A model-informed approach to radioligand therapy dosing under Project Optimus

In recent years, radioligand therapy programs have entered the clinic following a familiar path: borrowing organ dose limits from external beam radiotherapy (EBRT), and escalating activity levels until reaching or slightly exceeding these EBRT thresholds in organs such as kidneys. In August 2025, the FDA signaled that these approaches alone are no longer sufficient. Its draft guidance on dosage optimization formally brought therapeutic radiopharmaceuticals under Project Optimus.¹

Project Optimus is often interpreted as a push toward dose optimization rather than using maximum tolerated dose. In practice, it raises the bar for how dose and regimen decisions are justified. Sponsors are expected to support those decisions using the totality of available evidence, including dosimetry, pharmacokinetics, biodistribution, imaging, efficacy, and safety, rather than relying on tolerability alone.²

For radioligand therapies, the question is no longer simply what activity to administer. It is whether the selected activity and specific activity can be supported by a clear understanding of exposure, response, and patient variability. The strongest scientific rationale comes from understanding the biological and pharmacological factors that drive clinical outcomes. Increasingly, that requires mechanistic modeling such as quantitative systems pharmacology (QSP).

Beyond maximum tolerated dose (MTD) and traditional dose finding

The maximum tolerated dose paradigm was developed for cytotoxic therapies and does not translate cleanly to radioligand therapies. Efficacy can plateau well below the tolerated ceiling, and the limiting factor is often not systemic tolerability, but the balance between radiation delivered to the tumor and radiation delivered to organs at risk. That is a therapeutic-index problem, which is exactly why Project Optimus moves from MTD toward exposure–response and benefit–risk.

Radiopharmaceutical developers are being asked to answer a different question. Rather than identifying a single activity level, they must understand how efficacy and safety change across patients, target expression levels, and disease states. A single average answer can obscure the very factors that influence efficacy and safety. Addressing those questions requires incorporating the underlying biology directly into the QSP model.

QSP: Modeling the biology behind the therapeutic window

QSP provides a way to connect the biological drivers of response with the measurements used to guide development decisions. QSP models integrate target expression, binding, internalization, biodistribution, and radiation dosimetry to characterize tumor and organ exposure, therapeutic index, and dose response. They can also evaluate these relationships across diverse patient populations and varying levels of target expression, rather than relying on a single representative patient.

A QSP analysis of ¹⁷⁷Lu-PSMA-617 by Minucci and colleagues at Certara illustrates this approach.⁹The model incorporated blood, tumor, kidney, and salivary gland compartments, along with PSMA binding, internalization, and radioactive decay. After scaling from mouse to human, it successfully reproduced observed blood, tumor, and kidney exposure. More importantly, it challenged a long-held assumption: salivary gland uptake could not be fully explained by target expression alone. Even a hundredfold increase in modeled PSMA expression failed to account for the observed uptake. Whether the underlying mechanism is truly non-PSMA mediated remains an active area of investigation. Clinicopathologic studies showing reduced uptake in PSMA-/- mice suggest it may be a PSMA-mediated mechanism, while more recent analyses failed to establish a relationship between PSMA expression and salivary gland uptake.6,7 Regardless of the outcome, the model’s value lies in its ability to test hypotheses against data rather than rely on assumptions. This is particularly important when the interplay between the target and dose-limiting off-target organs influences the therapeutic window.

Because QSP captures the underlying targeting mechanisms, it can also evaluate how the therapeutic index changes as target expression varies between patients and evolves throughout treatment. This has important implications for radioligand therapy. As tumor burden decreases and target expression on tumor cells diminishes over successive treatment cycles, the tumor sink effect may diminish, increasing uptake in dose-limiting organs such as the salivary glands. Efforts to reduce exposure to those organs can, in turn, reduce tumor exposure in the patients who are responding to therapy.8 This effect has been demonstrated with QSP modeling, highlighting a key risk of increasing mass dose to reduce off-tumor uptake.9 A fixed dose limit cannot account for these dynamics. Characterizing them is essential for dose optimization.

Connecting biology to clinical outcomes

A mechanistic model is only part of the evidence. Population pharmacokinetic (PK) analyses that incorporate blood PK and time-activity-curves in organs and tumor from imaging are equally important.

One of the most common assumptions in radioligand therapies development is the use of organ dose limits derived from EBRT. The challenge is that dose rate matters. EBRT delivers a high dose over a short period of time, whereas radioligand therapies deliver radiation gradually over days, allowing healthy tissue additional time for repair and cell cycles to redistribute. As a result, the same absorbed dose may not produce the same biologically effective dose (BED).

This distinction has practical implications. A population PK analysis of a ¹⁷⁷Lu-PSMA radioligand demonstrated that kidney and salivary gland BED estimates differed substantially from values predicted using EBRT assumptions.³,⁴ The magnitude and direction of these deviations differed across organs, reflecting their distinct uptake, retention, and clearance kinetics, so a single correction factor applied uniformly across tissues would not capture the true biologically effective dose.

The complexity increases further with alpha-emitting therapies such as ²²⁵Ac. In these cases, absorbed dose estimates depend not only on the kinetics of the radioligand, but also on how radioactive decay daughters redistribute after potential liberation from the ligand.⁵Mechanistic and QSP models can address this directly by representing each decay daughter as a distinct species with its own kinetics, making assumptions about daughter redistribution explicit and testable against data rather than leaving them implicit in a single absorbed-dose estimate.

Those exposure estimates only matter once they are linked to outcomes. On the efficacy side, that means relating tumor-absorbed dose or biologically effective dose (BED) to measures of response such as PSA decline, radiographic response, or progression-free survival; on the safety side, relating absorbed dose or BED in organs at risk to the toxicities that limit therapy, such as bone marrow absorbed dose or BED to hematotoxicity or salivary glands absorbed dose or BED to xerostomia. Because population PK supplies individual exposure estimates across a heterogeneous population, these exposure-response relationships can be characterized patient by patient rather than as a single average, which is exactly what a benefit-risk assessment under Project Optimus requires.

The key point is that QSP and population PK are complementary. Understanding where radiation is delivered is essential, but so is understanding how that exposure translates into efficacy and safety. Project Optimus expects these approaches to be integrated into a single exposure-response framework rather than treated as separate analyses.

Dose optimization starts before the clinic

Many of the most consequential dose optimization decisions are made long before a patient enters a clinical trial. Mechanistic models can help evaluate target feasibility, determine whether target expression and accessibility are sufficient to support a viable therapeutic index, prioritize or inform candidate selection criteria, and inform first-in-human dose selection.

These models can also help compare how different constructs balance tumor uptake against off-target exposure, allowing teams to make more informed portfolio decisions earlier in development.

Identifying an unfavorable therapeutic window before committing radionuclide supply, manufacturing resources, and clinical trial capacity can reduce development risk, prevent costly late-stage decisions, and improve candidate selection.

From data to decisions

For radioligand therapy developers, the implications of Project Optimus extend beyond dose optimization. The guidance reflects a broader shift in regulatory expectations. Regulators want to understand not only what dosing regimen was selected, but why it was selected and how that decision is supported by the available evidence.

Radioligand therapies present unique challenges because efficacy, toxicity, target expression, biodistribution, and patient variability are closely interconnected. Addressing those challenges requires more than dosimetry alone. It requires an integrated understanding of how biology influences exposure and response.

Many of the decisions that shape a successful development program are made long before a pivotal study begins. Bringing together dosimetry, clinical pharmacology, pharmacometrics, QSP, and regulatory science early in development gives sponsors a stronger scientific foundation for dose selection and a clearer rationale to support regulatory interactions.

Dose selection will always remain a clinical decision. Demonstrating why that dose is appropriate for a particular therapy, and patient population increasingly depends on quantitative evidence. For radiopharmaceutical developers, the goal is not simply to identify the right dose. It is to build a clear, scientifically defensible rationale for the decisions that guide development from the earliest stages through registration.

Key takeaways

• Project Optimus requires a stronger scientific rationale for dose selection.
• Integrating dosimetry, QSP, and population PK provides a more complete understanding of exposure and response.
• Patient variability should be incorporated into dose optimization, not treated as an afterthought.
• Model-informed drug development (MIDD) approaches help answer critical development questions before pivotal studies begin.
• A well integrated quantitative strategy strengthens both development decisions and regulatory interactions.