Cell-type–specific intracellular payload delivery is desired for antibody-based–targeted therapy development. However, tumor-specific internalizing antigens are rare to find, and even rarer for those that are expressed at uniformly high levels. We constructed a bispecific antibody that is composed of a rapidly internalizing antibody binding to a tumor-associated antigen, ephrin receptor A2 (EphA2), and a noninternalizing antibody binding to a highly expressed tumor-associated antigen, activated leukocyte cell adhesion molecule (ALCAM). We found that the overall internalization property of the bispecific is profoundly impacted by the relative surface expression level (antigen density ratio) of EphA2 versus ALCAM. When the EphA2-to-ALCAM ratio is greater than a threshold level (1:5), the amount of the bispecific taken into the tumor cell exceeds what is achieved by either the monoclonal internalizing antibody or a mixture of the two antibodies, showing a bispecific-dependent amplification effect where a small amount of the internalizing antigen EphA2 induces internalization of a larger amount of the noninternalizing antigen ALCAM. When the ratio is below the threshold, EphA2 can be rendered noninternalizing by the presence of excess ALCAM on the same cell surface. We constructed a bispecific antibody–drug conjugate (ADC) based on the above bispecific design and found that the bispecific ADC is more potent than monospecific ADCs in tumor cell killing both in vitro and in vivo. Thus, the internalizing property of a cell surface antigen can be manipulated in either direction by a neighboring antigen, and this phenomenon can be exploited for therapeutic targeting.

The high specificity of mAb is often exploited for targeted therapy development. Ideally, a potent cytotoxic attached to a cell-type–specific antibody can route the cytotoxic to the target cell and preferentially accumulate in the target tissue. Antibody–drug conjugate (ADC) is a class of targeted therapeutic that has shown effectiveness in the clinic (1, 2). Internalizing antibody is often desired to achieve efficient intracellular payload delivery and tumor killing (1), although the requirement is not absolute for certain drugs such as MMAE that can diffuse through cell membrane to cause a bystander effect (3).

Although conceptually straightforward, target selection in ADC is hindered by the fact that it is rare to find the so-called tumor-specific antigens, and even rarer to find those that bear features desired for therapeutic targeting, that is uniformly expressed at high levels by cancer cells, and efficiently internalizing. Several approaches have been developed to improve antibody internalization and ADC efficacy. For example, through the biparatopic design and the consequent cross-linking effect, HER2 antigen has been targeted for improved ADC internalization (4). In another example, a bispecific composed of a moderately internalizing antibody arm (anti-HER2) and an internalization-inducing antibody arm (anti-CD63, anti-PRLR, or anti-APLP2) was constructed and used to improve ADC uptake (5–7). However, despite those efforts, the bispecific ADC only showed limited improvement over the parental monospecific anti-HER2 ADC, suggesting that key parameters regarding this design remains to be delineated.

The key question is whether the internalization propensity of a given cell surface antigen can be impacted by its neighboring surface antigens, and if so what are the parameters that govern the conversion of a noninternalizing antigen to an internalizing antigen, and vice versa. We have previously developed a guide-effector bispecific system to achieve cell-type–specific signaling modulation (8). Key to our design is the guide-to-effector ratio and the threshold of guide antigen expression. We hypothesize that internalization can be manipulated by our guide-effector–based bispecific. We developed a bispecific antibody targeting ephrin receptor A2 (EphA2), a rapidly internalizing antigen, and activated leukocyte cell adhesion molecule (ALCAM), a non- or slowly internalizing antigen, and found that the bispecific becomes internalized when the ratio of EphA2 to ALCAM is greater than approximately 1:5. We further showed that the bispecific effect is different from that of a simple mixture of the two mAbs: the number of bispecific molecules delivered into the tumor cell is greater than that of the antibody mix. Therefore, the guide-effector design can cause an amplification effect, starting with a small number of seed internalizing antigen and propagating the internalizing effect cross the more abundant noninternalizing antigen. Interestingly, when the ratio of EphA2 to ALCAM is below the threshold (1:5), the internalizing EphA2 can be rendered non- or slowly internalizing by ALCAM, demonstrating that the conversion is reciprocal depending on the ratio of the guide-to-effector antigen. Thus when targeted by a bispecific, internalization of a cell surface antigen can be readily manipulated by its neighboring antigen, resulting in either amplified intracellular uptake or greatly retarded internalization depending on the relative abundance of the two antigens, providing an opportunity for therapeutic exploitation of cell membrane dynamics induced by our guide/effector-based bispecific.

Cell lines and plasmids

Human embryonic kidney (HEK) lines HEK293 and HEK293A; prostate cancer cell lines DU145 and PC3; and pancreatic cancer cell lines Capan-1, Panc-1, and MIA PaCa2 were obtained from American Type Culture Collection. The L3.6pl line was obtained from Dr. Isaiah Fidler (MD Anderson Cancer Center, Houston, TX; ref. 9). The LNCap-C4-2B was originally obtained from UroCor Inc. and maintained in the laboratory (10). Cells were maintained in DMEM or RPMI1640 supplemented with 10% FBS (Thermo Fisher Scientific), 100 μg/mL penicillin/streptomycin (Axenia BioLogix) at 37°C 5% CO2. Full-length human EphA2 cDNA cloned into pCMV-Entry (OriGene) or pLV202 (OriGene) was used for transient or stable expression of EphA2, respectively.

Generation of anti-ALCAM single-chain variable fragment antibodies

A naïve single-chain variable fragment (scFv)-phagemid display library (11) was used for antibody selection. A recombinant human IgG–like V1-V2 domain of ALCAM fused with human IgG2 Fc (A-V-Fc) was produced from HEK293A cells and utilized as the target antigen for selection. A-V-Fc was coated on SPHERO polystyrene magnetic particles (Spherotech) at 4°C overnight. Phage library was depleted with uncoated beads in PBS/2% milk, and unbound phage particles were allowed to bind to the A-V-Fc–coated beads. After washing, bound phage were eluted and propagated as described previously (8). Individual phage binders were screened by FACS using the ALCAM-expressing DU145 cell line, and DNA sequence of scFvs was analyzed by IgAT tool (12).

Recombinant antibody production

VH and VL antibody genes were amplified from candidate scFv-phagemid by PCR and subcloned into Abvec Igγ and -λ expression vectors, respectively (13, 14). To produce bispecific IgG-scFvs, the anti-ALCAM 3F1 or a nonbinding control C10 was utilized as the IgG backbone, and the internalizing scFv was introduced at C-terminus of the λ light chain constant region by fusion with a (Gly4Ser)3 linker (8, 15). HEK293A cells were transfected with antibody expression plasmids mixed with polyethylenimine (Sigma Aldrich) in Opti-MEM (Life Technologies) for 24 hours. Transfection medium was changed to Freestyle 293 (Gibco) and the cells were further cultured up to 8 days. Secreted antibodies were purified from culture supernatants on protein A agarose (Thermo Fisher Scientific) and analyzed on SDS-PAGE gradient gels (4%–20%).

Generation of stable HEK293-EphA2 cell line

HEK293 cells were transduced with EpAh2-expresssing lentivirus and maintained in regular growth medium containing G418 (Sigma). Stable EphA2-expressing clones were identified by FACS using human anti-EphA2 antibody followed by Alexa Fluor 647–labeled goat anti-human IgG (Jackson ImmunoResearch Laboratories). Stable clones were further screened by FACS to obtain those that express varying levels of EphA2.

Cell surface antigen copy-number measurement

Cell surface antigen copy number (or antigen density) was measured as described previously (8, 14). Briefly, cells were dissociated by 0.25% trypsin digestion, washed, and resuspended in FACS assay buffer (PBS, 1% FBS, pH 7.4), incubated with anti-EphA2 or ALCAM antibodies that were conjugated with Alexa Fluor 647 by Monoclonal Antibody Labeling Kit (Molecular Probes) to detect EphA2 or ALCAM, respectively, and analyzed by BD Accuri C6 (BD Biosciences). Median fluorescence intensity (MFI) was converted into antibody-binding capacity using Quantum Alexa Fluor 647 MESF and Quantum Simple Cellular anti-human IgG (Bangs Laboratory) according to manufacturer's recommendations. EphA2/ALCAM ratios were calculated by dividing the copy number of EphA2 by that of ALCAM for each cell model studied.

Apparent Kd determination

Dissociated cells (∼2 × 105) were incubated with varying concentrations of human IgGs for 16 hours at 4°C. Following three washes with ice-cold PBS, cell-bound IgG was detected by Alexa Fluor 647–labeled goat anti-human IgG (Jackson ImmunoResearch Laboratories) and analyzed by FACS (8). Apparent Kd value was calculated by a curve-fitting method using GraphPad Prism software (8, 14).

Cell surface antigen depletion

Mono- or bispecific antibodies (100 nmol/L) were incubated with cells cultured in 24-well plates (∼80% confluence) for 24 hours, and EphA2 or ALCAM remaining on cell surface was determined using Alexa Fluor 647–labeled L1A1 anti-EphA2 human IgG (16) or L50 anti-ALCAM mouse IgG (Thermo Fisher Scientific), respectively. Cell surface copy number was calculated using methods described above and normalized against a control group without antibody treatment.

Immunofluorescence confocal microscopy

Antibodies were incubated with cells seeded in 8-well culture chamber slides (Thermo Fisher Scientific) for the indicated amount of time. To assess pathway of internalization (macropinocytosis), cells were coincubated with Texas Red–conjugated 70-kDa neutral dextran (ND70-TR, Life Technologies), a marker for macropinocytosis (16). Post incubation, cells were fixed with 4% paraformaldehyde and permeabilized with PBS/1% FBS/0.2% Triton-X100. Cell-associated antibodies were stained with Alexa Fluor 488- or 647-labeled goat anti-human IgG (Jackson ImmunoResearch Laboratories) for 1 hour at room temperature. Lysosomes were detected by rabbit anti-lysosomal–associated membrane protein 1 (LAMP1) antibody (Cell Signaling Technology) followed by incubation with Alexa Fluor 647–labeled goat anti-rabbit IgG (Jackson ImmunoResearch Laboratories). For analysis of antibody localization in tumorspheres, spheres were collected by centrifugation at 500 × g for 5 minutes, washed, fixed, permeabilized, and immunolabeled using antibodies described above. CyGEL (Abcam) was used to immobilize spheres in 8-well chamber slide for microscope analysis. For imaging, cells or spheres were counterstained using Hoechst33342 (Thermo Fisher Scientific) and imaged by FluoView FV10i Laser Confocal Microscope (Olympus) with an Olympus 60X phase contrast water-immersion objective.

Tumorsphere formation

Tumorspheres were generated by culturing suspended tumor cells from monolayer culture in serum-free medium (SFM) containing DMEM/F12 (Gibco), 20 ng/mL EGF, 10 ng/mL bFGF, 10 ng/mL IGF, and 2% B27 supplement (Gibco) in ultralow attachment 24-well plates (Corning) at 37°C/5% CO2. For sphere propagate assay, the first-generation spheres were trypsinized and sieved through a 40-μm nylon mesh cell strainer (Thermo Fisher Scientific) to obtain a single-cell population. A total of 200 cells per well were resuspended in 500 μL SFM, seeded in ultralow attachment 24-well plates (Corning) at 37°C/5% CO2 for 24 hours, and treated with indicated antibodies for 2 weeks. Cells were fed with 100 μL SFM every 3–4 days. Each well was sectionally scanned using the BIOREVO Digital Microscope (BZ-9000; Keyence) and merged to display the whole-well image. Spheres > 100 μm in diameter were counted.

Site-specific ADC generation

A cysteine residue was introduced to heavy chain position 116 (T116C) of IgG or bispecific IgG (bsIgG), and site-specific ADCs were generated as described previously (17) with modifications. Briefly, antibodies in PBS were reduced by incubation with 10-fold molar excess of Tris(2-carboxyethyl)phosphine hydrochloride (Thermo Fisher Scientific) at 37°C for 2 hours, purified by Zeba Spin Desalting Column (Thermo Fisher Scientific) and buffer exchanged in PBS/5 mmol/L EDTA. To reoxidize interchain disulfide bonds, reduced antibodies were incubated with 20-fold molar excess of dehydroascorbic acid (Sigma) at 25°C for 3 hours. After buffer exchange with PBS/5 mmol/L EDTA, the antibody was incubated at 25°C for 1 hour with 3-fold molar excess of maleimidocaproyl-valine-citrulline-p-aminobenzoyloxycarbonyl monomethyl auristatin F (MC-vc-PAB-MMAF) that was synthesized as described previously (14). The final conjugation product was purified by running twice through the Zeba Spin Desalting Column (Thermo Fisher Scientific) and analyzed by hydrophobic interaction chromatography (HIC)-high-performance liquid chromatography (HPLC) using the infinity 1220 LC System (Agilent). Drug-to-antibody ratio (DAR) was estimated from area integration using the OpenLab CDS Software (Agilent; ref. 14).

ADC cytotoxicity

Cells were seeded at 2 × 103/well in 96-well cell culture plates overnight, and incubated with varying concentrations of ADCs for 96 hours. Cell viability was determined using a Calcein-AM Cell Viability Assay Kit (Biotium Inc.; ref. 10).

In vivo xenograft study

All animal studies were approved by the UCSF Animal Care and Use Committee (AN092211) and conducted in adherence to the NIH Guide for the Care and Use of Laboratory Animals. NOD/SCID/IL-2Rγ−/− (NSG) female mice were engrafted with 1 × 106 Capan-1 cells, randomized into four groups at day 5 (n = 6 for each group). Mice were treated intravenously with the vehicle PBS or mono- or bispecific ADCs at 3 mg/kg every 4 days for a total of four injections. Tumor size was measured by a caliper, and tumor volume was calculated using the formula V = (Width2 × Length)/2. Body weight was monitored during the course of the study.

Identification of a high-affinity ALCAM antibody and generation of the ALCAMxEphA2 bispecific

To identify human antibodies against ALCAM, we performed scFv phage display library selection against N-terminal Ig-like V1-V2 domain of ALCAM (Supplementary Fig. S1A). We identified a panel of binding phage by FACS screening on the ALCAMHigh DU145 prostate cancer cell line (Supplementary Fig. S1B), and further identified an antibody 3F1 that binds with high affinity as an IgG1 (apparent Kd = 20.6 pmol/L) to DU145 cells (Supplementary Fig. S1C). We studied internalization of the 3F1 IgG on a panel of tumor cell lines by confocal microscopy and found that this antibody is non- or slowly internalizing (Supplementary Fig. S1D).

We constructed a tetravalent bispecific IgG-scFv (bsIgG-scFv) composed of the noninternalizing anti-ALCAM 3F1 IgG backbone and an internalizing anti-EphA2 scFv (RYR) fused to the C terminus of the 3F1 light chain (Fig. 1A). The anti-EphA2 scFv (RYR) was identified from our previous study, where we used high-content analysis to identify macropinocytosing antibodies (16). For control, a nonbinding C10 IgG was used to construct the control C10/RYR bsIgG (binding to EphA2 only). SDS-PAGE analysis showed the expected electrophoresis pattern of monoclonal and bispecific antibodies (Supplementary Fig. S2A). We next studied binding specificity of bsIgGs using the HEK293 cell line that expresses ALCAM, and an engineered HEK293 cell line that stably expresses a high level of EphA2 (HEK293-EphA2#2). As shown in Supplementary Fig. S2B, the anti-ALCAM 3F1 IgG bound to both HEK293 and HEK293-EphA2#2 cells as expected. The 3F1/RYR bsIgG bound at a higher level to HEK293-EphA2#2 (ALCAMhighEphA2high) compared with HEK293 (ALCAMhighEphA2low). The control C10/RYR bsIgG that binds to EphA2 only showed specific binding to HEK293-EphA2#2 but not HEK293 cells. Using these two cell line models, internalization activity of the 3F1/RYR bsIgG was studied by confocal microscopy. As shown in Fig. 1B, the 3F1/RYR bsIgG acquired effective internalization capacity in an EphA2-dependent manner, being internalized by the HEK293-EphA2#2 but not the HEK293 cell line. The control C10/RYR bsIgG is internalized by HEK293-EphA2#2 but not HEK293. The result shows that in our guide-effector bispecific design, the internalizing arm (EphA2, the guide) can impart the noninternalizing arm (ALCAM, the effector) and the bispecific as a whole with internalizing properties.

Figure 1.

A bispecific based on the guide-effector design can profoundly affect internalization dynamics of cell surface antigen. A, Illustration of the tetravalent ALCAMxEphA2 bsIgG. The IgG backbone is based on the noninternalizing anti-ALCAM antibody 3F1. The internalizing anti-EphA2 scFv is fused to the end of light chain C-terminus. B, Confocal microscopy study of antibody internalization. HEK293 or HEK293-EphA2#2 cells were incubated with indicated IgG or bsIgG (100 nmol/L) at 37°C for 2 hours. Antibodies (red) were detected using Alexa 647–labeled anti-human IgG secondary antibody, and cell images were analyzed using a digital laser confocal microscope. Scale bar, 20 μm. C, Kinetics of ALCAM cell surface removal by the bispecific. HEK293-EphA2#2 cells were incubated with indicated IgG or bsIgG for 1, 4, and 24 hours and surface ALCAM levels determined by FACS. The noninternalizing ALCAM is removed from cell surface by the bispecific (3F1/RYR) but not mAbs. D, Correlation between surface antigen (ALCAM) removal efficiency and EphA2/ALCAM expression ratio. HEK293 cell models with varying EphA2/ALCAM ratios were incubated with 3F1, 3F1/RYR, and C10/RYR (all at 100 nmol/L), and antigens remaining on the cell surface were determined by anti-ALCAM antibodies that bind to a different epitope than 3F1. Pearson correlation coefficient (r) was calculated (0.3266, −0.7550, and −0.1896 for 3F1, 3F1/RYR, and C10/RYR, respectively), and trend lines were depicted according to linear regression analysis. Data represent mean ± SD (duplicate). E, Illustration depicting bispecific-induced ALCAM internalization when the guide-to-effector ratio > threshold. CM, cell membrane. F, Significant retardation of EphA2 internalization by the bispecific 3F1/RYR when guide-to-effector ratio falls below threshold level. HEK293 cells that possess a low EphA2/ALCAM ratio (<0.2) were incubated with indicated antibodies (100 nmol/L), and surface EphA2 levels were measured by FACS. P values were determined using two-tailed Student t test. *, P < 0.05; ***, P < 0.001. G, Illustration of the phenomenon shown in F where EphA2 internalization is retarded when the EphA2-to-ALCAM ratio falls below a threshold.

Figure 1.

A bispecific based on the guide-effector design can profoundly affect internalization dynamics of cell surface antigen. A, Illustration of the tetravalent ALCAMxEphA2 bsIgG. The IgG backbone is based on the noninternalizing anti-ALCAM antibody 3F1. The internalizing anti-EphA2 scFv is fused to the end of light chain C-terminus. B, Confocal microscopy study of antibody internalization. HEK293 or HEK293-EphA2#2 cells were incubated with indicated IgG or bsIgG (100 nmol/L) at 37°C for 2 hours. Antibodies (red) were detected using Alexa 647–labeled anti-human IgG secondary antibody, and cell images were analyzed using a digital laser confocal microscope. Scale bar, 20 μm. C, Kinetics of ALCAM cell surface removal by the bispecific. HEK293-EphA2#2 cells were incubated with indicated IgG or bsIgG for 1, 4, and 24 hours and surface ALCAM levels determined by FACS. The noninternalizing ALCAM is removed from cell surface by the bispecific (3F1/RYR) but not mAbs. D, Correlation between surface antigen (ALCAM) removal efficiency and EphA2/ALCAM expression ratio. HEK293 cell models with varying EphA2/ALCAM ratios were incubated with 3F1, 3F1/RYR, and C10/RYR (all at 100 nmol/L), and antigens remaining on the cell surface were determined by anti-ALCAM antibodies that bind to a different epitope than 3F1. Pearson correlation coefficient (r) was calculated (0.3266, −0.7550, and −0.1896 for 3F1, 3F1/RYR, and C10/RYR, respectively), and trend lines were depicted according to linear regression analysis. Data represent mean ± SD (duplicate). E, Illustration depicting bispecific-induced ALCAM internalization when the guide-to-effector ratio > threshold. CM, cell membrane. F, Significant retardation of EphA2 internalization by the bispecific 3F1/RYR when guide-to-effector ratio falls below threshold level. HEK293 cells that possess a low EphA2/ALCAM ratio (<0.2) were incubated with indicated antibodies (100 nmol/L), and surface EphA2 levels were measured by FACS. P values were determined using two-tailed Student t test. *, P < 0.05; ***, P < 0.001. G, Illustration of the phenomenon shown in F where EphA2 internalization is retarded when the EphA2-to-ALCAM ratio falls below a threshold.

Close modal

The noninternalizing antigen can be rendered internalizing by the bispecific in a time- and guide-to-effector ratio–dependent manner

To quantitatively investigate cell surface antigen removal by antibody-induced antigen internalization, quantitative FACS analysis was conducted to measure ALCAM and EphA2 copy numbers on the cell surface (herein referred to as antigen density). As shown in Fig. 1C, the ALCAM level on HEK293-EphA2#2 cells incubated with the bispecific 3F1/RYR was decreased by approximately 90% within the first 4 hours of incubation. There was no significant change of the surface ALCAM level after treatment with monoclonal anti-ALCAM 3F1, the control monoclonal C10, or the control bispecific C10/RYR (Fig. 1C). Efficient ALCAM removal from cell surface by 3F1/RYR was only observed in HEK293-EphA2#2 (ALCAMhighEphA2high) but not HEK293 (ALCAMhighEphA2low; Supplementary Fig. S3A). We further sought to determine whether the antigen removal efficiency is influenced by the guide-to-effector ratio (EphA2/ALCAM). To generate HEK293-based cell models with varying EphA2/ALCAM ratios, EphA2 and/or ALCAM levels were manipulated by three ways: (i) transient transfection of EphA2-expressing plasmid, (ii) transient cotransfection of EphA2-expressing plasmid and ALCAM-siRNA, and (iii) lentiviral transduction of the EphA2 gene to achieve stable EphA2 expression. These cells showed varying EphA2/ALCAM ratios and varying patterns of surface antigen removal following the bispecific 3F1/RYR treatment. For example, the monoclonal anti-ALCAM 3F1 IgG or the control C10/RYR bsIgG did not remove ALCAM from the cell surface, whereas the 3F1/RYR bsIgG efficiently removed surface ALCAM (Fig. 1D). By Pearson correlation coefficient analysis, the effect significantly increases as the EphA2/ALCAM ratio increases (Fig. 1D). With regard to EphA2, the anti-ALCAM 3F1 IgG did not reduce surface EphA2 as expected, but the 3F1/RYR and the control C10/RYR that binds to EphA2 removed EphA2 efficiently from the cell surface (Supplementary Fig. S3B). The ability of the bispecific 3F1/RYR to remove surface ALCAM is affected by the ratio of EphA2 to ALCAM (guide-to-effector antigen ratio, outlined in Fig. 1E). As summarized in Table 1, when the ratio is <1:5 (0.2), only a small fraction of ALCAM is removed (20%–35%). When the ratio is between 0.9–3.5, 45%–65% surface ALCAM is removed. When the ratio is >3.5, greater than 70% of surface ALCAM is removed.

Table 1.

Summary of antigen density, guide-to-effector ratio, and ALCAM turnover efficiency on various cell line models

Antigen density and ratio (mean ± SD, n = 2)ALCAM Internalization (% over control, mean, n = 2)
Cell modelsEphA2ALCAMEphA2/ALCAM3F13F1/RYRC10/RYR
HEK293  5,064 ± 858 335,719 ± 2,475 0.02 6.6 19.6 10.8 
Transiently EphA2-transfected HEK293 HEK293-EphA2-0.02 357,994 ± 83,557 286,531 ± 6,581 1.25 29.4 68.9 28.6 
 HEK293-EphA2-0.1 1,081,566 ± 89,489 325,339 ± 17,985 3.32 7.7 68.9 −22.4 
 HEK293-EphA2-0.5 2,130,189 ± 210,485 311,259 ± 42,142 6.84 3.5 70.1 −36.3 
 HEK293-EphA2-2.5 3,386,996 ± 236,937 301,854 ± 3,913 11.22 21.3 71.2 7.1 
Stably EphA2-expressing HEK293 HEK293-EphA2#C9 18,398 ± 1,606 320,850 ± 36,242 0.06 2.8 20.7 −9.6 
 HEK293-EphA2#42 23,508 ± 2,955 202,617 ± 11,141 0.12 3.7 16.5 −16.5 
 HEK293-EphA2#8 1,876,422 ± 224,373 279,389 ± 11,756 6.72 −17.9 87.2 0.5 
 HEK293-EphA2#2 2,864,233 ± 409,871 295,696 ± 9,053 9.69 −7.8 89.1 14.6 
 HEK293-EphA2#31 4,608,433 ± 50,555 278,564 ± 2,880 16.54 −11.7 87.6 9.2 
Transiently ALCAM-siRNA/EphA2–transfected HEK293 HEK293-siRNA1 466,203 ± 82,224 296,677 ± 63,749 1.57 −7.5 70.2 −3.4 
 HEK293-siRNA2 377,816 ± 95,592 208,304 ± 23,759 1.81 −14.0 69.7 −6.0 
 HEK293-siRNA3 507,760 ± 17,793 60,512 ± 5,047 8.39 −6.6 87.5 −4.4 
Antigen density and ratio (mean ± SD, n = 2)ALCAM Internalization (% over control, mean, n = 2)
Cell modelsEphA2ALCAMEphA2/ALCAM3F13F1/RYRC10/RYR
HEK293  5,064 ± 858 335,719 ± 2,475 0.02 6.6 19.6 10.8 
Transiently EphA2-transfected HEK293 HEK293-EphA2-0.02 357,994 ± 83,557 286,531 ± 6,581 1.25 29.4 68.9 28.6 
 HEK293-EphA2-0.1 1,081,566 ± 89,489 325,339 ± 17,985 3.32 7.7 68.9 −22.4 
 HEK293-EphA2-0.5 2,130,189 ± 210,485 311,259 ± 42,142 6.84 3.5 70.1 −36.3 
 HEK293-EphA2-2.5 3,386,996 ± 236,937 301,854 ± 3,913 11.22 21.3 71.2 7.1 
Stably EphA2-expressing HEK293 HEK293-EphA2#C9 18,398 ± 1,606 320,850 ± 36,242 0.06 2.8 20.7 −9.6 
 HEK293-EphA2#42 23,508 ± 2,955 202,617 ± 11,141 0.12 3.7 16.5 −16.5 
 HEK293-EphA2#8 1,876,422 ± 224,373 279,389 ± 11,756 6.72 −17.9 87.2 0.5 
 HEK293-EphA2#2 2,864,233 ± 409,871 295,696 ± 9,053 9.69 −7.8 89.1 14.6 
 HEK293-EphA2#31 4,608,433 ± 50,555 278,564 ± 2,880 16.54 −11.7 87.6 9.2 
Transiently ALCAM-siRNA/EphA2–transfected HEK293 HEK293-siRNA1 466,203 ± 82,224 296,677 ± 63,749 1.57 −7.5 70.2 −3.4 
 HEK293-siRNA2 377,816 ± 95,592 208,304 ± 23,759 1.81 −14.0 69.7 −6.0 
 HEK293-siRNA3 507,760 ± 17,793 60,512 ± 5,047 8.39 −6.6 87.5 −4.4 

Internalization and noninternalization are interconvertible properties in a bispecific design

Although we have shown that a noninternalizing antigen (ALCAM) can be induced to internalize by the anti-EphA2/ALCAM bispecific when the EphA2-to-ALCAM ratio is above a threshold, we studied whether the rapidly internalizing EphA2 can be rendered slowly or noninternalizing by the presence of ALCAM at certain EphA2-to-ALCAM ratio. Using HEK293 cell line (ALCAMhighEphA2low) as the model, we found that when the EphA2-to-ALCAM ratio is <0.2, EphA2 internalization was greatly retarded, resulting in a higher fraction of surface bound EphA2 when targeted by the 3F1/RYR bsIgG but not the control C10/RYR (Fig. 1F), suggesting that internalization and noninternalization are interconvertible properties and the relative abundance of internalizing versus noninternalizing antigen profoundly impacts cell surface antigen turn over when targeted by bispecific antibodies (as outlined in Fig. 1G).

Expanding beyond model cell lines: regulation of internalization dynamics in tumor cell by a guide/effector-based bispecific

We next studied bispecific-induced surface antigen dynamics in a panel of pancreatic cancer cell lines with varying guide-to-effector ratios. We first determined cell surface antigen density of ALCAM and EphA2 by quantitative FACS (Supplementary Table S1). ALCAM is highly expressed by those cells, and the guide-to-effector (EphA2 to ALCAM) ratio for L3.6pl, Capan-1, and Panc-1 was estimated to be 0.31, 0.23, and 0.08, respectively. We next performed two sets of experiments to determine (i) how the noninternalizing ALCAM is converted into an internalizing antigen by the bispecific with an EphA2-to-ALCAM ratio above the threshold (>0.2); and (ii) how the rapidly internalizing EphA2 is rendered slowly internalizing by the bispecific with an EphA2-to-ALCAM ratio below the threshold (<0.2). The internalization dynamics of EphA2 and ALCAM were studied by measuring surface antigen level by FACS following antibody treatment. With regard to ALCAM, the bispecific 3F1/RYR was effective in removing about 60% cell surface ALCAM in both L3.6pl and Capan-1 cells where the guide-to-effector ratios are >0.2, but ineffective in Panc-1 cells where the ratio is 0.08, suggesting a cell-type selectivity based on the guide-to-effector ratio (Fig. 2A). The noninternalizing monoclonal anti-ALCAM antibody 3F1 did not remove any ALCAM antigen from the cell surface. The control C10/RYR or an antibody mixture of 3F1 and C10/RYR removed about 85% of surface EphA2 (Supplementary Fig. S3C) but failed to remove ALCAM (Fig. 2A), suggesting that ALCAM removal is a bispecific-dependent phenomenon not achievable by oligoclonal antibody mix. The above bispecific effect on antigen internalization was also studied by confocal microscopy. As shown in Fig. 2B, in L3.6pl cells where the EphA2/ALCAM ratio is >0.2 (∼0.31), the anti-ALCAM 3F1 was mostly detected on the cell surface, whereas the bispecific 3F1/RYR was detected mainly in the cytoplasm with some staining of the cell membrane. The control C10/RYR that binds to EphA2 was detected mainly in the cytoplasm, consistent with its ability to induce rapid EphA2 internalization. In contrast, for the Panc-1 cell line that has a EphA2/ALCAM ratio <0.2 (∼0.08), the bispecific 3F1/RYR was detected mainly on the cell surface (Fig. 2B), again suggesting that internalization of the bispecific depends on the EphA2/ALCAM ratio. These data confirm that in our guide-effector bispecific design, the ability of the bispecific to convert a noninternalizing to internalizing effector antigen depends on the guide-to-effector ratio.

Figure 2.

The bispecific 3F1/RYR effectively removes the noninternalizing antigen ALCAM from pancreatic cancer cell surface. A, ALCAM cell surface level post antibody treatment. Pancreatic cancer cell lines (L3.6pl, Capan-1, and Panc-1) were incubated with 3F1, 3F1/RYR, C10/RYR, or a mixture of 3F1 and C10/RYR. Following wash posttreatment, cell surface ALCAM level was determined using an Alexa 647–labeled IgG that binds to a different epitope on ALCAM than 3F1. MFI values were normalized against cells without antibody treatment. **, P < 0.01; ***, P < 0.001. Duplicates. B, Confocal microscopy study of cell-type selective internalization mediated by the bispecific. L3.6pl (EphA2-to-ALCAM ratio > 0.2) and Panc-1 (EphA2-to-ALCAM ratio < 0.2) cells were incubated with 3F1, 3F1/RYR, or C10/RYR, and internalizing antibodies were stained with FITC-labeled anti-human IgG. Scale bar, 20 μm. C, Colocalization of antibodies and macropinocytotic vesicles. L3.6pl cells were incubated with 3F1, 3F1/RYR, or C10/RYR at 100 nmol/L and ND70-TR (TR-Dextran, red) for 2 hours. Antibodies (Ab) were detected by FITC-labeled anti-human IgG (green). Nuclei were labeled with Hoechst 33342 (blue). Scale bar, 10 μm. D, Lysosomal trafficking post internalization. L3.6pl cells were incubated with indicated antibodies (100 nmol/L) for 2 hours. Internalized antibodies (green) and nuclei (blue) were stained as described in C, and lysosomes were detected using rabbit anti-LAMP1 primary IgG, followed by Alexa 647–labeled anti-rabbit IgG (red). Scale bar, 10 μm. E, Retarded EphA2 internalization on Panc-1 cell when targeted by the bispecific. **, P < 0.01; ***, P < 0.001. Duplicates. F, A time course of EphA2 removal from Panc-1 cell surface at 0.5, 1, and 4 hours post antibody treatment.

Figure 2.

The bispecific 3F1/RYR effectively removes the noninternalizing antigen ALCAM from pancreatic cancer cell surface. A, ALCAM cell surface level post antibody treatment. Pancreatic cancer cell lines (L3.6pl, Capan-1, and Panc-1) were incubated with 3F1, 3F1/RYR, C10/RYR, or a mixture of 3F1 and C10/RYR. Following wash posttreatment, cell surface ALCAM level was determined using an Alexa 647–labeled IgG that binds to a different epitope on ALCAM than 3F1. MFI values were normalized against cells without antibody treatment. **, P < 0.01; ***, P < 0.001. Duplicates. B, Confocal microscopy study of cell-type selective internalization mediated by the bispecific. L3.6pl (EphA2-to-ALCAM ratio > 0.2) and Panc-1 (EphA2-to-ALCAM ratio < 0.2) cells were incubated with 3F1, 3F1/RYR, or C10/RYR, and internalizing antibodies were stained with FITC-labeled anti-human IgG. Scale bar, 20 μm. C, Colocalization of antibodies and macropinocytotic vesicles. L3.6pl cells were incubated with 3F1, 3F1/RYR, or C10/RYR at 100 nmol/L and ND70-TR (TR-Dextran, red) for 2 hours. Antibodies (Ab) were detected by FITC-labeled anti-human IgG (green). Nuclei were labeled with Hoechst 33342 (blue). Scale bar, 10 μm. D, Lysosomal trafficking post internalization. L3.6pl cells were incubated with indicated antibodies (100 nmol/L) for 2 hours. Internalized antibodies (green) and nuclei (blue) were stained as described in C, and lysosomes were detected using rabbit anti-LAMP1 primary IgG, followed by Alexa 647–labeled anti-rabbit IgG (red). Scale bar, 10 μm. E, Retarded EphA2 internalization on Panc-1 cell when targeted by the bispecific. **, P < 0.01; ***, P < 0.001. Duplicates. F, A time course of EphA2 removal from Panc-1 cell surface at 0.5, 1, and 4 hours post antibody treatment.

Close modal

To assess pathway of internalization and trafficking to the lysosome, we performed confocal microscopy studies using L3.6pl cells. As shown in Fig. 2C, the bispecific 3F1/RYR was colocalized with the macropinocytosis marker, 70 kDa Neutral Dextran (ND70), suggesting the mode of internalization is macropinocytosis. Post internalization, 3F1/RYR and the control C10/RYR were colocalized with the lysosomal marker LAMP1 (Fig. 2D), suggesting that the anti-EphA2 antibody-guided bispecific traffics to the lysosome.

To investigate the other direction of the interconversion between internalization and noninternalization, that is, conversion of a rapidly internalizing antibody into a slowly or noninternalizing antibody, we studied surface removal of EphA2 by the bispecific in the presence of the neighboring noninternalizing antigen ALCAM. As shown in Fig. 2E, on Panc-1 cell where the EphA2-to-ALCAM ratio is <0.2 (∼0.08), EphA2 remains mainly on the cell surface when targeted by the bispecific 3F1/RYR. The control C10/RYR removes EphA2 from the cell surface. The C10/RYR plus 3F1 mixture does not retard EphA2 internalization, suggesting that the phenomenon is dependent on the bispecific. A time course internalization study shows that the 3F1/RYR greatly retarded EphA2 internalization kinetics, whereas the control C10/RYR induced rapid EphA2 internalization (Fig. 2F). These studies demonstrate that EphA2 internalization can be significantly retarded by the bispecific when ALCAM is present on the same cell with amounts driving the EphA2/ALCAM ratio below the threshold level.

The bispecific 3F1/RYR inhibits pancreatic tumorsphere formation

To assess functional consequences of surface antigen removal, we studied effects of the anti-EphA2–guided bsIgG (3F1/RYR) on survival and expansion of pancreatic tumorspheres. Previous studies have shown that cancer cells overexpressing ALCAM aggressively form tumorspheres, suggesting that ALCAM plays a role in tumor clonogenicity (18–20). We therefore studied whether ALCAM removal by our ALCAMxEphA2 bsIgG can inhibit pancreatic tumorsphere formation. We first assessed antigen expression in L3.6pl tumorspheres and found that ALCAM was significantly upregulated on the surface of sphere-forming tumor cells (Fig. 3A). There was no difference of EphA2 surface level between L3.6pl cells in monolayer versus sphere states (Fig. 3A). Following incubation of L3.6pl tumorspheres with antibodies for 2 weeks, the bispecific 3F1/RYR reduced ALCAM surface density by 70%, whereas the anti-ALCAM mAb 3F1 or the control C10/RYR that binds to EphA2 showed no effect on ALCAM surface level (Fig. 3B). We next performed confocal microscopy study to confirm antibody internalization. As shown in Fig. 3C, 3F1/RYR was effectively internalized in L3.6pl sphere–forming cells. In contrast, the monoclonal anti-ALCAM antibody 3F1 showed mainly surface staining (Fig. 3C). The control bispecific C10/RYR was internalized, consistent with its monospecific binding to EphA2 (Fig. 3C). Interestingly, based on the fluorescence signal intensity per cell, a greater amount of the 3F1/RYR was taken up by tumorsphere-forming cells compared with the control bispecific C10/RYR (Fig. 3C, right), suggesting an amplification effect unique to the bispecific. With regard to functional effect on tumor clonogenic activity, we found that L3.6pl sphere number (Fig. 3D) and size (Fig. 3E) was significantly decreased by treatment with 3F1/RYR but not 3F1 or C10/RYR, consistent with previous studies of the role of ALCAM in tumorsphere formation and growth. Thus the bispecific antibody with one arm binding to the internalizing antigen EphA2 can effectively remove the noninternalizing antigen ALCAM from tumor cell surface, resulting in inhibition of pancreatic tumorsphere growth.

Figure 3.

Bispecific-induced cell surface ALCAM removal has an anticlonogenic effect on pancreatic tumorspheres. A, Significant ALCAM upregulation on L3.6pl sphere cells compared with nonsphere tumor cells. Adherent- or sphere-cultured L3.6pl cells were separated into single cells and antigen expression was measured using 3F1 or RYR IgG, followed by Alexa 647–labeled anti-human IgG. The EphA2/ALCAM ratio is approximately 0.15 for sphere and approximately 0.305 for monolayer cells. ***, P < 0.001; N.S., nonsignificant. B, ALCAM removal from the surface of sphere-forming cells by 3F1/RYR. Single-cell population of L3.6pl (200 cells/well) was incubated with indicated antibodies (100 nmol/L) for 2 weeks in an ultralow attachment well plate. Cell surface levels of ALCAM post antibody treatment were determined by FACS. MFI values were normalized against control (no antibody treatment). **, P < 0.01. Duplicates. C, Antibody internalization into L3.6pl spheres. Tumorspheres incubated with indicated antibodies were collected by centrifugation, fixed, and permeabilized for confocal microscopy analysis. Antibodies and nuclei were stained with Alexa 647–labeled anti-human IgG (red) and Hoechst 33342 (cyan), respectively. Scale bar, 10 μm. Intracellular antibody fluorescence intensity was quantified by Image J and shown in the right panel. ***, P < 0.001. D, Inhibition of L3.6pl tumorsphere formation by 3F1/RYR, reduction in number. Tumorsphere numbers (>100 μm) were counted 14 days post antibody treatment (left) with representative well images shown (right). Error bars represent SD of a duplicate. *, P < 0.05. E, Inhibition of L3.6pl tumorsphere formation by 3F1/RYR, reduction in size. **, P < 0.01. Duplicates. Scale bar, 100 μm; Ab, antibody.

Figure 3.

Bispecific-induced cell surface ALCAM removal has an anticlonogenic effect on pancreatic tumorspheres. A, Significant ALCAM upregulation on L3.6pl sphere cells compared with nonsphere tumor cells. Adherent- or sphere-cultured L3.6pl cells were separated into single cells and antigen expression was measured using 3F1 or RYR IgG, followed by Alexa 647–labeled anti-human IgG. The EphA2/ALCAM ratio is approximately 0.15 for sphere and approximately 0.305 for monolayer cells. ***, P < 0.001; N.S., nonsignificant. B, ALCAM removal from the surface of sphere-forming cells by 3F1/RYR. Single-cell population of L3.6pl (200 cells/well) was incubated with indicated antibodies (100 nmol/L) for 2 weeks in an ultralow attachment well plate. Cell surface levels of ALCAM post antibody treatment were determined by FACS. MFI values were normalized against control (no antibody treatment). **, P < 0.01. Duplicates. C, Antibody internalization into L3.6pl spheres. Tumorspheres incubated with indicated antibodies were collected by centrifugation, fixed, and permeabilized for confocal microscopy analysis. Antibodies and nuclei were stained with Alexa 647–labeled anti-human IgG (red) and Hoechst 33342 (cyan), respectively. Scale bar, 10 μm. Intracellular antibody fluorescence intensity was quantified by Image J and shown in the right panel. ***, P < 0.001. D, Inhibition of L3.6pl tumorsphere formation by 3F1/RYR, reduction in number. Tumorsphere numbers (>100 μm) were counted 14 days post antibody treatment (left) with representative well images shown (right). Error bars represent SD of a duplicate. *, P < 0.05. E, Inhibition of L3.6pl tumorsphere formation by 3F1/RYR, reduction in size. **, P < 0.01. Duplicates. Scale bar, 100 μm; Ab, antibody.

Close modal

Potent and cell-type selective in vitro tumor cell killing by bispecific ADC

To explore therapeutic potential of the bispecific-induced amplification of intracellular uptake, we generated monospecific and bispecific ADCs by site-specific conjugation of MC-VC-pab-MMAF (Supplementary Fig. S4), analyzed conjugation products by HIC-HPLC, and determined the DAR (∼1.9; refs. 14, 17, 21). ADCs were tested on a panel of cancer cell lines that display different levels of cell surface EphA2 and ALCAM, and EphA2-to-ALCAM ratios (Supplementary Table S1). The 3F1/RYR ADC showed potent cytotoxicity with EC50 of 23 pmol/L on L3.6pl and 22 pmol/L on Capan-1 cells (Fig. 4A and B; Supplementary Table S2). These two cell lines have EphA2-to-ALCAM ratios above the threshold (0.2, Supplementary Table S1), resulting in more efficient internalization. In contrast, both 3F1 and C10/RYR ADCs showed lower potency. EC50 values of 3F1 and C10/RYR ADCs on L3.6pl are 2.37 nmol/L and 0.35 nmol/L, respectively, and on Capan-1 are 0.87 nmol/L and 0.18 nmol/L, respectively (Fig. 4A and B). Most importantly, the bispecific ADC is more potent than the mix of monoclonal ADCs (C10/RYR ADC + 3F1 ADC, Fig. 4A and B), suggesting again that this enhanced potency is unique to the bispecific. We also studied cytotoxicity of ADCs on Panc-1 cells with a low guide-to-effector (EphA2 to ALCAM) ratio and MIA PaCa2 cells without the effector antigen (ALCAM) expression. On Panc-1 cells with a low EphA2/ALCAM ratio (0.08), the bispecific 3F1/RYR ADC showed reduced potency (EC50 = 0.46 nmol/L) but is still more potent than 3F1 (EC50 = 9.3 nmol/L) and C10/RYR ADCs (EC50 > 100 nmol/L; Fig. 4C; Supplementary Table S2). Again, the cytotoxic potency of 3F1/RYR ADC was higher than a mixture of the 3F1 and C10/RYR ADCs (EC50 = 0.46 nmol/L vs. 7.14 nmol/L for the mix; Supplementary Table S2). On the ALCAM-negative MIA PaCa2 cell line, the 3F1 ADC showed little cytotoxicity as expected (Fig. 4D). 3F1/RYR and C10/RYR ADCs showed similarly low cytotoxicity due to the lack of expression of ALCAM and low expression level of EphA2 (Fig. 4D). To further evaluate cell-type selectivity, we studied LNCaP-C4-2B and HEK293 cell lines that express very low levels of EphA2. On LNCaP-C4-2B, as shown in Fig. 4E, the bispecific 3F1/RYR ADC did not show enhanced cytotoxicity versus the monoclonal 3F1 ADC (EC50 = 1.25 nmol/L vs. 1.64 nmol/L; Supplementary Table S2), due to the lack of the guide antigen EphA2 expression, showing a guide antigen–dependent cell-type selectivity. Similar results were obtained from studies using HEK293 cells that lack the guide antigen expression (Supplementary Fig. S5). Taken together, these data show that the bispecific ADC is more potent than monospecific ADCs or their mixture, and exhibits a cell-type selective potency enhancement depending on the guide-to-effector antigen ratio.

Figure 4.

In vitro potency and selectivity of ADCs with site-specific conjugation on tumor cell lines with varying EphA2/ALCAM ratios. Cytotoxicity of indicated ADCs or a mixture was studied on L3.6pl (A) and Capan-1 (B) cell lines with relatively high EphA2/ALCAM ratios, and Panc-1 (C) cell line that has a low EphA2/ALCAM ratio. MIA PaCa2 (D) and C4-2B (E) cell lines were utilized as an ALCAM-low/negative and EphA2-low/negative cancer cell model, respectively. Cell viability (%) was normalized against a control group without ADC treatment.

Figure 4.

In vitro potency and selectivity of ADCs with site-specific conjugation on tumor cell lines with varying EphA2/ALCAM ratios. Cytotoxicity of indicated ADCs or a mixture was studied on L3.6pl (A) and Capan-1 (B) cell lines with relatively high EphA2/ALCAM ratios, and Panc-1 (C) cell line that has a low EphA2/ALCAM ratio. MIA PaCa2 (D) and C4-2B (E) cell lines were utilized as an ALCAM-low/negative and EphA2-low/negative cancer cell model, respectively. Cell viability (%) was normalized against a control group without ADC treatment.

Close modal

In vivo antitumor efficacy of the ALCAMxEphA2 bispecific ADC

We next studied in vivo efficacy of the bispecific 3F1/RYR ADC along with control ADCs on pancreatic cancer xenografts. Capan-1 cells were implanted subcutaneously into NSG mice. When the tumor reached an average volume of 110 mm3, 3F1/RYR, 3F1, or C10/RYR ADCs were injected at 3 mg/kg every 4 days for a total of four times. Tumor status was monitored by caliper measurement. Overt toxicity was monitored by body weight loss. As shown in Fig. 5A, the bispecific 3F1/RYR ADC significantly inhibited tumor growth, whereas the monoclonal 3F1 ADC or the control bispecific C10/RYR ADC had only a moderate effect on tumor size reduction. There was no significant change of body weights during the course of the study for any of the ADCs studied (Fig. 5B). These data demonstrate that in our guide-effector bispecific design, the rapidly internalizing anti-guide (EphA2) scFv can induce internalization of the otherwise noninternalizing effector antigen (ALCAM), resulting in greater amount of bispecific ADCs taken into the tumor cells compared with monospecific ADCs, thus showing enhanced antitumor efficacy in vivo.

Figure 5.

Antitumor efficacy of the bispecific 3F1/RYR ADC in a pancreatic cancer xenograft model. A, Effect on tumor growth. Mice were inoculated subcutaneously with 1 × 106 Capan-1 cells and randomly divided into four groups (6 mice/group) with similar average tumor size. Vehicle (PBS) or ADCs (3 mg/kg) were intravenously injected at indicated time points (arrowheads). The mean tumor volumes ± SEM (mm3) were plotted. B, Body weight was monitored and plotted to assess toxicity of ADC treatment. No significant body weight loss (e.g., >15%) was seen for any of the groups studied.

Figure 5.

Antitumor efficacy of the bispecific 3F1/RYR ADC in a pancreatic cancer xenograft model. A, Effect on tumor growth. Mice were inoculated subcutaneously with 1 × 106 Capan-1 cells and randomly divided into four groups (6 mice/group) with similar average tumor size. Vehicle (PBS) or ADCs (3 mg/kg) were intravenously injected at indicated time points (arrowheads). The mean tumor volumes ± SEM (mm3) were plotted. B, Body weight was monitored and plotted to assess toxicity of ADC treatment. No significant body weight loss (e.g., >15%) was seen for any of the groups studied.

Close modal

In this study we adopted the guide-effector bispecific antibody design that we developed previously for cell-type–specific signaling modulation (8), and achieved cell-type–specific modulation of internalization. We show that when the guide-to-effector ratio crosses over a threshold (in our EphA2/ALCAM example, 1:5), a noninternalizing antigen (ALCAM) can be rendered internalizing by the bispecific. When the guide-to-effector ratio falls below the threshold, an internalizing antigen (EphA2) can be rendered non- or slowly internalizing by the bispecific. Thus in the context of bispecific targeting, the internalization behavior of a cell surface antigen is significantly impacted by its neighboring antigens, and can be readily manipulated in either direction through bispecific-based targeting of appropriately selected guide/effector pairs.

Our study has relevance to therapeutic development. In the direction of converting a noninternalizing into an internalizing antigen, our study has direct relevance to ADC development. ADC is a class of anticancer agents that utilize the specificity of the antibody to deliver cytotoxic drug to tumor cells. Although appealing in concept, clinical development of this class of anticancer agents has encountered various challenges. So far, there are only four ADCs that are approved by the FDA for clinical use. Early problems such as drug and linker stability have been addressed but other issues remain (1). Although very potent drugs such as DNA chelators have been used for ADC generation, those drugs cause accumulative toxicity, restricting the therapeutic window (22). Microtubule inhibitors such as auristatin derivatives are less potent compared with DNA chelators and their toxicity is not accumulative except for peripheral nerve damage. Because of the reduced potency of auristatin and the limited amount of drug delivered into tumor cell, the therapeutic window remains narrow (1). Increasing DAR can result in more drug molecules delivered to a tumor cell in vitro, but in vivo ADCs with high DARs are rapidly cleared from the circulation, thus reducing efficacy and increasing toxicity. Site-specific conjugation achieves near uniform DAR (n = 2), improves pharmacokinetics, but the total number of drug molecules delivered into a tumor cell remains limited (21, 23). In principle, ways to improve the therapeutic window of ADC include: (i) increase cell surface target density; and (ii) improve target internalization. Both should result in a higher number of ADCs delivered intracellularly into the tumor cell. We have previously explored macropinocytosing antibodies for ADC construction to improve internalization (16, 24). In this article, we describe our approach to increase target density through a guide-effector bispecific design.

We used a rapidly internalizing macropinocytosing anti-EphA2 (the guide) antibody and a non/slowly internalizing anti-ALCAM (the effector) antibody as a model system to study the bispecific effect. We found that when the antigen density ratio of EphA2/ALCAM is greater than a threshold (1:5 in our experimental system), a bispecific anti-ALCAMxEphA2 antibody can induce internalization of both EphA2 and ALCAM. In other words, the bispecific can turn a noninternalizing antigen (ALCAM) to an internalizing antigen. We found that the bispecific ADC is more potent than either of the monospecific ADCs and furthermore the mixture of these ADCs in in vitro cytotoxicity assays, consistent with increased amount of internalized ADC delivered by the bispecific. Thus there is an amplification effect that is unique to the bispecific but not the monospecific antibodies or their mix, where a small number of internalizing antigen (the guide, EphA2), upon targeted by the bispecific, can induce the internalization of a large number of noninternalizing antigen (the effector, ALCAM), resulting in greater amount of ADC and drug molecule delivered to the tumor cell compared with monoclonal ADCs and their mixture.

In addition to enhanced potency through amplified internalization, our study has implication for expanding the range and type of cell surface targets for ADC. A key challenge for current ADC is how to deliver payload specifically and in high amount to target cells. In a mAb setting, the target antigen needs to be expressed both specifically and at a uniformly high level on the tumor surface. In practice, however, antigen with both absolute specificity and uniformly high level of expression is rare to find. As such, lineage markers, which are expressed by the tissue from which the tumor is derived, have often been used for tumor targeting (1). There are two limitations for those lineage markers: (i) they tend to show decreased or heterogeneous expression in late-stage cancers as they are not functionally required for tumor survival. For example, PSMA expression in late-stage prostate cancer is heterogeneous and is downregulated in androgen-signaling inhibitor-resistant small cell-type (10). (ii) They are often expressed in more than one normal tissue type. For example, while mesothelin is expressed by a number of tumors such as mesothelioma, ovarian cancer, and pancreatic cancer, it is also expressed by the normal mesothelium (25). PSMA is expressed by prostate tumor but also by a number of normal tissues (26, 27). Likewise, CD19 is expressed by normal tissues other than B cells (28). It seems that by the mAb approach, the target selection is rather restricted or suboptimal. In the context of ADC, efforts have been directed to increase potency of the payload but the therapeutic window remains narrow, as discussed above. An alternative approach is to identify targets that expand the difference between payloads delivered to tumor versus normal cells. Our work is particularly relevant to this approach as our guide-effector bispecific design allows a large number of noninternalizing tumor-associated antigens to become internalized and thus contributing to increased intracellular delivery of ADC. The amplification effect is unique to tumor cells due to coexpression of both the guide and effector antigens.

We have previously described the guide-effector bispecific design for cell-type–specific signaling pathway modulation (8). This study expands the applicability of the bispecific approach to antigen internalization and ADC. The essence of guide-effector bispecific system is that the behavior of a given antigen (effector) can be shaped by the neighboring antigen (guide) when the ratio of guide-to-effector exceeds a threshold level. In the Wnt signaling study, when the guide/effector ratio exceeds 5–10:1, there is a 1,000-fold increase in potency of the bispecific over the mAb and the enhancement is cell-type specific. In this study, we showed that when the guide/effector ratio exceeds 1:5, a small number of the guide antigen (internalizing) can convert a large number of effector antigen (noninternalizing) to internalizing antigen. In both of our studies, the guide-to-effector ratio (or more precisely, the occupied guide-to-effector ratio) has the most significant impact, although other factors such as antibody affinity, epitope geometry, and membrane microdomain all could affect the outcome depending on the exact antibody and guide/effector pairs chosen.

Interestingly, although our study is focused on ADC and conversion of a noninternalizing antibody into an internalizing antibody, we showed that the reverse is true: when the ratio of internalizing to noninternalizing antigen is below the threshold level (in our system 1:5), the internalizing antigen EphA2 is rendered slowly internalizing by the presence of the noninternalizing ALCAM. This could be useful for applications where it is desirable to have antigen remain on the cell surface to prevent degradation, and to prolong signaling functions.

There are a few recent reports of bispecific ADC with the internalizing arm binding to either a lysosomal protein or an antigen that rapidly traffics to the lysosome (5–7). In most cases, the observation is empirical and the bispecific effect is rather moderate, suggesting that key parameters affecting bispecific-induced internalization have not been fully delineated. For example, it is unclear whether lysosomal antigen is required for this phenomenon. It is also unclear why the bispecific works on some cells but not the other. Our study shows that a key variable in the bispecific design is the ratio of the guide-to-effector antigen, and there is no special feature beyond internalization that is required for the internalizing arm. The guide antigen (the internalizing arm) needs not to be a lysosomal protein to induce internalization and lysosomal trafficking. For example, in this study we exploited macropinocytosis and selected the macropinocytosing antibody against the cell surface antigen, EphA2, as the guide to route the bispecific to the lysosomal compartment.

In summary, we show that in the context of bispecific targeting, internalization is no longer an intrinsic property of a given antigen. Instead, antigen internalization is heavily influenced by its neighboring antigens and can be readily manipulated in either direction in a cell-type–specific manner using properly selected guide/effector pairs. This bispecific-induced plasticity of cell surface dynamics can be exploited for therapeutic development.

B. Liu has ownership interest (including stock, patents, etc.) in Vivace Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N.-K. Lee, Y. Su, S. Bidlingmaier, B. Liu

Development of methodology: N.-K. Lee, Y. Su, S. Bidlingmaier, B. Liu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N.-K. Lee, Y. Su, S. Bidlingmaier, B. Liu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N.-K. Lee, Y. Su, S. Bidlingmaier, B. Liu

Writing, review, and/or revision of the manuscript: N.-K. Lee, Y. Su, S. Bidlingmaier, B. Liu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N.-K. Lee, Y. Su, B. Liu

Study supervision: B. Liu

We thank Dr. Byron C. Hann for advice and Donghui Wang, Paul Phojanakong, Fernando Salangsang, and Julia Malato at the UCSF Preclinical Therapeutic Core for performing animal studies. This research was supported by the NIH (R01 CA118919, CA129491, and CA171315 to B. Liu) and a fellowship grant to N.-K. Lee from the Basic Science Research Program through the National Research Foundation of Korea funded by the Ministry of Education, Science and Technology (2013R1A6A3A03060495).

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.

1.
de Goeij
BE
,
Lambert
JM
. 
New developments for antibody-drug conjugate-based therapeutic approaches
.
Curr Opin Immunol
2016
;
40
:
14
23
.
2.
Polakis
P
. 
Antibody drug conjugates for cancer therapy
.
Pharmacol Rev
2016
;
68
:
3
19
.
3.
Gebleux
R
,
Stringhini
M
,
Casanova
R
,
Soltermann
A
,
Neri
D
. 
Non-internalizing antibody-drug conjugates display potent anti-cancer activity upon proteolytic release of monomethyl auristatin E in the subendothelial extracellular matrix
.
Int J Cancer
2017
;
140
:
1670
9
.
4.
Li
JY
,
Perry
SR
,
Muniz-Medina
V
,
Wang
X
,
Wetzel
LK
,
Rebelatto
MC
, et al
A biparatopic HER2-targeting antibody-drug conjugate induces tumor regression in primary models refractory to or ineligible for HER2-targeted therapy
.
Cancer Cell
2016
;
29
:
117
29
.
5.
de Goeij
BE
,
Vink
T
,
Ten Napel
H
,
Breij
EC
,
Satijn
D
,
Wubbolts
R
, et al
Efficient payload delivery by a bispecific antibody-drug conjugate targeting HER2 and CD63
.
Mol Cancer Ther
2016
;
15
:
2688
97
.
6.
Andreev
J
,
Thambi
N
,
Perez Bay
AE
,
Delfino
F
,
Martin
J
,
Kelly
MP
, et al
Bispecific antibodies and antibody-drug conjugates (ADCs) bridging HER2 and prolactin receptor improve efficacy of HER2 ADCs
.
Mol Cancer Ther
2017
;
16
:
681
93
.
7.
DeVay
RM
,
Delaria
K
,
Zhu
G
,
Holz
C
,
Foletti
D
,
Sutton
J
, et al
Improved lysosomal trafficking can modulate the potency of antibody drug conjugates
.
Bioconjug Chem
2017
;
28
:
1102
14
.
8.
Lee
NK
,
Zhang
Y
,
Su
Y
,
Bidlingmaier
S
,
Sherbenou
DW
,
Ha
KD
, et al
Cell-type specific potent Wnt signaling blockade by bispecific antibody
.
Sci Rep
2018
;
8
:
766
.
9.
Bruns
CJ
,
Harbison
MT
,
Kuniyasu
H
,
Eue
I
,
Fidler
IJ
. 
In vivo selection and characterization of metastatic variants from human pancreatic adenocarcinoma by using orthotopic implantation in nude mice
.
Neoplasia
1999
;
1
:
50
62
.
10.
Su
Y
,
Liu
Y
,
Behrens
CR
,
Bidlingmaier
S
,
Lee
NK
,
Aggarwal
R
, et al
Targeting CD46 for both adenocarcinoma and neuroendocrine prostate cancer
.
JCI Insight
2018
;
3
:
e121497
.
11.
Lee
NK
,
Bidlingmaier
S
,
Su
Y
,
Liu
B
. 
Modular construction of large non-immune human antibody phage-display libraries from variable heavy and light chain gene cassettes
.
Methods Mol Biol
2018
;
1701
:
61
82
.
12.
Rogosch
T
,
Kerzel
S
,
Hoi
KH
,
Zhang
Z
,
Maier
RF
,
Ippolito
GC
, et al
Immunoglobulin analysis tool: a novel tool for the analysis of human and mouse heavy and light chain transcripts
.
Front Immunol
2012
;
3
:
176
.
13.
Smith
K
,
Garman
L
,
Wrammert
J
,
Zheng
NY
,
Capra
JD
,
Ahmed
R
, et al
Rapid generation of fully human monoclonal antibodies specific to a vaccinating antigen
.
Nat Protoc
2009
;
4
:
372
84
.
14.
Sherbenou
DW
,
Aftab
BT
,
Su
Y
,
Behrens
CR
,
Wiita
A
,
Logan
AC
, et al
Antibody-drug conjugate targeting CD46 eliminates multiple myeloma cells
.
J Clin Invest
2016
;
126
:
4640
53
.
15.
Orcutt
KD
,
Ackerman
ME
,
Cieslewicz
M
,
Quiroz
E
,
Slusarczyk
AL
,
Frangioni
JV
, et al
A modular IgG-scFv bispecific antibody topology
.
Protein Eng Des Sel
2010
;
23
:
221
8
.
16.
Ha
KD
,
Bidlingmaier
SM
,
Zhang
Y
,
Su
Y
,
Liu
B
. 
High-content analysis of antibody phage-display library selection outputs identifies tumor selective macropinocytosis-dependent rapidly internalizing antibodies
.
Mol Cell Proteomics
2014
;
13
:
3320
31
.
17.
Junutula
JR
,
Raab
H
,
Clark
S
,
Bhakta
S
,
Leipold
DD
,
Weir
S
, et al
Site-specific conjugation of a cytotoxic drug to an antibody improves the therapeutic index
.
Nat Biotechnol
2008
;
26
:
925
32
.
18.
Jiao
J
,
Hindoyan
A
,
Wang
S
,
Tran
LM
,
Goldstein
AS
,
Lawson
D
, et al
Identification of CD166 as a surface marker for enriching prostate stem/progenitor and cancer initiating cells
.
PLoS One
2012
;
7
:
e42564
.
19.
Yan
M
,
Yang
X
,
Wang
L
,
Clark
D
,
Zuo
H
,
Ye
D
, et al
Plasma membrane proteomics of tumor spheres identify CD166 as a novel marker for cancer stem-like cells in head and neck squamous cell carcinoma
.
Mol Cell Proteomics
2013
;
12
:
3271
84
.
20.
Moon
BS
,
Jeong
WJ
,
Park
J
,
Kim
TI
,
Min
do S
,
Choi
KY
. 
Role of oncogenic K-Ras in cancer stem cell activation by aberrant Wnt/beta-catenin signaling
.
J Natl Cancer Inst
2014
;
106
:
djt373
.
21.
Behrens
CR
,
Liu
B
. 
Methods for site-specific drug conjugation to antibodies
.
MAbs
2014
;
6
:
46
53
.
22.
Fu
Y
,
Ho
M
. 
DNA damaging agent-based antibody-drug conjugates for cancer therapy
.
Antib Ther
2018
;
1
:
33
43
.
23.
Panowski
S
,
Bhakta
S
,
Raab
H
,
Polakis
P
,
Junutula
JR
. 
Site-specific antibody drug conjugates for cancer therapy
.
MAbs
2014
;
6
:
34
45
.
24.
Ha
KD
,
Bidlingmaier
SM
,
Su
Y
,
Lee
NK
,
Liu
B
. 
Identification of novel macropinocytosing human antibodies by phage display and high-content analysis
.
Methods Enzymol
2017
;
585
:
91
110
.
25.
Chang
K
,
Pastan
I
. 
Molecular cloning of mesothelin, a differentiation antigen present on mesothelium, mesotheliomas, and ovarian cancers
.
Proc Natl Acad Sci U S A
1996
;
93
:
136
40
.
26.
Silver
DA
,
Pellicer
I
,
Fair
WR
,
Heston
WD
,
Cordon-Cardo
C
. 
Prostate-specific membrane antigen expression in normal and malignant human tissues
.
Clin Cancer Res
1997
;
3
:
81
5
.
27.
Kinoshita
Y
,
Kuratsukuri
K
,
Landas
S
,
Imaida
K
,
Rovito
PM
 Jr
,
Wang
CY
, et al
Expression of prostate-specific membrane antigen in normal and malignant human tissues
.
World J Surg
2006
;
30
:
628
36
.
28.
Wang
K
,
Wei
G
,
Liu
D
. 
CD19: a biomarker for B cell development, lymphoma diagnosis and therapy
.
Exp Hematol Oncol
2012
;
1
:
36
.