Mechanistic Modeling of Antibody–Drug Conjugate Internalization at the Cellular Level Reveals Inefficient Processing Steps

This article is featured in Highlights of This Issue, p. 1145

The EGFR (also known as ErbB1) is part of a family of ErbB receptor tyrosine kinases, which are often overexpressed in cancers and exhibit dysfunctional signaling previously associated with poor patient outcomes (1–3). EGFR can bind a diverse number of ligands, with eight different ligands s activate the receptor (4, 5), triggering signaling and endocytosis of the receptor–ligand complex (6). Once internalized, ligand-bound EGFR can either be recycled to the cell surface or sent for degradation down the lysosomal pathway (7).

Targeting antibodies have been successful as anti-cancer agents against tumor cells with specific antigen expression. To combat aberrant EGFR expression and signaling, two anti-EGFR antibodies have been approved for the treatment of various cancer indications (8). However, because many antibodies lack the requisite efficacy to render positive patient outcomes, research into arming antibodies with cytotoxic “warheads” has been a popular strategy for drug development, as reflected by the >50 antibody–drug conjugate (ADC) clinical trials currently in progress (9). In theory, by targeting antigens expressed only on tumor cells (or at significantly higher levels than normal tissue), more efficacious and less toxic therapies can be devised through enhanced delivery of drugs to target cells and concurrent decreases in the levels of potent drugs reaching normal tissue. On a cellular level, the process of drug delivery often begins with the target antigen on the cell surface being bound by the ADC and internalized through endocytosis (10). Antibodies directed against EGFR can increase the rate of internalization through specific epitope binding (11), suggesting that EGFR may be a good ADC target. Also, similar to EGFR trafficking, the internalized antigen–ADC complex can be either recycled to the cell surface or moved along the lysosomal pathway for enzymatic degradation and intracellular drug release. Because cellular trafficking of internalized EGFR is influenced by the type of ligand and where it binds the receptor (e.g., EGF induces degradation, TGFα induces recycling; ref. 12), if an EGFR-targeting antibody can mediate efficient trafficking of the receptor–antibody complex to the lysosome, conjugated drug could be effectively released inside tumor cells.

ADC processing has been studied in the past by a variety of experimental methods. Fluorescence-based instruments such as microscopy and high-content readers are often used to visualize internalization; these tools are also able to localize antibodies to specific intracellular compartments (13–15). Newer technology, imaging flow cytometry, has recently been used for assessing antibody internalization differences (16). Imaging flow cytometry obtains high-resolution, phenotypic images typically reserved for microscopy and quantitative population data of flow cytometry for a powerful analytical combination yielding cellular images with statistical-backing (17). For other quantitative antibody measurements, internalization assessments with radio- or fluorescently labeled antibodies are preferred methods. Furthermore, liquid chromatography-tandem mass spectrometry (LC-MS/MS) has great utility for the quantitation of released intracellular drug concentrations following ADC treatment (18). Here, imaging flow cytometry and mass spectrometry technologies were used to quantitatively assess cellular processing of antibodies and ADCs.

Understanding of the processing of an ADC on the cellular level is important in the context of drug design. Once the various parts are well-defined, improvements can be made to address specific limiting parameters to produce better binding, internalization, trafficking, or drug release depending on the need. In this study, the processing of an anti-EGFR ADC was detailed, with kinetic measurements of internalization, trafficking, and drug accumulation input into a mechanistic model that provided insights into the cellular underpinnings of an EGFR-targeting antibody and learnings that can be applied toward future ADC drug development efforts.

Alexa Fluor 488 tetrafluorophenyl ester (Invitrogen) was used to fluorescently label antibodies of interest on free primary amines. The antibodies to be labeled were in PBS pH 7.4. A 1/10 volume of 1 mol/L sodium bicarbonate was added to the antibody solution to raise the pH of the solution to 8. A 100 μg vial of AlexaFluor 488 was resuspened in 100 μL of DMSO. Approximately 20 μL of the 488 mixture was added for each 1 mg of antibody to be labeled. The samples were incubated at room temperature (RT) in the dark with shaking at 700 RPM for 1 to 1.5 hours. Following incubation, 1 mol/L Tris-HCl pH 7.4 was added to the mixture to quench the reaction before transferring to a 100 kDa molecular weight cutoff filter (EMD Millipore). The filter was spun at 14,000 × g for 10 minutes and then washed with 500 μL PBS pH 7.4 twice. The labeled antibodies were eluted by inverting the filter and centrifuging at 200 × g for 2 minutes. Concentration and degree of labeling were measured by a NanoDrop 2000 (Thermo Scientific). The average labeling was between 1 and 2 fluorophores per antibody. Labeling with pH sensitive dye (Promega) was performed in a similar fashion.

Ab033, a chimeric human-mouse monoclonal IgG antibody against EGFR, was used for all experiments. Ab033 was expressed by transient transfection in HEK 293-6E cells (NRC Canada) using a Wave Bioreactor (GE Healthcare) at 25 L scale. The cells were grown in Freestyle 293 expression media (Life Technologies) supplemented with 0.05% pluronic-F68 and 5 mL/L penicillin–streptomycin to a density of 1.2 × 10⁶ cells per mL. Cells were transfected with plasmid DNA encoding the AB033 HC and LC (in 2:3 ratio) using 0.5 mg DNA/l of expression mixed with 1 mg/mL PEI (Polyethylenimine) in a 1:4 ratio of DNA:PEI for 10 minutes in Freestyle media. Cells were grown at 37°C with 8% CO₂ and 26 rpm at an angle of 7°. Clarified media were harvested 12 days post-transfection by constant flow centrifugation at 6,000 × g and filtered through a 30-inch 0.6 μm ULTA Cap GF filter (GE) and a 20-inch 0.2 μm ULTA Cap CG filter (GE). AB033 was captured from the supernatant by loading at 100 mL/min onto a 300 mL MabSelectSure Protein A column (GE), which was equilibrated and washed with PBS pH 7.4. The column was eluted isocratically with 50 mmol/L Glycine, 50 mmol/L NaCl pH 3.5. The eluted pool was neutralized with 0.5 mol/L dibasic sodium phosphate to pH 7.0 and passed through a 7 mL Sartobind Q column (Sartorius) in flow through mode. The purity was determined to be >99% monomer by analytical SEC on an ACQUITY Protein BEH SEC column (Waters 186005225) using a Waters UPLC. The protein was then diafiltered into 1X PBS pH 7.4 using a Spectrum Labs KrosFlo system with a 50 kDa 1,100 cm² GE Kvick Lab slice (GE) at 450 mL/minute. Final protein was concentrated to 10.7 mg/mL using an Amicon 50 kDa concentrator and sterile filtered through 0.22 μm Stericup (Millipore). The binding kinetics of Ab033 to soluble wild-type recombinant EGFR was determined by surface plasmon resonance as previously described (19).

To form a drug conjugate, Ab033 was incubated in DPBS with two equivalents of tris(2-carboxyethyl)phosphine hydrochloride (TCEP) on ice for 5 hours and then conjugated with maleimidocaproyl-valine-citrulline-p-aminobenzoyloxycarbonyl-monomethyl auristatin E (mc-vc-PABC-MMAE) at RT for 30 minutes. The resulting conjugation mixture was subjected to hydrophobic interaction chromatography (HIC) purification to yield Ab033 with two mc-vc-PABC-MMAE conjugated (Ab033-E2). Reverse-phase liquid chromatography (RPLC) analysis of DTT-reduced Ab033-E2 showed the product contained mc-vc-PABC-MMAE on approximately 50% of the light chains (LC) and 50% of the heavy chains (HC).

The purified Ab033-E2 fraction was buffer exchanged with DPBS and incubated over night with 1.1 equivalents of TCEP on ice. The linker-drug mc-MMAF was then added for conjugation to Ab033-E2. The resulting mixture was subjected to HIC purification to obtain Ab033-E2 containing two mc-MMAF linker–drug payloads (Ab033-E2/F2). RPLC analysis of DTT-reduced Ab033-E2/F2 showed the majority of ADC product contained equal amounts of the toxins and 50% of the LC and HC conjugated with one mc-vc-PABC-MMAE each and 50% of the LC and HC conjugated with one mc-MMAF each.

The EGFR-expressing human epidermoid carcinoma derived A431 cells and human lung adenocarcinoma derived H441 epithelial cells were used for in vitro experiments. A431 and H441 cells were obtained from the ATCC and were propagated in growth medium consisting of RPMI-1640 (Sigma) supplemented with 10% FBS. The cells were incubated at 37°C with 5% CO₂ and 90% relative humidity. Both cell types were passaged with 0.05% trypsin-EDTA (Invitrogen) and used for experimentation within 10 passages from thawing. No cell authentication or Mycoplasma testing was performed, although EGFR expression levels were confirmed.

A431 and H441 cells were trypsinized from cell culture plates using trypsin-EDTA at RT for 2 minutes. Growth media were added to quench trypsin activity and samples were centrifuged at 250 × g. The trypsinization step was confirmed to not have a negative impact on EGFR levels. The supernatant was removed, and the cells were resuspended in fresh cold media on ice for 5 minutes before 15 μg/mL labeled Ab033 was added. Following a 45 minutes incubation on ice, cells were washed with PBS pH 7.4 containing 0.7% BSA to remove excess, unbound Ab033. The samples were then moved to a water bath set to 37°C. Upon the conclusion of each time point, cells were moved back to ice. Anti-human IgG secondary antibody conjugated with Alexa Fluor 647 was added at 1:300 for 20 minutes. Cells were washed with PBS and analyzed by imaging flow cytometry. Using minute intervals from zero to five minutes, the internalization rate was determined as previously described (20). Briefly, the percentage of internalized antibody was plotted against the time integral of the surface fluorescence, which was calculated using the trapezoidal rule. The integral serves to normalize the surface fluorescence across the data points to account for decreases in extracellular levels over time.

A titration of anti-Alexa Fluor 488 antibody concentration was performed to determine the optimal concentration for maximum fluorescence quenching of Alexa Fluor 488. First, A431 cells were incubated on ice with saturating levels of Alexa Fluor 488 labeled Ab033 for 15 minutes and then washed with PBS. Cells were then incubated with varied concentrations (5 μg/mL to 200 μg/mL) of anti-Alexa Fluor 488 antibody for 30 minutes on ice before proceeding with analysis of surface Alexa Fluor 488 signal.

A recycling assay was performed similarly to previously described methods (20, 21). Briefly, A431 and H441 cells were pulsed with 15 μg/mL Alexa Fluor 488-labeled Ab033 for 10, 20, or 60 minutes at 37°C. Cells were then transferred to ice and washed with cold PBS to remove excess Ab033. Anti-Alexa Fluor 488 antibody was added to the cells at 50 μg/mL for at least 20 minutes to quench cell surface signal. The cells were returned to 37°C while still in the presence of quenching antibody and chased for 0, 5, 10, 15, or 30 minutes before analysis by imaging flow cytometry.

Cells were plated on 6-well dishes at a density where the cells were nearly confluent after 2 days of growth. The cells were allowed to adhere overnight in fresh growth media and then treated with 15 μg/mL Ab033 labeled with AlexaFluor488 at defined time points. Cells were then washed twice with PBS and trypsinized from the plate with 0.05% trypsin-EDTA. Cells were washed with media, centrifuged, and washed once with cold PBS and 0.7% BSA. Following the first analysis by imaging flow cytometry, anti-Alexa Fluor 488 antibody was added at 100 μg/mL final concentration. The samples were then incubated on ice for at least 20 minutes with the quenching antibody before imaging flow-cytometry reanalysis.

Quantum beads (Bangs Laboratories, Inc.) were used to normalize the median Alexa Fluor 488 intensity of each cell event from imaging flow cytometry into median equivalents of standardized fluorescence (MESF). A standard curve was created using the beads, which had five populations with defined number of fluorophores. The standard curve was then used to normalize intensity values into MESF values. The number of fluorophores per antibody was obtained by NanoDrop and MESF data converted into the number of antibodies per cell.

An Amnis ImageStream Mark II (Amnis Corp.) was used for the imaging flow-cytometry data collection. For each experiment, the sample with the highest signal containing all fluorophores was analyzed first to set up an acquisition template with optimal laser power settings. Events were only collected if they were located within a gate set according to the size and aspect ratio from the brighfield channel of each event. The aspect ratio was the ratio of the height to width. We set the minimum aspect ratio value to >0.6 to collect single cells, because doublets often have aspect ratios around 0.5. All experiments included single fluorophore samples for color compensation during data analysis.

Raw imaging flow-cytometry data were analyzed by ImageStream Data Exploration and Analysis Software (IDEAS, Amnis Corp.). Compensation files were made from single color files accompanying each experiment. These adjustments were necessary to determine the amount of signal overlap from each fluorophore used in the experiment across other channels.

Single cells were gated based on the area and aspect ratio of the brightfield images (22), with a more selective gate applied to the collected events than the gate used during data acquisition. Additional gating was performed on the gradient RMS features to use optimally focused cells. The gates were further refined by manual analysis of the cells until only single cells were included before batch processing of the data files using a defined analysis template. Internalization and colocalization scores were calculated using algorithms in the IDEAS software. For each set of images included in this work, the same intensity settings were used for all cells. In addition, representative cells at the median intensity values were chosen.

Cells were plated on 6-well plates and allowed to adhere overnight before incubation with 15 μg/mL Ab033-E2/F2. Following the conclusion of treatment, cells were washed on the plate with PBS, trypsinized, centrifuged, aspirated, and frozen at −20°C. Before analysis of drug levels, 90 μL 95:5 acetonitrile/methanol mixture containing 50 nmol/L carbutamide internal standard was added to the cell pellets to normalize run-to-run variability resulting from loading and injection differences. Samples were centrifuged to pellet precipitated protein and the organic solution was added to 20 μL water and 5 μL DMSO. Standard curves were also prepared with varying concentrations of the main released products from Ab033-E2/F2 (unconjugated MMAE and cys-mc-MMAF) added to precipitated control cells in the same organic solution with the same internal standard used for the treated samples. Analysis was performed by LC-MS/MS with a Sciex 5500 QTrap mass spectrometer using selective reaction monitoring (SRM). The SRM transitions were setup with precursor and fragment mass to charge (m/z) pairs of 718.5 m/z /152.1 m/z for MMAE and 1046.6 m/z /428.1 m/z for cys-mc-MMAF. The SRM transition for carbutamide was 272.2 m/z /108.1 m/z for the precursor and fragment ions.

Mechanistic modeling and simulations were performed with MATLAB SimBiology (MathWorks, Inc.). The model was designed to mimic an in vitro cell culture environment with a defined number of cells in a limited volume. The conditions were set to 1 × 10⁶ cells per mL media to roughly recreate a 6-well dish setup. To avoid mass limiting conditions, the drug concentration was set to 100 nmol/L Ab033-vcMMAE to put the ADC in approximately 60-fold and 460-fold excess compared with the number of receptors present for A431 and H441 cells, respectively. Model parameters for binding, internalization, and recycling were determined experimentally and other parameters were found through fitting rates to measured levels of antibody and drug associated with cells. Equations to model the cellular kinetics were composed of ordinary differential equations using mass action. The detailed equations are listed in Supplementary Table S1. Sensitivity analyses were conducted to assess the effect of parameter changes on the model output of the AUC of intracellular MMAE. Normalized sensitivity coefficients were calculated using a standard forward difference approach where parameter values were varied by ±10%.

Ab033 is an IgG antibody with nmol/L binding affinity of EGFR driven by a fast on-rate [equilibrium dissociation constant, K_d, of 2.1 × 10⁻⁹ mol/L with on-rate, k_on, of 5.6 × 10⁵/mol/L × s, Supplementary Fig. S1]. However, binding does not necessarily translate into efficient internalization as epitope choice can play a major factor in internalization. Epitopes with excellent binding affinity have at times demonstrated slower internalization kinetics (23). Therefore, to assess the internalization potential of Ab033 in cancer cells, the EGFR-expressing cell lines A431 and H441 were incubated with saturating levels of AlexaFluor488-labeled Ab033. These cell types were chosen because of their high EGFR expression and their nearly 10-fold differential in EGFR expression between each other; H441 cells were determined to have 1.3 × 10⁵ bound Ab033 molecules on the cell surface and A431 to have 1.2 × 10⁶ Ab033 bound (Supplementary Fig. S2; ref. 24).

Cells were incubated for 1 hour in the presence of antibody at both 37°C and on ice, with the ice able to limit cellular internalization processes and provide a population of surface bound cells. A431 cells exhibited cell surface binding by imaging flow cytometry for both temperature conditions (Fig. 1A, top). For the samples incubated on ice, nearly all of the Ab033 and anti-IgG antibody staining was localized to the outer edge of the cell, observable as a ring around the cell. For cells incubated at 37°C, there was also extracellular staining and, in contrast with the cells incubated on ice, significant intracellular signal in the form of punctate staining (Fig. 1A, top). H441 cells displayed similar results as seen for A431, albeit with less signal intensity due to lower overall EGFR expression (Fig. 1A, bottom). A colocalization scoring algorithm quantified the degree of spatial overlap between labeled Ab033 antibody and extracellular secondary antibody. The median colocalization score decreased from 3.0 on ice to 1.3 at 37°C for A431 cells (Fig. 1B, top) and from 2.6 to 0.7 for H441 cells (Fig. 1B, bottom), suggesting reduced Ab033 in proximity to surface bound secondary antibody.

The internalization capacity of the cell represents a significant factor in the amount of ADC able to be accumulated intracellularly. As receptor internalization increases, a commensurate number of drugs can be internalized. The internalization rate can drive intracellular drug accumulation to the extent that a cell with low antigen concentration cell and high internalization can yield a higher net intracellular toxin concentration than a cell with higher antigen expression but lower internalization (25). To see where the Ab033–EGFR complex fell on the spectrum of internalization capacity, the internalization rate of surface bound anti-EGFR antibody was measured in A431 and H441 cells (Supplementary Fig. S3). The amount of surface bound secondary antibody decreased over time after the initial antibody binding on ice (Fig. 2A and B). Compared with the cells on ice, A431 cells had a decrease in surface antibody signal of 46% after 15 minutes and 67% after 60 minutes (Fig. 2A). H441 cells showed a 54% loss in signal after 15 minutes that stayed approximately the same over the remainder of the time course (Fig. 2B). Both cell types displayed internalization kinetics featuring marked initial internalization followed by either slower internalization (A431) or steady state (H441). The overall kinetic profile of H441 was more extreme than the profile for A431 as nearly all of the observed internalization for H441 occurred within the first 10 minutes and then stayed roughly unchanged in an equilibrium state. In contrast, A431 cells displayed a less pronounced initial internalization period for 15 minutes, followed by 45 minutes of slower, but steady decreases in surface bound levels. The endocytosis rate, k_e, was calculated using the fraction of internalized antibody and the time integral of fluorescence on the cell surface. The k_e for A431 was 0.047/min (Fig. 2A, inset) and for H441 was 0.15/min (Fig. 2B, inset). Ab033 conjugated with mc-vc-PABC-MMAE and mc-MMAF was also assessed and found to have similar internalization rates as naked Ab033 in A431 cells (Supplementary Fig. S4). Images from the first 5 minutes corroborate these significant decreases in surface staining (Fig. 2C; Supplementary Fig. S5). Over the remainder of the time course, intracellular Ab033 accumulation was observed (Supplementary Fig. S6). Although decreases in extracellular signal are likely the result of internalized antibody, dissociation of antibody would also produce similar decreases. However, imaging flow-cytometry data confirmed time-dependent decreases in colocalization between surface-localized secondary antibody and Ab033 along with increases to internalization metrics, lending further evidence of antibody internalization (Supplementary Fig. S7A and S7B).

After receptor-mediated endocytosis, internalized EGFR proceeds along the lysosomal degradation pathway or is returned to the cell surface by recycling endosomes (26). The slowing of internalization suggests involvement of recycling processes (Fig. 2). To investigate further, the amount of recycling was measured by first pulsing H441 and A431 cells with fluorescently labeled antibody followed by a chase in the presence of fluorescence-quenching antibody and quantitation of the decrease in total cellular fluorescence. By keeping quenching antibody in the media, recycled antibody was bound and its AlexaFluor488 fluorescence neutralized. The efficiency of cellular fluorescence quenching was measured to be 93% (Supplementary Fig. S8A and S8B). For the recycling experiment, cells were pulsed for 10, 20, or 60 minutes to accumulate intracellular antibody (Fig. 3A, 0 minutes images). Less of the internalized pool of antibody for A431 cells pulsed for 10 minutes was in the lysosome when compared with cells pulsed for 60 minutes (Fig. 3A, 0 minutes cells and Supplementary Fig. S9). However, over time the colocalization score increased for the 10 and 20 minutes pulses, indicating trafficking of intracellular Ab033 into the lysosome. In addition to lysosomal accumulation, the degree of recycling decreased as the pulse was lengthened; 10 minutes pulsed samples decreased overall intracellular levels by 45%/29%, 20 minutes pulsed samples decreased 42%/27%, and 60 minutes pulsed samples decreased 28%/10% after 30 minutes of chase for H441 and A431 cells, respectively (Fig. 3B and C). Because of higher levels of observed lysosomal Ab033 and lessened recycling for longer pulsed cells, these data suggest a reduced pool of “recyclable” antibody as the incubation progresses.

Although initial internalization of Ab033 was rapid, the interplay of internalization, recycling, and efflux determines drug concentration over time, which plays a significant role in overall drug efficacy. As such, the amount of cell-associated antibody was tracked across time while in the presence of saturating concentrations of labeled antibody. Over a 24 hours time course, levels in both A431 and H441 cells increased (Fig. 4A). The total antibody values shown refer to total amount of antibody taken into the cell over time and not the current amount present. Because of the low permeability of Alexa Fluor 488, fluorophore from degraded antibody can accumulate in the lysosome. In total, A431 cells accumulated an average of 2.5 × 10⁶ Ab033 molecules per cell after 24 hours and H441 cells accumulated a mean of 4.63 × 10⁵ Ab033 molecules per cell. The uptake of a labeled non-targeting antibody in both cell types was substantially less than Ab033 uptake with 5.2 × 10⁴ molecules per H441 cell and 1.6 × 10⁴ molecules per A431 cell (Fig. 4A, inset).

Quenching antibody was used to determine the intracellular component of the Ab033 signal and subtraction of the intracellular Ab033 levels from the total Ab033 levels yielded the amount of surface bound Ab033. The levels of Ab033 on the surface of H441 cells changed from 1.30 × 10⁵ Ab033 bound receptors to 1.54 × 10⁵ receptors, an 18% increase over the duration of the time course (Fig. 4B). In contrast, A431 surface Ab033 levels decreased by 31%, from 1.08 × 10⁶ to 8.23 × 10⁵ bound receptors (Fig. 4C and D). After 24 hours, 2.6 × 10⁵ Ab033 molecules were located inside H441 cells and 1.7 × 10⁶ Ab033 molecules were located inside A431 cells.

To follow intracellular trafficking of Ab033, AlexaFluor488-labeled Ab033 was conjugated with a pH sensitive dye exhibiting bright fluorescence in the acidic pH environment of endocytic and lysosomal compartments and limited fluorescence at neutral physiological pH values (27), allowing for great discrimination between intra- and extracellular antibody (14). Here, low levels of signal from the pH dye were present at 1 hour, despite significant amounts of antibody bound to the surface of both A431 and H441 cells (Supplementary Fig. S10A and S10B). Over time, the pH dye signal increased in a manner correlative to the intracellular Ab033 levels whereas limited pH dye fluorescence of extracellular bound Ab033 persisted. After 24 hours, nearly all pH sensitive dye signal was colocalized with intracellular Ab033 signal, indicating that Ab033 was located nearly exclusively within endosomes or lysosomes.

The trafficking of antibody to lysosome is favorable for drug release using lysosome cleavable linkers, such as mc-vc-MMAE lysosome–mediated cleavage shown previously (28, 29). However, because each drug has a defined cellular potency where a threshold of drug molecules must be met to elicit cell death (30), intracellular disposition data on free intracellular drug concentration is needed to determine whether ADC delivery of drug will achieve the levels needed to mediate the intended effect. The levels of released drug over time were therefore measured and LC-MS/MS was used to obtain absolute quantitation of intracellular drug levels.

The ADC analyzed here was a dual toxin ADC containing two cleavable mc-vc-PABC-MMAE linker–drug payloads and two non-cleavable mc-MMAF drug payloads on each Ab033 antibody (Ab033-E2/F2). The schematic and preparation results for Ab033-E2/F2 are shown in Supplementary Fig. S11. By adding both types of payload to the same antibody, a direct comparison of release differences between non-cleavable and cleavable attachments to auristatin toxins could be made independent of potential differences in ADC binding, internalization, or trafficking. In both A431 and H441 cells, the release of the non-cleavable product, cysteine-mc-MMAF (cys-mc-MMAF), slightly lagged the release of the cleavable product, MMAE, after 1 hour (Fig. 5A and B). However, the rate of appearance of both drugs was similar for the few hours following the first hour of the incubation. In total, A431 cells accumulated 5.1 × 10⁶ MMAE and 2.6 × 10⁶ cys-mc-MMAF molecules over 24 hours. Accounting for the MMAE drug-to-antibody ratio (DAR) of 2, the number of MMAE molecules closely tracks with the flow-cytometry data of accumulated intracellular Ab033 antibody molecules. H441 cells contained 5.6 × 10⁵ MMAE molecules per cell on average, which equals 2.8 × 10⁵ ADC DAR 2 equivalents, a value close to the 2.6 × 10⁵ Ab033 molecules found by flow cytometry. Media concentrations of both released drug products were also measured and the free concentrations of the two drugs in media increased over the incubation period (Supplementary Fig. S12).

Experimental measurements were integrated into a model designed to represent a cellular in vitro system (Supplementary Fig. S13). Cellular processing rates of Ab033-vcMMAE were derived from A431 and H441 data on cell binding, internalization, recycling, trafficking, accumulation, drug release, and efflux (Supplementary Table S2). Levels of receptor–ADC complex, internalized ADC, and free intracellular drug concentrations over time were simulated and plotted alongside measured values (Fig. 6A and B). Notably, because this model was developed to determine cellular processing rates, pharmacodynamic effects were not integrated into the model. Without pharmacodynamics considerations needing to be taken into account, intracellular drug concentrations were treated as one parameter, free MMAE, regardless of intracellular location and whether MMAE was free in the cytosol, was bound to intracellular proteins, or remained in the lysosome. Sensitivity analyses were conducted to assess the effect of ±10% changes in rates of binding (k_on and k_off), internalization (k_in), recycling (k_out), trafficking (k_lys), release (k_release), and drug efflux (k_effluxdrug) on intracellular MMAE accumulation. The overall net effect on AUC of intracellular MMAE caused by each parameter change was used to determine the sensitivity of each parameter compared with each other (Fig. 6C and D). Overall, fluctuations in internalization, recycling, and lysosomal trafficking rates had the largest impact on intracellular drug concentrations for both cell types. Alteration of unconjugated drug efflux rate also significantly changed the AUC of MMAE while binding differences resulted in little change to cellular MMAE levels. An additional simulation with modifications to k_lys between ±25% was performed to visualize the change to several parameters at the same time (Fig. 6E). When k_lys was increased, corresponding increases were found for levels of intracellular ADC, lysosomal ADC, and intracellular MMAE with decreases to surface EGFR–ADC complex levels.

Development of more efficacious ADCs requires improved delivery of payload to target cells. Increased understanding of ADC cellular processing (e.g., internalization kinetics) will facilitate the design of better ADCs. Furthermore, processing is largely driven by properties of the target antigen. Although enhanced expression of EGFR in many cancers makes it an attractive target for ADC approaches, ultimately the unique internalization and degradation characteristics of EGFR in different tumor types will influence whether ADC EGFR therapeutics will be successful. To improve our understanding of cellular processing of EGFR-directed ADCs, the internalization and trafficking of Ab033 were assessed in the EGFR-expressing cancer cells A431 and H441. This work was centered on the cellular processing of ADCs. Cytotoxicity data were therefore not included to keep the focus on the determination of inefficient cellular processing steps of ADCs. A431 cells have approximately 10-fold higher cell surface EGFR expression than H441 cells (Supplementary Fig. S2), allowing for comparison of EGFR kinetics in two systems with substantially different EGFR levels. Internalization occurred quickly upon Ab033 binding to EGFR; H441 cells internalized 52% of surface bound Ab033 within 10 minutes (Fig. 2). The calculated endocytosis rate of 0.147/min for H441 cells was approximately 3-fold higher than for A431 cells and approached the reported 0.2 to 0.6/min internalization rates measured for EGFR in the presence of EGF ligand, an inducer of EGFR internalization (31, 32). In addition, higher EGFR turnover rates have previously been seen in cell lines with lower EGFR expression (33). The A431 endocytosis rate of 0.047/min was similar to other reported literature values of monoclonal antibody-induced EGFR internalization (34). To benchmark against another popular ADC target, the rates measured here are >50-fold higher than those previously reported for ErbB2 (also known as HER2), which is not induced to internalize upon antibody binding of certain epitopes and is also subject to extensive recycling (20, 35). These EGFR endocytosis rates give EGFR an internalization t_1/2 on the order of minutes compared with several hours for HER2.

In order for ADCs to mediate their intended cytotoxic effect, drug must be brought into the target cells at a sufficient concentration. High recycling rates can prevent such drug accumulation. Significant portions of Ab033 were indeed recycled, but the amount being recycled decreased over time (Fig. 3) and cells were able to continually accumulate antibody across longer time courses (Fig. 4B and C). These observations agree with previously proposed cellular timelines for recycling and degradation of ligand-bound EGFR (36). Quantitation of released drug from Ab033-E2/F2 revealed levels of intracellular accumulation of MMAE to be closely aligned with total Ab033 uptake data measured by flow cytometry. These data indicated that after 24 hours of incubation with Ab033-E2/F2, the majority of MMAE brought into the cell by the ADC had been released and was presiding within the cell as unconjugated drug. In contrast, the cys-mc-MMAF levels were lower than the MMAE levels, likely due to the lengthened time needed for antibody catabolism to release payload. Overall, despite substantial recycling, internalized Ab033 accumulated intracellularly and trafficked to the lysosome (Supplementary Fig. S10), making Ab033 appear to be a reasonable candidate as the delivery agent for an ADC containing a linker-payload with lysosomal-mediated drug release.

The binding of Ab033 to EGFR significantly increased EGFR turnover, which has been shown to have a t_1/2 nearing 24 hours in the absence of ligand (37). The large number of internalized molecules after 24 hours in the presence of saturating amounts of Ab033 (Fig. 4B and C) represented internalized Ab033 levels 2.0- and 1.6-fold of initial EGFR cell surface expression for H441 and A431 cells, respectively. Despite such internalization, improvement in antibody internalization could be envisioned. For example, surface-bound EGFR levels were fairly consistent on A431 and H441 cells, even after hundreds of thousands of internalization events per cell and extended incubation times out to 24 hours (Fig. 4B and C). In addition, recycling appeared to limit the total amount of accumulation within the cells by returning a portion of internalized antibody back to the cell surface (Fig. 3). H441 cells showed nearly 3-fold higher internalization rates than A431 cells, but higher recycling rates led to nearly equal net MMAE drug accumulation after normalizing for starting receptor counts (4.30 MMAE/receptor for A431 and 4.25 MMAE/receptor for H441 after 24 hours). All data were integrated into a quantitative systems model to better describe ADCs in a cellular context (Fig. 6A and B). Previous modeling approaches have been used to simulate cellular processing functions and conduct sensitivity analyses (35, 38). The mechanistic modeling here features similar elements but, in an effort to fully describe ADC processing and not rely on estimated or literature values, has incorporated more quantitative measurements to better define key model parameters, including recycling, free drug in media, and trafficking as well as intracellular drug levels, arguably the most important parameter for ADCs. The model confirmed the overall accumulation of free drug to be rate-limited by the internalization, recycling, and trafficking parameters (Fig. 6C and D). Several of the parameters from the sensitivity analysis gave similar results as a previous trastuzumab–emtansine ADC model, including limited effect of altering binding affinity and the highest impact for altering internalization rates (38). Future iterations of EGFR-targeting antibodies able to improve upon any of these properties would lead to the most productive gains in drug delivery to target cells (Fig. 6E). Several approaches have been shown to increase internalization, promote receptor degradation, and improve lysosomal trafficking; these include the use of biparatopic antibodies (13), bispecific antibodies against different antigens (39), and mixtures of antibodies for differential epitope coverage (11, 40).

The sensitivity analysis performed here was specific for Ab033-vcMMAE; sensitivity analyses of other ADC molecules could yield different conclusions. To illustrate one area where this may be true, changes to k_release for vc-MMAE made little difference in overall intracellular drug levels, likely due to efficient cleavage of vc-MMAE. Conversely, a linker with slow release kinetics might produce much larger differences in drug concentrations with small cleavage rate increases. Another key point about sensitivity analysis is that parameter changes are all relative. Thus, small changes of one parameter may lead to larger net effects than for the same percentage change of another parameter. Furthermore, certain rates may be easier to improve upon than others. Affinity maturation of antibody binding for low affinity antibodies could net multiple-fold changes in binding affinity. On the other hand, finding ways to increase internalization, such as changing the binding epitope, may be more difficult to accomplish. Another important aspect to consider apart from antibody- and linker-related parameters is drug properties. The sensitivity analysis of the model here demonstrates changing drug uptake and efflux parameters leads to significant altering of drug disposition. Structure–activity relationship (SAR) modifications of drug properties could change permeability or transporter protein activity, substantially impacting intracellular drug levels.

One of the most impactful components of the modeling is the overall amount of drug delivery. If intracellular drug levels are less than the concentration requirements for efficacy, then either a more potent drug must be selected, or properties of the ADC need to be changed until levels are above the efficacy threshold. Through quantitative measurements of the mechanistic disposition of antibodies and ADCs, rates of cellular processes related to drug functionality can be defined and input into cellular models to maximize learnings and better define aspects of cellular processing with potential to be optimized. These learnings can then inform subsequent ADC SAR to improve limiting components of drug delivery, whether from antibody, linker, or drug. By focusing the scope of optimization to certain parameters, the drug development process can be streamlined and the probability of success increased.

No potential conflicts of interest were disclosed.

Conception and design: K.R. Durbin, X. Liao

Development of methodology: K.R. Durbin, C. Phipps, X. Liao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.R. Durbin, X. Liao

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.R. Durbin, C. Phipps, X. Liao

Writing, review, and/or revision of the manuscript: K.R. Durbin, C. Phipps, X. Liao

All authors are employees of AbbVie. The design, study conduct, and financial support for this research were provided by AbbVie. AbbVie participated in the interpretation of data, review, and approval of the publication. The authors would like to thank Axel Hernandez, Edit Tarcsa, Gary Jenkins, and Anthony Haight for helpful comments, discussions, and article review as well as Enrico Digiammarino for Biacore binding data.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.