Improved Tumor Penetration and Single-Cell Targeting of Antibody Drug Conjugates Increases Anticancer Efficacy and Host Survival

Abstract

Current antibody-drug conjugates (ADC) have made advances in engineering the antibody, linker, conjugation site, small molecule payload and drug-to-antibody ratio (DAR). However, the relationship between heterogeneous intratumoral distribution and efficacy of ADC is poorly understood. Here we compared trastuzumab and ado-trastuzumab emtansine (T-DM1) to study the impact of ADC tumor distribution on efficacy. In a mouse xenograft model insensitive to trastuzumab, co-administration of trastuzumab with a fixed dose of T-DM1 at 3:1 and 8:1 ratios dramatically improved ADC tumor penetration and resulted in twice the improvement in median survival compared to T-DM1 alone. In this setting, the effective DAR was lowered, decreasing the amount of payload delivered to each targeted cell but increasing the number of cells that received payload. This result is counterintuitive because trastuzumab acts as an antagonist in vitro and has no single-agent efficacy in vivo, yet improves the effectiveness of T-DM1 in vivo. Novel dual-channel fluorescence ratios quantified single-cell ADC uptake and metabolism and confirmed that the in vivo cellular dose of T-DM1 alone exceeded the minimum required for efficacy in this model. Additionally, this technique characterized cellular pharmacokinetics with heterogeneous delivery after one day, degradation and payload release by two days, and in vitro cell killing and in vivo tumor shrinkage 2-3 days later. This work demonstrates that the intratumoral distribution of ADC - independent of payload dose or plasma clearance - plays a major role in ADC efficacy.

Introduction

Antibodies and antibody-drug conjugates (ADCs) make up the largest portion of the growing biologics market. Currently, there are over 50 FDA approved antibodies, and nearly 500 in the various stages of the clinical pipeline (1). Although there are currently over 70 ADCs in the clinical pipeline (2,3) only four, Adcetris, Besponsa, Mylotarg, and Kadcyla (T-DM1), are currently approved by the FDA. While these ADCs have had clinical success, the factorial optimization of the antibody, linker, conjugation site, small molecule payload, drug loading, and target selection make development of each ADC a unique challenge. Although there have been advances in engineering the biophysical characteristics to improve safety, stability, and develop more homogeneous products, ADCs continue to be limited by toxicity, which is typically driven by the toxicity of the small molecule payload(4,5). In particular, several of the recent ADC failures may have been prevented by marginal gains in tolerability(2).

It is widely known that antibodies/ADCs exhibit heterogeneous distribution in solid tumors(6-11); however, it is not well understood how the heterogeneous tissue distribution of ADCs impacts their overall efficacy. ADC efficacy requires a multi-step process, which includes the distribution of intact ADC in the tumor, cellular uptake and degradation of the antibody, release of the small molecule payload, induction of apoptosis by the cytotoxin, and potentially bystander effects on neighboring cells(12-14). Therefore, there is a need to understand the ADC's effects from the subcellular scale (e.g. how many ADCs are required to achieve cell death in vivo) to the tissue level (e.g. how many cells in the tumor are receiving a therapeutic dose) to whole organ biodistribution (e.g. what is the healthy tissue exposure and resulting toxicity) in order to develop effective therapeutics.

Previously we developed a combined tissue and physiologically-based pharmacokinetic (PBPK) model to describe both the tumor and systemic distribution of T-DM1(15). We found that co-administration of trastuzumab with T-DM1 (trastuzumab linked to the payload DM1, which therefore competes for the same binding epitope of HER2) dramatically improved tumor penetration, but total tumor uptake of ADC was unchanged. Co-administration of trastuzumab with T-DM1 lowers the effective drug to antibody ratio (DAR) while competing with T-DM1 for HER2 receptors and driving penetration deeper into the tissue. The higher the trastuzumab dose, the farther the ADC will penetrate into the tumor; however, the average DM1 payload concentration in each targeted cell will be lower. Several studies in the literature demonstrate that cohorts of mice treated with ADCs having different DAR but the same overall payload dose (i.e. DAR2 given at 2 mg/kg vs. DAR4 given at 1 mg/kg) had better outcomes with the lower DAR (and therefore higher antibody dose which correlates with better tumor penetration)(15). These results are consistent with tumor penetration playing a major role in efficacy independent of the target antigen, antibody, linker, payload, or bystander effects. However, these were not prospective studies, and although certain mechanisms, such as DAR-dependent clearance, did not appear to play a role in the analysis, they could have affected the data interpretation. Therefore, we wanted to design an experimental study to isolate the impact of distributionon efficacy.

Here, we demonstrate that co-administration of trastuzumab with T-DM1 improves efficacy (tumor growth reduction and overall survival as measured by tumor volume endpoint) in a trastuzumab-insensitive mouse xenograft model. Against trastuzumab-insensitive cell lines the addition of trastuzumab was antagonistic in vitro (i.e. lowered efficacy by blocking T-DM1 uptake), as expected. Counterintuitively, co-administration of trastuzumab (which acts as an antagonist in vitro and has no single-agent efficacy in this animal model in vivo) with T-DM1 showed a significant improvement in efficacy in a mouse xenograft model despite the same small molecule dose and tumor uptake as T-DM1 alone. In fact, the combination of trastuzumab, which had no single agent efficacy, and T-DM1 was synergistic, meaning the net improvement was greater than additive (since trastuzumab alone had no impact on efficacy in the absence of T-DM1). Histological imaging showed a significant increase in T-DM1 tumor penetration with the co-administered trastuzumab. Additionally, we present a novel near-infrared (NIR) fluorescence ratio technique with dually labeled ADCs to track the metabolism and distribution of ADCs at the single-cell level. Applying this technique to single agent T-DM1 therapy showed the delivery of ADC to cells within the targeted population in vivo was higher than the threshold required for cell death, while the majority of tumor cells did not receive any ADC. These results demonstrate that the intratumoral distribution of ADCs in tumor tissue plays a major role in determining their efficacy independent of the amount of total tumor payload delivered. To our knowledge, this is the first time that the distribution itself, independent of the other parameters that affect efficacy and tumor penetration such as dose, plasma clearance, and molecular weight, significantly impacted survival.

Materials and Methods

Antibodies and NIR Imaging Agents for Ratio Measurements

Herceptin (trastuzumab, Roche) and Kadcyla (T-DM1, Roche) were obtained from the University of Michigan Pharmacy. Alexa Fluor 680 NHS Ester (AF680, ThermoFisher Scientific, A37567), IRDye 800CW NHS Ester (IRDye, LI-COR, 929-70020), and CellTrace™ Far Red DDAO-SE (DDAO, ThermoFisher Scientific, C34553) were conjugated to the antibodies following the manufacturer's instructions as previously described(15,16). Antibody/ADC at 2mg/mL supplemented with 10% sodium bicarbonate (v/v) was reacted with dye at molar ratios of 0.5 (AF680, IRDye) and 1.5 (DDAO) for 2 hours at room temperature and purified using P6 Biogel (1g gel/10mL PBS) resulting in dye to protein ratios of approximately 0.3 (AF680, IRDye) and 0.7 (DDAO). Our previous work has shown that the distribution of T-DM1 is unchanged after labeling with AF680 at dye to protein ratio of 0.3 or less(17). Antibody/ADC dye conjugates were run on SDS-PAGE and scanned on the Odyssey CLx Scanner (LI-COR) to ensure free dye was removed. For fluorescence histology, antimouse CD31 (BioLegend, 102402) was conjugated with Alexa Fluor 555 (ThermoFisher Scientific, A37571), mouse antihuman IgG Fc antibody (BioLegend, 409302) was conjugated with Alexa Fluor 488 (ThermoFisher Scientific, A20000), and trastuzumab was conjugated with Alexa Fluor 750 (ThermoFisher Scientific, A20011) at dye to protein ratios of 1.5.

Cell Lines and In Vitro Toxicity

NCI-N87 and HCC1954 cells were purchased from ATCC in May 2015 and June 2016, respectively. Cell line authentication was performed by ATCC using cytochrome C oxidase (COI) testing and short tandem repeat (STR) profiling. Cells were grown at 37°C with 5% CO₂ in RPMI 1640 growth medium supplemented with 10% (v/v) FBS, 50 U/ml penicillin, and 50 μg/ml streptomycin. Mycoplasma testing was performed yearly using the Mycoalert testing kit (ThermoFisher Scientific, NC9719283). Cells were cultured 2-3 times per week up to passage number 50 (approximately 3-4 months). For cell viability assays, 5,000 cells were plated in 96 well plates. Titrations of T-DM1 or T-DM1 and trastuzumab were replaced daily for 6 days and viability was measured using the PrestoBlue Cell Viability Reagent (ThermoFisher Scientific, A13261). Briefly, cells were washed twice with media and a 1:10 dilution of PrestoBlue in media was incubated for 25 minutes at 37°C. After incubation, the fluorescence (560/590, Ex/Em) of each well was measured using a Biotek Synergy plate reader. Background signal from wells without cells was subtracted from all samples and then viability was normalized to untreated cells.

In Vitro NIR Fluorescence Ratio Measurements and Fluorescence Microscopy

ADC metabolism was studied by dually labeling T-DM1 with DDAO and IRDye as described above. As the labeled ADC binds to the cell surface receptor, gets internalized, and subsequently degraded, the low molecular weight and more lipophilic DDAO diffuses out of the cell while the IRDye remains trapped(16). DDAO therefore approximates the intact protein (since it is cleared upon degradation), while IRDye approximates the cumulative uptake in the cell(18). Unlike pH effects(19) or quenching/FRET, this provides an irreversible measurement of both intact protein and payload delivery without requiring a high degree of labeling (self-quenching approach) or larger dye-quencher conjugates. NCI-N87 and HCC1954 cells were plate in 96 well plates. Cells were labeled for 30 minutes at 37°C at different times over a 48 hour period. After each labeling cells were washed twice to remove excess media. After 48 hours cells were washed three times, then harvested using Cellstripper (Corning, 25-056-CI), a non-enzymatic cell dissociation solution, and fluorescence intensity was quantified using an Attune Acoustic Focusing Cytometer (Life Technologies). The signal for each dye was normalized to the initial time point, and the normalized ratio of DDAO divided by IRDye was plotted to show the ratio of intact ADC to cumulative uptake. Alternatively, cells were imaged using fluorescence confocal microscopy (Olympus) with a 635 nm laser for DDAO and a 748 nm laser for IRDye.

Tumor Growth Studies

All animal studies were conducted in accordance with the University of Michigan Institutional Animal Care and Use Committee (IACUC). For all tumor xenografts studies, 5×10⁶ NCI-N87 cells were inoculated in the rear flanks of mice (one flank for tumor growth studies, both flanks for all others). For tumor growth studies, tumor volume was measured using calipers every other day using the formula volume = 0.5*length*width². Trastuzumab, T-DM1, both trastuzumab and T-DM1 (all unlabeled), or saline were injected via tail vein once tumors reached 250mm³. For tumor growth studies 10 animals were used for all treated and untreated cohorts. Kaplan-Meier survival curves were generated in PRISM and were analyzed by log-rank test at significance level of p ≤ 0.05.

In Vivo NIR Fluorescence Ratio Measurements and Fluorescence Histology

To study the cellular uptake and metabolism kinetics in vivo, tumor xenografts were treated with DDAO and IRDye labeled T-DM1, and the tumors were resected, digested into a single cell suspension, and analyzed on flow cytometry (similar to the in vitro assay). Once the longest axis of the tumor reached 9-10mm, 100μg of dually labeled T-DM1 was injected via tail-vein and animals were sacrificed at 24, 48, and 72 hours. After sacrifice, tumors were resected and sliced before being placed in a collagenase IV solution (5mg/mL). The tissue was digested for 25 minutes before centrifugation (5 min, 300 g). The cell pellet was resuspended in media, washed twice, and filtered through a40μm filter. Cells were then analyzed by flow cytometry. Uninjected negative control tumor digests were used to establish gates for DDAO and IRDye fluorescence. To determine the percent intact (from the DDAO/IRDye ratio) we examined cells that were targeted with T-DM1 (IRDye+). The background mean fluorescence intensity in RL1 and RL2 from negative control tumors was subtracted from the mean fluorescence intensity of DDAO (RL1) and IRDye (RL2) to get fluorescence per targeted cell. Then the DDAO signal was divided by IRDye to get the DDAO/IRDye ratio. To get the percent intact, this ratio was normalized to the initial intact ratio, which was determined by harvesting in vitro cells, labeling on ice for 25 minutes, washing twice, and analyzing by flow cytometry. In addition to running tumor cells on flow cytometry, part of the tumor was used for fluorescence biodistribution. The low autofluorescence in the near-IR makes IRDye a suitable fluorophore to determine organ uptake(17,20,21), and fluorescence biodistribution was performed as previously described(15,17,22). Fluorescence histology was performed as previously described(15,17). A detailed description is provided in the Supplementary Material.

Results

Co-Administration of Trastuzumab with T-DM1 Improves T-DM1 Tumor Penetration

ADC tumor distribution is dependent on many parameters including dose, DAR, systemic clearance, antigen expression and internalization rate in tumor (and healthy tissue), and the surface to volume (S/V) ratio of the vasculature. The clinical dose of T-DM1-AF680 (3.6 mg/kg) shows a heterogeneous, perivascular distribution in NCI-N87 tumor xenografts (Fig. 1A, Supplementary Fig. S1) consistent with high affinity antibodies that target highly expressed receptors and are dosed at subsaturating levels(9,23,24). Co-administration of unlabeled trastuzumab at 10.8 or 28.8 mg/kg (3:1 or 8:1, respectively) dramatically increases tumor penetration of a constant T-DM1 dose (Fig. 1B, 1C), allowing more cells to receive the cytotoxic payload. However, adding trastuzumab lowers the effective DAR, increasing the number of targeted cells while reducing the average number of payload molecules delivered per cell. Immunofluorescence staining with antihuman IgG Fc-488 shows antibody distribution is more homogeneous with increasing doses of trastuzumab (Fig. 1A-C). Similarly, increasing the dose of T-DM1 alone also improved tumor penetration (Supplementary Fig. S2), but this exceeds the maximum tolerated dose in humans. HER2 was stained ex vivo with trastuzumab-AF750 to ensure the heterogeneous distribution is not from lack of available antigen (Supplementary Fig. S3). T-DM1 binding affinity was unchanged by fluorophore conjugation (Supplementary Fig. S4).

Figure 1. Improving T-DM1 tumor distribution through co-administration of trastuzumab.

Administration of T-DM1 at 3.6 mg/kg (single agent) results in a heterogeneous, perivascular distribution due to rapid binding relative to transport in the tissue (A). The tumor penetration of a constant dose of T-DM1 is improved when co-administered with a subsaturating (B) or saturating (C) dose of trastuzumab. Trastuzumab competes for binding sites, increasing T-DM1 penetration. The middle column shows distribution of AF680-labeled T-DM1 (green) at 3.6 mg/kg with unlabeled trastuzumab at 0:1, 3:1, and 8:1 trastuzumab:T-DM1 ratios (0, 10.8, and 28.8 mg/kg, respectively). Immunofluorescence staining with CD31-AF488 (red) shows the tumor vasculature. The right column shows immunofluorescence staining with antihuman IgG Fc-AF555 (gray). The window leveling between images is different since the intensity of the T-DM1 decreases with an increasing ratio while the anti-Fc staining labels both trastuzumab and T-DM1 thereby maintaining a constant intensity while the penetration increases (see Supplementary Material). Scale bar is 200μm.

T-DM1 is Effective In Vitro When Occupying Only a Fraction of HER2 Receptors Even Under Saturating Antibody Conditions

The increasing doses of trastuzumab improve T-DM1 penetration into tumor tissue but lower the DM1 payload delivery by competing with HER2 receptors. To determine if the lower payload delivery was still sufficient to kill cells in vitro, we measured efficacy with T-DM1 alone or a saturating combination of total antibody (T-DM1 + trastuzumab) while varying the T-DM1 to trastuzumab ratio. Toxicity assays with trastuzumab alone showed only slight growth inhibition at the highest concentrations (Supplementary Fig. S5), consistent with literature reports that the NCI-N87 cell line is sensitive to T-DM1, while trastuzumab has a slight growth inhibitory effect in vitro(25). We measured the toxicity of T-DM1 in vitro with the NCI-N87 and HCC1954 cell lines and we found IC50 values of 82 and 33 pM for NCI-N87 and HCC1954 (Fig. 2A, 2B), respectively, consistent with other reports(26,27). Next, we performed toxicity assays with fluorescently tagged T-DM1-AF680 and examined cellular uptake of T-DM1 by flow cytometry. The IC50 for each cell line was much less than the concentration needed for half of the normalized uptake, indicating that complete surface receptor saturation is not needed for T-DM1 cytotoxicity in vitro. Additionally, we found that fluorophore conjugation had no impact on T-DM1 cytotoxicity in vitro (Supplementary Fig. S6). The maximum cellular uptake occurred at concentrations below the K_d (1.8 nM(17)), indicating treatment likely impacts uptake over this time scale. To mimic the effect of co-administering trastuzumab in vivo, we performed toxicity assays where the T-DM1 concentration was varied, but the total antibody concentration was kept constant at 10 nM (saturating) to see how competition for receptors would affect toxicity (Fig. 2C, 2D). T-DM1 still showed toxicity below the antibody K_d against both cells lines with the addition of trastuzumab, although the IC50 was higher likely due to competition from the trastuzumab. Using a simple competitive inhibition binding model, we found that T-DM1 cytotoxicity was similar when adjusting for the fraction of receptors bound by T-DM1 (Supplementary Fig. S7). Additionally, a similar number of molecules of T-DM1 were required to achieve 50% cell death in both cases (Supplementary Fig. S8). Because trastuzumab increases the penetration (while lowering the single-cell delivery) of T-DM1 and T-DM1 remains toxic at sub-saturating conditions, we tested the efficacy of this combination in a trastuzumab-insensitive mouse xenograft model.

Figure 2. In Vitro Cytotoxicity.

(A, B) in vitro cytotoxicity of T-DM1 against NCI-N87 (A) and HCC1954 (B) cell lines. Assays were performed in triplicate and average IC50's were 82 ± 10 pM and 33 ± 20 pM for NCI-N87 and HCC1954 cell lines, respectively. The number of DM1 molecules per cell (gray) was estimated by measuring T-DM1-680 uptake with flow cytometry and quantitative beads. (C, D) in vitro cytotoxicity varying the T-DM1 concentration, while keeping total antibody concentration (T-DM1 + trastuzumab) constant at 10 nM. Cells were incubated with T-DM1 and/or trastuzumab in media for 6 days and media was replaced daily for all in vitro cytotoxicity assays. Data plotted as mean ± standard deviation.

Addition of Trastuzumab (an In Vitro Antagonist) to T-DM1 Therapy Improves In Vivo Efficacy in a Trastuzumab Resistant Xenograft Model

To examine the impact of ADC tumor distribution on efficacy, we administered a fixed dose (3.6 mg/kg) of T-DM1 with varying ratios of trastuzumab in an NCI-N87 tumor xenograft mouse model. Non-fluorescently labeled (clinical) T-DM1 and trastuzumab were used for in vivo efficacy studies. We selected the NCI-N87 cell line because a) it was less sensitive than the HCC1954 cell line to T-DM1 in vitro (Fig. 2) and in vivo(27) providing more room to detect improvements in efficacy, and b) other groups have shown that at moderate doses, like the ones used in this study, trastuzumab treatment did not significantly alter tumor growth from control(25,28). Trastuzumab treatment results in modest (but statistically significant) growth inhibition at higher dosages (>60 mg/kg total dose) but does not result in tumor reduction even at highest doses of 280 mg/kg total dose over several weeks (28-30). Additionally, we chose to use large established tumorsthat were 250mm³ or greater. Others have shown antibody-dependent cell-mediated cytotoxicity (ADCC) is reduced in larger established tumors, albeit with a different HER2 expressing cell line(31). Expectedly, the clinically approved T-DM1 therapy alone showed significant improvement over control (Fig.3). Although others have shown complete tumor regression using T-DM1(26), the larger tumors and single administration prevented consistent cures. Addition of trastuzumab to T-DM1 resulted in slower tumor growth than T-DM1 alone for all dosage levels (Fig. 3A), with 3:1 and 8:1 (T:T-DM1) ratios having a statistically significant effect. The 3:1 and 8:1 dosing levels had several more partial responses and exhibited a statistically significant increase in survival with increasing trastuzumab doses (Fig. 3B). Kaplan-Meier survival plots with 95% confidence intervals calculated by PRISM and individual tumor growth curves are shown in the Supplementary Material (Supplementary Fig.S9 and Fig.S10, respectively). Animals receiving trastuzumab only were similar to saline control demonstrating that trastuzumab has no direct effect on efficacy at these doses. Additionally, mice receiving treatment were weighed over the course of the study and showed comparable tolerability (Supplementary Fig. S11), consistent with the payload and not the antibody dose driving toxicity. Since trastuzumab is tolerated at much higher doses than T-DM1(32) and the same T-DM1 dose was given, all treatments were consistently well tolerated.

Figure 3. Co-administration of trastuzumab with T-DM1 results in a significant reduction in tumor growth compared to T-DM1 alone.

(A) Tumor growth curves for mice bearing NCI-N87 tumor xenografts following treatment with a single administration of saline, single-agent trastuzumab (10.8 mg/kg), single-agent T-DM1 (3.6 mg/kg), or co-administration of trastuzumab and T-DM1 at 1:1, 3:1, and 8:1 dosage levels (T-DM1 constant at 3.6 mg/kg, trastuzumab varied at 3.6, 10.8, and 28.8 mg/kg). Non-fluorescently labeled T-DM1 and trastuzumab were used for tumor growth experiments. Data plotted as mean ± standard error. The number of partial responses, complete responses, and durable complete responses is tabulated. (B) Kaplan-Meier survival curves of time to progression to 1000 mm³. Survival curves were analyzed by log-rank test (significance level of p ≤ 0.05). All treatments except single-agent trastuzumab resulted in statistically significant improvements in survival (p = 0.0442, 0.0021, 0.0006, 0.001 for 0:1, 1:1, 3:1, 8:1, respectively). The 3:1 and 8:1 treatments significantly improved survival over the single-agent T-DM1 (p = 0.0486, 0.0484 for 3:1 and 8:1, respectively). N = 10 for each treatment. T, trastuzumab; PR, partial response; CR, complete response; DCR, durable complete response.

Single-Cell T-DM1 Uptake and Metabolism In Vivo Highlights a Fraction of Cells with Higher Delivery than Needed for Cell Killing While Other Cells Receive Negligible T-DM1

The fluorescence histology images (Fig. 1) qualitatively show better penetration, but they cannot be used to quantify payload delivery. We utilized a novel NIR fluorescence ratio technique to determine the absolute uptake and payload delivery per cell in vivo. In particular, we applied the technique to single agent T-DM1 therapy to confirm that targeted tumor cells were receiving more payload than necessary to achieve cell death and the lack of tumor penetration was limiting efficacy in vivo. Fig.4A shows a graphic depiction of the technique. Two NIR fluorescent dyes, DDAO and IRDye, were chosen because of their widely differing residualization rates(16). The non-residualizing dye rapidly leaks out of the cell upon degradation, thereby approximating intact protein, while the residualizing dye is trapped in the cell, approximating cumulative ADC uptake. This approach is analogous to radiolabeling with ¹²⁵I as the non-residualizing probe and ¹¹¹In as the residualizing probe(33).

Figure 4. In vitro T-DM1 metabolism and pharmacodynamics.

(A) Agraphic depiction of the NIR fluorescence ratio technique. The dually labeled antibody binds the target, is internalized, anddegraded. The non-residualizing DDAO (red star) leaks out of the cell, while the residualizing IRDye 800CW (green star) is trapped within in the cell. (B) Representative confocal images of dually labeled T-DM1. DDAO (red) shows cell surface labeling with a loss of signal over time. IRDye (green) shows initial cell surface labeling followed by the formation of punctate spots as it is trapped in the lysosomes. Scale bar is 10 μm. (C) T-DM1 metabolism. Fraction of intact ADC following pulse of dually labeled T-DM1 for HCC1954 (black) and NCI-N87 (gray) cells. Data plotted as mean ± standard deviation. (D) Timing of T-DM1 pharmacodynamics. The fraction of viable cells over time for HCC1954 (black) and NCI-N87 (gray) cells when treated with a constant 5 nM T-DM1. Non-fluorescently labeled (clinical) T-DM1 was used for this assay. Data plotted as mean ± standard deviation. (E) Representative flow cytometry plots of dually labeled T-DM1 gated on cells. Intact dually labeled T-DM1 appears in DDAO(+)/IRDye(+) quadrant. Over time as ADC is degraded there is a gradual shift towards DDAO(-)/IRDye(+).

To visualize the change in fluorescence ratio and validate the method in vitro, cells were imaged using a confocal microscope after pulsing with labeled ADC for 30 minutes at different times over 48 hours. Fig.4B shows separate and merged channels for DDAO (red) and IRDye (green) for dually labeled T-DM1. Initially, DDAO and IRDye are only seen on the surface. As ADC is internalized and degraded, however, the DDAO signal gradually decreases, while IRDye forms punctate spots in lysosomes. Fig. 4C shows the fraction intact of T-DM1 for HCC1954 (black) and NCI-N87 (grey) cell lines over time. Although both express the same antigen, the HCC1954 cell line degraded T-DM1 slightly faster than the NCI-N87 cell line. Using the residualizing IRDye signal, we quantified the single-cell uptake of T-DM1 and estimated the release of DM1 payload (Supplementary Tables S1 and S2). To quantify the kinetics of cell death following cell targeting, we measured the viability of both cell lines over six days (Fig. 4D). Consistent with the faster T-DM1 degradation of HCC1954 cells and published link between intracellular payload concentration and toxicity(13), their viability decreased more quickly than NCI-N87 cells. We used flow cytometry to quantitatively measure the cellular signal for DDAO and IRDye over time at the single-cell level (Fig. 4E). Initially the ADC is intact and the cell population is positive for DDAO and IRDye. Over time there is a gradual shift to DDAO(-)/IRDye(+), indicating that the ADC is degraded and DDAO has washed out.

After demonstrating the NIR fluorescence ratio technique in vitro, we applied the technique in vivo. Mice were injected via tail-vein with100μg (∼4 mg/kg) of dually labeled T-DM1 and euthanized at 24, 48, and 72 hours. Then tumors were resected and digested into a single cell suspension and analyzed by flow cytometry (Fig. 5). Representative flow cytometry plots of single cell tumor digests are shown in Fig. 5A. 24 hours post injection, targeted cells (IRDye (+)) show mostly intact protein (DDAO(+)/IRDye(+)) (Fig.5A, 5B) because there is a constant supply of intact ADC from the blood. However, by 48 and 72 hours there is a shift from DDAO(+)/IRDye(+) to DDAO(-)/IRDye(+) for the targeted cells, indicating that much of the surface bound ADC was internalized and degraded. Additionally, consistent with the tumor histology (Fig. 1), only a small fraction of cells in the tumor (around 10% according to flow cytometry and consistent with histology and a ∼10-fold higher dose for saturation) are targeted with ADC, despite administering the clinical dose. Because IRDye is a residualizing dye and approximates the cumulative uptake of ADC(16), we used quantitative beads to convert the IRDye signal from targeted cells into the number of ADCs per cell. Combining the number of ADCs targeted per cell with the percent intact, we estimated the number of DM1 payload molecules released in targeted cells (Fig. 5C). Consequently, the amount of DM1 released increased dramatically between 24 and 48 hours and started to plateau by 72 hours.

Figure 5. In vivo T-DM1 metabolism in tumor.

(A) Representative flow cytometry plots of single cell suspension from NCI-N87 tumors at 24, 48, and 72 hours post-injection of 3.6 mg/kg of dually labeled T-DM1. Intact dually labeled T-DM1 appears in DDAO(+)/IRDye(+) quadrant. Over time as ADC is degraded there is a gradual shift towards DDAO(-)/IRDye(+). Cells targeted with T-DM1 (IRDye+) were used to calculate percent intact as described in Materials and Methods. (B) T-DM1 degradation in tumor cells. The fraction of intact ADC (for the fraction of cells that are targeted by T-DM1 at this dose). Data plotted as mean ± standard deviation. (C) Molecules of DM1 payload released per target cell in vivo for the targeted cells calculated using the total cell uptake and fraction intact. Data plotted as mean ± standard deviation. (D) T-DM1 biodistribution. T-DM1 shows maximum uptake 24 hours post injection

In addition to single cell tumor metabolism from flow cytometry, the NIR fluorescence ratio technique was used to measure the biodistribution of ADC in the same animals (Fig. 5D) to verify the normal systemic distribution of the antibodies. The reduced autofluorescence in the near infrared window makes IRDye an appropriate dye for biodistribution studies(17,20,21). We have previously shown that at a dye to protein ratio of 0.3 or less there is no impact on protein pharmacokinetics over the first 3-4 days(17). The plasma clearance of dually labeled trastuzumab was similar to trastuzumab-IRDye, indicating the addition of the DDAO fluorophore did not impact clearance (Supplementary Fig. S12). The biodistribution of dually labeled T-DM1 shows primarily liver and tumor uptake and is similar to radiolabeling studies of trastuzumab, albeit with lower tumor uptake in this model(34). Consistent with the single cell flow cytometry data, the maximum tumor uptake was reached 24 hours after injection and there was a gradual decrease at 48 and 72 hours (Fig. 5D).

Untargeted Tumor Cells Can Sustain Tumor Growth Through Newly Functional Tumor Vessels

To better understand the relationship between the heterogeneous T-DM1 distribution and efficacy, we imaged the distribution of T-DM1 and trastuzumab at maximum uptake and during treatment. The NIR fluorescence ratio technique showed that the maximum uptake was reached 24 hours post injection and payload release appeared to plateau around 3 days. From the tumor growth curves it appeared that the maximum shrinkage was occurring several days after payload release, around 5 days after initial injection for T-DM1. Once tumor cells are killed, they are no longer able to internalize and degrade the drug, potentially allowing ADCs to penetrate deeper into the tissue. However, the tumor distribution of T-DM1-680 and trastuzumab-680 (3.6 mg/kg) at one and five days post-injection remained heterogeneous and perivascular with a significant fraction of the tumor untargeted (Fig. 6A).15 minutes prior to sacrifice we injected Hoechst 33342 to stain functional tumor vasculature and then stained histology slices with antimouse CD31 ex vivo to show all (functional and nonfunctional) vasculature. Using an automated image analysis algorithm, we calculated the absolute vessel surface area to tumor volume ratio (S/V) along with the fraction of these vessels that had Hoechst and/or T-DM1 signal around them (Fig. 6B) (See Supplementary Material).24 hours after injection, T-DM1 distribution was localized to functional vessels (Fig. 6A, arrows). By five days after injection there were several regions of the tumor that had functional vessels but no perivascular ADC (Fig. 6A, arrowheads), and the image analysis indicated a significant increase (Student's t test, p < 0.0001) in the fraction of functional vessels (Hoechst and CD31) lacking ADC (Fig. 6B), consistent with angiogenesis and/or opening of collapsed vessels. Collapsed vessels (CD31 vessels stained with ADC but not Hoechst, indicating they are no longer functional) were also present. These phenomena were present in both T-DM1 and trastuzumab treated tumors, indicating that they are not necessarily a result of T-DM1 efficacy. However, the formation of newly functional vessels in regions untargeted by T-DM1 could play a role in rescuing the tumor, further supporting the importance of ADC distribution on efficacy.

Figure 6. Immunofluorescence histology after treatment.

(A) Tumor distribution of 3.6 mg/kg of T-DM1-680 (top, green) or trastuzumab-680 (bottom, green) at 24 hours (left) or 5 days (right) after tail-vein injection. 15 minutes prior to sacrifice 15 mg/kg Hoechst 33342 (blue) was administered via tail-vein to label functional vasculature. CD31-555 (red) was stained ex vivo to label all vasculature (functional and nonfunctional). Arrows highlight examples of functional vessels that contain perivascular T-DM1 labeling, which dominate at early times after treatment. By 5 days, a significant fraction of functional vessels (CD31 and Hoechst labeled) lack perivascular T-DM1 (arrowheads). Since some vessels still contain perivascular T-DM1 at this time point, presumably these vessels are newly formed (or became functional) once a significant fraction of T-DM1 cleared. The window leveling is different for each image (qualitative distribution only). Scale bar is 200 μm. (B) Image analysis of histological samples. The fraction of functional vessels (Hoechst and CD31) containing perivascular T-DM1 at 5 days after treatment was significantly less than at 1 day (p < 0.0001), indicating angiogenesis and/or opening of collapsed vessels.

Discussion

The efficacy of ADCs is determined by a complex interplay between tumor uptake, distribution, cellular targeting, internalization, antibody degradation, and release of the small molecule payload. Here, we show that improving tumor penetration by co-administering trastuzumab enhances the efficacy of T-DM1 in a trastuzumab-insensitive mouse xenograft model. These results have significant implications for the development of ADCs. Although substantial efforts have been made in optimizing the drug itself (high affinity antibodies, stable linkers, and highly potent small molecules), these data demonstrate the intratumoral distribution (independent of the payload dose) plays a major role in determining efficacy. Given that the ADC dose is often limited by the small molecule payload dose and not the amount of antibody, matching the potency of the ADC with delivery (rather than trying to maximize potency) may provide a way to improve the efficacy of ADCs while maintaining tolerability(5).

The rapid binding of antibodies relative to their tissue penetration results in receptor saturation of perivascular cells (8,10). To penetrate deeper into the tissue, additional antibody must enter the tissue, but the toxicity of ADCs generally prevents the administration of higher ADC doses. Although there are other strategies to improve tumor penetration, such as decreasing protein size (i.e. F(ab) or F(ab′)₂ fragments) or lowering affinity, increasing the antibody dose has the potential to improve multiple mechanisms of action. The co-administration of unconjugated antibody improves penetration (Fig. 1) and is generally well tolerated relative to the cytotoxic payload. For example, trastuzumab is well tolerated even at high doses, such as an intensive loading schedule totaling 18 mg/kg given over 15 days(32). This does not increase (or decrease) the amount of ADC uptake in the tumor; it only changes the distribution as long as the dose remains sub-saturating(15). (In the clinic, saturating doses can require multiple grams of antibody for highly expressed and/or rapidly internalized antigens(35), possibly higher with heavy tumor burdens due to target mediated binding(36).) More uniformly delivering the ADC could potentially lower efficacy on the perivascular cells that typically receive a high concentration of ADC. However, ADCs tend to have low IC50's, often over an order of magnitude below the K_d (Fig. 2)(4,37,38), indicating sub-saturating concentrations can result in cell death. Therefore, the heterogeneous delivery of ADC in the tumor can result in “overkill” of perivascular cells, where they receive more therapeutic than what is needed, while other cells receive none. When co-administering trastuzumab with T-DM1, the perivascular cells receive a smaller payload dose; however, more cells overall receive therapeutic levels of payload. At super-saturating doses (e.g. 60-120 mg/kg in this high expression model), it is anticipated efficacy would decrease since there would be no increase in penetration for a saturated tumor and payload uptake would decrease. Likewise, administering a saturating antibody dose a day before the ADC (e.g.(39)) can decrease efficacy.

These data show that the addition of an antagonist (trastuzumab, which antagonizes T-DM1 at high concentrations in vitro, Fig. 2) with no single-agent efficacy (Fig. 3) can improve in vivo efficacy and survival. The converse of this concept must also be considered. Newer and more potent payloads may be required for targets with low to moderate receptor expression and/or slow internalization rates(40). However, an ADC with higher in vitro potency may actually be less efficacious in vivo for some targets that are highly expressed. The increased potency (whether from higher DAR or a more toxic payload) could lower the maximum tolerated dose. This lower dose reduces the number of cells that can be targeted, thereby lowering the overall efficacy. When developing new ADCs, these results indicate that neither the maximum cellular potency nor the maximum antibody dose is optimal. Rather, this work emphasizes the need to match the single-cell potency with single-cell delivery to maximize efficacy.

In previous work(15), we identified studies that used a constant small molecule dose with different DAR/antibody doses and demonstrated the higher antibody dose (and correspondingly lower DAR) exhibited better efficacy. These studies included multiple targets, antibodies, linkers, and payloads (with and without bystander effects, e.g. (13,41), see Supplemental Material) indicating that the impact of tissue penetration is important across all ADCs studied to date. Since this publication(15) two other studies reported the same result –keeping the small molecule dose the same and increasing the antibody dose improved efficacy(37,42). However, a potentially confounding factor in these studies was that higher DAR ADCs tend to have faster clearance (DAR-dependent clearance), although the difference in payload AUC was less than 25%for these cases (42,43). The current work avoids potential DAR-dependent clearance by only using T-DM1. Another possible explanation could be that adding trastuzumab resulted in a dose dependent slower clearance of T-DM1(44). However, the plasma clearance rates are similar with or without trastuzumab(15).

Using the NIR fluorescence ratio technique, the fraction of tumor cells targeted by ADC, the number of ADC molecules delivered per cell, and the fraction of intact ADC versus degraded were measured (Fig. 5). The slowly clearing T-DM1 showed mostly intact protein (∼85%) at 24 hours post-injection due to continuous delivery from the blood during this period with the highest plasma concentrations. Conversely, over half of the ADC was degraded by 24 hours in vitro when the ADC was pulsed (Fig. 4). After 3 days, once the tumor had surpassed maximal uptake and plasma concentrations decreased, the majority of the ADC in the tumor was degraded (Fig. 5). Consistent with the histology images, only ∼10% of the tumor cells are targeted by a 3.6 mg/kg dose of T-DM1 at maximum uptake. This is also in agreement with the 9-fold higher antibody dose required for saturation of the tumor (Fig. 1). The number of DM1 molecules delivered per targeted cell is estimated at 1.7 +/- 0.3 million. Given the 2-3 day residualization half-life of IRDye(16), this measurement at 3 days is likely lower than the actual payload delivery but significantly higher than the uptake at the IC50 in vitro (Fig. 2). These results are consistent with a rapid targeting of ADC (∼1 day) in perivascular cells at a higher concentration than needed for cell death. A large fraction of cells within the tumor (∼90%) are not exposed to T-DM1 even after 48-72 hrs when maximum payload delivery is achieved within the tumor. Therefore, a significant fraction of cells receives more drug than needed for cell killing, while a large fraction of cells completely escape therapy, lowering the overall tumor efficacy despite efficient targeting (15% ID/g in these300-400 mm³ tumors).

The fluorescently tagged ADC was used to image distribution during tumor response (the nadir in the tumor growth curves occurs around 5-7 days, Fig. 3). The tumor distribution of both trastuzumab and T-DM1 5 days after injection shows both functional (CD31 vessels labeled with intravenous Hoechst) and nonfunctional vessels with signal. Additionally, there are functional vessels that do not have detectable ADC signal, indicating that after maximum uptake in the tumor is reached there may be new vessels that form, which won't receive significant payload until a high plasma concentration is achieved with the next dose (every 3 weeks in the case of T-DM1). The irregular, dynamic vasculature has important implications for ADC treatment when a significant fraction of cells are untargeted by ADC (Fig. 6). New vessels deliver both oxygen/nutrients for survival and drugs for cell killing. However, the ADC requires approximately 1 day for uptake, 1 day for complete metabolism, and several days for cell killing, while oxygen and nutrients can rescue the cells more quickly. Therefore, the dynamic vasculature within the tumor has the potential to repeatedly rescue untreated regions of the tumor even with continuous ADC in the plasma, which may stymie attempts to kill cells layer by layer with successive treatments.

Although our results demonstrate that ADC tumor distribution has a significant role in efficacy, there remain several challenges to clinical implementation. First, matching the single-cell potency to single-cell delivery is challenging given that many targets are measured using immunohistochemistry rather than a more quantitative method capable of reporting receptors/cell. Second, selecting the optimum potency could be challenging given intra- and inter-patient variability where one could perfectly ‘match’ potency and delivery of an ADC to a primary tumor but not a metastasis with much higher or lower expression. It is unknown whether it is better to err on the side of higher potency (targeting fewer cells with a higher dose than necessary for cell killing which could help avoid mechanisms of drug resistance(45)) or higher delivery (increased tumor penetration at a sub-toxic dose). In this model system, higher penetration appears to be more beneficial. The use of higher antibody doses to increase ADC penetration has additional potential benefits, such as maximizing other mechanisms of action, including receptor signaling blockade and/or immune cell interactions(46). Sacituzumab govite can, an ADC that has received Breakthrough Therapy designation from the FDA, takes this approach using a lower potency payload with much higher antibody doses (8-10 mg/kg, Clinical Trial: NCT01631552)(47). Additionally, our modeling work shows that healthy tissue with low target expression would have less uptake when ADC and antibody are administered together compared to ADC alone(15). Finally, imaging may play a useful role in identifying optimal treatment regimens(48). The ZEPHIR trial looks to combine pretreatment and early metabolic response molecular imaging to select patients that respond best to T-DM1 therapy(49). Combining molecular imaging and pharmacokinetic models(17) to determine the optimum antibody/ADC dosage could provide an individualized treatment with potentially better outcome. Nonetheless, some patients that are HER2 positive may not have trastuzumab uptake(49), while other patients that have HER2 negative primary cancers may have metastases that are HER positive(50), making individualized treatment challenging.

In conclusion, we have shown that improving tumor penetration of a constant dose of T-DM1 by co-administration of trastuzumab (in a trastuzumab-insensitive xenograft model) results in significantly better efficacy than T-DM1 alone. Maximizing tumor penetration of ADCs in addition to optimizing the antibody, linker, and payload during development may help improve the efficacy of future ADCs in the clinic.