Introduction

MER proto-oncogen, tyrosine kinase (MerTK) is a receptor tyrosine kinase mainly expressed on macrophages (1). MerTK participates in normal tissue homeostasis by mediating efferocytosis, a process in which apoptotic cells are phagocytosed by macrophages (2, 3). In tumor tissues, macrophages overexpress MerTK and activation of MerTK can lead to secretion of anti-inflammatory cytokines which contribute to immunosuppression (4, 5). In mice, targeting MerTK on macrophages has been shown to reduce tumor growth (6, 7) and enhance the efficacy of immune checkpoint blockade (ICB) by augmenting anti-tumor immunity (8, 9). In addition to macrophages, MerTK is expressed on retinal pigment epithelial cells (RPEs), where it functions to facilitate the phagocytosis of shed photoreceptor outer segments (POS) (10, 11). Removal of shed POS by RPEs is critical for hemostasis of photoreceptors, as rodents and patients carrying MerTK mutations are predisposed to retinal dystrophy (12,13,14). Recent studies showed that the reduction of MerTK activity by pharmaceutical inhibitors can lead to retinal degeneration in mice (15, 16), which underscores the importance of targeting MerTK in tumor tissues and sparing MerTK in retinal tissues.

Bispecific antibodies (bsAb), antibodies that are designed to target two different antigens (17,18,19,20), can be used to selectively target MerTK on tumor tissues. BsAbs have been used to improve on-tumor efficacy and reduce off-tumor toxicity of therapeutics that target antigens expressed on both tumor and healthy tissues (21,22,23). The enhanced specificity of bsAbs is achieved by designing the bsAbs to contain two arms: (1) a therapeutic arm that binds to a target co-expressed on cells in tumors as well as healthy tissues, and (2) a targeting arm that binds to a target over-expressed on cells in the tumor tissues. The preferential binding of the targeting arm to tumor tissues will enhance, through avidity effect, the binding of the therapeutic arm to its target in tumors. Therefore, the targeting arm directs the pharmacological effect of the therapeutic arm toward the tumor tissues and increases the therapeutic index (TI) of the drug. Although a bsAb could be designed to maximize the efficacy and minimize the toxicity of MerTK-targeted antibodies, it is unclear what the key determinants are for an optimal design of MerTK bsAb.

We present a modeling framework, based on target-mediated drug disposition (TMDD) model (24,25,26,27), in vivo pharmacokinetics (PK), and in vitro receptor occupancy (RO) data, to predict the TI and optimal design of anti-MerTK antibodies. We first utilized the TMDD model (26) to predict the efficacious and toxic doses of an anti-MerTK monospecific antibody (msAb). This prediction was made by comparing the simulated tumor and retina PK of the drug with the experimentally determined in vitro effective concentrations (EC) of the drug that would induce 90% and 20% MerTK RO on macrophages and RPEs (28,29,30,31). Our prediction was in-line with in vivo efficacy and toxicity data, indicating that in vitro RO thresholds could be used to predict in vivo efficacy and toxicity. To widen the therapeutic window of the anti-MerTK antibody, we generated bsAbs targeting both MerTK and Programmed Death-Ligand 1 (PD-L1). Since PD-L1 is overexpressed on macrophages but not on RPEs, and both PD-L1 and MerTK are highly expressed on macrophages (8, 32), the PD-L1-targeting arm would improve the binding of the anti-MerTK arm to macrophages. We then utilized a bispecific TMDD model, in combination with in vitro RO data and RO thresholds defined from msAb data, to predict the efficacious doses, toxic doses, and TI of anti-MerTK bsAbs. Our model predicted the highest TI for the bsAb with an attenuated MerTK binding arm, which was in-line with in vivo observations. With the validated modeling framework, we utilized sensitivity analyses and parameter scans to suggest a theoretical molecular design for the next generation anti-MerTK bsAb to optimize TI in humans.

Materials and Methods

Generation and Characterization of Therapeutic Antibodies

Bivalent msAbs targeting MerTK and PD-L1 were generated, expressed, and purified, as described previously (33). Two variants of anti-MerTK/PD-L1 bsAbs were generated—one with a potent (wild type/WT) and one with an attenuated (Atten) anti-MerTK binding arm. Anti-MerTKWT/PD-L1 and anti-MerTKAtten/PD-L1 bsAbs were assembled via “knobs-into-holes” technology (34, 35). All therapeutic antibodies contained a mIgG2a backbone. The monovalent binding affinity (KD), association rate constant (kon), and dissociation rate constant (koff) of binding arms were measured by surface plasmon resonance (SPR) (8).

In Vitro Characterization of Antibodies

In vitro RO assay was performed to evaluate the binding of antibodies to MerTK on macrophages and RPEs. Splenic macrophages and RPEs were isolated from C57BL/6 mice (36, 37). Macrophages and RPEs were pre-treated with various concentrations of anti-MerTK msAb or bsAbs for 6 h. Following pre-treatment, the cells were fixed with 4% paraformaldehyde (4% PFA) and subsequently incubated with PE-conjugated anti-MerTK staining antibody targeting the same epitope for 30 min. The degree of anti-MerTK fluorescent staining, representing the amount of MerTK not occupied by therapeutic antibodies, was quantified using flow cytometric analysis. Percent receptor occupancy (RO) was calculated by normalizing the degree of fluorescent staining for the drug-treated sample to that of isotype control. RO versus drug concentration curves were constructed, and the effective concentrations for 20% or 90% MerTK RO in RPE or macrophages (ECRO20RPE and ECRO90MΦ) were calculated from the curve using non-linear regression analysis performed with Prism (GraphPad). In vitro Gas6-MerTK binding and efferocytosis assays were performed, as detailed in Supplementary Information (SI) (8).

In Vivo PK, Toxicity, and Efficacy Studies

Mouse studies were performed to evaluate the in vivo PK, efficacy, and toxicity of anti-MerTK msAb, anti-MerTKWT/PD-L1 bsAb, and anti-MerTKAtten/PD-L1 bsAb. Detailed description of these in vivo studies can be found in SI.

Analysis of PK/TK Parameters and Statistical Analysis

Detailed procedures for statistical analysis and analysis of PK/TK parameters can be found in SI.

Development of TMDD Model and Prediction of Efficacious and Toxic Doses

TMDD models were utilized to describe the serum PK of unbound anti-MerTK msAb and bsAbs in mice and to predict the efficacious and toxic doses of these antibodies. Model structures are presented in figures (Figs. 2a and 4a). Equations and parameters used in the model can be found in SI (Table S1–S3). Predicted tumor or retina PK of the unbound drugs was calculated by assuming a tumor:serum partition of 5–10% (38, 39) or retina:serum partition of 0.25–1% (40), respectively. Toxic doses were assumed to be doses in which retina concentration was higher than or equal to ECRO20RPE (≥ ECRO20RPE), with the highest dose below the predicted toxic range deemed predicted maximum tolerable dose (MTD). Efficacious doses were assumed to be doses in which tumor concentration was greater than or equal to ECRO90MΦ (≥ ECRO90MΦ), with the lowest dose in the efficacious range deemed predicted minimum efficacious dose (MED). About 90% RO was assumed to be the threshold for efficacy since previous studies reported that > 90% RO is needed for antagonists to completely inhibit their targets (29, 31). About 20% RO was assumed to be the threshold for toxicity since this is the RO threshold commonly used in industry for determining clinically safe first-in-human dose (28, 30). Predicted TI was calculated as predicted MTD divided by predicted MED.

A two-compartment bispecific TMDD model was used to describe the PK of anti-MerTK bsAbs, where the drugs were distributed from the central compartment to the peripheral compartment with distribution clearance rate (CLd) and eliminated from the central compartment with non-specific clearance rate (CLns). Since both MerTK and PD-L1 are expressed in multiple tissues including blood and liver, the targets were lumped into the central compartment to facilitate the parameter identifiability (7, 41, 42). Free drug binding to either target can form a dimer and the subsequent binding of the other target will form a trimer. Since both targets are on the same cell, the binding of one arm of the drug to one target can enhance the binding of the other arm to the other target. This enhancement in binding is represented by an avidity constant (y) which is applied to the association rate constants for the formation of trimer (43). An in vitro bispecific model was used to estimate the avidity constant by fitting the model to the in vitro RO data (details in SI). For the anti-MerTK msAb, a monospecific two-compartment TMDD model was used to describe the serum PK of the drug. The parameters in the models were estimated by fitting the model to experimental serum PK data with MATLAB Simbiology software (v2020a, Mathwork, MA) using particle swarm global fitting and combined error model. The parameter estimates were validated by comparing the serum TK prediction with experimental TK data. Goodness-of-fit and residual plots were constructed. To reduce the number of parameters that require model fitting and estimation, no explicit representation of drug binding in tumor or retina compartment was included in the model.

TMDD Model Informs Antibody Design with Optimal TI in Humans

To simulate the PK profiles of anti-MerTK bsAbs in humans, key PK parameters in the bispecific TMDD model were assumed to be the typical values for mAbs in humans (44). Derivative-based sensitivity analysis was performed with MATLAB Simbiology software to identify parameters in the model with the largest impact on the PK of bsAbs in humans. Parameter scans were then conducted by varying these parameters (from 5 to 500% of the original values) to calculate predicted MTD, MED, and TI for each of the parameter values. The values that led to maximum predicted TI were identified. To obtain the optimal monovalent KD for each arm of the bsAbs, the in vitro bispecific model was used to systematically alter the KD values to find combinations that would result in high predicted ECRO20RPE and low predicted ECRO90MΦ, while keeping the avidity constant unchanged. Promising KD combinations were further evaluated using the bispecific TMDD model to identify an optimal combination that would maximize the predicted TI.

Results

Targeting MerTK Augmented Anti-tumor Activity of Anti-PD-L1 Antibody, but Resulted in Toxicity in a Preclinical Mouse Model

An anti-MerTK msAb with high affinity (KD against mouse MerTK = 0.6 nM) was produced (Fig. 1a and Table I). Anti-MerTK msAb efficiently blocked MerTK binding to its target, Gas6, and inhibited macrophage efferocytosis in vitro (Fig. 1b and c). We also evaluated the ability of anti-MerTK msAb to bind to MerTK on both macrophages and RPEs, as inhibiting MerTK on RPEs can lead to retinal degeneration (15, 16). Using an in vitro RO assay, we found that anti-MerTK mAb binds to MerTK on macrophages and RPEs with a similar potency (Fig. 1d and e, ECRO50MΦ = 0.2 μg/mL vs. ECRO50RPE = 0.12 μg/mL).

Fig. 1
figure 1

In vitro and in vivo pharmacology of a bivalent monospecific anti-MerTK antibody. a SPR sensorgram of anti-MerTK monospecific bivalent antibody (anti-MerTK msAb). b Percent inhibition of the binding of mouse Gas6 to mouse MerTK in the presence of varying concentrations of an anti-MerTK msAb. IC50 of this inhibition (0.04 nM) was calculated. c The effects of PBS, control IgG, or anti-MerTK msAb on macrophage efferocytosis were quantified as the percent of total macrophages that took up dexamethasone-treated apoptotic T cells in an in vitro cell culture assay (mean ± SEM). d–e Receptor occupancy of MerTK on the surface of macrophages (d) or RPEs (e) following the treatment with varying concentrations of anti-MerTK msAb. f Serum concentration–time PK profile of anti-MerTK msAb (mean ± SEM) in mice following single IV dosing of the antibody at 1, 10, or 30 mg/kg. g Tumor growth curves of MC38 subcutaneous tumors treated with anti-MerTK msAb (30 mg/kg BIW), anti-PD-L1 msAb (5 mg/kg BIW), and/or anti-gp120 control in mice

Full size image
Table I Binding Properties of Wild-type Anti-MerTK msAb
Full size table

The PK of the anti-MerTK msAb was assessed in healthy C57BL/6 mice following single IV administration (Fig. 1f). PK parameters demonstrated an increase in AUC and a decrease in clearance (CL) with increasing dose, indicating TMDD of anti-MerTK msAb in mice (Table II). Next, C57BL/6 mice bearing subcutaneous MC38 tumors were treated with anti-MerTK msAb at 30 mg/kg twice weekly (BIW) and/or anti-PD-L1 msAb at 5 mg/kg BIW. Anti-MerTK msAb treatment significantly reduced CD206 expression and enhanced CD86 expression on CD45 + immune cells in the tumors, indicating polarization of immune cells to pro-inflammatory phenotypes (Fig. S1). Combination treatment of anti-MerTK msAb and anti-PD-L1 msAb significantly reduced tumor growth (Fig. 1g), whereas single agent anti-PD-L1 msAb did not demonstrate anti-tumor activity. These data suggest that anti-MerTK msAb enhances the efficacy of immune checkpoint blockade by promoting a pro-inflammatory phenotype of tumor-associated immune cells.

Table II Serum PK Parameters of Anti-MerTK msAb in C57BL/6 Following a Single IV Injection of the Antibody
Full size table

To assess the tolerability of anti-MerTK msAb, a 4-week toxicity study was performed with healthy BALB/c mice dosed at 5, 30, and 45 mg/kg BIW. BALB/c mice were selected as a conservative safety assessment model, as these mice are predisposed to the development of retinal lesions (45). TK analysis showed sustained serum exposure of the drug over the duration of 24 ~ 28 days at all dose levels (Fig. S2 and Table S4). Additionally, histopathology analysis revealed retinal lesions at a dose of 5 mg/kg and significant damage to retinal structures at 30 mg/kg and 45 mg/kg (Table S5). These data demonstrate that exposure to anti-MerTK msAb results in retinal toxicity in mice at all dose levels tested.

Two-Compartmental TMDD Models, Combined With In Vitro RO Data, Adequately Described the Dose/PK/Efficacy/Toxicity Relationship of Anti-MerTK msAb in Vivo

We next assessed whether an optimal dosing level exists for anti-MerTK msAb, such that we can achieve anti-tumor efficacy without inducing retinal toxicity. To address this question, we built a modeling framework to predict the efficacious and toxic doses of anti-MerTK msAb based on in vivo PK and in vitro RO data. A two-compartment monospecific TMDD model was used to describe the serum PK of the anti-MerTK msAb (Fig. 2a). Parameters in the TMDD models were estimated by fitting the model to the in vivo serum PK profile of anti-MerTK msAb in mice (Table S6). The model fit was able to adequately capture the experimental serum PK data (Fig. 2b; S3a, and S3b). We then validated the model against the TK data for mice receiving 30 mg/kg anti-MerTK msAb BIW. The predicted serum TK profile matches well with the experimental observation (Fig. 2c and S3c–d).

Fig. 2
figure 2

Two compartment target-mediated drug disposition (TMDD) PK model for anti-MerTK monospecific antibody enables the prediction of in vivo efficacy and toxicity. a Schematic depicting the two-compartment TMDD model for the anti-MerTK msAb PK. b Model fit (line) and individual experimental observation (dot) of serum concentration–time PK profile of anti-MerTK msAb in mice following a single IV injection of the antibody at a dose of 1, 10, or 30 mg/kg. c The model prediction (line) and experimental observation (dot) of serum concentration–time TK profile of an anti-MerTK msAb in mice following IV injection of 30 mg/kg of the antibody twice weekly. d Predicted concentration–time profile of an anti-MerTK msAb in retina (left) or tumor (right) in mice following various dosing regimens. Shaded area represents predicted toxic (red) or efficacious (purple) ranges based upon in vitro MerTK receptor occupancy data. e Percent complete responder quantification for MC38 subcutaneous tumors treated with anti-PD-L1 msAb or anti-MerTK msAb at various dosing levels as indicated (0–45 mg/kg)

Full size image

We used the TMDD model to simulate the serum PKs of the mice receiving BIW dosing of 1–60 mg/kg anti-MerTK msAb (Fig. S4). Since tumors and retina are the active sites for anti-MerTK antibodies, we further simulated the antibody concentration in these two tissues by assuming a tumor:serum drug partition of 5% (38) and a retina:serum drug partition of 0.25% (40) (Fig. 2d). By comparing the simulated tumor and retina PK to ECRO90MΦ (0.6 μg/mL) and ECRO20RPE (0.08 μg /mL) for anti-MerTK msAb, our model predicted that anti-MerTK msAb is efficacious at doses ≥ 2.5 mg/kg and toxic at doses > 1 mg/kg (Fig. 2d). These predictions are consistent with experimental observations, since a dose-ranging in vivo study demonstrated an MED of 2.5 mg/kg for anti-MerTK msAb (Fig. 2e and S5), while a dose as low as 5 mg/kg induces retinal lesion in subjects (Table S5). This consistency of the model prediction with the experimental results indicates that our modeling framework and in vitro RO threshold of 20% MerTK RO on RPEs and 90% MerTK RO on macrophages could be used to predict efficacious and toxic doses.

Anti-MerTK/PD-L1 Bispecific Antibodies Enhance Targeting to Macrophages While Reducing the Targeting to Retinal Cells

To mitigate the retinal toxicity of anti-MerTK antibodies, we developed bsAbs targeting both MerTK and PD-L1 (Fig. 3a). Since PD-L1 is highly expressed on macrophages (Fig. 3b) but not on RPEs (Fig. 3c), a targeting arm for PD-L1 was hypothesized to drive the binding of the anti-MerTK therapeutic arm to macrophages. Two bsAb variants were produced, one with a potent anti-MerTK Fab (anti-MerTKWT/PD-L1 bsAb, monovalent KD against mouse MerTK = 1.33 nM) and the other with an affinity-attenuated anti-MerTK Fab (anti-MerTKAtten/PD-L1 bsAb, monovalent KD against mouse MerTK = 37.6 nM), with the majority of the difference in KD attributed to dissociation rate constant (koff) (Table III). An in vitro RO assay was performed to assess the binding of these bsAbs to MerTK expressed on macrophages and RPEs. Both bsAbs demonstrated ~ sixfold decrease in ECRO90MΦ compared to anti-MerTK msAb. In addition, anti-MerTKWT/PD-L1 and anti-MerTKAtten/PD-L1 showed a 5- and a 56-fold increase, respectively, in ECRO20RPE compared to anti-MerTK msAb (Fig. 3d). Serum concentration–time profiles of both bsAbs (Fig. 3e) and PK parameters obtained from NCA analysis (Table IV) demonstrated non-linear dose-exposure relationships reflected in CL and AUCinf values, suggesting target engagement and TMDD of the bsAbs in mice.

Fig. 3
figure 3

In vitro and in vivo pharmacology of bispecific anti-MerTK/PD-L1 antibodies. a Schematic of the anti-MerTK/PD-L1 bispecific antibody. b–c Flow cytometric analysis of PD-L1 and MerTK expression on mouse macrophages (b) or mouse retina pigment epithelial cells (RPEs) (c). Numbers indicate the percentage of cells that express PD-L1 or MerTK. d Receptor occupancy of MerTK on the surface of macrophages (left) or RPEs (right) following the treatment with varying concentrations of anti-MerTK msAb or anti-MerTK/PD-L1 bsAbs. e Serum concentration–time PK profile of anti-MerTKWT/PD-L1 bsAb (left, mean ± SD) or anti-MerTKAtten/PD-L1 bsAb (right, mean ± SD) in mice following a single IV bolus injection of the antibody at a dose of 1, 10, or 30 mg/kg

Full size image
Table III Monovalent Binding Properties of Wild-type Anti-MerTK (Anti-MerTKWT), Attenuated Anti-MerTK (Anti-MerTKAtten), and Anti-PD-L1 Binding Arms
Full size table
Table IV Serum PK Parameters of Anti-MerTK bsAb in C57BL/6 Following Single IV Injection of the Antibody
Full size table

Bispecific TMDD Models Adequately Described the PK of Anti-MerTK bsAbs and Predicted the Dose/Efficacy/Toxicity Relationship

We used a bispecific TMDD model and in vitro RO data to predict the efficacious and toxic dose of the anti-MerTK/PD-L1 bsAbs. To accomplish this, we utilized the bispecific TMDD model to simulate serum PK of bsAbs and compared the simulated tumor and retina PK to the ECRO90MΦ and ECRO20RPE. The predicted efficacious and toxic doses for bsAb were calculated in the same manner as for the anti-MerTK msAb. Predicted MTD was calculated as the maximum dose that resulted in simulated retina concentrations just below ECRO20RPE (< ECRO20RPE), while predicted MED was calculated as minimal dose that resulted in simulated tumor concentrations that are higher than or equal to ECRO90MΦ (≥ ECRO90MΦ).

The bispecific TMDD model included the binding of the bsAb to MerTK (Target A) or to PD-L1 (Target B), as well as the binding of single target-drug complex to the other target (i.e., Complex A to Target B). Because binding of the bsAb to Target A can increase the binding of the bsAb to Target B on the same cell surface, an avidity constant (y) was used to represent this increase in affinity (Fig. 4a). Using this bispecific modeling framework, as well as the measured monospecific kon and koff for each Fab (Table III), we performed parameter estimation by fitting the model to experimental serum PK data of anti-MerTKWT/PD-L1 and anti-MerTKAtten/PD-L1 bsAbs (Tables S7 and S8). The avidity constant was estimated by using the bispecific TMDD model and in vitro MerTK RO data (Fig. S6; Table S9). The resulting model fits adequately described the serum PK data for both bsAbs, as confirmed by residual and goodness-of-fit plots (Fig. 4b and c and S7).

Fig. 4
figure 4

Two-compartment bispecific TMDD PK model for the anti-MerTK/PD-L1 bispecific antibody enable the prediction of in vivo efficacy and toxicity. a Schematic depicting the two-compartment TMDD model for anti-MerTK bsAb PK. b–c Model fit (line) and individual experimental observation (dot) of serum concentration–time PK profile of an anti-MerTKWT/PD-L1 bsAb (b) or anti-MerTKAtten/PD-L1 bsAb (c) in mice following a single IV injection of the antibody. d–e Predicted concentration–time profile of an anti-MerTKWT/PD-L1 bsAb (d) and anti-MerTKAtten/PD-L1 bsAb (e) in retina (left) and tumor (right) in mice following various dosing regimens. Shaded area represents predicted toxic (red) and efficacious (purple) ranges based upon in vitro MerTK receptor occupancy data. f Ranges of predicted TI for anti-MerTK msAb and bsAbs based upon various retina:serum or tumor:serum partition coefficients of the antibody. g–h The model prediction (line) and experimental observation (dot) of serum concentration–time TK profile of an anti-MerTKWT/PD-L1 bsAb (g) or an anti-MerTKAtten/PD-L1 bsAb (h) in mice following twice a week IV injection of 30 mg/kg of the antibody

Full size image

To predict efficacious and toxic doses, we first used the bispecific TMDD model to simulate the serum, tumor, and retinal PKs in the mice receiving BIW dosing of 1–60 mg/kg anti-MerTK bsAb (Fig. S8; Fig. 4d and e). Simulated tumor PK predicted that both bsAbs should be efficacious at doses ≥ 3.5 mg/kg (predicted MED = 3.5 mg/kg). Simulated retina PK predicted that anti-MerTKWT/PD-L1 bsAb would be toxic at doses > 5 mg/kg (predicted MTD = 5 mg/kg), while anti-MerTKAtten/PD-L1 bsAb would be toxic at doses > 30 mg/kg (predicted MTD = 30 mg/kg). With these definitions, the simulation predicted a TI of 1.43 for anti-MerTKWT/PD-L1 bsAb and a TI of 8.6 for anti-MerTKAtten/PD-L1 bsAb. To test how variabilities in tumor:serum or retina:serum partitions impacted MTD/MED/TI predictions, we performed the simulation by varying the retina:serum partition between 0.25 and 1% and the tumor:serum partition between 5 and 15%. Even when accounting for the variability in the drug partitions, these results predicted higher TIs for anti-MerTKAtten/PD-L1 bsAb compared to anti-MerTKWT/PD-L1 bsAb or anti-MerTK msAb (Fig. 4f and Table S10).

We then compared the simulation results with the in vivo efficacy and toxicity results. Both anti-MerTK/PD-L1 bsAbs had comparable anti-tumor efficacy in MC38 tumor models (Fig. S9). A 4-week toxicity study was also performed with healthy BALB/c mice to evaluate the tolerability of anti-MerTK bsAbs, with both bsAbs dosed at 30 mg/kg twice weekly. Serum concentration–time TK profile (Fig. 4g and h) and parameters (Table S11) revealed sustained exposure of both bsAbs throughout the course of the study. Model-predicted serum TK profiles of both bsAbs match well with experimental serum TK data (Fig. 4g and h; Fig. S10). Histopathology revealed significantly reduced retinal lesions in animals that received bsAbs as compared to animals that were treated with 30 mg/kg of anti-MerTK msAb. Anti-MerTKAtten/PD-L1 bsAb was better tolerated than anti- MerTKWT/PD-L1 bsAb. Specifically, the attenuated variant of the bsAb resulted in no retinal lesions with the exception of one animal that displayed minimal vacuolation of the retina photoreceptor layer (Table S12). Taken together, these in vivo data revealed that anti-MerTKAtten/PD-L1 bsAb was better tolerated and produced equivalent efficacy compared to the other anti-MerTK therapeutic antibodies, matching well with our prediction that anti-MerTKAtten/PD-L1 has the highest TI.

TMDD Model Suggested Molecular Attributes of Anti-MerTK bsAb that Could Maximize the Therapeutic Index in Humans

Next, we evaluated the molecule design parameters to maximize TI of anti-MerTK bsAb. First, we conducted a sensitivity analysis at doses of 1, 10, and 30 mg/kg, using the bispecific model and typical human PK parameters (Vc, Vp, and CLd), to identify the parameters in the model that significantly contributed to serum PK of anti-MerTK bsAbs in humans. MerTK and PD-L1 concentrations in humans were assumed to be the same as those in mice. As expected, serum PK is the most sensitive to changes in non-specific clearance (CLns) of the antibody at 30 mg/kg. In contrast, at 1 ~ 10 mg/kg, serum PK is more sensitive to the expression level of antigen that the targeting arm binds to (RB, Target B) and the turnover rate of that antigen (kdegB) (Fig. 5a and S11). Altogether, these results indicated that selecting an antibody with an appropriate CLns and/or identifying an antigen with appropriate properties that the targeting arm binds to (i.e., RB, kdegB) is the most effective method to modify the exposure of anti-MerTK bsAbs (Fig. S11).

Fig. 5
figure 5

Sensitivity analysis and parameter scans reveal a molecular design for anti-MerTK bsAb that maximizes TI in human. Serum, retina, and tumor PKs in human were simulated using the bispecific TMDD model and human parameters. Sensitivity analysis and parameter scans were performed to assess how changes in model parameters affect MTD, MED, and TI in human. Based on the parameter scan, an optimal molecular design for anti-MerTK bsAb with parameters that maximize TI in human is then suggested. a Results of the sensitivity analysis showing the parameters in the bispecific TMDD model that contribute significantly to serum PK profile of anti-MerTKAtten bsAb in humans. b–d Variations in MTD, MED, and TI of anti-MerTKAtten bsAb as the results of changes in Target B concentration (RB) (b), non-specific clearance (CLns) (c), and turnover rate of target B (kdegB) (d). e–f Variations in ECRO20RPE (e) and ECRO90MΦ (f) as the results of changes in monovalent KD of MerTK binding arm (KD_MerTK) and targeting arm (KD_targeting arm). g Retina (top) and tumor (bottom) PK of the anti-MerTK bsAb with optimal parameters that maximize the TI. Shaded area represents predicted toxic (red) and efficacious (purple) ranges based on MerTK receptor occupancy data

Full size image

To obtain desirable RB, CLns, and kdegB that could lead to optimal TI, we evaluated how changes in these parameters affected predicted MTD, MED, and TI of the drug in humans by performing parameter scans using the bispecific TMDD model. Our results showed that while decreasing the concentration of Target B (RB) can significantly reduce MED, the impact on MTD was minimal, leading to high TI at low RB (Fig. 5b). In contrast, increasing CLns greatly increased MTD but had a modest effect on MED, resulting in a local maximum TI at ~ 13.2 mL/day/kg (Fig. 5c). Furthermore, decreasing kdegB slightly decreased both MTD and MED, leading to a maximum TI at low target turnover rate (Fig. 5d).

Next, we performed parameter scans using the in vitro bispecific model (Fig. S6a) to evaluate how monovalent KD of the anti-MerTK arm (KD_MerTK) and the targeting arm (KD_Targeting arm) impact ECRO20RPE and ECRO90MΦ. We found that decreasing KD_MerTK significantly decreased both ECRO20RPE and ECRO90MΦ, while decreasing the KD_Targeting arm significantly decreased ECRO90MΦ, but not ECRO20RPE (Fig. S12a–d). The ratio of ECRO20RPE to ECRO90MΦ is maximized at a KD_MerTK range of 25–500 nM and a KD_Targeting arm range of 20–100 pM (Fig S12e and f). We then performed a two-dimensional sensitivity analysis to evaluate how simultaneous variation in both affinities in these ranges affected predicted MTD, MED, and TI. We found that at a low KD_Target arm of 26 pM and an intermediate KD_MerTK of 300 nM, the bispecific TMDD model predicted a high ECRO20RPE and a low ECRO90MΦ that maximized TI (Fig. 5e, f, and S13). Simulation performed using these two optimal KD values, combined with previously optimized CLns, kdegB, and RB values, predicted retina and tumor PKs that resulted in predicted MTD of 120 mg/kg, MED of 0.5 mg/kg, and TI of 240 in humans (Fig. 5g and Table S13).

Discussion

In the present study, we used a TMDD modeling framework to predict the toxic doses, efficacious doses, and TI for anti-MerTK msAb and anti-MerTK/PD-L1 bsAbs using in vivo PK and in vitro RO data. Our model recapitulated key in vivo efficacy and toxicity observation. Specifically, the model revealed a low TI for anti-MerTK msAb in mice, which agreed with in vivo observations. The model also predicted that anti-MerTKAtten/PD-L1 bsAbs would be better tolerated than both anti-MerTKWT/PD-L1 bsAb and anti-MerTK msAb without any impact on efficacy, which matched with the experimental results. Finally, we used our modeling framework to explore a theoretical optimal molecular design for the anti-MerTK bsAb that could maximize TI in humans.

We observed that targeting MerTK with a msAb resulted in non-reversible retinal toxicity in mice, which is in alignment with previous reports showing that blocking MerTK led to pathophysiological changes in mouse retina (15, 16). It is previously assumed that since antibodies have a ~ 5–20-fold higher partitioning to tumor tissues compared to retinal tissues (38, 40), the large differences in tumor and retina concentrations could contribute to a high TI for msAb. However, our simulation shows that this disparity in drug partition to retinal and tumor tissues alone cannot qualify anti-MerTK msAb as a feasible drug candidate, as dosing levels that produce efficacious tumor concentrations result in retinal concentrations higher than the toxicity threshold. Simulations performed using the TMDD model predicted that there is no dose level which we can safely administer anti-MerTK msAb while maintaining efficacy. Even though tumor:serum partition as high as 15% (38) can boost tumor concentrations, our model predicted that a lower-bound MED was reached at 2.5 mg/kg, since a high tumor:serum partition cannot compensate for the rapid decline in serum concentration due to target-mediated clearance. Therefore, to widen the therapeutic window, we utilized anti-MerTK/PD-L1 bsAbs.

Our model prediction and experimental results revealed that anti-MerTKAtten/PD-L1 bsAb had a significantly improved retinal safety profile compared to anti-MerTKWT/PD-L1 bsAb, which in turn was better tolerated than anti-MerTK msAb, even in mice that are predisposed to retinal dystrophy (45). This improvement in safety was due to the reduction in bsAb binding to MerTK on RPEs. In vitro RO data revealed that anti-MerTKAtten/PD-L1 bsAb resulted in a 56-fold increase in ECRO20RPE compared to anti-MerTK msAb. This was to be expected as PD-L1 is minimally expressed on RPEs and single-arm MerTK binding was less potent than bivalent binding. In contrast, both bsAbs showed sixfold decreases in ECRO90MΦ compared to the anti-MerTK msAb, indicating that binding of bsAb to macrophages was influenced by avidity. Our bispecific model estimates a high avidity constant of ~ 100 based on in vitro RO data, similar to typical literature reported values for avidity constants (43).

Assumptions were made in this model to improve parameter identifiability. For instance, the internalization rates of drug-target complexes were assumed to be identical, since similar internalization rates of antibody-PD-L1 and antibody-MerTK complexes were identified from literature report and our experimental data (46, 47). We did not apply quasi-equilibrium assumption in the model structure since we were able to experimentally obtain kon and koff that adequately described the in vivo PK data, especially at early time-points. We further assumed that the avidity constant obtained from in vitro study can apply to in vivo settings. In addition, receptor occupancy threshold for efficacy and toxicity was obtained from in vitro study and assumed to be 90% and 20%, respectively, which could be different in vivo. We further assumed that the binding of bsAbs to PD-L1 on macrophages is more important than PD-L1 on tumor cells, since the expression of PD-L1 on macrophages is higher than that of tumor cells (48). Additionally, we noticed our model slightly over-predicted the TK of bsAbs at 2/9 time-point sampled, which could result in an inaccurate prediction of MTD. However, since model prediction adequately captured the majority of the TK data, the impact of this over-prediction on the conclusion should be minimal. Finally, tissue:serum partition method was used to estimate tumor and retina PK. Although this method does not include the prediction of accumulation phase in the tissue PK, the impact of this phase on model conclusion should be minimal, since the duration of this phase (~ 3–6 h) (49) is much shorter than the frequency of the treatment (every 3–4 days). Furthermore, this partition method could overpredict retina PK and contribute to inaccuracy in sensitivity analysis, as targets in the retina can deplete the already low drug concentration in this tissue. Yet, since the majority of MerTK and PD-L1 are expressed in leaky tissues and blood (7, 41, 42), the impacts of TMDD on drug concentration from these tissues are expected to out-weight the impact from the retina. To improve the prediction of tissue PK and the sensitivity analysis, separate tumor and retina compartments could be included in the future iteration of this model. However, experimentally measured tumor and retina PK are needed to fully identify all the parameters in such a complex model. Despite these assumptions, results from the in vivo efficacy and toxicity studies were consistent with our in silico MTD/MED/TI prediction using the simplified model structure.

Lastly, we used the bispecific TMDD model to suggest theoretical design parameters for a new anti-MerTK bsAbs that have 30-fold higher theoretical TI in humans compared to our current anti-MerTKAtten/PD-L1 bsAb. Our in silico sensitivity analysis revealed that CLns affects MTD more than MED, as CLns influences the bsAb PK more at a high-dose level when the targets are saturated. An optimal design space for CLns was identified to be 10–15 mL/day/kg. On the other hand, kdegB and RB had a large impact on MED, since both parameters control the PK of bsAbs at a low-dose level when the clearance of the drug is primarily due to TMDD. Parameter scan suggests that to increase the TI of the anti-MerTK bsAb, the targeting arm should be designed to bind to low abundance targets that are specifically expressed on macrophages (low RB), many of which have been described in literature (50,51,52). This design effectively decreases the theoretical MED of the drug, as it reduces the possibility of a large pool of targets that would serve as drug sinks. Finally, simulation demonstrates that a high affinity targeting arm can increase TI, as it allows the bsAb to tightly bind to macrophages for more effective MerTK inhibition. In contrast, an optimal KD for the therapeutic arm is predicted to be ~ 300 nM, and KD_MerTK at both extremes result in low TI. This prediction is reasonable since a low affinity therapeutic arm would have impaired binding to macrophage MerTK thus limiting efficacy, while a high affinity therapeutic arm would have enhanced binding to MerTK on RPEs thus increasing retinal toxicity. Although additional in vitro RO experiments with optimized bsAb are needed to validate these simulation results, our in silico analyses demonstrated how modeling approaches can be used to guide the rational design of bsAbs.

Conclusion

We developed a TMDD model framework to predict PK, efficacious doses, toxic doses, and TI of anti-MerTK msAb and bsAbs using in vitro RO and in vivo PK data. We leveraged the model to suggest improvements in drug design to further enhance the TI of the anti-MerTK bsAb. This modeling framework could aid the rational design of other bsAbs that aim to enhance on-tumor efficacy and reduce off-tumor toxicity.