Development of a Tetravalent Anti-GPA33/Anti-CD3 Bispecific Antibody for Colorectal Cancers

This article is featured in Highlights of This Issue, p. 2077

Colorectal cancer is the third leading cause of death among cancers in the United States (1) and accounts for 10% of all cancers in men and 9.2% in women worldwide (2). Although there has been steady yearly 3% decline of incidences from 2004 to 2013, around 135,000 new cases are expected annually in the United States alone. Although localized and regional diseases are often curable, the prognosis of metastatic colorectal cancers (mCRC) is poor, with a 5-year survival rate of only 14% (1).

The standard treatment of mCRC consists of chemotherapy in combination with monoclonal antibodies that block tumor signaling or angiogenesis. Currently, four monoclonal antibody drugs have been approved by the FDA for colorectal cancer treatment. Bevacizumab and ramucirumab target the VEGF–VEGFR angiogenesis pathway, whereas cetuximab and panitumumab target the EGFR pathway. However, cetuximab and panitumumab do not provide clinical benefits in mCRC patients with RAS mutations (3), which account for around 40% of mCRC (4, 5). Moreover, improvement in survival with these antibodies is generally modest and can be associated with severe side effects (5–7).

Given the clinical success in melanoma, renal cell carcinoma, and lung cancer, immune checkpoint inhibitors (ICI) have revolutionized modern immunotherapy and have become the backbones of numerous ongoing trials in a broad spectrum of adult cancers (8, 9). Unfortunately, for colorectal cancer, benefits so far have been restricted to the subset of patients with microsatellite instability (MSI), while the majority of mCRC are microsatellite stable (MSS) and are not expected to benefit from ICI monotherapy (10–12).

More effective strategies to take advantage of T cells to treat mCRC are needed. T-cell–engaging bispecific antibodies (T-BsAb), which function by actively redirecting polyclonal cytotoxic T cells to kill cancer cells, are promising candidates. Their efficacy in leukemia has been proven for blinatumomab, an anti-CD19 T-BsAb approved by the FDA in 2014. For solid tumors, numerous preclinical studies have documented their potent antitumor activities and several are in clinical trials.

We hypothesized that T-BsAbs could expand the potential of ICIs and overcome genetic barriers in the treatment of mCRC. We now report a T-BsAb that targets the protein GPA33, an antigen expressed in >95% of colon cancers and a subset of gastric cancers (13). This target has been extensively explored in radioimmunodiagnosis and radioimmunotherapy, notably using a murine-derived antibody A33 against GPA33 (14–17) and its humanized version, which have been relatively well tolerated (15, 18, 19), despite the presence of human anti-mouse antibodies (HAMA; refs. 18, 19), and human antihuman antibodies (HAHA; refs. 15, 20). Using a rehumanized version of the mouse A33, we successfully built a tetravalent T-BsAb and characterized its in vitro physicochemical properties and functional activities, as well as its in vivo antitumor properties in multiple xenograft models. We showed that its efficacy was independent of the MSI status of colorectal cancer tumors, a necessary attribute if this T-BsAb is meant to function against all mCRCs.

Cell lines used in this study and their sources were summarized in Supplementary Table S1. Luciferase-expressing cell lines were established using lentiviral transduction. All cells were authenticated by STR typing within 6 months of experiments. Cells were maintained in RPMI medium supplemented with 10% FBS (Sigma), 0.03% l-glutamine (Gibco), and penicillin–streptomycin (Gibco). Buffy coats from healthy donors were purchased from New York Blood Center, and human peripheral blood mononuclear cells (PBMC) were isolated by Ficoll gradient of Buffy coats.

Size purity of T-BsAb was analyzed using HPLC system (Shimadzu Scientific Instruments Inc.). Monomeric species was identified using molecular weight standard (Bio-Rad) and percentage of monomers was calculated based on the relative area under curve (AUC) of different non-buffer peaks.

Using CDR grafting, mouse A33 was humanized as IgG1. Two different heavy chain variable (VH) and light chain variable (VL) sequences were combined to generate 4 different humanized A33 antibodies. Binding kinetics was compared with that of chimeric antibody chA33 using surface plasmon resonance (SPR) analysis. The heavy chain of selected sequence is

EVQLVESGGGLVKPGGSLRLSCAASGFAFSTYDMSWVRQAPGKRLEWVATISSGGSYTYYLDSVKGRFTISRDNAKNSLYLQMNSLRAEDTAVYYCAPTTVVPFAYWGQGTLVTVSS

and the light chain sequence is

DIQMTQSQSSLSTSVGDRVTITCKASQNVRTVVAWYQQKPGKSPKTLIYLASNRHTGVPSRFSGSGSGTEFTLTISNVQPEDFADYFCLQHWSYPLTFGSGTKLEIKR.

HuA33-BsAb was constructed by fusing the humanized OKT3 scFv onto the C-terminus of the light chain of huA33 antibody via a (G₄S)₃ linker as previously described (21, 22). The DNA construct was then transfected into CHO-S cells and stable clones were selected for high levels of antibody production. For larger-scale antibody purification, selected stable clone was expanded in shaker flasks. Bispecific antibody was purified from supernatant using one-step protein A affinity chromatography.

Human GPA33 (Novoprotein) was immobilized on CM5 chips. Five concentrations of 2-fold serially diluted huA33 IgG1 or T-BsAbs (starting at 20 nmol/L) were flowed over the chip using a Biacore T100 system. Binding kinetics of huA33 was measured at 25°C and those of T-BsAbs at both 25°C and 37°C. The sensorgrams were fitted with a 1:1 binding model for both to derive kinetic parameters.

IHC staining of human tissue was done using ABC immunoperoxidase (Vector Laboratories). Five-micrometer thick sections were fixed in cold acetone for 30 minutes at −20°C, followed by rinsing in phosphate-buffered saline (PBS), and immersed in 0.1% hydrogen peroxide for 15 minutes at room temperature (RT). All the following steps were performed at RT. Avidin and biotin solutions were added sequentially for 20 minutes each. After blocking with 10% horse serum in PBS for 1 hour, each subsequent step was followed by washing of slides with PBS. Primary huA33-BsAb or control T-BsAb was added to the tissue sections at 1.0 μg/mL for 45 minutes, followed by anti-OKT3-mFc at 0.1 μg/mL for 30 minutes. Lastly, biotinylated horse anti-mouse IgG (H + L) antibody at 1:500 dilution was added for 30 minutes, followed by 100 μL of ABC for 30 minutes. Peroxidase substrate was then added to each tissue section for 2 minutes, rinsed with running tap water for 5 minutes, and counterstained with Myer's hematoxylin. Stained slides were dehydrated by dipping sequentially in 75%, 95%, and 100% ethanol, and finally in xylene. Dehydrated slides were mounted with one drop of Cytoseal (Richard-Allan Scientific) for examination.

Human CD3 IHC was performed at the Molecular Cytology Core Facility of Memorial Sloan Kettering Cancer Center (MSKCC) using the Discovery XT processor (Ventana Medical Systems). Anti-CD3 antibody (DAKO; cat. #A0452, 1.2 μg/mL) was applied to deparaffinized sections and incubated for 5 hours, followed by 60 minutes of incubation with biotinylated goat anti-rabbit IgG (Vector Labs, cat. #PK6101) and detected with DAB detection kit (Ventana Medical Systems) according to the manufacturer's instruction. Slides were counterstained with hematoxylin and coverslipped with Permount (Fisher Scientific). All images were captured using Nikon ECLIPSE Ni-U microscope and NIS-Elements 4.0 imaging software (Nikon Instruments, Inc.).

Cytotoxicity assays were performed using both ⁵¹Cr-release assay and LDH-release assay (Pierce). For both assays, we used anti-CD3/anti-CD28 Dynabeads activated T cells (ATC) for 14 days as effector cells, except for sorted cells from PBMCs, which were used for TDCC assay without prior stimulation. ⁵¹Cr assay was performed as previously described (23). LDH assay was done following the manufacturer's instructions. EC₅₀ values were calculated by fitting the curves to a 4-parameter nonlinear regression model using GraphPad Prism.

T cells were purified from PBMCs using pan–T-cell isolation kit (Miltenyi Biotec). Target and control tumor cells were incubated with 2 × 10⁶ cells/well at an effector-to-target ratio of 5:1 in 24-well plates in triplicates, with 2 mL total volume per well. Culture supernatants were collected at 24, 48, 72, and 96 hours, and cytokine levels were measured with flow cytometry using LEGENDplex human CD8/NK Panel (BioLegend), following the manufacturer's protocol.

Fresh PBMCs from healthy donors were labeled with 2.5 μmol/L CFSE (Life Technologies) for 5 minutes at RT, followed by neutralization using PBS containing 5% FBS. Target cells were then incubated with labeled PBMCs under different conditions at 37°C before analyzing expression of surface activation markers and T-cell proliferation at 24 and 96 hours.

All experiments have been conducted in accordance with and approved by the Institutional Animal Care and Use Committee in MSKCC. For subcutaneous (s.c.) tumor models, 5 × 10⁶ LS174T, 3 × 10⁶ Colo205, or 5 × 10⁶ SNU16 cells were combined with fresh PBMCs at a 1:1 ratio, mixed with Matrigel (volume of cell:gel = 1:2), and implanted with 100 μL/mouse at the flank of Balb/c Rag2^−/−IL2Rγ^−/− (DKO) mice (derived from colony of Dr. Mamoru Ito, CIEA, Kawasaki, Japan, and now commercially available from Taconic as CIEA BRG mice). Treatment started after confirmation of tumor presence with a twice a week (BIW) schedule at 100 μg/mouse and continued for 3 to 4 weeks. Tumor growth was monitored by weekly measurement of tumor volume using a caliper or a digital device Peira TM900 Scanner (Peira Scientific Instruments). When caliper was used, the tumor volume was calculated as length × width × width/2.

For intraperitoneal (i.p.) tumor models, 3 × 10⁶ or 1 × 10⁶ luciferase-expressing LS174T-luc or SW1222-luc cells, respectively, were resuspended in RPMI medium and injected i.p. into DKO mice. Human effector cells were injected i.v. as 20 × 10⁶ ATC per dose for LS174T-luc tumors or 5 × 10⁶ ATC per dose for SW1222-luc tumors, whereas antibody was injected at 50 μg/mouse/dose for LS174T-luc tumors or 100 μg/mouse/dose for SW1222-luc tumors. Mice were treated with 2 to 3 weekly cycles, with each component separated by 3 to 4 days. Growth of tumors was followed weekly by measuring luminescence signals on an IVIS Spectrum in vivo imaging system (PerkinElmer) after injection of 3 mg/mouse of luciferin. Luminescence signals were analyzed and quantified using Living Image Software (PerkinElmer).

Anti-hCD25-PE, anti-CD69-PE, anti-hCD8-APC, anti-hCD45-PECy7, anti-hCD4-PE, anti-hCD4-BV421, anti-hCD62L-PercpCy5.5, anti-hPD1-BV421, and anti-hCD45RO-PECy7 were from BioLegend. Goat antihuman IgG-PE was from SouthernBiotech. Streptavidin-PE, anti-hCD4-APC, and anti-hCD25-APC were from BD Biosciences. All FACS analyses were carried out using FACSCalibur or LSRII system (BD Biosciences) and analyzed using FlowJo (FlowJo, Inc.).

Parental huA33 was converted into scFv format with a 20–amino acid (G₄S)₄ linker and cloned into a yeast display vector. HuA33 scFv was randomly mutated using GeneMorph II mutagenesis kit (Agilent Technologies). PCR product was electroporated together with linearized vector into yeast and the library was subjected to 4 rounds of sorting using biotinylated GPA33. Individual clones from the last round were PCR amplified and sequenced to analyze the mutation pattern. Conversion of selected scFv clones into T-BsAb format was done using a one-step 4-fragment ligation method with 50 ng linearized vector and a 1:3 vector to insert molar ratio for the other 3 components. Ligation was done with a Rapid DNA ligation kit (Thermo Fisher Scientific) at RT for 1 hour. Type II restriction enzyme SapI (NEB) was used to ensure seamless linkage among the different components (Supplementary Fig. S3). Selected clones were transiently expressed using the Expi293 expression system (Thermo Fisher Scientific) following the manufacturer's instructions. Supernatant from Expi293 cells after 4 to 5 days of culture in shaking flasks was used to purify antibodies using MabSelect SuRe (GE Healthcare) and dialyzed against pH8.0 citrate buffer in dialysis membrane (SpectrumLabs.com).

Differences in tumor sizes were tested using two-tailed Student t test. Kaplan–Meier survival analysis and log-rank test were used to compare survival curves. When P < 0.05, the differences were considered statistically significant.

When compared with chimeric A33, all four newly humanized A33 antibodies have slightly improved k_off in binding to immobilized GPA33 in SPR analysis (Fig. 1A). Based on the K_D, stability at 37°C and T20 humanness score (24), one huA33 clone was chosen for further development.

The huA33 antibody was reformatted into the 2+2 bispecific format (25) by fusing scFv of humanized OKT3 to the C-terminus of light chain via a flexible GS linker (Fig. 1B). The DNA construct was used to establish a CHO-S stable cell line producing 50 mg/L to 100 mg/L of protein without extensive optimization. Slightly lower yields were observed using Expi293 transient expression system (around 33 mg/L). One-step protein A purification routinely produced protein with purity above 90%, as measured by SEC-HPLC. After incubating the protein at 37°C for 4 weeks, there was only minimal decrease in the percentage of monomers, as shown in Fig. 1C. These data suggest that huA33-BsAb had good solubility, purity, and thermal stability, all of which are critical characteristics for further downstream development.

We measured the avidities of huA33-BsAb toward GPA33 at both 25°C and 37°C using GPA33-immobilized CM5 chips. As shown in Fig. 1D, huA33-BsAb bound GPA33 with a high apparent affinity of around 0.2 nmol/L, which is slightly lower than 0.13 nmol/L obtained for parental huA33. FACS analysis of a panel of cell lines derived from different cancers showed that huA33-BsAb stained colon cancer cell lines and one gastric cancer cell line but not GPA33⁻ neuroblastoma cell line IMR32, osteosarcoma cell line TC32 or melanoma cell line SKMEL5 (Fig. 1E; Supplementary Table S1), suggesting that huA33-BsAb retained the specificity of parental antibody A33 in binding to target antigens on colon cancer cells and a subset of gastric cancer cells. Specific expression of GPA33 on colon tissues was also confirmed by IHC (Supplementary Fig. S1). Staining of ATC also showed that huA33-BsAb bound to CD3 on T-cell surface (Fig. 1E).

To test the ability of huA33-BsAb to activate unstimulated T cells, CFSE-labeled PBMCs were mixed with Colo205 cells at an effector-to-target ratio of 5:1 (E:T = 5:1) and cultured in the presence of huA33-BsAb (1 μg/mL). As negative controls, we used huA33-C825 that carried an irrelevant scFv (26) instead of the anti-CD3 scFv, as well as a control T-BsAb antibody that did not bind to Colo205 by FACS. After 24 and 96 hours, cells were stained with different T-cell activation markers to assess T-cell activation status and proliferation. As early as 24 hours, huA33-BsAb caused activation of both CD4⁺ and CD8⁺ T cells, as shown by the upregulation of CD25 and CD69 markers on cell surface (Fig. 2A). In contrast, huA33-C825 and control T-BsAb caused only minimal upregulation of CD25. Control T-BsAb did increase the expression of CD69, especially in CD4⁺ T cells. Similarly, PD-1 upregulation was observed after 24 hours and persisted until 96 hours (Supplementary Fig. S2A). Cell division, as measured by CFSE dye dilution, was observed in both CD4⁺ and CD8⁺ T cells after 96 hours (Fig. 2B). CD8⁺ T cells showed higher cell division cycles, suggesting that CD8⁺ T cells divided faster than CD4⁺ T cells (Fig. 2C). Control antibody huA33-C825 did not stimulate significant amount of cell division in either T-cell subset, confirming the requirement of CD3 activation for T-cell division. A low level of cell division was induced by control T-BsAb, consistent with the low level of activation observed above. When GPA33⁻ SKMEL5 targets were used, huA33-BsAb did not activate T cells (Supplementary Fig. S2B). These data suggested that activation of T cells by huA33-BsAb depended on the presence of cognate antigens on tumor cells. We also observed that cell division was associated with expansion of CD45RO⁺ effector/memory cells (Fig. 2D), suggesting the importance of this subset in mediating T-BsAb activity, which was confirmed below. These assays were repeated using a different colon cancer cell line LS174T with similar conclusions (Supplementary Fig. S2C).

To see if T cells can be activated in vivo, we conducted an in vivo proliferation assay by mixing CFSE-labeled PBMCs with Colo205 cells and implanted the mixture subcutaneously onto DKO mice. HuA33-BsAb was injected the next day, and the tumors were isolated after another 4 days and analyzed by FACS. As shown in Fig. 2E, around 25% of CD4⁺ and CD8⁺ T cells upregulated CD25 expression while undergoing cell division (progressive halving of CFSE fluorescence), suggesting that huA33-BsAb was able to stimulate T-cell activation and proliferation in vivo as well.

We next examined the secreted cytokine profile of T cells activated by huA33-BsAb in the presence of target tumors. Total T cells were purified from PBMCs and cultured in the presence of Colo205 tumor cells at E:T = 5:1; to estimate nonspecific activation, we used SKMEL5 as a negative control. Cell culture supernatants were collected daily over 4 days and the levels of cytokines and cytotoxic molecules were measured using a flow cytometry–based multiplex method. As shown in Fig. 3, both Th1 cytokines (IL2, IFNγ, and TNFα) and Th2 cytokines (IL4 and IL10) were secreted by ATCs, although IL4 was secreted at a much lower level than other cytokines. Similarly, significant amount of IL6 was secreted by ATCs. Interestingly, Th17 cytokine IL17α was also secreted at high levels. As expected, cytotoxic components sFasL, Granzyme A, Granzyme B, Perforin, and Granulysin were all released in the supernatant.

We next tested if huA33-BsAb could redirect T cells to kill cancer cells. Based on the survey of GPA33 expression on multiple human cancer cell lines (Supplementary Table S1), we selected three colon cancer cell lines (LS174T, SW1222, and Colo205) and one gastric cancer cell line (SNU16) for cytotoxicity studies. These cell lines were classified as MSI (LS174T) subtype or MSS (SW1222, Colo205, SNU16) subtype based on their MSI profile (27–29). Moreover, LS174T cells carried a KRAS G12D mutation while the others carried TP53 mutations or deletions (30–33). Cancer cells were incubated with ATC at E:T = 10:1 in the presence of 10-fold serial dilutions of huA33-BsAb. GPA33⁻ melanoma cell line SKMEL5 and osteosarcoma cell line TC32 were used as negative controls. As shown in Fig. 4A, huA33-BsAb redirected T cells to specifically kill all GPA33-expressing cancer cells regardless of their genetic backgrounds, while sparing SKMEL5 and TC32, confirming the antigen specificity of huA33-BsAb–mediated TDCC. Maximal level of cytotoxicity seemed to correlate with the level of FACS staining (Figs. 1E and 4A). Moreover, TDCC induced by huA33-BsAb was potent, with EC₅₀ values in the picomolar range. We also confirmed the specificity of TDCC mediated by huA33-BsAb by replacing either the huA33 arm with control BsAb or OKT3 arm with irrelevant C825; replacement of either arm completely abolished TDCC activity (Supplementary Fig. S3).

To see which subsets of T cells were mobilized by huA33-BsAb, we sorted CD45RA⁺CD62L⁺ and CD45RA⁻CD45RO⁺CD62L⁻ memory subsets from both CD4⁺ and CD8⁺ T cells. Sorted cells were cultured in the presence of Colo205 tumors cells at E:T = 5:1 for 48 hours before measuring cytotoxicity. As shown in Fig. 4B, both CD4⁺ and CD8⁺ memory T-cell subsets were capable of inducing cytotoxicity, with CD8⁺ memory T cells mediating more efficient killing at higher concentration. Interestingly, CD45RA⁺CD62L⁺ subsets of both CD4⁺ and CD8⁺ populations, with the majority being naïve T cells, were capable of inducing cytotoxicity after 48 hours, although the potency was less compared with memory T cells. When T cells from the TDCC assay were stained with CD45RO and CD25, it was found that CD25 expression was upregulated in the presence of huA33-BsAb, in both CD45RO⁺ and CD45RO⁻ fractions (Supplementary Fig. S4), confirming that both naïve and memory T cells could be activated by T-BsAb. The majority of CD45RA⁺CD62L⁺ T cells stayed CD45RO⁻ after incubation, with a small but significant population increasing their CD45RO expression, especially among the CD8⁺ cells. However, this population could have been derived from either the maturation of naïve T cells or from expansion of rare CD45RO⁺ cells present in the initial culture. Overall, these data demonstrated that huA33-BsAb could induce potent cytotoxicity against colon and gastric cancer cells in a GPA33 dependent manner, by mobilizing both CD4 and CD8 T cells, especially those of the memory phenotype.

In an attempt to further improve the potency of huA33-BsAb, we used yeast display method to affinity mature scFv derived from huA33. Because scFv tends to aggregate easily, we developed a method to rapidly reformat scFv to T-BsAb (Supplementary Fig. S5A), which is a more relevant format and could be readily produced in high yield and purity.

From sequence analysis of 60 single clones, we selected 7 clones (Supplementary Fig. S5B) for further characterization by SPR and TDCC assay. All 7 clones showed increased binding affinity (4.3- to 51-fold) as compared with parental antibody (Supplementary Fig. S5C). The majority of improvement was contributed by slower off-rate. When tested in the TDCC assay, all clones showed slight improvement in maximal killing (Supplementary Fig. S5D); however, no significant enhancement in EC₅₀ was observed (Supplementary Fig. S5D).

Efficacy of huA33-BsAb was tested in vivo with two xenograft models in DKO mice adoptively transferred with human T cells: (i) s.c. tumor plus s.c. effectors and (ii) i.p. tumor plus i.v. effectors.

LS174T cells were mixed with PBMCs at a 1:1 ratio and implanted subcutaneously in DKO mice. As shown in Fig. 5A, without i.v. antibody treatment, both tumor-only and tumor + PBMC groups showed rapid tumor growth. Mice from tumor + PBMC group developed tumor ulceration and had to be euthanized. In contrast, 6 doses of i.v. huA33-BsAb over 3 weeks effectively cured the mice, which remained tumor free for at least 120 days.

To simulate malignant ascites, a common occurrence in colon cancers, luciferase-expressing LS174T cells were planted i.p. into DKO mice. When tumor growth was confirmed by luminescence, mice were randomized into different treatment groups: ATC only, ATC + huA33-C825, ATC + control BsAb, and ATC + huA33-BsAb. Antibody treatment started on day 8 after tumor implantation and consisted of 6 doses i.p. over 3 weeks, whereas T-cell treatment consisted of weekly injection of ATC through retro-orbital route over 3 weeks. As shown in Fig. 5B and C, all three control groups showed exponential growth of tumor in the abdomen. In contrast, i.p. huA33-BsAb significantly suppressed metastatic growth of LS174T tumor in the abdominal areas of treated mice, accompanied by substantial amount of T-cell tumor infiltration (Fig. 5D). All mice in this group remained alive until at least 55 days without further treatment, while all mice in control groups succumbed to tumors by day 43 (Fig. 5B, right). This is consistent with our previous results showing that huA33-C825 had no direct antitumor effect on LS174T tumor (26, 34).

These data suggested that huA33-BsAb was effective against colorectal cancer tumors with an MSI genotype. However, as mentioned before, MSI tumors account for only a minority of colorectal cancer patients. A majority of colorectal cancer patients are MSS. Therefore, efficacy of huA33-BsAb was tested further using MSS tumors.

Two MSS colon cancer cell lines, Colo205 and SW1222, were tested. For Colo205 cells, tumors were implanted subcutaneously with PBMCs as effector cells. In this model, 4 doses of i.v. huA33-BsAb were enough to completely eradicate Colo205 tumors and the mice remained tumor free for at least 4 months after a total of 6 doses of antibody treatment (Fig. 6A and B).

For the SW1222 cell line, tumors were planted i.p. and treated with 6 doses of i.p. antibody over 3 weeks, and i.v. T cell weekly over 2 weeks. As shown in Fig. 6C and D, the two control groups had tumor growth spread throughout the whole abdominal area, whereas treatment with huA33-BsAb suppressed tumor growth and significantly prolonged mice survival.

These results together suggested that huA33-BsAb had similar efficacy in MSS tumors as that in MSI tumors. We expected more treatment cycles could eradicate the growth of tumors more efficiently, as shown in subcutaneous models.

GPA33 is expressed in a subset of gastric cancer cells (13). By FACS analysis and in vitro TDCC assays, SNU16 expressed GPA33 and was sensitive to huA33-BsAb redirected T cell killing. SNU16 cells were xenografted subcutaneously after mixing with human PBMCs. As expected, i.v. huA33-BsAb effectively cured the mice of s.c. tumors (Supplementary Fig. S6) and significantly prolonged survival.

Curing mCRC is an urgent unmet medical need. In this study, we demonstrated a potential strategy using a T-BsAb, huA33-BsAb, in vitro and in vivo, to redirect polyclonal cytotoxic T cells to kill tumors that expressed GPA33 antigen. Besides systemic spread, peritoneal carcinomatosis has also been difficult to control. Unlike metastasis to liver and lung, it is usually unresectable and unresponsive to chemotherapy and radiation, resulting in significant morbidity. Current treatment is mostly palliative, consisting of cytoreductive surgery or hyperthermic chemotherapy (HIPEC), which are effective in only a small percentage of patients with small volume disease. Development of effective immunotherapy using i.p. T-BsAbs for malignant ascites, as demonstrated in this study, is an attractive alternative for several reasons. First, the peritoneal cavity is an immunocompetent compartment harboring immune cells, the majority (>40%) being T cells and monocytes/macrophage (>40%; ref. 35). Second, i.p. administered antibodies are placed in immediate contact with cancer cells, facilitating tumor binding through high local drug concentrations. Third, rapid uptake of antibodies by tumors reduces systemic drug exposure, hence limiting the toxicity of T-cell activating biologics commonly seen with i.v. injections. It was not surprising that the first T-BsAb drug approved for solid tumor, the anti-EpCAM antibody catumaxomab, was administered i.p. for malignant ascites. Besides targeting the peritoneal tumor compartment, huA33-BsAb could suppress tumor growth and prolong survival in subcutaneous xenograft models of aggressive mCRC. Furthermore, huA33-BsAb could recruit cytotoxic T cells from the systemic circulation into the peritoneal cavity for effective tumor kill—an important first step in T-cell–based immunotherapy. Unlike the robust response of s.c. models, i.p. tumors were not cured in all mice. However, considering the suboptimal engraftment of human cells in Balb/c derived DKO mice without exogenous cytokines (36, 37), the efficacy of huA33-BsAb could be underestimated. One could even propose combining injections of i.p. huA33-BsAb with i.p. T cells or with i.p. PBMC to enhance the antitumor effects.

T-BsAbs generally have higher tumor killing potency than ADCC-dependent IgG antibodies, possibly because of the presence of competing IgGs present in the medium and/or the intrinsic differences between cytotoxic T cells and NK cells. Our in vitro TDCC assay showed that huA33-BsAb killed colon cancer cell lines with an EC₅₀ of around 4 to 66 pmol/L and 25 pmol/L for purified CD4⁺ memory T cells (Fig. 4B), which was 2 to 40 times lower than the apparent affinity of the molecule. This potency is comparable to MEDI-565 (38), which has an EC₅₀ of 91 pmol/L and higher than RG7802, which has an EC₅₀ of 0.24 to 6.28 nmol/L (39). One of the most potent T-BsAbs reported so far by Xu and colleagues (21) that targeted GD2 has EC₅₀s in the fM range. These differences may be attributed to the differences in the intrinsic properties of antigens or in the particular epitopes targeted by T-BsAb. Nevertheless, it is still not clear if high potency translates into clinical benefit because more efficient killing of tumor cells can be accompanied by more on-target off-tumor toxicities, unless the therapeutic window is improved (21). A tradeoff between potency and toxicity has to be carefully considered, especially for tumor antigens that are also expressed in normal tissues.

Our analyses of cytokines and cytotoxic molecules in culture supernatants provided a mechanistic basis for the action of huA33-BsAb. Cytokines belonging to Th1, Th2, and Th17 types were all involved, although with different kinetics. Cytokines play a major role in modifying the immunologic landscape of the tumor microenvironment and may be important for the long-term efficacy and toxicity of T cell–based immunotherapy. An extreme example is the uncontrolled simultaneous release of a myriad of cytokines that results in “cytokine storm,” a potentially fatal condition, of which IL6 is one of the main components. Understanding how a tumoricidal cytokine pattern transitions into systematic toxicity will undoubtedly help improve current T cell–based immunotherapy.

We also tested if increasing affinity of huA33-BsAb to 3.2 pmol/L could improve T-BsAb potency. Although the facile yeast system was efficient in identifying novel mutations to improve affinity, no significant improvement in EC₅₀ among these variants was observed in TDCC assay. There appears to be an affinity ceiling of subnanomolar, beyond which further increase in affinity will not significantly improve the potency of T-BsAbs, as was also observed using other T cell–engaging molecules (40). The molecular basis of this phenomenon is unclear. It is also possible that antibodies with these extremely high-affinity molecules became irreversibly trapped by dead cells and no longer available during cytotoxicity assays.

Besides GPA33, both EpCAM and CEA are being tested as therapeutic targets for mCRC. EpCAM has the advantage of being restricted to epithelial cells, which are not present in healthy peritoneal cavity because it is of mesenchymal origin. CEA has the advantage of having a polarized expression pattern with expression restricted to the apical surface of normal colon tissues. However, both proteins are expressed in a wide variety of normal tissues and could potentially raise toxicity concerns. In contrast, the expression of GPA33 is specifically restricted to colon tissues and no other organs, which could be advantageous in limiting its toxicity. Because the huA33 antibody is not cross-reactive with mouse GPA33, it is impossible for us to test systemic toxicity in the murine xenograft models used. However, we reasoned that healthy normal colon cells could be shielded from the peritoneal cavity by the physical barriers of serosa and muscular layers wrapping around colons. In addition, on the luminal side, the continuous turnover of the gut epithelium carrying with it GPA33 could partially alleviate this cross-reactivity issue (41). It is conceivable that removing FcR(n) affinity could further reduce antibody transport into the gut lumen (42). More careful investigations using antibodies against this antigen in nonhuman primates or in humans will be needed to address these issues. The results from ongoing Macrogenics' phase I clinical trial of MGD007 (Clinicaltrials.gov, NCT02248805), a bivalent (1+1) T-BsAb directed at GPA33, with antitumor efficacy in a Colo205 subcutaneous xenograft model, should provide valuable information on the clinical safety of such approaches.

Z. Wu, H. Xu, and NK Cheung have ownership interest as co-inventors on a patent application for humanized anti-GPA33 antibodies huA33 filed by Memorial Sloan Kettering Cancer Center. NK Cheung reports receiving other commercial research support from Ymabs and Abpro, has ownership interest (including stock, patents, etc.) in Ymabs, Abpro, and Eureka, and is a consultant/advisory board member for Abpro and Eureka. No potential conflicts of interest were disclosed by the other author.

Conception and design: Z. Wu, H. Xu, NK Cheung

Development of methodology: Z. Wu, H. Xu, NK Cheung

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z. Wu, H.-F. Guo, H. Xu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Z. Wu, H.-F. Guo, H. Xu, NK Cheung

Writing, review, and/or revision of the manuscript: Z. Wu, H. Xu, NK Cheung

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Z. Wu, H.–F. Guo, H. Xu, NK Cheung

Study supervision: NK Cheung

We thank Dr. Mamoru Ito of Central Institute for Experimental Animals, Kawasaki, Japan, for providing the DKO mice. Z. Wu, H.F. Guo, H. Xu, and NK Cheung were partly supported by Enid A. Haupt Endowed Chair and the Robert Steel Foundation. Technical service provided by the MSK Animal Imaging Core Facility was supported in part by the NCI Cancer Center Support Grant P30 CA008748.

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.