Oncology Dose Optimization: Tailored Approaches to Different Molecular Classes

Abstract

Oncology dose optimization during the era of chemotherapy focused on identifying the maximum tolerated dose (MTD) for registrational trials, often resulting in significant toxicity. The advent of molecular targeted drugs and immunotherapies offers the potential to achieve similar efficacy with lower doses and fewer side effects, as maximal efficacy is often reached at doses below the MTD. Recent FDA guidance outlines expectations for improving dose optimization in oncology drug development. This review presents a framework for tailored dose optimization by categorizing oncology molecules into four distinct classes based on their mechanisms of action and clinical activities: small molecule targeted therapies and antibody‐drug conjugates (Class 1), large molecule antagonists (Class 2), cancer immunotherapy agonists (Class 3), and molecules with limited or no single‐agent activity (Class 4). Unique dose optimization considerations for each class are discussed, supported by illustrative case examples. To enhance robust dose decision‐making and optimize patient resource utilization, we propose using proof of activity as a gate for initiating dose expansion with one or multiple dose levels. This review emphasizes the importance of integrating all relevant preclinical data, disease knowledge, and clinical measurements and highlights the essential role of quantitative pharmacology and statistical modeling in optimizing doses.

During the era of chemotherapy, oncology dose optimization focused on identifying the maximum tolerated dose (MTD) for registrational trials, often with additive toxicity to patients. Newer treatments, such as molecular targeted drugs and immunotherapies, often reach maximal efficacy at doses below the MTD, allowing for lower doses with similar effectiveness and fewer side effects. ¹ , ² Overly high doses can negatively impact the patient's quality of life, leading to dose modifications or discontinuations, which ultimately affect the patient's ability to receive the therapy. An optimal dose finds a balance between ensuring the drug is effective, while being safe and tolerable for patients in long‐term treatment. Health authorities, including the FDA and EMA, advocate for reforms in dose optimization for oncology drugs, with FDA guidance outlining general expectations. ³ , ⁴ Numerous publications advocate for a case‐by‐case approach to dose optimization, recognizing the multi‐dimensional complexity of oncology drug development. ⁵ , ⁶

Despite ongoing efforts, drug developers often face challenges in devising dose‐optimization strategies for new molecules and call for tangible solutions to address these gaps. This paper aims to provide a tailored dose optimization framework as a foundation for developing practical solutions. We propose categorizing oncology molecules into four distinct classes based on their mechanisms of action (MOAs) and clinical activities: small molecule (SM) targeted therapies and antibody‐drug conjugates (ADCs) (Class 1), large molecule antagonists (Class 2), cancer immunotherapy agonists (Class 3), and molecules with limited or no single‐agent activity (Class 4) (Figure 1 ). Unique dose optimization considerations for each class in initial indications are discussed, supported by illustrative case examples. Emerging new modalities, such as cell therapies and cancer vaccines, are excluded as they may require distinct approaches.

Figure 1.

Molecular classes based on their respective mechanisms of action and clinical activities. ADCs, antibody‐drug conjugates.

GENERAL CLINICAL DEVELOPMENT CONSIDERATIONS FOR ONCOLOGY DOSE OPTIMIZATION

Dose optimization encompasses a holistic plan spanning all stages of drug development. This includes integrating results from preclinical studies, first‐in‐human Phase I trials involving dose‐ranging and dose expansion, and continuing through the initiation of the first registrational trial in the initial clinical indication (Figure 2 ). Two important milestones in clinical development include proof of activity (POA) and proof of concept (POC) (Figure 2 ). POC is defined as meeting predefined safety and efficacy targets when compared to the standard of care (SOC). These criteria often serve as the gate to initiating a registrational trial. In contrast, POA is defined in this review as demonstrating clinically measurable anti‐tumor activity (e.g., tumor response, tumor size reduction, PSA or M protein reduction, ct‐DNA clearance), which depends on the molecule, disease setting, and/or development goals. Here, we propose to use POA to gate the decision to advance development from the “dose‐ranging” phase (dose escalation/backfills) to the “dose expansion” phase (dose expansion with one or multiple dose levels, potentially randomized).

Figure 2.

Key clinical development considerations for dose optimization in the initial oncology indication. POA, proof of activity; POC, proof of concept.

Preclinical studies set the foundation for the initial entry into human dose and aid in identifying potential safety risk(s) and relevant pharmacodynamic (PD) biomarkers, key for clinical assessment. It is therefore critical to place careful consideration into the preclinical‐to‐clinical translation of pharmacokinetics and pharmacodynamics (PK‐PD), toxicity, and efficacy to better estimate the therapeutic window.

The initial Phase I study (the dose‐ranging phase [i.e., dose escalation and backfill cohorts]) is designed to estimate the upper (i.e., maximal tolerated or administered dose to obtain exposure/safety information) and lower (i.e., minimally pharmacologically active dose) boundaries of both the therapeutic and pharmacologically active dose range, while determining acceptable safety and tolerability (Figure 2 ). Peripheral and/or tumor‐associated PD biomarkers aid in demonstrating proof of mechanism (POM) and characterizing the pharmacologically active dose range. Additionally, backfilling the dose cohort with a more homogenous population (i.e., specific tumor type or biomarker selected patients) may be considered to generate additional evidence of POA (Figure 2 ).

For molecules with single agent activities, POA is expected to be established during the dose‐ranging phase. Once POA is established, dose selection can be achieved through single or multiple dose expansion(s) in a select patient population (Figure 2 ). When more than one dose is considered, randomized dose comparison cohorts could be used to control selection bias and potential confounders. These cohorts are typically not powered to demonstrate statistically significant differences in efficacy or safety between dose levels, which also aligns with the FDA guidance. ³ Nonetheless, these cohorts should provide sufficient evidence to support the recommended Phase II or III dose, based on a quantitative and qualitative assessment of the collected clinical data.

To better define safety and tolerability, recent recommendations from cross‐industry encourage incorporating patient reported outcomes (PROs) into the initial Phase I study to assess tolerability and symptomatic toxicities. ⁷ Excluding PROs from Phase I may result in a failure to understand the patients' perspective on tolerability. ⁷ Therefore, inclusion of PROs in both the dose‐ranging and dose‐expansion phases of Phase I and II clinical trials should be considered.

Overall, dose decision should be based on the totality of evidence (e.g. safety, efficacy, biomarkers, PK/PD, nonclinical, and PROs), which involves strategic planning to generate and quantitatively integrate relevant translational and clinical data ⁸ (Figure 2 ).

MOLECULAR CLASS UNIQUE CONSIDERATIONS FOR DOSE OPTIMIZATION

Class 1: Small molecule targeted therapies and antibody‐drug conjugates

Class 1 comprises SM targeted therapies and ADCs (e.g., asciminib, sotorasib, trastuzumab deruxtecan). These molecules primarily modulate distinct oncogenic pathways or target surface antigens on cancer cells with high selectivity. The availability of predictive in vitro or animal models typically results in a better translatability of efficacy from preclinical data to humans compared to other classes ⁹ (Figure 3 ). Target efficacious dose ranges can often be projected from these data to inform the dose escalation strategy. However, the therapeutic window for Class 1 tends to be narrower than that of Class 2 (large molecule antagonists), attributed to potential off‐target effects or on‐target and off‐tumor toxicities associated with SMs and ADCs. ² An analysis of FDA initial approvals for SM targeted therapies or ADCs from 2019 to 2021 revealed frequent dose modifications at the recommended dose, ¹⁰ underscoring the necessity of characterizing both early and late onset toxicities as well as patient tolerability, as assessed by PROs. Class 1 SMs/ADCs are targeted therapies typically designated for select patient populations (e.g., patients with the KRAS G12C mutation for sotorasib, or those with HER2+ tumors for trastuzumab deruxtecan). Thus, dose‐finding studies in a select patient population could be considered for Class 1 molecules.

Figure 3.

Unique clinical development considerations for dose optimization across molecular classes. POA, proof of activity; POC, proof of concept.

Asciminib: Preclinical pharmacology guided dose finding

Asciminib is an allosteric SM tyrosine kinase inhibitor that exhibits potent activity against both wild‐type (WT) and mutated BCR‐ABL1, including the T315I mutant. Preclinical studies showed that asciminib is approximately 4 to13 times more potent against WT tumors than tumors harboring the T315I mutation, suggesting higher asciminib exposure was necessary to achieve anti‐tumor activities in tumors harboring the T315I mutation. Based on these preclinical pharmacology findings, the Sponsor conducted a Phase I study with two separate dose escalations and expansions for WT patients and T315I‐mutant patients, which informed the dose(s) used in the respective pivotal study or cohort. ¹¹ FDA approved asciminib with distinct dosages for WT (40 mg BID or 80 mg QD) and T315I‐mutant (200 mg BID) populations. ¹² Asciminib illustrates the unique dose‐finding considerations for this class; notably, similar considerations have been reported for many ADCs. ¹³

Class 2: Large molecule antagonists

Class 2 includes large molecules with antagonistic activities (e.g., monoclonal antibodies [mAbs]: pembrolizumab and isatuximab). Class 2 molecules usually bind to their target antigens with high specificity and are often associated with a lower risk of off‐target toxic effects, except for immune‐mediated reactions for immunotherapies. Compared with Class 1, Class 2 molecules often exhibit a wider therapeutic window. Between 2000 and 2013, out of 82 first‐in‐human clinical trials involving monoclonal antibodies (mAbs), 69 studies (84%) did not reach the maximum tolerated dose (MTD). In many cases, the recommended Phase II dose (RP2D) was determined in the absence or independently of the MTD. ¹⁴ Consequently, it is important to leverage preclinical pharmacology data, clinical PK and PD, instead of MTD, to guide dose selection. For molecules with a novel MOA, it is often challenging to use pathway PD markers to support dose selection due to a lack of understanding of the clinical relevance of these biomarkers. Considering the pharmacological effect of Class 2 molecules is mainly driven via target antigen engagement, saturation of target engagement at the site of action (i.e., tumor) can be used as a surrogate for pharmacology saturation (i.e., maximal effect on downstream signaling) (Figure 3 ). Escalation of a dose well above tumor target saturation may not be warranted because it is unlikely to provide additional antitumor activity and may lead to increased toxicity. Given this caveat, long‐term tolerability, in addition to acute safety, needs to be taken into consideration before deciding an optimal dose.

Pembrolizumab: Intratumoral target engagement guided dose selection

Pembrolizumab is a highly selective anti‐PD‐1 humanized mAb, approved for over 40 oncology indications worldwide. A pragmatic dose selection strategy was adopted by leveraging intratumoral target engagement along with randomized dose comparisons to select the RP2D of pembrolizumab. ¹⁵ The RP2D was estimated based on Phase I PK data and intratumoral PD‐1 target saturation, which is a surrogate for maximal pharmacological effect for antagonist mAbs. Given the challenges and limitations associated with intratumoral target engagement measurement, a physiologically based PK model (PBPK) was developed to predict intratumoral target engagement. The RP2D was estimated as 200 mg every 3 weeks, at which PD‐1 saturation was predicted to be achieved in the tumor and systemic circulation. ¹⁵ The appropriateness of the predicted RP2D was confirmed by randomized dose comparison studies, which showed similar efficacy at the projected RP2D and at a higher dose with nonoverlapping PK exposures (i.e., 5‐fold higher dose/exposure). ¹⁵

Class 3: Cancer immunotherapy agonists

Class 3 includes molecules with cancer immunotherapy agonistic activities, including T‐cell engagers (TCEs, e.g., teclistamab and mosunetuzumab) and immunocytokines (e.g., aldesleukin). ¹⁶ The clinical translatability of preclinical models is limited because the complexity of the human immune system is not fully represented in these models. Therefore, few insights can be generated from preclinical data for dose optimization for Class 3 compared to Class 1. ⁹ Designing an optimal dose regimen is challenging for Class 3 due to the differentiated agonistic mode of action and distinct mitigation strategies for acute toxicities, which need to be considered to improve the therapeutic window. Additionally, a bell‐shaped efficacy relationship is theoretically possible for TCEs, which may render the dose optimization process more complicated ¹⁷ Notably, cytokine release syndrome (CRS) is a common on‐target adverse event (AE) of concern for TCEs, which occurs acutely following the first cycle and dissipates with time. Given the unique time dependency of CRS, implementation of step‐up dosing (i.e., giving one or more incremental dose prior to the target dose) may dampen cytokine release with each subsequent step, allowing higher target doses to be reached and leading to better efficacy. ¹⁷ Similarly, for immunocytokines, splitting a large dose into several small doses over time is often considered to mitigate the risk of Cmax‐driven toxicities. As a result, each of the doses used in a fractionated or step‐up dosing approach may require optimization during the dose‐ranging phase (Figure 3 ). The first aims at identifying the optimal step‐up dosing regimen or the dose fractionation scheme. The second focuses on defining the optimal target dose to ensure the right balance toward determining benefit–risk. Given the complex MOA and limitations of preclinical data, mechanistic pharmacology modeling approaches (e.g., PK/PD or QSP) have been increasingly implemented to support dose optimization for this class of molecules. ¹⁷ , ¹⁸

Mosunetuzumab: Model‐guided dose escalations (step‐up and target dose optimization)

Mosunetuzumab, an anti‐CD20/CD3 T‐cell engaging bispecific antibody, is approved for treating relapsed/refractory (R/R) follicular lymphoma (FL) patients who have received at least two prior systemic therapies. As a conditional agonist, target B‐cell killing is expected only upon simultaneous binding of mosunetuzumab to CD20 on B‐cells and CD3 on T‐cells. Given the potent T‐cell activation induced by mosunetuzumab, on‐target acute toxicity such as CRS could impact the therapeutic window, potentially limiting its dose and utility. To understand the mechanisms behind safety and efficacy and explore safety mitigation strategies, a novel mechanistic QSP model was developed to capture the dynamics of B and T‐lymphocytes and their interactions in multiple physiological compartments along with mosunetuzumab pharmacology. ¹⁹ The model simulation results suggested that a single‐step or double‐step fractionated dosing would mitigate peak cytokine levels with minimal impact on antitumor response. This provided a strong rationale and quantitative guidance for the clinical Phase I dose escalation study design of mosunetuzumab. ²⁰ Instead of having one dose escalation, the Sponsor strategically planned two different dose escalations: fixed dosing and cycle 1 step‐up dose escalations. ²⁰ Emerging results from the trial confirmed the predictions and have enabled a higher target dose with acceptable CRS risk. ²⁰

Class 4: Molecules with limit or NO single agent activity

Class 4 includes molecules with limited to no single agent (SA) activity in the clinical setting. Like Class 3, insights generated based on the preclinical models are often limited. Given that Class 4 has limited or no single agent activity, single agent backfill evaluations typically do not inform dose selection; instead, the goal is to move quickly to the combination setting to establish POA. This assumes that the safety profile of the single agent is tolerable, and ideally there is supportive PD modulation in line with expected MOA. ²¹ The risk assessment of overlapping toxicity guides the dose escalation strategy for the combination and appropriate safety management guidelines. As efficacy and safety assessments are often confounded by the contributions of the combination partner, backfill cohorts in more homogenous populations might be used to further characterize the PK/PD relationship but usually not POA. As a result, POA of the combination with an active agent may only be demonstrated at the time of the first randomized experiment to isolate an individual contribution of the components. Thus, even in the absence of the POA, dose randomization may be opportunistically considered for inclusion in the first POC‐enabling randomized experiment in early clinical development, based on the strategic priorities and evidence from preclinical and dose escalation experiments (Figure 3 ). Alternatively, dose optimization for Class 4 molecules could be pursued, if warranted, after establishing POC at a “no‐regret” dose. ⁵

Relatlimab

There are limited examples of Class 4 approved molecules, with relatlimab being a recent example. ²² While the flat dose‐ and exposure‐response relationship for safety and efficacy, along with an overall positive clinical benefit–risk profile, justifies the approved dose regimen of relatlimab, ²² available publications and regulatory review documents provide limited insights into its unique dose optimization considerations. Specifically, detailed considerations for dose‐ranging and dose‐expansion study design in monotherapy and combination therapy within the context of POM, POA, and POC are not well documented.

CONCLUSION

Oncology drug development is a complex endeavor with multi‐dimensional optimization requiring strategic trade‐offs and sequencing of POM, POA, POC, patient population selection, combination strategies, and dose optimization within a broader strategic framework. Due to these complexities, there is no simple one‐size‐fits‐all solution for oncology dose optimization. ⁵

This review highlights a tailored dose optimization framework for each molecule class as a foundation for developing tangible solutions. Key considerations include the translatability of preclinical data, the relevance of target engagement and PD biomarker, the nature of AEs, the therapeutic window, and distinct MOA, all of which are critical for developing effective dose optimization strategies for each class (Figure 3 ).

Classes 1, 2, and 3 are expected to demonstrate single‐agent antitumor activity, such as overall response rate (ORR) during the dose escalation/backfill (i.e., dose‐ranging phase), POA should be established to gate dose expansion with one or multiple dose levels (Figure 2 ). It is not the best use of patient resources to expand at a dose that is not expected to have any antitumor activities. ²³ Furthermore, it may be challenging to make the dose decision only based on safety/tolerability data without any anti‐tumor activity when expanding at an ineffective dose, especially for the molecules lacking reliable and clinically relevant biomarkers for efficacy. POA could be more challenging to be established for Class 4 as any observed antitumor activities could be contributed by the combination partner. As discussed earlier, POA for Class 4 may only be demonstrated in combination at the time of the first randomized experiment to account for the individual contribution of the components.

Randomized studies (or cohorts) with multiple dosages can support dose optimization by eliminating selection bias when comparing doses. Building in flexibility in the study design earlier in the clinical development plan enables flexible infrastructure with the ability to turn on randomization as needed and adapt to emerging data for a molecule (i.e. adding an additional dose level or removing a dose level). All molecular classes should assess the need for randomized dose comparison once the safety profile is sufficiently characterized, and preliminary evidence of anti‐tumor activity is shown for a single agent or a combination to make best use of patient resources.

Empirical and semi‐mechanistic modeling, including population PK modeling, dose/exposure‐response analyses, and PK/PD modeling, has proven to be invaluable tools for gaining in‐depth understanding of drug exposure, efficacy, and toxicity. ⁸ Due to the complexity and extensive data and knowledge requirements for model development, multiscale mechanistic models (e.g., PBPK and QSP models) are typically employed in specific scenarios. If target engagement is important for dose selection per MOA of the molecule of interest (e.g., Class 2), PBPK modeling could be a valuable tool to predict the drug concentration and target engagement at the site of action (i.e., tumor). For molecules with novel and complex MOA (e.g., ADCs, TCEs), there is an increasing trend for these molecules to adopt multiscale mechanistic QSP models to support dosage selection. ¹⁸ , ²⁴ These models play a pivotal role in guiding data‐driven dose decisions throughout development, from first‐in‐human starting doses to expansion doses and recommended Phase II/III doses. By integrating diverse data sources and adopting flexible study designs, these approaches optimize patient resource use and support robust dose optimization across oncology drug classes.

FUNDING

This work was funded by F. Hoffman‐La Roche, Ltd.

CONFLICT OF INTEREST

J.Z., A.S., S.F., C.B., C.J., N.K., K.K., E.P., J.A., P.T., P.G., J.F., C.L. are employed by Genentech, Inc., F. Hoffmann‐La Roche Ltd, or Roche Innovation Center, and hold company stocks.