Pharmacokinetics and Pharmacodynamics of TAK-164 Antibody Drug Conjugate Coadministered with Unconjugated Antibody

Abstract

The development of antibody-drug conjugates (ADCs) has led to the approval of 7 ADCs by the FDA in four years. Given the impact of intratumoral distribution on efficacy of these therapeutics, coadministration of unconjugated antibody with ADC has been shown to improve distribution and efficacy of several ADCs in high and moderately expressed tumor target systems by increasing tissue penetration. However, the benefit of coadministration in low expression systems is less clear. TAK-164, an ADC composed of an anti-GCC antibody (5F9) conjugated to a DGN-549 payload, has demonstrated heterogeneous distribution and bystander killing. Here, we evaluated the impact of 5F9 coadministration on distribution and efficacy of TAK-164 in a human tumor xenograft mouse model. Coadministration was found to improve the distribution of TAK-164 within the tumor but had no significant impact (increase or decrease) on efficacy. Experimental and computational evidence indicates that this was not a result of tumor saturation, increased binding to perivascular cells, or compensatory bystander effects. Rather, the cellular potency of DGN549 was matched with the single-cell uptake of TAK-164 making its IC₅₀ close to its equilibrium binding affinity (K_d), and as such, coadministration dilutes total DGN549 in cells below the maximum cytotoxic concentration, thereby offsetting an increased number of targeted cells with decreased ability to kill each cell. These results provide new insights on matching payload potency to ADC delivery to help identify when increasing tumor penetration is beneficial for improving ADC efficacy and demonstrate how mechanistic simulations can be leveraged to design clinically effective ADCs.

Introduction

Antibody-drug conjugates (ADC) have seen significant growth in the past few years, with a total of 11 FDA-approved ADCs and many more in the pipeline. Although ADCs have been explored in pursuit of the proverbial cancer ‘magic bullet’ for over 30 years, progress has been slow until recently, with 7 of the 11 ADCs approved just within the past four years.(1) However, the development of ADCs for solid tumors is still a challenge that depends on many physicochemical properties of its components and the complex tumor microenvironment. ADCs are composed of 1) an antibody that targets receptors overexpressed on cancer cells, 2) a small molecule payload that leads to cytotoxicity, and 3) a linker that connects the payload to the antibody(2-4). Optimizing each of these ADC components to function in concert for a given target is critical for the development of more efficacious ADCs.(5)

Many ADCs show heterogeneous distribution in solid tumors at clinical doses. Once administered intravenously, ADCs extravasate from blood vessels, diffuse into the tumor interstitium, bind to receptors on cancer cells, and undergo internalization. After the linker is cleaved in the cell lysosome, the payload is released to bind to the intracellular target and kill the cell. However, due to their large size, ADCs extravasate into the tumor slowly (the rate limiting step) which results in only a fraction of the injected dose accumulating in the tumor. In addition, ADC-receptor binding rates are much faster than ADC diffusion in the interstitium, resulting in a ‘binding site barrier’, where most of the ADC, even when administrated at clinical doses, is localized on tumor cells close to the blood vessels. Several strategies can be used to increase tissue penetration, including the use of smaller formats(6, 7), temporarily blocking the antibody binding site to slow the effective binding rate(8), reducing the affinity(9, 10), and administering higher antibody doses, each with strengths and limitations. While these approaches often deliver similar amounts of payload at their maximum tolerated dose, one notable difference is that the three former approaches result in targets being bound by ADC or remaining unoccupied, while the latter approach results in targets being bound by an ADC or an antibody. This antibody may provide additional efficacy in some systems. As reported previously, coadministration of ADC and its unconjugated antibody, even in systems that are insensitive to the antibody, has the potential to improve distribution and increase efficacy in many systems.(11-15) However, most of these targets have moderate to high expression levels on tumor cells (10⁵ – 10⁶ receptors per cell), and few studies focus on targets with less than 10⁵ receptors per cell, making the impact of coadministration in lower expression systems less clear.(16)

To study the impact of coadministration in low-moderate expression systems, we utilized TAK-164, an ADC that targets guanylyl-cyclase C (GCC) receptors in colorectal and gastrointestinal cancers.(17, 18) It consists of a 5F9 antibody backbone and a DNA-alkylating DGN-549 payload. DGN-549 can achieve bystander killing, where the payload can leave the cell targeted by the ADC and enter adjacent cells, and this effect has been a focus of next generation ADCs. The efficiency of bystander payload killing depends on its ability to transport across cell membranes (lipophilicity) and kill nearby cells (potency) prior to washing out of the tumor.(19-22) The highly lipophilic DGN-549 payload is predicted to possess optimal properties that balance its ability to diffuse through the tumor tissue while still accumulating inside cells.(23, 24) However, this class of payloads is often ultra-potent, resulting in a lower maximum tolerated dose (MTD) in the clinic, e.g. ~ 0.2mg/kg or lower for PBD-based ADCs(25) compared to 3.6mg/kg for Kadcyla^® (microtubule inhibitor payload(26)) or 10mg/kg for Trodelvy^® (topoisomerase inhibitor(27)).

ADC efficacy can also depend on the tumor microenvironment, which is composed of malignant cancer cells, immune cells, tumor vasculature, stromal cells (e.g. fibroblasts) and extracellular matrix proteins.(28) Primary human tumor xenograft (PHTX) models are useful for preclinical screening, given their greater tissue complexity compared to cell-derived tumor xenografts.(29) In this work, we utilized a PHTX model to incorporate these more complex tumor environments for greater clinical relevance. Previous evaluation of TAK-164 revealed its heterogeneous distribution in PHTX-11C tumors (low-moderate GCC expression(30)) at doses of 0.4mg/kg (similar to clinical tested doses up to 0.32mg/kg(31)). Additionally, TAK-164 has also been shown to exhibit strong bystander effects that could potentially compensate for this heterogeneous ADC distribution.(32) This combination provides the opportunity to study the impact of antibody coadministration on efficacy with an ADC exhibiting heterogeneous distribution and bystander killing in a low expression and clinically relevant PHTX mouse model.

In this study, we combined experimental and computational work using multichannel fluorescence imaging, mouse tumor growth studies, and an agent-based predictive tumor efficacy model to evaluate how antibody coadministration impacts the distribution of TAK-164, its released payload, and the corresponding efficacy in a PHTX model. We observed experimentally that coadministration of antibody with the ADC improved distribution but, unlike previous studies using DGN-549(16), had a negligible impact on efficacy. Computational simulations are valuable tools to predict optimal ADC properties, scale preclinical animal results to the clinic(23, 33-41), and provide complete control over individual variables, facilitating studies on individual mechanisms. We therefore utilized both molecular imaging and simulations to investigate the mechanism driving the observed outcome. Specifically, we examined whether the insensitivity to a carrier dose was due to i) tumor saturation that reduced total tumor ADC uptake, ii) increased ADC/antibody binding by perivascular cells that limited tissue penetration, iii) over-dilution of intracellular payload concentration in targeted cells reducing efficacy, or iv) compensation of heterogeneous ADC distribution by payload bystander effects.

Methods

Antibody Constructs

Production of the recombinant anti-GCC antibody 5F9(30) and its conjugation to sulfonated-DGN549 to generate TAK-164 has been described previously.(17) Both constructs were provided by Takeda Development Center Americas- Inc.. AlexaFluor 647 (AF647) conjugation of 5F9 and TAK-164 have been described previously.(32) Antibodies for immunohistochemistry were fluorescently labeled via Lysine-NHS chemistry.(42)

In vitro binding affinity and cell viability

Transfected HEK-293-GCC (generated at Millennium Pharmaceuticals Inc.) were cultured at 37°C and 5% CO₂ in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 10 μg/mL blasticidin (Invitrogen)(32). For binding affinity measurements, cells were incubated in triplicate with titrations of 5F9 conjugated to the fluorophore Alexa Fluor 647 (AF647-5F9) on ice for 3 hours (to achieve equilibrium) and analyzed via flow cytometry. For cell viability measurements, 5000 cells per well were plated in 96 well black-walled clear bottom plates and allowed to adhere for 24 hours. Titrations of TAK-164 were replaced daily for 6 days to avoid depletion at low concentrations, followed by incubation with 1:10 dilution of PrestoBlue in media for 25 minutes at 37°C. Fluorescence (560/590, Ex/Em) of was measured using a plate reader (Biotek Synergy) and viability was calculated by normalizing background subtracted signal of treated cells to untreated cells.

In vivo tumor xenograft studies

All animal studies were approved and performed in accordance with the Institutional Animal Care and Use Committee (IACUC) of the University of Michigan and Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC) guidelines. Primary human tumor xenografts (PHTX-11C) tumors were initiated in 6-7-week-old female CB17 SCID mice, and subsequently propagated in 8-week-old female Fox1^nu/nu nude mice under ABSL2 conditions, as described previously(32).

To study TAK-164 intratumoral distribution, mice bearing ~100-250mm³ tumors were injected intravenously with 0.4mg/kg or 1.5mg/kg of TAK-164 conjugated to the fluorophore Alexa Fluor 647 (AF647-TAK-164) and euthanized for tumor resection 72 hours post-injection. Ten minutes prior to euthanasia, the mice were administered 15mg/kg Hoechst-33342 intravenously to mark functional blood vessels. Resected tumors were flash frozen in OCT using isopentane chilled on dry ice. As detailed previously(32), slices of tumors were stained ex-vivo with AlexaFluor555-anti-mouse CD31 antibody (BioLegend, 102402), AlexaFluor488-anti-Mac3 antibody (BD Biosciences, 553322) and AlexaFluor750-5F9, and imaged using an upright Olympus FV1200 confocal microscope.

To measure cellular uptake, mice bearing ~250mm³ tumors were injected intravenously with 0.75mg/kg, 1.5mg/kg, or 3mg/kg of AF647-5F9 and euthanized for tumor resection 24 hours post-injection. Resected tumors were digested using a human tumor dissociation kit (Miltenyi Biotech), and the digests were filtered through a 40μm filter to remove tumor clumps and form a single-cell suspension. Cells were analyzed via flow cytometry (Attune NxT, ThermoFisher) and the fluorescent signal was compared to a calibration curve generated using Quantum Simply Cellular anti-human beads (Bangs Laboratories) labeled with AF647-5F9 to obtain an antibody molecules/cell measurement.

In vivo tumor growth curves

Tumor bearing Fox1^nu/nu nude mice were assigned into five treatment groups: PBS vehicle control (n = 9), 1.6mg/kg 5F9-only control (n = 9), 0.4mg/kg TAK-164 (“0:1” group; n = 10), 0.4mg/kg TAK-164 coadministered with 0.4mg/kg 5F9 (“1:1” group with a total antibody dose of 0.8mg/kg; n = 9) or 0.4mg/kg TAK-164 coadministered with 1.2mg/kg 5F9 (“3:1” group with a total antibody dose of 1.6mg/kg; n = 9). Treatments were administered as a single intravenous dose when tumors reached a volume of ~ 100 mm³. Tumor volumes and mice weights were monitored every other day until the tumor reached the study protocol maximum end point or until the end of the study. Kaplan–Meier survival curves were generated in GraphPad Prism.

Computational model

We extended our hybrid agent-based model (ABM)(13, 43) to incorporate additional physical and biological phenomena that occur for TAK-164, its payload DGN549, and the tumor environment.(13) Briefly, our model is comprised of cells and blood vessels that behave based on probabilistic rules and their microenvironment. The simulation environment is a 2D representation of a tumor section that contains cells and blood vessels overlayed with binding and diffusion of ADCs and payloads.(13) The cells and blood vessels behave as agents, occupy specific positions on the simulation grid, and have different states, i.e., alive or dead for cells and functional or non-functional for vessels. Each cell occupies a volume of 2 x 10⁻¹² L (12.6 μm side grid), and the initial tumor (~1,940 cells) is assumed to be representative of an initial tumor volume range of 75-150 mm³. ADCs enter the tumor through active blood vessels, the density of which changes based on the tumor size(43). Cells change their state from alive to dead based on the concentration of payload bound to DNA inside the cell.

The previous model included plasma dynamics (clearance), drug dynamics (ADC mechanism for bystander payloads), cell dynamics (e.g., cell division, death, expression level), and blood vessel dynamics that impact the tumor volume in our simulations. Here, we extended the model to include: 1) the non-linear internalization of the antibody 5F9 and TAK-164, 2) tumor uptake dynamics in vivo, and 3) the payload with bystander effects that binds to DNA irreversibly.

Modeling In Vitro Internalization of 5F9

Internalization of 5F9 produces a rapid downregulation followed by recovery of the GCC receptors.(32). The kinetics are best described with a model that includes two phases: an initial phase with internalization rate constants k_e and k_eBound for free and bound receptors, respectively, and a second phase with lowered values for these rate constants k_e2, k_eBound2) that occurs following significant binding (fb) for a period of time (t_change). These parameters were fit to the in vitro data prior to implementing in the ABM(32) (Figure S.2).

In Vivo Tumor Uptake

For comparison to simulations, average drug uptake (molecules per cell) is defined as:

$$\text{Uptake}=\frac{{\sum }_1^n({5F9}_{bound}+{5F9}_{int}+\text{dye}){.V}_{cell}.Av}{n}$$ (1)

${5\mathrm{F}9}_{bound}$, ${5\mathrm{F}9}_{int}$, and dye are the concentrations in nM of 5F9 bound on the surface of the cell, 5F9 internalized, and residual dye, respectively. V_cell is the volume of the cell (2x10⁻¹²L), Av is Avogadro’s number, and n is the number of cells.

Results

In Vitro and In Vivo Uptake, Distribution, and Efficacy of TAK-164

Mice bearing PHTX-11C tumors were intravenously administrated at 0.4 mg/kg or 1.5mg/kg of AF647-TAK-164. Stromal macrophages were stained ex vivo using AF488-anti-Mac-3 antibody. The distribution of TAK-164 in histology tumor slices demonstrated better penetration at the higher dose (1.5mg/kg compared to 0.4mg/kg). As seen in Figure 1.A, for the lower dose (0.4 mg/kg on the left panel), TAK-164 was distributed heterogeneously around blood vessels that are located within the stromal tissue. The higher dose (1.5 mg/kg on the right panel) had higher total uptake and penetrated deeper in the tumor tissue, reaching most cells as seen previously.(32)

Figure 1: Increased distribution of TAK-164 with increased dose.

(A) Increasing the ADC dose increased tissue penetration at sub-saturating doses. Blood vessels (not shown) are exclusively in regions of macrophages (Mac 3 in red). The TAK-164-Alexafluor 647 (green) distributes perivascularly, reaching most cells at a 1.5 mg/kg dose at 3 days. (B) 5F9 conjugated with Alexa Fluor 647 was administered via tail vein injection to the PHTX-11C xenograft model at 0.75, 1.5, and 3 mg/kg. Tumors were excised, digested, and run via quantitative flow cytometry to measure the average number of antibodies per cell. Doses close to 3 mg/kg approached saturation. (C) Viability and binding affinity of TAK-164 in HEK-293-GCC cells revealed a high potency versus affinity in HEK-293-GCC cells in vitro, while potency was expected to be lower in the lower expressing PHTX models. (D) Mice with PHTX-11C xenografts were treated with 0.4 mg/kg of TAK-164 via tail vein with or without 0.4 mg/kg or 1.2 mg/kg of 5F9 antibody. Controls included untreated mice or those treated with 1.6 mg/kg 5F9 antibody alone. (E) Survival curves of mice treated in (D). Coadministration had a negligible effect on efficacy. All panels represent mean±S.D.

Uptake of the 5F9 antibody in PHTX mouse tumors was quantified by flow cytometry. After AF647-5F9 was injected intravenously at 0.75, 1.5, and 3 mg/kg into PHTX-11C tumor bearing mice, the mice were euthanized 24 hours post injection, tumors were digested into a single cell suspension and analyzed by flow cytometer. The median number of antibodies per cell was estimated as shown in Figure 1.B. As the dose increased, the average uptake of antibodies per cell also increased. However, while doubling the dose from 0.75 to 1.5 mg/kg almost doubled the number of molecules per cell, a subsequent doubling of the dose from 1.5 mg/kg to 3mg/kg increased the number of molecules per cell only by ~25%. This data suggests that tumor saturation occurs when approaching doses of 3mg/kg. Notably, tumor cells stained ex vivo with fluorescent 5F9 following a single cell digest showed higher levels of receptor expression (20,000 to 30,000 per cell(32)) than antibody uptake in vivo (12,000 per cell). This combined with the persistence of acinar pockets with high densities of receptors(32) suggests that some fraction of receptors in cells in clinical tumors may have reduced accessibility to the in vivo administered dose, either due to physical barriers in the form of tight junctions or a dynamic binding site barrier. ((17), Figure S1)

Since the PHTX models cannot be propagated in vitro, a transfected HEK-293-GCC line with ~10⁵ GCC receptors per cell was used to test TAK-164 cytotoxicity and 5F9 binding affinity to the antigen (Figure 1.C). The IC₅₀ in these cells was found to be 4.6 +/− 1.3 pM while the binding affinity of unlabeled TAK-164 was 51 +/− 10 pM. (Note that the average measured K_D for AF680-TAK164, unlabeled TAK-164, and AF680-5F9 was ~ 60pM, and this averaged value was used for the simulations described later in the manuscript.) Since potency is typically proportional to payload delivery,(44, 45) a 10-fold lower expression would be expected to result in ~10-fold lower potency (~46 pM).

Coadministration of 5F9 and TAK-164 Has a Negligible Impact on Efficacy

Given the deeper tissue penetration at higher antibody doses, we evaluated the benefit of the coadministration of 5F9 antibody and TAK-164 in primary human tumor xenograft PHTX-11C model. When tumors reached ~ 100mm³, the mice were injected with either 1.5 mg/kg dose of 5F9 (4:0), 0.4 mg/kg of TAK-164 (0:1), coadministration of 1:1 and 3:1 of 5F9 with TAK-164, or saline-only for the control. As seen in Figure 1D, coadministration of 5F9 and TAK-164 did not increase or decrease the efficacy (no statistical significance). Consistent with this lack of impact, survival was also similar across all cohorts treated with TAK-164, but all significantly improved survival relative to untreated mice (Figure 1E). Given the multiple demonstrations of improved distribution increasing efficacy (either via antibody coadministration or ADC protein engineering),(8, 10-12, 16, 46) we sought to determine why the increased ADC penetration from antibody coadministration resulted in similar efficacy. Establishing a mechanistic understanding of the ADC response here can provide important insights for appropriate dosing of TAK-164 and other ADCs currently in development.

Four different mechanisms were hypothesized as to why a carrier dose impacted the tissue distribution but not efficacy (Figure 2). The first is due to tumor saturation. With a saturating dose, tissue penetration is no longer limiting efficacy, and results more closely mirror in vitro experiments where a carrier dose reduces efficacy(12). Doses near saturation could have competing effects resulting in no change in efficacy (Figure 2A). The second possibility is due to slower binding with low target expression. Antibodies typically bind much faster than they diffuse, resulting in a saturation front moving through the tissue.(47) For low expression targets, however, the ratio between binding to receptors (k_on[Ag]/ε) and diffusion, represented by the dimensionless Damkohler number (Da), is slower due to a lower effective antigen concentration, and antibody can diffuse deeper into the tissue without saturating the first cell layer.(48) Essentially, the first layer of cells is not saturated, so the addition of a carrier dose continues to bind within this region without increasing penetration (Figure 2B). This does not necessarily mean the antibody is distributed uniformly in the tumor, just that the binding front has a shallower gradient and does not fulfill the ‘shrinking core’ criterion.(47) A third possibility is that the potency of the payload could be matched to the cellular delivery such that any increase in tissue penetration of the ADC is offset by a reduction in payload concentration per cell, thus resulting in reduced killing of the targeted cells (Figure 2C). Finally, the bystander payload could diffuse deeper into the tissue, minimizing the impact of heterogeneous ADC distribution on efficacy (Figure 2D).

Figure 2: Diagram with Hypothesized Mechanisms of Carrier Dose Impact.

Four different mechanisms were tested that could reduce the impact of a carrier dose on efficacy. (A) Tumor saturation at the tested doses would prevent heterogeneous distribution, and a carrier dose might only compete with the ADC dose, potentially leading to a decrease in potency. (B) Slower binding in the tissue due to lower receptor expression could prevent saturation on perivascular cells such that a carrier does simply saturates these regions without increasing tissue penetration. (C) A reduction in payloads per cell could reduce the cell killing to offset the increase in targeting. (D) Bystander payloads can diffuse to adjacent cells, improving payload distribution independent of ADC heterogeneity.

It is difficult to test each hypothesis exclusively with animal experiments given the challenges with isolating the independent impact of each variable in vivo, limitations on time and resources, and ethical implications of excessive animal use. However, multiscale modeling can capture the effects across this wide range of spatial and temporal scales(49, 50). Therefore, we leveraged our previously validated (13, 43) computational platform (Figure 3) in conjunction with experimental data to determine the mechanism behind similar efficacy despite improved tissue penetration. This model captures systemic clearance, tissue penetration, cellular uptake and processing, bystander payload distribution, and efficacy at the mechanistic level. By comparing the simulation results with experimental data and manipulating individual parameters, each hypothesis was examined in detail.

Figure 3: Simulation Schematic and Distribution In Silico.

(A) Plasma Dynamics for TAK-164 and 5F9 determine tumor uptake. (B) Drug dynamics include irreversible reaction of the DGN-549 payload with DNA. (C) Cells and blood vessels act as agents to capture the dynamics and efficacy of TAK-164 in a PHTX-11C model. (D) The distribution of ADC in the tumor is compared with (E) the hybrid ABM that reproduces a similar level of heterogeneity at 0.4 mg/kg dose while showing higher and more uniform uptake at the 1.5 mg/kg dose. Images in D are from Figure 1 with modified color scheme and Gaussian blur to remove noise for comparison, and the simulations (E) show bound ADC at 72 hrs. Scale bar = 100 μm

We first updated our previously published computational model (13, 43) to include the parameters relevant for TAK-164 in the PHTX-11C model. All the parameters and rates constants for the model dynamics are shown in Table 1. Note, the calibration of TAK-164 non-linear internalization rate was performed to match the observed kinetics in receptor recovery in vitro and penetration depth in tumor spheroids from pulse-chase experiments(32). Sensitivity analysis of parameters beyond GCC receptor expression and payload potency was not performed since parameters were fit to data from individual experiments.

Table 1.

Model Parameters

Parameter	Value	Unit	Description	Reference
kTT	1.47 x 10−5	s−1	Total antibody clearance^	calibrated to measured plasma clearance data
k12	4.29 x 10−4	s−1	Antibody clearance from compartment 1 to 2^
k21	4.81 x 10−4	s−1	Antibody clearance from compartment 2 to 1^
V1	1.2 x 10−3	L	Volume central compartment	(13)
S/V	14	cm−1	Average blood vessel surface area per tumor volume	measured
kTT	9.67 x 10−6	s−1	Total antibody clearance*	calibrated to measured plasma clearance data
k12	4.33 x 10−4	s−1	Antibody clearance from compartment 1 to 2*
k21	4.79 x 10−4	s−1	Antibody clearance from compartment 2 to 1*
Kd	60	pM	TAK-164 dissociation constant	measured (32)
ke	3x10−4	s−1	5F9 Internalization rate constant free Rec (phase 1)	calibrated (32)
keBound	4.6x10−4	s−1	5F9 Internalization rate constant bound Ab (phase 1)
ke2	4.8x10−5	s−1	5F9 Internalization rate constant free Rec (phase 2)
keBound2	4.8x10−5	s−1	5F9 Internalization rate constant bound Ab (phase 2)
kdeg	8.0x10−6	s−1	TAK-164 lysosome degradation rate constant	measured (32)
kin	1.05x10−3	s−1	DGN549 Payload rate constant entering cell membrane	estimated (32)
kout	5.53x10−5	s−1	DGN Payload rate constant leaving cell membrane	estimated (32)
kon	3.2x10−11	nMs−1	DGN Payload binding rate to microtubules	calibrated (32)
koff	0	s−1	DGN Payload unbinding rate to microtubules	assumed from literature (32)
DAR	2.9	-	Drug to antibody ratio	(32)
Rs	ke*[Ag]0	M/s	Target synthesis	-
ε	0.24	-	Intracellular void fraction for ADC	estimated (32)
εp	0.44	-	Intracellular void fraction for payload
Dp	1.4 x 10−11	m2/s	Diffusivity for payload
D	1 x 10−11	m2/s	Diffusivity for ADC
P	3 x 10−9	m /s	Vascular permeability for ADC
Pp	1 x 10−6	m /s	Vascular permeability payload	estimated (51)
R	18	m /s	Payload partition coefficient	calibrated (23)
GCCo	3 x 103 - 3 x 105	rec/cell	Total number of targets per cell	measured/ calibrated
fb	0.4	-	Fraction of receptors bound	calibrated
tchange	6	hrs	Time for switch from phase 1 to phase 2	calibrated
Pmax	0.011	-	Maximum probability for cell killing	calibrated
Km	3.5	nM	Michaelis-Menten constant	calibrated
td (in vivo)	7-10	days	In vivo doubling time	calibrated
td (in vitro)	1-2.5	days	In vitro doubling time	calibrated

^{^}for dose 0.4 mg/kg

^*for dose 0.75 mg/kg or higher

Model Calibration to Experimental Data

Care was taken to fit as few parameters as possible and independently validate the values. When available, experimentally measured parameter values were used in the model; others were estimated based on observation (e.g. in vitro doubling time). Remaining values, including model PK and PD parameters, were calibrated to experimental data in vitro and in vivo using our published calibration protocol, CaliPro(52). In vitro measurements using HEK-293-GCC cells were used to determine the non-linear internalization rate parameters (t_change and fb, Figure S2), and in vivo uptake was used to determine the effective expression in PHTX-11C since in vitro measurements could not be collected with this model. The best fit values for target expression, time to change internalization phases (t_change), and fraction of receptors bound (fb) were 3,000 receptors per cell, 6 hours, and 0.4 respectively. The heterogeneous distribution seen in vivo (Figure 3D) was captured by the model (Figure 3E), indicating a good match with the tissue-level pharmacokinetics.

For capturing the pharmacodynamics, the tumor growth rate was fit to untreated tumors and the cell killing rate was fit to xenografts treated with TAK-164 alone. The tumor growth corresponded to a cell proliferation rate doubling time range of 7 to 10 days (Figure 4.A). P_max and K_M, which were fit to the TAK-164 monotherapy, resulted in a P_max and K_m of 0.011 and 3.5nM respectively as shown in Figure 4.B. The model was next validated by comparing the predicted efficacy using these fixed parameters to the efficacy observed with coadministration of 1:1 and 3:1 antibody to TAK-164 as shown in Figure 4.C-E. The same probability of cell killing was used in vitro and compared to the experimental data for further validation (Figure 4.F). The close agreement between the treated tumor regimens and in vitro results demonstrated that the model was appropriately calibrated.

Figure 4. Calibration of in vitro and in vivo pharmacodynamics and validation.

(A) In vivo proliferation and (B) In vivo cell killing rates were fit to experimental data for calibration after a single administration of 0.4mg/kg of TAK-164. (C -D) The fitted parameters were then used to predict efficacy and compared to experiments for in vivo validation following coadministration 1:1 and 3:1 with 5F9 to TAK-164. (E) Model predictions of the impact on efficacy with 0:1, 1:1, and 3:1 ratios of 5F9 antibody with 0.4 mg/kg TAK-164. (F) In vitro toxicity curves were compared to the model predictions using the same fitted parameters as in vivo. All panels represent mean±S.D

Testing Hypothesis 1 and 2: Low dose TAK-164 is Non-Saturating and Exhibits a Perivascular Binding Front in PHTX-11C

The first hypothesis and most intuitive reason for no improvement in efficacy from antibody coadministration in low receptor expression systems is tumor saturation (Figure 2A). This occurs when the initial dose of ADC is already sufficient to saturate the tumor, and subsequent addition of antibody results in competition with the ADC for receptors. Consistent with Khera et al.(32) Figure 1A and 1B show that 0.4mg/kg TAK-164 does not saturate the PHTX-11C tumors.

The second hypothesis suggests that the binding site barrier effect is reduced with TAK-164 in PHTX-11C tumors. For low receptor expression systems, the Damkohler number, which is a function of the target concentration, could become slow enough to enable antibodies/ADCs to diffuse deeper into the tumor without saturating the receptors on the initial cell layers (Figure 2B). In this scenario, addition of antibody coadministration may not affect the overall distribution of the ADC, thereby resulting in no change in penetration, and therefore no change in efficacy until perivascular cell saturation is achieved. To test if this effect can explain the behavior of TAK-164 in PHTX-11C tumors, we compared immunohistology images of PHTX-11C tumors treated with 0.4mg/kg and 1.5mg/kg 647-TAK-164 or 647-5F9 (Figure 1A) and found that increasing the dose in the sub-saturating range increased tumor penetration. This was confirmed by comparing to simulated ADC distribution at the same doses, which showed qualitatively similar behavior as the experimental data.

Testing Hypothesis 3: Expression Level Determines Impact of Antibody Coadministration

To determine if the insensitivity of efficacy to coadministration was related to the cellular delivery versus potency (hypothesis 3), we simulated tumors with different numbers of GCC per cell. Increased expression would deliver more payload per cell than required for cell death, reducing penetration and reaching fewer cells. A carrier dose would then benefit these tumors with higher expression by increasing tumor penetration while still maintaining a toxic dose. In Figure 5A, we show the distribution of the payload DGN549 and efficacy of the TAK-164 with different coadministration regimens of 0:1, 1:1, 3:1, and 8:1. Although the ADC distributes heterogeneously at an effective receptor expression of 3,000 (3K) GCC per cell (Figure S.3), the bystander ability of DGN549 helps the payload to reach most cells in the tumor by day 7. When the receptor expression increases to 30,000 (30K) GCC per cell, we see an increased benefit from coadministration leading to the best efficacy at the 8:1 regimen (Figure 5B). This is consistent with hypothesis 3 – at 30,000 GCC per cell, more payload is delivered to perivascular cells than needed (black color), so a carrier dose spreads out the payload for improved efficacy. At 3,000 GCC per cell, the payload per cell is already low at 0:1 (blue color), so an increase in ADC tissue penetration results in even lower payload delivery and efficacy per cell. When the average receptor expression is increased to 300,000 (300K) GCC per cell, coadministration shows no significant changes in distribution or efficacy. At this expression level and ADC dose (0.4 mg/kg), the carrier dose is not sufficient to improve tissue penetration; the initial dose does not saturate the perivascular cells, so the increase in antibody dose simply increases antibody cell binding within the saturation front on these cells (a theoretical example of hypothesis 2).

Figure 5. Prediction of distribution and efficacy in tumors varying receptor expression:

Distribution of the payload DGN549 at day 7 (A) and Efficacy at day 50 (B) of the coadministrations of 5F9 and TAK-164 (0:1, 1:1, 3:1, 8:1) in tumors with an average of 3K, 30K, and 300K receptors/cell. All figures represent mean±S.D

Testing Hypothesis 4: Bystander Effects Do Not Change the Impact of a Carrier Dose

To isolate the impact of bystander killing during coadministration with TAK-164 and determine if this was the reason for a lack of effect (hypothesis 4), we compared simulations with bystander effects to those that computationally eliminate the ability of the payload to enter adjacent cells (by setting the payload uptake rate constant to zero (k_inp = 0)). Simulations were run with 0:1, 1:1, 3:1, and 8:1 coadministered doses at 3,000 and 30,000 receptors/cell. Note that the ability of the payload to diffuse out of targeted cells was maintained (unlike a ‘traditional’ non-bystander payload that typically cannot cross membranes to escape), but free payload uptake was artificially set to zero to eliminate payload tissue penetration effects. If bystander effects eliminated the benefit of a carrier dose, then removing the bystander effect should restore a benefit of a carrier dose. As seen on Figure 6A, a carrier dose does not increase efficacy at 3K receptors, even where bystander effects are computationally eliminated (inconsistent with hypothesis 4). This demonstrates that the concentration of DGN549 retained in the directly targeted cells is just enough to mediate efficacy in PHTX-11C tumors (represented by the 3K receptor group), so further ‘dilution’ results in a substantial decrease in efficacy (hypothesis 3). Computationally, all dosing regimens are sub-saturating in the tumor (similar %ID/g) except for the 8:1 ratio with 3K GCC/cell (Figure 6B). The dose of the ADC alone exhibited some target-mediated drug disposition (TMDD), lowering the %ID/g(17, 53), but this had a negligible impact on the simulated efficacy.

Figure 6: Predictive simulations of regimens with and without bystander effects for tumors with different receptor expression.

(A) Average tumor volume at 40 days for tumors with 3K and 30K receptors per cell with and without bystander effects. Payloads had the bystander effect computationally removed by setting the payload internalization rate to adjacent cells to zero (k_inp=0). (B) %ID/g of TAK-164 (with bystander effects) for coadministration regimens in tumors with 3K and 30K receptors per cell. All panels represent mean±S.D

When comparing tumors with moderate receptor expression (30K receptors/cell), the k_inp = 0 group shows the anticipated effect of coadministration, where increasing antibody dose increases the efficacy. Furthermore, a comparison between dosing regimens where bystander killing is retained shows a similar trend in increasing efficacy with a higher antibody coadministered dose. In these tumors with more heterogeneous TAK-164 distribution, the bystander payloads can diffuse deeper in the tissue to reach some untargeted cells, resulting in a more considerable improvement in efficacy from the bystander effect. However, bystander effects tend to be less efficient than direct cell targeting(23, 54), so coadministration is able to further improve DGN549 access to distant untargeted tumor cells even when bystander effects are retained (Figure 5A).

Together, the results indicate that a carrier dose of 5F9 has little impact on efficacy in a low expression model like PHTX-11C due to the matching of TAK-164 potency (payload and DAR) with cellular delivery (target expression and internalization rate). Any increase in the antibody dose that increases tissue penetration is offset by a reduction in cell killing on targeted cells. A higher expression level (30K receptors/cell) brings about an imbalance by delivering more payload than needed for cell death to perivascular cells, and a carrier dose then helps improve efficacy. Under this hypothesis, the corollary should also be true. An increase in the payload potency brings about an imbalance, again delivering more payload than needed to perivascular cells with the increased potency. Such a scenario was simulated (Figure S4), and indeed a carrier dose then improves efficacy. Increased potency results in ‘overkill’ of the targeted cells due to a mismatch between cellular delivery and cellular potency, and a carrier dose then reduces uptake to restore the balance and improve efficacy.

Discussion

The development of new ADCs involves customizing and coordinating the different components of the ADC and ensuring they work in concert to maximize efficacy for a particular target at clinically tolerated doses. Recently, the negative impact of heterogeneous delivery of ADCs at clinical doses has resulted in strategies to improve tissue penetration, including the use of bystander payloads, lower potency payloads or DAR to enable larger dosing, and/or antibody coadministration. For antibody coadministration, the ADC dose, target expression, and receptor internalization rate all affect the potential benefit of a carrier dose, and the impact in systems with low expression is incompletely understood. Previous work has shown that coadministration of the ADC T-DM1 with its unconjugated antibody, trastuzumab, in a model that is insensitive to trastuzumab alone, increases efficacy by improving distribution.(55) Here, we tested antibody coadministration for the pair 5F9 and TAK-164 in a low expression (10K-20K receptors/cell) system to measure the impact on tissue distribution, cellular uptake, and efficacy to better understand which conditions benefit from a carrier dose.

TAK-164 is an anti-GCC targeting ADC designed to treat patients with gastrointestinal cancers. GCC is considered an excellent target due to its specificity for GI cancers.(30) However, GCC receptors can exhibit inter-patient expression heterogeneity that can vary up to 100-fold in some cell lines.(56) This variability can strongly influence the pharmacokinetics and pharmacodynamics of the ADC making it unclear if a carrier dose would increase or decrease efficacy. Antibody coadministration is known to improve efficacy with high receptor expression tumors (e.g. HER2 and others(11, 12, 15)), and recent work by Ponte et al. shows that antibody coadministration can also improve efficacy in moderate expressing systems, such as Mirvetuximab-DGN549 ADC, even with expression levels as low as 40,000 receptors per cell.(16) However, additional work with IMGN-853 (Mirvetuximab soravtansine) shows that with low receptor expression tumors, antibody coadministration can be detrimental to efficacy.(16) Therefore, we evaluated whether a carrier antibody dose could also improve the efficacy of TAK-164 in patient derived xenograft tumors.

Despite the improvement of tissue penetration with the addition of 5F9, which does not have efficacy alone, to the TAK-164 treatment, a carrier dose did not change the efficacy in the PHTX 11C model (Figure 1).(16) Several possible mechanisms could contribute to this unexpected disconnect between distribution and efficacy (Figure 2), including 1) saturation of the tumor, 2) additional binding on perivascular cells, 3) matched payload potency with payload delivery, or 4) bystander effects. We utilized computational simulations (Figure 3) in conjunction with experimental data to determine the tissue and cellular pharmacokinetics and pharmacodynamics of the system and systematically evaluate each hypothesis.

The first hypothesis focuses on tumor saturation, where super-saturating doses of a carrier dose block enough GCC receptors to reduce total ADC uptake and therefore payload delivery. However, both experimental and computational evidence is not consistent with this result. Regions of the tumor remained untargeted at 0.4 mg/kg (Figure 1A), the average number of antibodies per cell increased with increasing doses from 0.75 to 3 mg/kg (Figure 1B), and the %ID/g in the tumor was roughly constant for 3K receptors/cell in Figure 6B. This ruled out the hypothesis that super-saturation of tumor receptors prevented an increase in efficacy.

The second hypothesis involves antibody doses well below saturation, where even the perivascular cells are not saturated. With low expression levels, the effective binding rate is slower, so antibodies can diffuse deeper into the tissue without saturating these initial cell layers. An increase in dose therefore does not penetrate deeper but rather continues to bind to the perivascular cells. However, the imaging data shows increased tissue penetration with increasing doses (Figure 1,3), ruling out the 2^nd hypothesis. In contrast, simulations with 300K receptors per cell showed this unsaturated binding front, resulting in the 5F9 carrier dose binding GCC receptors on the perivascular cell layers without competing with TAK-164. Even a 3.6 mg/kg total dose (8:1) does not saturate beyond the first few cell layers close to the blood vessel (Figure S3), leading to no change in efficacy across the different dosing regimens (Figure 5).

A third possible explanation for the lack of impact from a carrier dose is efficient bystander payload penetration compensating for the heterogeneous ADC delivery. This is difficult to directly test experimentally since changes in payload properties impact multiple aspects of efficacy. However, it can readily be tested computationally using validated models(54). We prevented escaped payload from entering adjacent cells in simulations, and compared it to control. In Figure 6, increased efficacy from bystander effects is more evident for tumors with moderate receptor expression than for tumors with lower receptor expression. These data indicate that in tumors with moderate receptor expression, cells that are not directly targeted by the ADC can benefit from optimized bystander payloads.(23) However, in tumors with lower expression, the impact of the bystander killing is less important for tissue penetration. Notably, bystander effects can still be important for killing antigen negative cells (cellular heterogeneity), but they are less important for tissue penetration (tissue heterogeneity) under these conditions.(38)

Overall, the experimental and computational evidence shows that the cellular potency of the ADC is matched with the cellular delivery; there is no ‘overkill’ with the ADC where more payloads are being delivered than needed for cell death. When the potency is matched, the IC₅₀ is close to the equilibrium binding affinity (K_D) of the ADC such that cell killing is proportional to surface receptor bound ADC. Therefore, an increase in antibody dose reaching more cells is countered by a decrease in killing of the targeted cells. The in vitro data, adjusted for the expression level of the patient-derived xenograft versus the cell line, indicated that the IC₅₀ and K_d of TAK-164 are well-matched, supporting this hypothesis.

To further corroborate this hypothesis, simulations were used to look at alternative scenarios with an imbalance between potency and delivery. With 30K receptors/cell, cell death occurs without saturating the receptors on the cell surface (Figure 1B), so a carrier dose can distribute the ADC to reach more cells without an equivalent drop in cell killing. This can be seen with a larger fraction of cells receiving payload concentrations in the therapeutic range (> 2.5 nM) for 30K versus a larger fraction of cells receiving a sub-therapeutic dose (< 2.5 nM) for cells with 3K receptors (Figure 5). Likewise, with the same expression/cell delivery and an increase in the payload potency of the ADC, improved efficacy was seen when adding a carrier dose (Figure S4). In other words, maximum efficacy (for a given ADC dose) is achieved when the cellular ADC delivery and payload potency are matched.

Conclusion

Understanding ADC and payload tissue penetration can help develop better mechanistic approaches to increase the therapeutic window of ADCs. Here, we explored coadministration of TAK-164 with its antibody 5F9 and tested the impact on tumor distribution and efficacy using a combination of experimental measurements and computational modeling with a specific focus on the role of receptor expression, bystander efficiency, distribution, and DGN549 potency in non-clinical models. The results showed that TAK-164 coadministration would only improve efficacy in tumors with expression levels 10-fold higher than found in patient-derived xenografts (e.g. 30,000 receptors/cell). More generally, coadministration of ADC with a carrier antibody is most effective when it can match the cellular delivery (expression/internalization) with the ADC potency (payload potency and DAR). In practice, this occurs more frequently in moderate to high receptor expression systems. At the expression levels found in the PHTX-11C tumor model, the cellular delivery of TAK-164 was ideally matched with the potency of DGN549 without the coadministration of additional 5F9. The mechanisms examined and observations in this study highlight the potential utility of integrating computational modeling and characterizing therapeutic attributes in a variety of non-clinical models to help clarify when a carrier dose can improve the therapeutic window of ADCs.