A Generalized Minimal PBPK‐PD Model of Bispecific Antibodies: Case Studies and Applications in Drug Development

ABSTRACT

Bispecific antibodies (bsAbs), for which each arm binds a distinct molecular target, are developed to engage soluble and cell surface targets in different therapeutic indications. Three key examples of mechanisms of action (MoA) for bsAbs are (1) immune cell engagers that foster immune cell interactions with target cells, (2) bispecifics that use one arm to increase specificity/localization to desired tissues to reduce on‐target off‐tissue toxicity, and (3) bispecifics that use two different arms to neutralize different disease targets. Understanding the pharmacokinetic (PK) profiles and target engagement of bsAbs poses unique challenges due to the existence of multiple targets with distinct biological properties. Here, we present a generalized minimal physiologically based pharmacokinetic (mPBPK) model to capture and predict the PK and target engagement of bsAbs across multiple tissues. First, we model the clinical PK and pharmacodynamic (PD) data for an anti‐IL‐13/IL‐17 bsAb to capture and explain the PD response in the soluble cytokine target levels. Second, we model and simulate the PK‐PD of the anti‐CD20/CD3 T cell engaging antibody, mosunetuzumab, which acts via trans‐binding between cell targets on B‐ and T‐lymphocytes. Third, we use the model to explore case studies for other bsAb approaches to demonstrate the impact of binding affinity and avidity on both PK and target engagement and to provide insights into drug design. Overall, our work yields a model with example applications to advance the use of mechanistic modeling for early PK and target engagement predictions as well as for optimization of bsAb design.

Study Highlights

• What is the current knowledge on the topic?

  • ○
    Bispecific antibodies (bsAbs) are a versatile class of therapeutics that engage multiple targets with a broad range of mechanisms. However, predicting pharmacokinetic (PK) profiles and target engagement remains a major challenge.
• What question did this study address?

  • ○
    This study addressed the need for a generalized model to support early PK/PD predictions to assess critical drug development and design questions such as soluble target neutralization, affinity/avidity requirements, preclinical and clinical dose selection, and target engagement.
• What does this study add to our knowledge?

  • ○
    This study documents a generalized bispecific model with wide applicability, useful for soluble and/or cell surface targets, T‐cell engagers, and tumor‐targeted bsAbs. The model offers insights into PKPD relationships and drug design.
• How might this change drug discovery, development, and/or therapeutics?

  • ○
    The generalized bispecific model is positioned to accelerate early and translational bispecific antibody therapeutic development by identifying promising PK/PD characteristics of new targets, reducing the need for expensive preclinical animal models, and projecting relevant clinical dose levels to support Phase I trials.

1. Introduction

Bispecific antibodies (bsAbs) represent a versatile class of therapeutics offering new avenues for treating disease by leveraging their ability to engage multiple targets. There are numerous clinical‐stage bispecific therapies [1, 2, 3] that generally fall into one of two categories: (1) the antibody binds to soluble targets, or (2) the antibody binds to cell surface targets. Of the molecules that bind to cell surface targets, they can either bind to two different targets on the same cell (cis‐binding) or two different targets on different cells (trans‐binding). Despite their potential advantages, bsAbs present unique research and development challenges. A key early drug design challenge common to all bsAbs is predicting the impact of affinity and target biology (particularly, target levels and turnover) on PK, target engagement, and level of complex formation in relevant tissues. For soluble targets, understanding the affinity, dose, and regimen needed to neutralize two targets simultaneously remains a challenge that can be informed by predictive modeling. For bispecific molecules that bind to cell targets, target binding at different affinities to distinct targets with different levels, turnovers, and tissue distribution profiles complicates PK and target engagement projections, as target‐mediated drug disposition (TMDD) through either or both target receptors can play a major role. For bsAbs that exhibit cis‐binding, avidity becomes an important factor in predicting receptor occupancy (RO) at the tissue of interest. These challenges crucially lend themselves to PK/RO modeling for predictions that inform decision making in the design and development of bispecific molecules.

Bispecific molecules can have different mechanisms of action. T‐cell engagers are a particular subset of trans‐binding bispecifics that promote synapse formation between T cells and target cancer cells to elicit T‐cell‐mediated target cell killing. Empirical models have been used to capture preclinical PK using model structures for time‐dependent and target‐mediated clearance terms [4]. However, the translation of such models for clinical predictions requires caution as it can be unclear how species‐dependent differences in the expression of each target influence PK. Mechanistic models with or without explicit target binding have been used to include the effects of such factors on PKPD, ideally improving translational capabilities [5, 6, 7, 8]. Tissue‐targeting bsAbs are a subset of bsAbs that preferentially target a therapeutic to specific tissue, typically a tumor [9]. These molecules contain one tissue‐targeting arm, which binds to a receptor highly expressed in the tissue of interest, and one therapeutic arm, which binds to the receptor implicated in disease. The intent is to localize the antibody to the tissue of interest via cis‐ or trans‐binding to the tissue‐targeting antigen and reduce potential safety concerns [10, 11]. The development of both T‐cell engagers and tissue‐targeted bsAbs can benefit from translational modeling to connect systemic PK to target engagement in the tissue(s) of interest.

Throughout development, the complexity of bsAb mechanisms of action poses unique challenges that can be investigated with mechanistic modeling [12, 13, 14, 15, 16]. While two‐ or three‐compartment empirical or semi‐mechanistic PK models can often capture PK of bispecific molecules, they do not address drug distribution into distinct tissues or physiological compartments which may differentially express drug targets and thereby impact PK, target engagement, and downstream pharmacodynamics. Minimal PBPK models, which explicitly account for blood, leaky tissue, tight tissue, and lymph compartments, offer a practical middle ground between full PBPK models, which may be too large and/or require many parameters in a rapid, early‐stage research environment, and simple compartmental models of blood and peripheral tissue [17, 18]. Minimal PBPK models are positioned to answer questions regarding PK, RO, and target neutralization in central and lumped tissue compartments at early stages of drug development when little information on the target may be available. Additionally, minimal PBPK models [19, 20] that incorporate target engagement can further be developed to link PK to downstream PD as a function of receptor occupancy (RO) or complex formation in a fit‐for‐purpose manner.

The design space for bsAbs can be quite large as affinity, avidity, and dosing strategies are all considerations that can impact the PK, RO, and subsequent efficacy and safety for these molecules. In this work, we provide a tool that can be used to directly address these considerations. We share a generalized minimal PBPK model that can be used to address bsAb PK and target engagement across multiple species, therapeutic areas, and receptor types. We further present applications of the model to illustrate its relevance to both soluble and cell surface target bispecifics. We specifically show that the model can be used to capture clinical PKPD profiles of a soluble‐cytokine targeted anti‐IL‐13/IL‐17 bsAb and preclinical cynomolgus monkey PK of the T cell engaging anti‐CD20/CD3 bsAb mosunetuzumab. We emphasize the impact of T‐cell mediated drug disposition and CD3 affinity on T‐cell engaging bsAb PK. Lastly, we use the model to explore the potential benefits of tumor‐targeted bsAbs over monospecific bivalent antibodies and provide insight into how to maximize the therapeutic window of tumor‐targeted bispecific molecules. This work highlights a generalized modeling framework with wide applicability, especially in the early research and development phase, that can impact the design and dosing strategies of different types of bsAb therapeutics and can be extended to incorporate the specific downstream MoA of molecules that advance into clinical development.

2. Methods

2.1. Model Details

We built a minimal PBPK model using the structure outlined previously [19, 20, 21]. Briefly, the model structure includes compartments for blood (central), leaky and tight tissue, lymphatics, and optional safety and efficacy compartments (Figure 1a). In all compartments except lymph, simulated drug binds to target(s), if present. The model can simulate the binding kinetics of a bsAb (Figure 1b, left) as well as a monospecific, bivalent antibody (Figure 1b, right). Targets are specified as either cell surface or soluble targets. We explore three different bispecific scenarios: (1) bispecific binding to two soluble targets, (2) bispecific trans‐binding to two different membrane targets expressed on different cells (e.g., bispecific T cell engagers), and (3) bispecific cis‐binding to the same or two different cell surface targets (Figure 1c).

FIGURE 1.

The generalized minimal PBPK model framework models bispecific antibody PK and target engagement of both targets. (a) Drug is dosed either intravenously or subcutaneously and transports through leaky, tight, safety, efficacy, and lymph compartments. Drug is cleared nonspecifically in the central compartment. (b) Binding to each target receptor can occur in all compartments except lymph. For bispecific antibodies, the first binding interaction is modeled using reversible binding kinetics. The second binding interaction includes an effective avidity term, χ. For bivalent, monospecific antibodies, the first binding interaction is modeled using reversible kinetics by incorporating the bivalency factor of 2. The second binding interaction is modeled using an effective avidity term, χ. (c) The model simulates three scenarios: Trans‐binding on membrane receptors, cis‐binding of membrane receptors for mono‐ and bispecific antibodies, and binding to soluble receptors. Each scenario accounts for avidity using the given equations.

The model was built using Simbiology, and all simulations and analyses were conducted in MATLAB 2022a (MathWorks, Natick, MA). Plotting was done in MATLAB 2022a. Model fitting was performed in Simbiology and gQSPSim [22]. The Simbiology project file as well as further technical details can be found in Data S1. Model diagrams were designed using an enterprise license for BioRender.

3. Results

We conducted multiple analyses to assess and illustrate the applicability of the generalized bispecific model to various scenarios. Here, we present examples where the model captured published PK profiles of bsAbs as well as two exploratory case studies demonstrating the utility of the model for PKPD predictions of soluble and cell surface targets. For soluble targets, we use the model to investigate the clinical PKPD of a bsAb targeting two cytokines, IL‐13 and IL‐17. For cell surface targets, we consider two scenarios: (1) trans‐binding, where we model the PK of a T cell dependent bispecific, mosunetuzumab, and (2) cis‐binding, where we explore the impact of affinity and avidity on RO achieved by a theoretical tumor‐targeted bispecific compared to a bivalent antibody.

3.1. Case Study 1: Soluble Targets: Generalized Bispecific Model Captures Clinical PKPD of Anti‐IL‐17/IL‐13 Bispecific Antibody in Healthy Volunteers

To model clinical PK and PD data from the anti‐IL‐17/IL‐13 bsAb (BITS7201A) in healthy volunteers, we included both soluble targets (IL‐17AA and IL‐13) in the leaky, tight, and blood compartments (Figure 2a) [23]. We assumed that the drug‐soluble target complexes can traffic through tissues at the same rate as free drug, allowing targets to accumulate in compartments where they were not expressed at baseline (Figure S1). In this healthy patient population, given the lack of tissue target data following drug administration, we assumed baseline IL‐17AA and IL‐13 levels in tissues were the same as in blood. We calibrated our model to the PKPD data from the single‐ascending dose (SAD) cohorts. Key model parameters are shown in Table S1. In the SAD cohorts, the study examined 30 mg, 90 mg, and 300 mg of drug dosed subcutaneously (SC) as well as 300 mg and 750 mg of drug dosed intravenously (IV). The model was calibrated to clinical PK data of BITS7201A across all of these dose levels (Figure 2b,c). While anti‐drug antibodies (ADAs) were detected in many healthy volunteers, there was no noticeable impact on exposure; therefore, all ADA+ and ADA‐ data were included for calibration [23]. BITS7201A binding to IL‐17AA and/or IL‐13 did not impact predicted PK because we made the assumption that all drug‐target complexes are cleared at the same rate as free drug.

FIGURE 2.

Model captures clinical PK/PD response to single‐ascending dose anti‐IL‐17/anti‐IL‐13 bispecific antibody data. (a) The generalized bispecific model is parameterized for a scenario such that IL‐13 and IL‐17 are synthesized and degraded as soluble receptors in the central compartment only. The model captures the total drug PK via (b) subcutaneous and (c) intravenous injection over 84 days from dose levels of 30–750 mg. (d) The model captures the range of clinical post‐dose serum IL‐17AA elevation across all dose levels by varying the baseline IL‐17AA levels between 0.04 and 0.3 pg/mL. (e) The model predicts a range of serum IL‐13 elevation across all dose levels by varying the baseline IL‐13 levels.

Following drug administration, total IL‐17AA levels in serum rose. We calibrated the model using the reported affinities and fit the baseline IL‐17AA concentration and turnover rate, reflecting that the cytokine level increase results from target stabilization through antibody binding. Due to the low (undetectable) and likely variable levels of IL‐17AA at baseline, we ran simulations over a range of baseline concentrations (0.04–0.3 pg/mL), which are consistent with other reports [24, 25], to capture the overall range of the observed PD response. The wide prediction bands that capture the data demonstrate that total serum IL‐17AA levels are sensitive to baseline IL‐17AA levels in the model (Figure 2d). Simulations also predicted that the peak IL‐13 levels post‐dose increased in a dose‐dependent manner and were also dependent on the baseline IL‐13 levels (Figure 2e). Data for IL‐13 levels post‐treatment with BITS7201A were not published in this study and therefore cannot be verified by our model; however, the model predicts dose‐dependent IL‐13 elevation consistent with clinical levels that have been observed with MEDI7836, a high affinity anti‐IL13 antibody treatment, as well as with IMA‐026 and IMA‐638, two humanized IgG1 antibodies targeting IL‐13 [26, 27].

We validated the clinical PKPD predictions from our model using the published multi‐ascending dose (MAD) cohort data in healthy volunteers who received three subcutaneous doses of BITS7201A every 4 weeks at dose levels 150 mg, 300 mg, and 600 mg [23]. The model captured the PK over the reported 140‐day time interval for all dose levels (Figure 3a). The model also captured the dose‐dependent serum IL‐17AA level elevation using the same baseline range of IL‐17AA as fit during calibration (Figure 3b). Similar to the SAD cohort, the model predicted IL‐13 levels in the absence of data in the MAD cohort (Figure 3c). We further analyzed our simulations to predict the percent target neutralization of IL‐17AA and IL‐13 in the blood as well as the leaky and tight tissue compartments, a key advantage of using a mPBPK model over a two‐compartment PK model, following intravenous doses (Figure 3d,e). The model predicted that the highest dose reported in the SAD study (750 mg) will neutralize ~80% of the IL‐17AA and ~90% of the IL‐13. We also show the predicted PK of BITS7201A in the leaky and tight compartments for both the SAD and MAD studies (Figure S2). Overall, the model captures bsAb PK and PD for BITS7201A in SAD and MAD cohorts, demonstrating the applicability of the model to bispecifics targeting soluble molecules.

FIGURE 3.

The model is validated by the multiple‐ascending dose clinical PK/PD data for the anti‐IL17/anti‐IL13 bispecific antibody. (a) The model captures the multi‐dose total PK data at dose levels 100–600 mg administered every 4 weeks. (b) The model captures the range of IL‐17AA responses following drug administration. One subject (dashed line) in the 150 mg SC q4w dose group was considered an outlier and removed from fitting. (c) The model predicts the range of IL‐13 responses following drug administration. (d) Following the administration of single, intravenous doses, the model predicts the target neutralization of IL‐17AA and (e) IL‐13 in both the serum and healthy tissue (leaky and tight). Target neutralization is defined as 100*(free target/baseline target). Values for target neutralization were taken at Day 28.

3.2. Case Study 2: T Cell‐Dependent Bispecific Antibodies: Generalized Bispecific Model Framework Was Calibrated and Validated to Capture Mosunetuzumab PK in Cynomolgus Monkeys

The generalized bispecific model can also be used to investigate drugs that bind to two cell surface targets on different cells (trans‐binding), such as T cell‐dependent bsAbs. We aimed to demonstrate the ability of the model to capture the PK of mosunetuzumab, an anti‐CD20/CD3 bispecific molecule, in cynomolgus monkeys, and while doing so, demonstrate a workflow for how to estimate drug PK and cell surface target capacity. We employed a workflow (Figure 4) to estimate the total target capacity of CD3 and CD20 to inform TMDD and PK predictions of other large molecules, and then leveraged these target capacities to estimate the nonspecific clearance of mosunetuzumab in cynomolgus monkeys. We assumed for simplicity that all targets can be represented in the central compartment, demonstrating the versatility of the mPBPK model framework in capturing dynamics that are possible by simpler models. Model diagrams outlining the difference in structure for all simulations are shown in Figure S3. To estimate the total CD3 capacity, we calibrated the model to published anti‐gD/CD3 PK data in cynomolgus monkeys, where three molecules were dosed at 1 mg/kg: anti‐gD, anti‐gD/CD3‐low affinity, and anti‐gD/CD3‐high affinity (Figure 5a) [28]. The model was fit to these data using the nonspecific clearance for anti‐gD. The fit estimate for total CD3 capacity is 19 nM. To account for heterogeneity in CD3 expression, total T cells, and potential uncertainty in other model parameters such as binding affinity or target turnover, we assume a ±30% range on the total CD3 capacity, giving a range of 13–25 nM. To estimate the total CD20 capacity, we leveraged published PK data from obinutuzumab, an anti‐CD20 bivalent antibody (Figure 5b) [29]. The fit estimate for total CD20 capacity is 4.4 nM. Similar to CD3, we apply a ±30% range on the total CD20 capacity, giving a range of 3.3–5.8 nM. To validate our CD3 and CD20 target capacity predictions, we simulated our model against published data from two tool molecules: anti‐CD20/CD3W (weak) and anti‐CD20/CD3M (moderate), as well as an isotype control (Figure 5c) [30]. Our simulations reasonably captured the data from these anti‐CD20/CD3 tool molecules in cynomolgus monkeys. Lastly, we used the reported in vitro binding affinities of mosunetuzumab and typical antibody parameters for tissue distribution [19] to capture mosunetuzumab PK from a multi‐dose PK study in cynomolgus monkeys across dose levels from 0.01 to 1 mg/kg, fitting only the nonspecific clearance to the PK data (Figure 5d) [8]. The final set of key parameters used in these simulations are provided in Table S2. The final calibration allows us to parse the nonspecific clearance of mosunetuzumab from the specific, receptor‐mediated clearance mechanisms through CD20 and CD3.

FIGURE 4.

Workflow of generalized minimal PBPK model calibration to mosunetuzumab PK data in cynomolgus monkeys. The model was calibrated to mosunetuzumab PK data by estimating total CD20 and CD3 levels in cynomolgus monkeys using published antibody PK from other molecules against these targets. First, total CD20 levels were estimated using PK data from Obinutuzumab. Second, CD3 capacity levels were estimated using anti‐gD/CD3 cynomolgus monkey PK data. The model calibration was validated against PK data from anti‐CD3/CD20 bispecific. Finally, mosunetuzumab PK was calibrated by incorporating CD20 and CD3 levels, as well as nonspecific clearance.

FIGURE 5.

Model was calibrated and validated against multiple anti‐CD3 and anti‐CD20 free PK datasets in cynomolgus monkeys to predict mosunetuzumab (anti‐CD20/anti‐CD3) PK. (a) The total CD3 capacity in the model was calibrated to anti‐gD/CD3 PK data of three molecules with varying CD3 binding affinities at 1 mg/kg. The best fit line is shown, with the shaded region representing ±30% from the best fit. (b) The total CD20 capacity in the model was calibrated to Obinutuzumab (anti‐CD20) PK data at dose levels 1 and 10 mg/kg. Again, the best fit line is shown, with the shaded region representing ±30% from the best fit. (c) The CD3 and CD20 capacities were validated against anti‐CD20/CD3 tool molecules at 0.1 mg/kg with weak (w) and moderate (m) affinities to CD3. The shaded region represents ±30% variation from the nominal simulation. (d) The model captures multi‐dose mosunetuzumab PK across the dose range 0.01–0.1 mg/kg in cynomolgus monkeys. (e) Time course of mosunetuzumab PK with (solid lines) and without (dashed lines) incorporating T cell‐mediated drug disposition demonstrates the loss of exposure due to internalization through T cells. (f) Time course of mosunetuzumab PK with (solid lines) and without (dashed lines) incorporating B cell‐mediated drug disposition demonstrates little loss of exposure due to internalization through B cells.

To understand the extent to which CD3 on T cells contributes to the overall TMDD of mosunetuzumab, we compared simulations with and without T cells present across the same dose range. Simulations demonstrate that T cell‐mediated drug disposition has a substantial impact on the exposure of mosunetuzumab even at the highest dose tested (Figure 5e). The model predicts that CD20‐mediated clearance also impacts overall PK, albeit to a smaller extent than CD3‐mediated clearance (Figure 5f). Overall, these simulations reveal that even at the highest dose level tested in this study, nonspecific clearance alone cannot account for mosunetuzumab's total clearance, and a portion of this clearance is likely accounted for by T‐cell and B‐cell binding and internalization in cynomolgus monkeys.

3.3. Case Study 3: CD3 Affinity of T Cell‐Dependent Bispecific Antibodies Impacts Drug PK

Because overall T cell capacity had a major impact on mosunetuzumab PK in cynomolgus monkeys, we next simulated a clinical scenario. In this case study, we consider a bispecific T‐cell engager where the target cell receptor (non‐CD3) is expressed in either low levels or is quickly depleted post‐treatment, such that we can reasonably assume that the amount of target‐mediated drug disposition through the non‐CD3 arm is negligible. We simulated a single intravenous dose at four dose levels (0.01, 0.1, 1, and 10 mg/kg) using typical clinical antibody distribution and clearance parameters (Figure 6a), and the same model setup as shown in Figure S3C. We assumed the same CD3 concentration in blood that was estimated in cynomolgus monkeys. At each dose level, we ran a simulation using a moderate (40 nM) and low (400 nM) CD3 affinity, as well as a simulation with no CD3 binding. Our results demonstrate that, for moderate affinity CD3 antibodies, CD3 binding significantly decreases drug exposure at low dose levels. Our simulations show that most of the drug clearance at dose levels below 10 mg/kg is driven by CD3‐mediated target disposition rather than nonspecific clearance (Figure 6b). By weakening the CD3 affinity, the overall clearance is driven by nonspecific mechanisms (Figure 6c). These data emphasize the importance of the CD3 binding affinity in controlling overall drug exposure of bispecific T‐cell engaging antibodies.

FIGURE 6.

CD3 affinity greatly impacts exposure and fraction of drug cleared by CD3‐mediated disposition for T cell‐engaging bispecific antibodies. (a) Case study of a T‐cell engaging bispecific antibody in cynomolgus monkeys demonstrates the impact of CD3 affinity on free PK over 21 days. Area plots illustrate the percent of total drug clearance at Day 21 via nonspecific clearance (NS), CD20‐mediated TMDD, and CD3‐mediated TMDD for a bispecific T‐cell engaging antibody with a (b) high or (c) low binding affinity for CD3.

3.4. Case Study 4: Tumor‐Targeted Bispecific Antibodies for Tumor‐Specific Target Engagement

Tumor‐targeted bsAbs are under investigation as therapeutics that can target differentially expressed antigens between normal and tumor tissues to limit systemic toxicity by creating large differences in therapeutic target engagement in the tumor compared to healthy tissues. We investigated how two properties of tumor‐targeted bsAbs impact target engagement of the cell surface therapeutic receptor: (1) the ratio of affinities of each arm, and (2) the effective avidity. In this case study, we considered cells that express two receptors, the targeting receptor and the therapeutic receptor. The targeting receptor is more highly expressed in the tumor than in systemic blood and tissue, whereas the therapeutic receptor is expressed at the same levels in the central, leaky, tight, and tumor compartments (Table S3). These cells are present in all compartments, and the bsAb can distribute between compartments (Figure 7a). We consider the blood to be the safety/toxicity compartment and the tumor to be the efficacy compartment. To investigate the optimal properties of a clinical tumor‐targeted bispecific that will generate a receptor occupancy differential between the blood and tumor compartment, we evaluated a range of affinity ratios between the targeting and therapeutic arm of the bispecific across dose levels 0.01 mg/kg to 10 mg/kg (Figure 7b). The arrows point directionally from the central to tumor receptor occupancy at Day 21.

FIGURE 7.

The model predicts that tumor‐targeted bispecific antibodies can achieve large differences in central versus tumor therapeutic receptor occupancy when the avidity interaction is high. (a) The bispecific model is parameterized for a scenario where the efficacy compartment represents a tumor. Both the therapeutic and targeting receptors are present in all tissues except lymph. The model can capture PK of either a targeted bispecific antibody or a bivalent, monospecific antibody. (b) Across a dose range of single doses between 0.01 and 10 mg/kg and at varying ratios of affinities of antibody for the targeting and therapeutic arm, the model predicts the difference in therapeutic receptor occupancy between the central and tumor compartments at Day 21. Arrows are directed from the data point corresponding to the central to that of the tumor RO, that is, arrows pointing right indicate higher RO in tumor than central, and arrows pointing left indicate higher RO in central than tumor. The avidity factor was held constant at 1000 in these simulations. (c) In simulations with constant therapeutic and target receptor affinities, but varying effective avidity (1‐1e6), the model predicts that the effective avidity has minimal impact on therapeutic receptor engagement in the blood at Day 21 across a range of dose levels. (d) In the tumor, moderate effective avidity factors can drive therapeutic receptor engagement at Day 21. The black outline highlights the simulations with high therapeutic windows, where the therapeutic receptor engagement in the blood is less than 40% and in the tumor is greater than 60%. (e) The model predicts that a monospecific, bivalent antibody with an affinity of 10 nM will drive a dose‐dependent increase in therapeutic receptor engagement in the blood and (f) tumor.

The simulations demonstrate four key points. First, as the targeting: therapeutic arm affinity ratio decreases, the differential in therapeutic receptor occupancy between blood and tumor is lost, indicating that targeting to the tissue of interest is not occurring. Second, due to the higher systemic versus tumor exposure, high doses of tumor‐targeted bispecifics can diminish and even reverse the differential, such that there is higher therapeutic receptor occupancy in the blood than in the tumor compartment. Third, the optimal ratio of affinities of the targeting and therapeutic arms of the bsAb depends on the desired tumor RO and acceptable RO in the blood compartment. Fourth, while the largest affinity ratio may achieve optimal target engagement in the tumor, the benefits of high affinity ratios taper off above 1000‐fold, suggesting further increases to the targeting arm affinity may have diminishing returns (Figure 7b).

We then used the model to compare the PK and RO of a tumor‐targeted bsAb with a bivalent, monospecific antibody in patients with solid tumors. We simulated clinical RO profiles at a range of bispecific avidity factors shown to be relevant for a wide panel of antibodies [31]. These simulations use an affinity of the bsAb for the targeting receptor that is 2000‐fold higher than for the therapeutic receptor to ensure maximal targeting, whereas the bivalent antibody has a 10 nM affinity for the therapeutic receptor (Table S4).

We examined the effect of avidity across multiple dose levels ranging from 0.01 mg/kg to 10 mg/kg. While avidity has minimal impact on the RO in the blood, it controls the RO in the tumor. At low dose levels, there are larger differences in tumor RO compared to blood RO for the bsAb (Figure 7c,d). In this case study, we consider the therapeutic window to be defined by a safety threshold of less than 40% RO and an efficacy threshold of greater than 60% RO on Day 21. Depending on the avidity factor, the optimal dose levels within this therapeutic window can vary widely, from 0.01 mg/kg to 3 mg/kg (Figure 7c,d, denoted by black outlines).

Additionally, we compared dosing with the bispecific and bivalent antibody across this dose range. In the blood, the bispecific shows lower RO due to the lower binding affinity to the target than the bivalent molecule (Figure 7e). In the tumor, the RO depends on the avidity factor, and can be more, less, or the same as that of the bivalent molecule across this dose range (Figure 7f). Our results demonstrate the importance of understanding the difference in on‐site, on‐target engagement that a bispecific molecule will attain over a bivalent antibody when developing tumor‐targeted bsAb therapeutics.

4. Discussion

We built a generalized bispecific model for preclinical and clinical contexts across many bispecific mechanisms of action, indications, and applications. The model includes mechanistic details beyond two‐ or three‐compartment PK models (but can be reduced to capture dynamics typically captured by these models), yet is simpler than a full PBPK model. One useful property of this model is that it operates at a level of detail desired by the user; it is highly versatile and can be easily simplified to capture only blood dynamics or expanded to represent targets in tissues of interest. The model is relevant for scenarios where the bsAb binds to soluble targets or cell surface targets (via cis‐ or trans‐binding) expressed in blood, healthy tissue, or diseased tissue. We demonstrated the utility of the model by applying it to study the PK of BITS7201A, an anti‐IL‐17/anti‐IL13 bispecific molecule, mosunetuzumab, a clinically approved anti‐CD20 bispecific T‐cell engager, and other prototypical bispecific T‐cell engagers and tumor‐targeted bsAbs.

The generalized bispecific model was parameterized to capture the PKPD of BITS7201A, an anti‐IL‐17/IL‐13 bsAb. We calibrated the model to published clinical data and predicted the extent of IL‐17 and IL‐13 neutralization in the central, leaky, and tight tissue compartments. While the overall aim of this model is to support preclinical and translational predictions, these results highlight how the model can also be used to support drug development in clinical scenarios.

From early preclinical development through Phase I clinical studies, it is difficult to predict PK for bsAbs. In this work, we presented an example of how to estimate PK of mosunetuzumab by analyzing PK data from other molecules targeting CD20 and CD3 individually. We first analyzed PK from a set of anti‐gD/CD3 molecules having high and low affinity to CD3 to estimate the CD3 capacity. We next analyzed PK from obinutuzumab to estimate the CD20 capacity. Lastly, we validated the CD3 and CD20 capacities against another set of anti‐CD20/CD3 tool molecules and used these capacities to narrow the parameter space for prediction of mosunetuzumab PK. This strategy can aid in a priori prediction to inform a cynomolgus monkey study, or it can be used after acquiring PK data to constrain the parameter ranges of target capacity, target turnover, and nonspecific clearance to fit the data.

Our model captured the PK of mosunetuzumab, the anti‐CD3/anti‐CD20 T‐cell engaging bsAb, following multi‐dose administration in cynomolgus monkeys. By incorporating both the nonspecific and specific clearance into the model, we can examine how much clearance is coming directly from T‐cell and B‐cell mediated disposition, and the timing of the onset of each clearance type, which can aid in understanding PK following repeat dosing. At dose levels up to 1 mg/kg, the model predicts that T‐cell mediated clearance plays a role in the clearance of mosunetuzumab, even though there is no obvious nonlinearity in the PK profiles. Adequately capturing both T cell‐mediated and nonspecific clearance for T‐cell engaging bsAbs preclinically will allow us to allometrically scale the nonspecific clearance and use a clinical CD3 target capacity to fully capture the clinical PK profile. Consistent with known CD3 expression profiles in vivo [32], our findings suggest that the impact on clearance of the CD3 capacity in cynomolgus monkeys is high and should be considered when translating PK from preclinical studies using mechanistic models of T‐cell engaging bsAbs.

T‐cell engaging bsAbs with low CD3 affinities may be favorable drug candidates due to their lower risk of eliciting cytokine release syndrome, which has been associated with excessive T‐cell engagement [33, 34]. Our simulations indicate reducing the CD3 affinity may drive higher exposure by minimizing target‐mediated drug disposition. Previous work estimates that few (< 100) drug‐CD3‐target trimers are necessary for the T‐cell engaging bsAb mechanism of action to achieve half maximal cell lysis [6, 35], suggesting that appropriately reduced CD3 affinity may not compromise efficacy. Additionally, in the development of T‐cell engaging bsAbs, a clone with reduced CD3 affinity can potentially rescue the PK of a molecule that has a higher nonspecific clearance compared to a high affinity CD3 clone [36]. The importance of CD3 affinity on bispecific T‐cell engaging antibody PK highlights the utility of mechanistic PK modeling in informing molecule design.

Tumor‐targeted bsAbs are being explored for therapeutic targets that have significant expression in non‐tumor tissues [10, 11]. We used the generalized bispecific model to simulate the PK/RO relationship for a prototypical tumor‐targeted bsAb, and we predicted that the relationship between dose and avidity is crucial for adequate but non‐excessive therapeutic target engagement. At high doses, on‐target off‐tumor binding is likely unavoidable, and toxicity poses a risk. At low doses, on‐target on‐tumor binding may be low, and efficacy may not be reached. These types of bsAbs may rely on high avidity to be more viable as drug candidates than their bivalent antibody counterparts. Effective avidity will depend on the expression level and number of cells expressing both the therapeutic target and the tumor‐specific antigen co‐target. Our findings support the need to fully characterize the strength of avidity between the tumor‐specific antigen and co‐target receptor if using a bsAb to increase tumor localization is the therapeutic strategy.

Our simulations examined the potential advantages of a tumor‐targeted bispecific over a bivalent, monospecific antibody. In this case study, we considered a prototypical tumor‐targeted bispecific with high affinity for the tumor‐specific antigen and low affinity for the therapeutic target, and a bivalent, monospecific antibody with high affinity for the therapeutic target. We found that despite having half the amount of binding arms as the bivalent antibody, a tumor‐targeted bispecific can exhibit increased or similar target engagement in the tumor while maintaining lower systemic target engagement. These findings suggest that the mechanism of action of the tumor‐targeted bispecific is advantageous in specific scenarios where systemic toxicity is a major concern and the tumor target does not need to be fully saturated for drug efficacy. Our modeling suggests it is difficult to leverage a tumor‐targeted bsAb if the safety threshold is relatively low (< 40% RO) and the efficacy threshold is relatively high (> 60% RO), because some off‐tumor on‐target engagement is inevitable.

To maximize potential for clinical success, three key aspects of the tumor‐targeted bispecific should be considered. First, there should be a large differential target expression between the tumor‐specific antigen and therapeutic target to drive binding to the tumor‐specific antigen instead of the therapeutic receptor. The generalized bispecific model is a useful tool to inform the necessary systemic‐to‐tumor expression level differential required because the model can bridge the knowledge gap between dose, affinity, avidity, and toxic or efficacious levels of target engagement. Second, the drug should have a high affinity for the tumor‐specific co‐target and a low affinity for the therapeutic target. Third, the avidity factor should be experimentally estimated and maximized. These key biological and drug properties allow for the largest therapeutic window by minimizing the risk of systemic toxicity while driving tumor‐selective target engagement.

The generalized bispecific model was designed to be repurposed for new bsAbs by updating relevant molecule‐, species‐, and biology‐related parameters. Future work on this model may also involve adding more mechanistic detail to capture the trafficking dynamics of certain cell types (and thus, the receptors expressed on their surface), or PKPD dynamics of engineered cytokines [37, 38]. We highlight that the generalized bispecific model can be used in early‐stage research to aid in molecule design as well as late‐stage research through clinical development and post‐approval to aid in dose selection.

Author Contributions

All authors wrote the manuscript. P.S., S.R., and I.H. designed the research. P.S. performed the research. All authors analyzed the data.

Funding

The authors have nothing to report.

Conflicts of Interest

P.S., L.J., K.G., and I.H. are full‐time employees at Genentech Inc. S.R. was a full‐time employee at Genentech Inc. during this work.

Supporting information

Data S1: psp470167‐sup‐0001‐DataS1.zip.

Spinosa P., Joslyn L., Ramanujan S., Gadkar K., and Hosseini I., “A Generalized Minimal PBPK‐PD Model of Bispecific Antibodies: Case Studies and Applications in Drug Development,” CPT: Pharmacometrics & Systems Pharmacology 15, no. 2 (2026): e70167, 10.1002/psp4.70167.