MM

Multiple myeloma

MTD

Maximum tolerated dose

NCI

National Cancer Institute

OBD

Optimal biological dose

PBMCs

Peripheral blood mononuclear cells

PBPK

Physiologically-based Pharmacokinetic

PTCL

Peripheral T-cell lymphoma

PK

Pharmacokinetic

PD

Pharmacodynamic

QD

every day

Q2W

once every two weeks

Q4W

once every four week