Cellular-Resolution Imaging of Bystander Payload Tissue Penetration from Antibody-Drug Conjugates

Abstract

After several notable clinical failures in early generations, antibody-drug conjugates (ADCs) have made significant gains with seven new FDA-approvals within the last 3 years. These successes have been driven by a shift towards mechanistically informed ADC design, where the payload, linker, drug-to-antibody ratio, and conjugation are increasingly tailored to a specific target and clinical indication. However, fundamental aspects needed for design, such as payload distribution, remain incompletely understood. Payloads are often classified as ‘bystander’ or ‘non-bystander’ depending on their ability to diffuse out of targeted cells into adjacent cells that may be antigen negative or more distant from tumor vessels, helping to overcome heterogeneous distribution. Seven of the eleven FDA-approved ADCs employ these bystander payloads, but the depth of penetration and cytotoxic effects as a function of physicochemical properties and mechanism of action have not been fully characterized. Here, we utilized tumor spheroids and pharmacodynamic marker staining to quantify tissue penetration of the three major classes of agents: microtubule inhibitors, DNA-damaging agents, and topoisomerase inhibitors. PAMPA data and co-culture assays were performed to compare to the 3D tissue culture data. The results demonstrate a spectrum in bystander potential and tissue penetration depending on the physicochemical properties and potency of the payload. Generally, directly targeted cells show a greater response even with bystander payloads, consistent with the benefit of deeper ADC penetration. These results are compared to computational simulations to help scale the data from in vitro and preclinical animal models to the clinic.

Introduction

Antibody-drug conjugates (ADCs) are a sophisticated class of cancer therapeutics that have gained significant attention for their ability to specifically target tumor cells(1). After decades of investigation but lagging clinical success, the field is gaining traction again, with the approval of seven new ADCs within the last 3 years, bringing the total to eleven FDA-approved ADCs. Antibody-drug conjugates are comprised of three distinct components – (1) an antibody backbone, which binds specific tumor antigens for targeted accumulation of the ADC in tumors, (2) a small molecule payload (typically a cytotoxin), which mediates efficient cell death, and (3) a chemical linker, which links the payload to the antibody backbone. The design of these agents can be tailored for a specific target and expression level, choice of backbone (e.g. antibody fragments(2) and small molecules(3)), optimized drug-to-antibody ratio (DAR), payload class and potency, and linker release mechanism and kinetics to maximize the therapeutic window. Despite these optimizations, ADCs designed for solid tumors can exhibit heterogeneous targeting, usually in the form of heterogeneous antibody distribution (i.e., binding site barrier effect) and/or antigen expression heterogeneity (i.e., antigen negative tumor cells). This can often leave a substantial fraction of the tumor untargeted by the ADC, impacting efficacy. However, in addition to ADC-directed delivery of the cytotoxic payload, some payloads are also known to exhibit a ‘bystander effect’ where the free payload released intracellularly can escape the ADC-targeted tumor cell and re-enter neighboring untargeted cells and mediate cell killing(4). Of the seven ADCs approved in the past three years, four are designed to target solid tumors and use payloads capable of bystander effects – Enhertu (HER2/DXd), Trodelvy (TROP-2/SN-38), Padcev (Nectin-4/MMAE), and Tivdak (Tissue factor/MMAE). Despite the increasing use of bystander payloads, little is known about the tissue penetration distance of these payloads in the tumor microenvironment, and it remains unclear how efficiently bystander penetration of released payloads can compensate for heterogeneous antibody distribution.

In general, bystander effects are presumed to improve ADC efficacy, but current methods of evaluating the cell killing efficiency of bystander effects provide insufficient or incomplete resolution for clinically translatable insights. Furthermore, these methods are more tailored towards heterogeneous (expression) bystander effects (HBE)(5) i.e., using relatively well-interspersed Ag+/Ag− co-culture systems. However, quantifying spatial bystander effects (SBE)(5) also has major clinical implications, not only in cases of limited ADC tissue penetration, but for tumors with heterogeneous antigen expression, since Ag+ and Ag− tumor cells are not always well-interspersed in a tumor(6,7). In such scenarios, efficient spatial penetration of the bystander payload becomes critical for targeting compartmentalized Ag− tumor cell populations. Previous modeling work shows that different bystander payloads exhibit different tumor penetration efficiencies based on subtle differences in their physicochemical properties(8), i.e. different payloads have a different quantitative “bystander potential” that influences their ability to mediate cell death. Although experimental tracking of payloads in tumors is challenging, we recently described a sensitive experimental platform using tumor spheroids and an established pharmacodynamic marker to quantitatively map the penetration of cytotoxic concentration of bystander payloads(9). Here, we combine this experimental platform, mechanistic simulations, and complementary assays for measurement of bystander effects to characterize the bystander killing efficiency of a panel of payloads, many of which are employed by clinical ADCs, with the aim of mechanistically understanding the driving factors that influence bystander tissue penetration and killing efficiency.

Materials and Methods

Trastuzumab-Payload Constructs and Fluorescent Labeling

All payloads (Supplementary Figure 1) and ADC constructs (Supplementary Figure 2-3), except T-DM1/Kadcyla (obtained through the University of Michigan Pharmacy), were provided by Synaffix and prepared according to previously reported protocols(10,11). The Supporting Information contains the analytical results for the prepared ADCs (Supplementary Figures 4-5, Supplementary Tables 1-2). The linkers used in all ADCs consisted of a protease cleavage and self-immolative moiety. The potency of all ADCs was similar to, or greater than, the free payloads (except for SN-38), consistent with efficient payload release. Lysine residues on trastuzumab/ADCs were conjugated to AlexaFluor680 (AF680) via NHS-amine chemistry as described previously(12). The final concentration and degree-of-labeling (DoL; number of dyes per antibody) for the fluorescent ADC stock typically ranged from 10–20 µM and ~0.4 DoL.

PAMPA Permeability Assay

Permeability measurements were performed using the Gentest pre-coated PAMPA plates (Corning) following the manufacturer’s protocol. Payloads were provided by Synaffix, except SG3199-PBD (Levena Biopharma, CA). Briefly, all payloads were solubilized in DMSO to a concentration of 10 mM. A working solution (200 μM) in 20% methanol in PBS was prepared immediately prior to performing the assay. Payload solutions were added to the donor wells (300 μL/well) and 20% methanol in PBS was added to the corresponding pre-coated filter (acceptor) well (200 μL/well). The filter plate was incubated undisturbed at room temperature for 2.5 hours or 5 hours, after which 100 μL solution from each well (donor and acceptor) was injected onto a C-18 column and analyzed via HPLC. The peak area was compared to a standard concentration curve. Mass balance completion was performed using the acceptor and donor well concentration, and permeability of each payload was calculated as listed in the PAMPA plate manual.

Cell Culture

HCC1954 (Ag+) and MDA-MB-468 (Ag-) cells were procured from ATCC (Manassas, VA). No further characterization was performed, and experiments were conducted within 3–4 months of thawing cells. Mycoplasma testing was performed using MycoAlert (Lonza) every 6 to 12 months and all tested negative. HCC1954 and MDA-MB-468 cells were cultured at 37°C and 5% CO₂ in RPMI-1640 and DMEM, respectively, supplemented with 10% fetal bovine serum (FBS) and 50 U/mL penicillin and 50 µg/mL streptomycin. Cells were passaged 2 to 3 times per week (~80–90% confluency).

Cytotoxicity Assay

The in vitro viability assay was performed according to a previously published protocol(12,13). Briefly, cells were seeded at 5000 cells/mL (in triplicate for each treatment condition) in 96 well black-walled, clear bottom plates (Corning) and allowed to adhere for 24 hours. Titrations of free payload or ADC were replaced daily for 6 days, and viability was measured by incubating wells with a 1:10 dilution of PrestoBlue Cell Viability Reagent (Thermo Fisher Scientific) in media for 45 minutes at 37°C. The fluorescence (560/590, Ex/Em) of each well was measured relative to PrestoBlue-only using a Biotek Synergy plate reader. The signal was then normalized to untreated cells.

Bystander Co-Culture Assay

Wu et al. (13) have described a comprehensive co-culture protocol that provides highly quantitative information about bystander cell death. We adopted this assay, modified to fit a flow cytometry protocol reported by Ogitani et al.(14), to gather bystander information on the payload panel. Briefly, cells were seeded in black walled, clear bottom 96-well plates (Corning) at (1) varying % of Ag+ (HCC9154) and Ag− (MDA-MB-468), ranging from 0% to 100%, with the total number of cells being fixed at 10,000 cells per well, or (2) varying number of Ag+ (HCC1954) cells for a fixed number (5000 cells/well) of Ag- (MDA-MB-468) cells. Adhered cells were incubated for 6 days with no media changes with media-only (untreated control) or with a fixed concentration of ADC selected to be above the IC₉₀ for Ag+ cells and below the IC₅₀ for Ag- cells where possible (Supplementary Figure 6). Total viability was measured using PrestoBlue as described above. Viability of the treated group was normalized to the corresponding Ag+/ Ag− untreated groups. After measuring total viability, Presto Blue was removed, wells were washed with PBS, and then incubated with 40 µL 0.05% Trypsin-EDTA for 5 minutes at room temperature to lift cells from the plate. Trypsin was neutralized by adding 100 µL of FBS containing media. Cells from each well were transferred to the corresponding wells in a V-bottom 96 well plate (Corning), incubated with 100 nM unlabeled trastuzumab on ice for 30 minutes to saturate any free HER2 receptors, washed, and incubated with a secondary anti-human Fc FITC antibody (BioLegend, 1:200 dilution) for 30 minutes on ice. Cells were washed in 1X PBS and analyzed via flow cytometry (Attune Nxt) to quantify number of Ag+ and Ag− cells remaining. Ag− cell viability was quantified by normalizing the number of remaining Ag− cells in treated groups to the number of Ag− cells in the corresponding untreated group.

Pharmacodynamic Immunofluorescence Staining

HCC1954 cells were plated overnight in chamber slides and incubated with free payload (50 nM, 16 hours) or fluorescent ADC (100 nM for microtubule inhibitors and topoisomerase inhibitors, 5 nM for T-CaliG and T-PBD, 50 pM for T-CaliD, 72 hours). Post-incubation, cells were washed in PBS, fixed with 4% formaldehyde (BD Cytofix), and permeabilized/blocked in 0.5% bovine serum albumin (BSA) in 1X perm buffer (BD Cytoperm) in PBS (0.5% BSA/perm) for 15 minutes each at room temperature. Phospho-histone H2A.X (Ser139) (20E3) rabbit primary antibody (Cell Signaling Technologies) or Phospho-histone H3 (pSer10) rabbit primary antibody (Sigma-Aldrich) was diluted to 0.15 µg/mL in 0.5% BSA/perm and incubated with permeabilized cells for 30 minutes in the dark at room temperature. Cells were washed twice for 5 min each with 0.5% BSA/perm. Secondary antibody incubation was performed with AF555-labeled goat anti-rabbit (Fab)₂ antibody (Cell Signaling Technologies) diluted to 2.5 µg/mL in 0.5% BSA/perm for 30 minutes in the dark at room temperature. Cells were washed twice for 5 min each with 0.5% BSA/perm, followed by 5-minute incubation with 0.5 mg/mL Hoechst 33342 in PBS. Microscopy was performed using an upright Olympus FV1200 confocal microscope using a 20x objective and 405 nm (Hoechst 33342; nuclei), 543 nm (AF555; PD signal), and 635 nm (AF680, fluorescent ADC) lasers. Image analysis was performed using ImageJ.

Tumor Spheroid Experiments

Tumor spheroids were cultured using custom 384-well plates described previously(9). Briefly, 3000 cells suspended in 25 μL of seeding media comprising complete RPMI culture media, 1.2% (w/v) methylcellulose (Dow-Corning), and Matrigel were added to a 384-well hanging drop plate. Spheroids were cultured for 7 days until they attained a diameter of 400–500 μm. To study distribution, the hanging drops were incubated with AF680-ADC or AF680-trastuzumab, at a final concentration of 25 nM for 24 hours, 48 hours, and 72 hours. The drug concentration in the media was assumed to be constant over the course of the incubation (i.e., no depletion effects). After incubation, the spheroids were extracted from the wells, washed 2X with PBS, fixed with 4% formaldehyde, frozen in OCT, and stored at −80°C until further processing. Frozen blocks were sectioned for histology (16-μm slices), and sections were processed for pharmacodynamic staining and imaging as described above. Image analysis was performed in MATLAB and ImageJ to generate Euclidean distance maps.

Computational Model and Damköhler Number Calculation

Computational simulations of ADC and payload distribution were performed using a previously published Krogh cylinder model but adapted to spherical geometry for tumor spheroids(8,9). Estimation of trastuzumab kinetic parameters and payload parameters have been described in detail previously(15) and in Supplementary Materials (Supplementary Table 3-4). Payload molecular weight and cLogP were calculated using MarvinSketch (ChemAxon). Details of the Damköhler number analysis has been reported previously in (8).

The data generated in this study are available upon request from the corresponding author.

Results

High resolution visualization of payload penetration

To obtain cellular resolution of payload distribution, we used a previously developed protocol for indirectly imaging payload distribution via pharmacodynamic response markers(9). Although payloads exert their cytotoxicity via different mechanisms of action, they often converge to one or more common downstream signals that can be used as quantification biomarkers. Phosphorylation of Ser 139 on the H2A.X histone protein (γH2A.X) is a well-established marker for double-stranded DNA breaks (DSBs) used to track the DNA-damaging payloads and topoisomerase inhibitors. Phosphorylation of the Ser 10 residue on the H3 histone protein (pH3), an established marker of mitotic chromosomal condensation(16) and apoptosis(17), was a proxy for microtubule inhibitor activity (in addition to some isolated healthy cells undergoing mitosis). ADCs and free payloads both resulted in detectable PD signal in monolayer HCC1954 cells (Supplementary Figure 7).

Since trastuzumab, the antibody backbone used as the model system for all ADCs in this study, is known to exhibit perivascular distribution in high HER2+ systems(12,15,18), the spheroids enabled physical separation between ADC-targeted and untargeted cells (referred to as “bystander cells” henceforth). This distinguished PD signal from direct payload delivery at the spheroid edge versus bystander payload accumulation in the spheroid center (Figure 1A). A time-course of treated spheroids was included to better capture temporal dynamics. Antibody and ADC penetration is highly heterogeneous even after 72 hours of continuous exposure due to constant internalization and degradation of the antibody(19), providing sufficient tissue depth to capture differences in payload penetration as it manifests as PD signal. The directly-targeted cells also provide a convenient ‘internal control’ to normalize the results relative to the DAR, dose, release efficiency, payload potency, and cellular response, making the method for quantifying the efficiency of bystander effects less dependent on these variables.

Figure 1. High resolution imaging of payload penetration in spheroids.

(A) Schematic of pharmacodynamic marker (PD) manifestation for bystander and non-bystander ADC payloads. (B) Immunofluorescence microscopy of spheroids treated with 25 nM AF680-trastuzumab (green) conjugated to three different microtubule inhibitors. The PD signal (red) shows different extents of penetration into the spheroids. Nuclei are labeled with Hoechst (blue).

Spheroids treated with microtubule inhibitor ADCs revealed a dynamic transition in the manifestation of pH3 PD signal from early to later time points. (Figure 1B). The overall appearance of pH3 signal occurs in the directly targeted cells by 24 hours. T-DM1, which releases the non-bystander payload Lys-SMCC-DM1, did not exhibit bystander killing even at 48 hours, while MMAE exhibited pH3 signal throughout the spheroid at 48 hours, consistent with its predicted rapid diffusion(20). T-Ahx-maytansine (T-May) resulted in bystander penetration in cells immediately adjacent to directly targeted cells, while spheroids treated with trastuzumab alone showed minimal spontaneous mitosis.

Spheroids treated with DNA damage inducing ADCs yielded similar variability in bystander penetration. At 24 hours, DNA-damage ADCs did not exhibit much PD signal (e.g., T-PBD in Supplementary Figure 8). However, T-calicheamicin D (T-CaliD) exhibited strong PD signal in both directly targeted cells and a considerable fraction of bystander cells (Figure 2A). T-CaliD showed PD signal all the way to the center of the spheroid by 48 hours, and treated spheroids completely disintegrated by 72 hours due to cell death. All other ADCs began exhibiting PD signal in directly targeted cells by 48 hours and measurable PD signal in bystander cells by 72 hours (Figure 2B). For the DNA-interacting payloads, calicheamicin D had the greatest bystander cell killing, followed by PBD and calicheamicin G (CaliG). For the topoisomerase inhibitors, exatecan exhibited the greatest PD signal at the spheroid center, followed by DXd and then SN-38.

Figure 2. High resolution imaging of DNA damage payload penetration in spheroids.

Immunofluorescence microscopy of spheroids treated with 25 nM AF680-trastuzumab (green) conjugated to DNA-interacting payloads (CaliD, PBD, CaliG) or topoisomerase inhibitors (exatecan, DXd, SN-38) (A) at early time points (B) at later time points. The PD marker is shown in red, and nuclei in blue (Hoechst). Note that the CaliD ADC time points are 24 hrs earlier than other payloads.

Insights on bystander killing efficiency from payload penetration quantification

To quantify payload penetration distance, we performed Euclidean distance map analysis(2,21) of the spheroid ADC and PD signal (Figure 3). Each row represents a different class of payloads categorized broadly by their mechanism of action: microtubule inhibitors (Figure 3A-C), DNA interacting payloads (Figure 3D-F), and topoisomerase I inhibitors (Figure 3G-I). The pattern of PD signal manifestation quantifies the bystander potential of these payloads, with columns on the left indicating more efficient penetration within each group than those on the right.

Figure 3. Quantification of payload penetration.

(A-C) Microtubule inhibitors, (D-F) DNA interacting agents, (G-I) Topoisomerase 1 inhibitors.

Mean ± SD across N=7 spheroids per group

Green = AF680 ADC; solid orange = 24 hours PD signal, dashed red = 48 hours PD signal, dashed purple = 72 hours PD signal, Dotted black = pH3 autofluorescence.

PD signal depicted here has been adjusted to non-specific signal from untreated spheroids (diffuse fluorescence not originating from cells).

Among the microtubule inhibitors (Figure 3A-C), Kadcyla (T-DM1) showed pharmacodynamic (pH3) signal in cells directly targeted by the ADC on the periphery of the spheroid (distance from spheroid edge ~75µm), but not in bystander cells, consistent with the established non-bystander behavior of Lys-SMCC-DM1(22). In contrast, T-MMAE shows pH3 signal in directly targeted cells as well as bystander cells increasing over time. At early times (24 hours), only directly targeted cells show pH3 signal when treated with T-May, similar to T-DM1. However, at 48 hours, bystander cells immediately adjacent (distance from spheroid edge ~ 125µm) to the directly targeted cells show considerable pH3 signal, while those at the spheroid center remain pH3 low, indicating that Ahx-maytansine exhibits detectable but localized bystander cell death. Both T-MMAE and T-DM1 showed a decrease in the average absolute pH3 intensity in the directly targeted cells at 48 hours. This trend was not observed for T-May, which showed slower manifestation of pH3 signal. The pH3 signal was lower for all microtubule inhibitors at 72 hours, even in directly targeted cells (e.g., T-DM1 in Supplementary Figure 8), limiting our analysis to the first 48 hours for this payload class.

Among the DNA-interacting payloads (Figure 3D-F), T-calicheamicin D (T-CaliD), exhibited the most striking pharmacodynamic response, with prominent intensity of γH2A.X in directly targeted cells and bystander cells as early as 24 hours. T-calicheamicin G (T-CaliG) also showed strong PD signal but was largely confined to directly targeted cells, consistent with other acetylated calicheamicins(23). Additionally, the PD signal intensity and pattern remained unchanged between 48- and 72-hour spheroids. T-PBD showed lower absolute PD signal compared to the calicheamicin ADCs, even in directly targeted cells. This difference is likely due to differences in mechanism of action between calicheamicins (direct strand scission leading to DSBs) and PBD (cross-linking DNA to interrupt replication, forming DSBs)(24), resulting in a differential rate of PD manifestation (Supplementary Figure 9). However, looking at the intra-spheroid pattern, T-PBD exhibits a considerable increase in PD signal across the radius of the spheroid, indicating substantial bystander penetration of the payload, consistent with literature reports of PBD bystander effects(25,26).

Among the topoisomerase I inhibitors (Figure 3G-I), T-exatecan exhibited the most efficient bystander penetration, exhibiting a steady increase in γH2A.X PD signal from 48 to 72 hours at the center of the spheroid. T-DXd was next in bystander efficiency, followed by T-SN-38.

A critical observation is that the pharmacodynamic response in bystander cells at the center of the spheroid is at most 50% of that in the directly targeted cells (Supplementary Figure 10). The potential exception is T-MMAE, which appears to show uniform pharmacodynamic response at 48 hours. However, this occurs mainly due to a drop in the signal for MMAE in directly targeted cells versus an increase in signal at the spheroid center that occurs for several other payloads. This exception aside, the general trend provides quantitative experimental support for previous predictions that direct cell targeting with an ADC is more efficient than bystander killing(8). Additionally, varying the dose in spheroids with T-CaliD (the most potent ADC in panel) shows that while the absolute PD response in both directly targeted and bystander cells increases with increasing ADC dose (Supplementary Figure 11A-B), the relative intra-spheroid normalized PD signal trend is similar across dose groups (Supplementary Figure 11C). Likewise, very high doses of T-CaliG still have low bystander killing (Supplementary Figure 12). This indicates the efficiency of bystander penetration for any payload is relatively independent of the ADC dose.

Permeability alone is a valuable but mechanistically incomplete predictor of bystander killing efficiency

Analysis of the spheroid data revealed many differences in bystander efficiency that were not apparent solely from structural or physicochemical differences (Supplementary Table 5). For example, calicheamicin D and calicheamicin G are nearly identical in structure and properties yet exhibit a striking difference in their bystander killing efficiency in spheroids. In contrast, exatecan (DX-8951) and DXd (DX-8951 derivative) also share structural similarities(14) but exhibit less dramatic differences in bystander penetration. The ability of the payload to pass lipid membranes, either via permeability measurements or free payload cytotoxicity assays, has previously been used as a predictor for bystander effects in vivo. We measured the cytotoxicity of the ADC (to measure cell death from direct ADC targeting), the permeability of the free payload via a PAMPA assay, and the cytotoxicity of the free payload (a proxy measure for payload permeability) (Figure 4) to capture the ability of the payload to achieve cytotoxic cellular concentration(27,28). Overall, the higher potency of the ADC relative to the free payload suggests efficient release of the active payload.

Figure 4. Cytotoxicity and permeability provide valuable but incomplete information for in vivo bystander killing efficiency.

(A) Cytotoxic IC₅₀ of trastuzumab-payload conjugates in HCC1954 cells (log scale), (B) Permeability of free payloads as measured using PAMPA, (C) Cytotoxic IC₅₀ of free payloads in HCC1954 cells (log scale).

*Ogitani et al. 2016(14)

Among the microtubule inhibitors, the ADC potency (IC₅₀) was nearly identical for all three (Figure 4A, Supplementary Figure 13), suggesting negligible differences in intracellular kinetics or intrinsic potency. However, free MMAE exhibited nearly 10-fold higher permeability (Figure 4B) and over 100-fold more potent free payload IC₅₀ than either Ahx-maytansine (Ahx-May) or Lys-SMCC-DM1 (Figure 4C, Supplementary Figure 13). Here, the PAMPA permeability and free payload IC₅₀ correlate well with the observed trend in bystander killing efficiency in spheroids, indicating that permeability measurements can be good indicators of bystander killing efficiency

Among the topoisomerase inhibitors, the cytotoxic IC₅₀ of T-exatecan and T-DXd were similar to each other (1.0 and 1.4 nM, though some literature reports suggest exatecan may be more potent in other cell systems(14,29)) and ~10-fold more potent than that of T-SN-38, consistent with literature evidence(30). This could explain why cells directly targeted by T-SN-38 did not show as strong a γH2A.X signal in spheroids (Figure 3G-I) compared to T-exatecan or T-DXd. The effect could also be driven by potential differences in linker processing and/or intracellular binding kinetics, since the ADC potency was lower than the free SN-38 potency, which, in part, can be explained by the fact that protease release of SN-38 appears less effective than the hydrolysis of a carbonate linker(31). Exatecan permeability was ~2-fold higher than DXd and ~5-fold higher than SN-38 (Figure 4C), indicating a reasonable correlation between permeability and observed bystander penetration. Since SN-38 showed measurable permeability and free payload IC₅₀ (i.e., it is capable of passing across the cell membrane), it is likely that SN-38 exhibits some bystander effects, but poor payload release/accumulation and/or lower cellular potency of SN-38 appears to be driving the lower PD signal in T-SN-38 spheroid bystander cells. This dataset indicates that while permeability and free payload IC₅₀ measurements correlate reasonably well with expected bystander killing, other factors such as potency can also be important to characterize bystander potential and interpretation of imaging results.

Finally, among the DNA-interacting payloads, the ADC cytotoxicity measurements revealed a striking 100-fold stronger potency for T-CaliD (sub-picomolar IC₅₀) compared to T-CaliG and T-PBD, indicating that CaliD is intrinsically more toxic. Additionally, free CaliD exhibited ~3-fold and ~2-fold higher permeability than PBD and CaliG, respectively. While this empirically matches the spheroid results, 2-fold differences in permeability between exatecan and DXd had much less of an impact on bystander killing. A further loss in correlation between permeability and bystander killing efficiency is observed when comparing PBD and CaliG. ADC cytotoxicity revealed T-PBD and T-CaliG have roughly similar IC₅₀’s after 6 days, but the rate of achieving maximum PD signal is slower for PBD (Supplementary Figure 9), likely due to different DNA damage mechanisms. Despite a slower mechanism and lower PAMPA permeability, PBD showed higher PD signal in cells at the center of the spheroid than CaliG (Figure 3, Supplementary Figure 10). While membrane permeability is critical for bystander effects, this data set emphasizes the need to account for other factors in assessing bystander payload tissue penetration.

In vitro co-culture assays do not capture bystander payload transport limitations

The current gold standard in identifying bystander effects is using in vitro co-culture and in vivo mosaic tumor systems consisting of a mixture of antigen-positive (Ag+) and antigen-negative (Ag−) cells. The efficiency of bystander killing is measured by the degree of Ag− cell death via free payload released from Ag+ cells. We performed an in vitro co-culture assay using HCC1954 (HER2+ cell line, ~3 million receptors per cell) and MDA-MB-468 (HER2- cell line (32,33), Supplementary Figure 14A).

Among microtubule inhibitors (Figure 5A-C), treatment with non-bystander T-DM1 (1 nM) resulted in complete Ag+ cell death but no Ag− toxicity even at a 90% Ag+ cell percentage. In fact, the absence of Ag+ cells competing for resources appeared to allow faster growth of Ag− cells compared to controls (Supplementary Figure 14B, C). A similar effect was observed with 1 nM T-May up to 75% Ag+ cells, but at 90% Ag+ cells, the Ag− cells begin to exhibit bystander toxicity. This observation matched tumor spheroid results, where only the cells immediately adjacent to directly targeted cells showed bystander killing activity. In contrast, T-MMAE demonstrated near complete Ag− cell death by 50% Ag+ cells despite being treated with the same total ADC/payload dose, consistent with a higher MMAE free payload toxicity and spheroid bystander effects.

Figure 5. In vitro bystander co-culture assay with HCC1954 (Ag+) and MDA-MB-468.

(Ag−). The fraction of Ag- cells in treated wells versus Ag- cells in untreated wells is shown for different ratios of Ag+ and Ag- cell coculture. ADC dose was fixed to be above the IC₉₀ for Ag+ cells and below the IC₅₀ for Ag- cells where possible (Supplementary Figure 6). The ADC is internalized, and the payload is released inside the Ag+ cells. For a non-bystander payload, the payload is unable to escape the Ag+ cells, and therefore incapable of killing the surrounding Ag- cells. Therefore, the fraction of surviving Ag- cells (y axis) is ≥ 1 regardless of the ratio of Ag+ to Ag- cells (e.g. T-DM1). On the other hand, a bystander payload can escape Ag+ cells and efficiently accumulate in surrounding Ag- cells, leading to fraction of surviving Ag- cells (y axis) < 1 in the presence of Ag+ cells, approaching 0 with increasing ratio of Ag+ to Ag- cells. The more efficient the bystander effect, the smaller the number of Ag+ cells needed to drive the y-axis to 0.

Among the DNA interacting agents (Figure 5D-F), T-CaliD (50 pM) and T-PBD (500 pM) exhibit bystander cell death with as low as 25% Ag+ cells. However, the 10-fold difference in ADC treatment concentration (to maintain selectivity on Ag+ versus Ag− cells) obscures the high bystander killing of CaliD in spheroids driven by its extreme intrinsic potency. T-CaliG (1 nM) exhibits modest bystander action in this assay.

The co-culture assay could not be applied to the topoisomerase inhibitors due to a 2–3 fold higher payload sensitivity of MDA-MB-468 cells to free SN-38 (IC₅₀ ~0.6nM) and DXd (IC₅₀ ~ 2.3nM) compared to HCC1954 with IC₅₀’s of ~1.8 nM and ~6.2 nM, respectively (Supplementary Figure 15). This made it impossible to select an ADC concentration that would mediate receptor-mediated cell death in Ag+ cells without also causing Ag− cell death due to non-specific pinocytosis of the ADC (Supplementary Figure 6). With further protocol modification, where a fixed number of Ag− cells are co-cultured with increasing number of Ag+ cells and treated with a fixed ADC concentration (Figure 5G-I), we were able to provide some resolution for exatecan and SN-38, with exatecan being more efficient. The extreme sensitivity of MDA-MB-468 to DXd prevented treatment of the co-culture with any concentration higher than 250 pM (resulting in 100% Ag− cell death), giving the erroneous impression that DXd is not a very efficient bystander payload.

Simulations of bystander ‘tissue penetration efficiency’ versus bystander ‘killing’

Simulations are playing an increasingly vital role in clinical design of ADCs where experimental acellular, in vitro, and in vivo data cannot fully overcome translational challenges (e.g. differences in payload sensitivity/toxicity(34), variable tumor saturation(12,34), measurement of efficacy and toxicity in different species (e.g. mice and primates)(35), etc.). The simulations can integrate all these disparate factors into a single framework for clinical prediction and more robust metrics for bystander payload design. To incorporate the multiple factors influencing the bystander effect, such as membrane permeability, payload distribution, payload potency, and reversible/irreversible payload target interaction, we utilized a previously developed computational framework to compare with the experimental results and enable scaling from in vitro to in vivo and the clinic. These simulations capture the predicted concentration profile of payload distribution (Figure 6A) and associated pharmacodynamic response (Figure 6B, Supplementary Figure 16)(8,9). For the microtubule inhibitor ADCs, the payload concentration profile adequately captures the bystander killing efficiencies quantified by the experimental tumor spheroids, namely that MMAE exhibits good penetration to the center of the spheroid, Ahx-May shows limited but detectable penetration, while Lys-SMCC-DM1 shows no bystander penetration. This trend is mirrored in the simulated pharmacodynamic response (calculated by estimating the intracellular payload concentration and measured PD signal in monolayer cells, Supplementary Figure 17), since the intrinsic potency of these payloads is similar. For the topoisomerase inhibitor ADCs, exatecan and DXd are more efficient at reaching the center of the spheroid than SN-38, which translates into higher PD signal at the center similar to that observed experimentally. T-DXd is predicted to be as efficient as T-exatecan, yet we see more efficient penetration with exatecan experimentally in the spheroid experiments (Supplementary Figure 10G-H). This could be due to the partial charge on exatecan or simplification of the intracellular binding kinetics to a single equilibrium reaction.

Figure 6. Predictive simulations to distinguish bystander effects from bystander killing.

(A) Spheroid payload concentration patterns highlighting bystander distribution/effects, along with corresponding Da (200µm), (B) Conversion of payload concentration to expected pharmacodynamic response, highlighting the pattern of bystander killing.

The predictions for DNA-interacting agents with significant intrinsic potency differences highlight the distinction between bystander payload distribution and killing. Both T-CaliD and T-CaliG are predicted to distribute similarly due to their similar molecular weight, lipophilicity, and membrane permeability (Figure 6A). T-CaliG has a slightly higher concentration due to accumulation over a longer period of time (72 hours vs. 48 hours for T-CaliD). However, the dramatically higher potency of T-CaliD results in much higher bystander killing for T-CaliD than T-CaliG (Supplementary Figure 18), consistent with experimental PD signal (Supplementary Figure 10D-F). CaliG is predicted to distribute efficiently (Figure 6A) in contrast to experimental quantification (and co-culture assay results). Accounting for the lower intrinsic potency of CaliG (compared to CaliD) does not completely explain the discrepancy. We speculate potential differences in activation (reduction and diradical formation(36)) or inactivation by e.g. “self-sacrifice” proteins(37,38) account for the poor bystander response with T-CaliG given CaliG’s reasonable permeability, strong response in directly targeted cells, and reasonable potency of the free payload (when replacing the payload/media every day). T-PBD is predicted to have lower penetration, but when accounting for difference in intrinsic potency (Supplementary Figure 17), the simulated PD signal matches the experimental trends well (Figure 6B). Note that the slower manifestation of PD signal with the PBD relative to calicheamicins (Supplementary Figure 9) is not included in the simulations.

Damköhler number as a descriptor of bystander efficiency

Membrane permeability is a key factor in enabling payload escape and bystander effects. However, there are cases where too high of a permeability can be detrimental to tissue penetration when the cellular uptake rate and target binding exceed the rate of diffusion(39). The dimensionless Damköhler number (Da) described in detail in (8) quantifies the theoretical bystander penetration of ADC payloads by capturing this fundamental ratio of cell uptake (described by k_in,P) versus tissue diffusion of the payload (described by D_eff,P), the two critical pharmacokinetic features influencing payload penetration, without the need for complex simulations. A ratio well below 1 indicates a payload that is slow to escape cells and may diffuse evenly but wash out of the tissue before entering bystander cells. A ratio much greater than 1 indicates the payload will rapidly enter adjacent cells before penetrating deeper into the tissue (dependent on the depth of tissue in that system).

$$Da=\frac{k_{in,P}{R}_{spheroid/Krogh}^2\left(1-{\epsilon }_P\right)}{4D_{eff,P}{\epsilon }_P}$$

The Da is calculated from: (i) the transient effective extracellular diffusion, D_eff,P - calculated from the steady state diffusion coefficient, D (a function of molecular weight, lipophilicity/cLogD, # of H-bond donors, and # of H-bond acceptors(40)) and ratio of bound to free drug in tissue (R), (ii) the rate of intracellular uptake of the payload (k_in,P or t_{½ ,uptake}, estimated from the PAMPA membrane permeability), and (iii) a characteristic length of distribution (spheroid radius, R_spheroid ~ 200µm or half the tumor intercapillary distance, R_Krogh ~ 75µm). A summary of the Da calculation steps is listed in Supplementary Information, and detailed description and steps can be found in a previous dedicated publication (8). Based on previous simulations, a payload with 1 ≤ Da ≤ 3 exhibits the most efficient bystander effects/distribution, with Da > 3 representing payloads with slow tissue penetration, and Da < 1 representing payloads exhibiting rapid extracellular penetration that wash out of the tumor/spheroid before substantial uptake by cells.

For payloads of similar potency (e.g. the microtubule inhibitors and topoisomerase inhibitors), Da reasonably predicts the trends in bystander killing (Supplementary Table 5). However, a key limitation of the Da is that it only predicts payload distribution and does not account for differences in potency or intracellular binding kinetics. Notably, the Da calculation presumes irreversible immobilization of the payload following cellular uptake. This may not hold true for payloads like calicheamicins, PBDs, and DGNs (e.g. DGN549), which exhibit a relatively slower rate of reaction to DNA(9,41). Therefore, the Da predicts slower penetration for CaliD (Da_200µm = 14.6) than the full simulations (Figure 6A).

The Damköhler number can also be used to predict optimal payload physicochemical properties. The transient diffusion coefficient, cellular uptake rate, and spheroid/tumor radius (i.e., physical distance payload would need to penetrate) were varied in a local sensitivity analysis to determine the impact on the Damköhler number as a function of physicochemical properties. The theoretical optimal Da of 1–3 generally corresponded to a lipophilicity (clogD) range of 2–3 when a large penetration distance is desired (maximum radius of penetration = 200 µm) at the cost of lower tumor retention. For shorter distances (e.g. well-vascularized tumors) a lipophilicity (clogD) range of 3–4 is optimal with a 75 µm penetration distance (Supplementary Figures 19-21).

Discussion

Mechanistic modeling of ADC payload transport reveals that payloads possess a ‘bystander potential’ that covers a spectrum from ‘non-bystander’ to efficient ‘bystander’ payloads and quantifies their ability to reach and kill untargeted tumor cells(8). However, the challenges with direct tracking of small molecule payloads makes experimental quantification of this bystander potential and tissue penetration difficult. Even subtle differences in the structure of a small molecule, such as a hydroxy vs. ethoxy functional group, can dramatically impact distribution(39). Using pharmacodynamic markers, a well-established tool to detect the cytotoxic effects of exogenous agents, ADC payload response can be tracked with cellular resolution(9). In this study, we utilized this experimental PD platform with mechanistic modeling to quantify the ‘bystander potential’ of a panel of ADC payloads, including the three major classes: DNA-damaging agents, microtubule inhibitors, and topoisomerase inhibitors, all of which are employed by FDA-approved ADCs. Because these payloads encompass a wide range in potency, this study was performed in a well characterized, high expression HER2 system so that all agents could be studied in the same spheroid system. The high expression also maximizes the spatial distinction between ADC-targeted and untargeted cells. Although lower Ag expression or small-fragment ADCs/small molecule drug conjugates often exhibit less heterogeneous intratumoral distribution (2,9), the purpose of this work was to quantify the efficiency of bystander payload tissue penetration. These approaches for improving ADC tissue penetration (separate from payload penetration) are also expected to increase efficacy, since targeting cells directly with the ADC is more efficient than bystander killing. However, bystander effects are also important for tumors with heterogeneous expression and/or antigen negative cells.

The tumor spheroid system using pharmacodynamic markers described here presents a viable system for direct assessment of anticipated in vivo tissue penetration of bystander payloads, . Importantly, the pharmacodynamic (PD) response in bystander cells was only ~50% of the maximum observed in peripheral cells targeted directly by the ADC. This supports previous predictions of higher cell killing efficiency on directly targeted cells versus bystander cells, and it explains the benefit of increasing ADC tissue penetration with higher antibody doses even for ADCs with bystander payloads(8). Most critically, the 3D spheroid system captures the mass transport limitations (e.g., intratumoral diffusion and cellular uptake) bystander payloads face in vivo, processes that are not completely captured by 2D in vitro or acellular assays alone. For example, in vitro co-culture assays (current gold standard) often have Ag+/ Ag− cells well interspersed among each other, providing released free payloads easy access to the Ag− cells. However, clinical tumors do not always have evenly interspersed Ag+/ Ag− cells, and often exhibit compartmentalized Ag+ and Ag− clusters that can be span several hundreds of microns(42). Therefore, payloads that exhibit bystander killing in well-mixed 2D in vitro co-culture assays may exhibit good Ag− cell bystander killing in well-interspersed heterogeneous tumors, but not necessarily in compartmentalized tumors . For example, in this study CaliG shows reasonable bystander killing in vitro, yet in spheroids the bystander killing response at the center of the spheroid is negligible (Figure 2–3), even with a higher ADC dose (Supplementary Figure 12). Additionally, co-culture assays that report only a narrow range of Ag+/ Ag− cell ratios yield binary “yes/no” assessments of bystander effects that may miss the quantitative spectrum in bystander response. Finally, designing co-culture experiments can present unanticipated challenges, such as differences in sensitivity of Ag− and Ag+ cells, and matched Ag+/ Ag− cell lines are not always available. For example, despite efficient receptor-mediated uptake in HER2+ HCC1954 cells, the IC₅₀ of two camptothecin-based ADCs was in fact lower (T-DXd) or similar (T-SN-38) in the Ag− cells, precluding the use of a selective ADC concentration. In comparison, quantification of PD signal in spheroids offers a straightforward method of bystander killing quantification within the same cell type by exploiting spatial bystander effects (SBE) instead of heterogeneous bystander effects (HBE)(5).

The spheroid system here provides a compromise between the simplicity of PAMPA and cellular assays and the complexity of in vivo tissue penetration studies. PAMPA, originally developed as a correlate of oral drug absorption, is a valuable assay that captures diffusion through lipids and aqueous compartments, thereby providing an estimate of cellular uptake rates. Permeability is frequently a good predictor of bystander effects – indeed a strong correlation can be expected between PAMPA permeability and bystander killing in 2D in vitro assays, as cellular uptake is the key payload transport requirement for efficacy. However, PAMPA does not capture cellular activation and/or intratumoral diffusion, which can impact bystander payload penetration in tissue. For example, CaliD and CaliG displayed a striking difference in bystander killing efficiency in tumor spheroids, despite only a 2-fold difference in permeability in the PAMPA assay. This could potentially occur due to a greater interaction with DNA(43) or delayed generation of the diradical (with a stability of ~3 seconds(36)), both of which are captured in the spheroid system. For high permeability causing penetration issues, one of the clearest examples is Hoechst 33342 vs. Hoechst 33258(39) (dyes with topoisomerase 1 activity(44) but originally developed as anti-parasitic drugs(45)). These DNA intercalating agents both have high PAMPA permeability (10 to 20 × 10⁻⁶ cm/s) with Hoechst 33342 being more permeable than 33258, but this more permeable agent has poor tissue distribution as predicted by simulations and seen in spheroids and animal models.

Although PAMPA permeability may not always independently correlate with in vivo bystander killing, it is critical to mechanistically understanding bystander payload penetration through the Damköhler number (Da) analysis. The Da, which is a ratio of extracellular diffusion to cell uptake (estimated from PAMPA permeability), is a useful tool for predicting the bystander distribution of a payload(8). This dimensionless number quantifies the distribution efficiency of the payload over a given distance. It does not provide insight on the degree of cell death in bystander cells, i.e., the magnitude of bystander ‘killing’, since it does not incorporate intrinsic payload potency. However, the Da provides a measure of penetration efficiency (distribution) while the potency and delivered dose determine the absolute killing (magnitude). As an analogy, the bystander potential determines the “slope of the shoreline” while the potency and tolerable dose determine the “water level” to yield the amount of “dry ground” (cell death).

It is important to distinguish distribution from potency, as a large dose of an extremely potent payload could exhibit substantial cell killing in spheroids even with less efficient penetration. The toxicity in vivo could limit the clinical dose to levels that may not achieve strong bystander killing due to low absolute concentrations. For example, the intrinsic payload potency of CaliD was about 100-fold higher than that of CaliG and the CaliD ADC was ~75-fold more potent. Mechanistically, CaliG has an aminoacetyl group similar to the clinical payload ozogamicin, which reduces the DNA interaction(46) likely by the loss of an ionic interaction(43). A comparison of PD response in spheroids with potency matched doses of T-CaliG (75nM) and T-CaliD (1nM) shows that CaliD is more efficient at reaching cells in the spheroid center when normalized to directly targeted cells. However, the absolute magnitude of PD response is higher with the larger dose of T-CaliG (Supplementary Figure 12). A typical MTD for a PBD-based ADC is often on the order of 0.1–0.3 mg/kg, while clinical doses of ozogamicin-based ADCs are 0.02–0.08 mg/kg(47,48). Given the higher potency, a CaliD-based ADC is likely to have an even lower MTD and therefore very limited total payload accumulation in the tumor. With this low payload uptake in the tumor, even the ultrahigh potency of CaliD may not compensate for its limited tolerability (Supplementary Figure 18). In contrast, free PBD maintains efficient tissue penetration with higher tolerable dosing. This highlights the need to integrate bystander potential data with other aspects of ADC design, such as toxicity, ADC disposition, and efficacy.

The transient distribution of released ADC payloads is complex, particularly when overlaid with the heterogeneous distribution of the ADCs themselves and tolerability/potency considerations. The work here presents quantitative insight into the tissue penetration efficiency of released payloads for all three major classes to help interpret animal data and scale these results to the clinic. The ideal scenario for isolating and measuring tissue penetration alone would have been to have ADCs matched by mechanism of action, ADC potency, and intrinsic payload potency. However, this is possible in silico and for some payloads (e.g., the microtubule inhibitors here), but not in most circumstances since the payload chemical structure drives both the potency and the tissue distribution. Most choices in ADC design involve selections among payloads that are not matched in all these parameters, and we wanted to measure the efficiency of bystander killing for each. However, there are limitations to this approach, particularly regarding the intrinsic payload potency, uncertainties with molecular kinetics, and impact of the linker and release mechanism (similar to other assays such as PAMPA and 2D co-culture). For example, the bystander effect of SN-38 was not readily apparent in the tumor spheroids despite the success of SN-38-bearing sacituzumab govitecan (Trodelvy). The simulations indicate this is partially a result of the fast intratumoral diffusion that results in washout of the payload, but there are likely benefits from the higher DAR of sacituzumab govitecan (DAR ~8) versus the DAR 4 tested here. Moreover, Trodelvy’s carbonate-based hydrolysable linker has demonstrated greater efficacy than protease-cleavable linkers for SN-38(31,49). Likewise, the difference in CaliD and CaliG bystander response in spheroids does not appear to be entirely due to the 100-fold difference in potency and could also be due to differences in metabolic processing of the linker and/or payload. For linker impact, the free sulfhydryl of the maytansine payload-linker DM4 can be methylated(50), resulting in a lipophilic payload with better bystander response (and correspondingly more optimal Damköhler number(8)). Nevertheless, this method for tracking the tissue penetration of ADC payloads can be applied to these and other linker-payload combinations to determine optimal properties for clinical candidates.

In conclusion, pharmacodynamic staining can be used to track the tissue penetration of ADC payloads with cellular resolution. In agreement with theoretical simulations, directly targeted cells show a stronger response than bystander cells regardless of the payload properties or mechanism of action. This highlights the need to improve ADC tissue penetration to maximize efficacy. All three classes of payloads demonstrate bystander killing to varying degrees, with MMAE, CaliD, and exatecan showing the greatest bystander killing of the agents measured here. The Damköhler number provides a measure of efficiency in payload penetration, and this predicts higher efficiency of PBD than CaliD at tolerable doses given the more optimal physicochemical properties. Overall, the data can be used to quantify the efficiency of novel payload/linkers and match optimal payloads for a given target.