We have developed MGD007 (anti-glycoprotein A33 x anti-CD3), a DART protein designed to redirect T cells to target gpA33 expressing colon cancer. The gpA33 target was selected on the basis of an antibody-based screen to identify cancer antigens universally expressed in both primary and metastatic colorectal cancer specimens, including putative cancer stem cell populations. MGD007 displays the anticipated-bispecific binding properties and mediates potent lysis of gpA33-positive cancer cell lines, including models of colorectal cancer stem cells, through recruitment of T cells. Xenograft studies showed tumor growth inhibition at doses as low as 4 μg/kg. Both CD8 and CD4 T cells mediated lysis of gpA33-expressing tumor cells, with activity accompanied by increases in granzyme and perforin. Notably, suppressive T-cell populations could also be leveraged to mediate lysis of gpA33-expressing tumor cells. Concomitant with CTL activity, both T-cell activation and expansion are observed in a gpA33-dependent manner. No cytokine activation was observed with human PBMC alone, consistent with the absence of gpA33 expression on peripheral blood cell populations. Following prolonged exposure to MGD007 and gpA33 positive tumor cells, T cells express PD-1 and LAG-3 and acquire a memory phenotype but retain ability to support potent cell killing. In cynomolgus monkeys, 4 weekly doses of 100 μg/kg were well tolerated, with prolonged PK consistent with that of an Fc-containing molecule. Taken together, MGD007 displays potent activity against colorectal cancer cells consistent with a mechanism of action endowed in its design and support further investigation of MGD007 as a potential novel therapeutic treatment for colorectal cancer. Mol Cancer Ther; 17(8); 1761–72. ©2018 AACR.
Colorectal cancer is the second leading cause of death due to cancer in the United States with 49,190 deaths expected in 2016 and 134,490 expected new cases. Approximately 50% to 60% of patients with colorectal cancer have metastatic disease at diagnosis with common sites of involvement, including liver, lymph nodes, lung, peritoneum, and soft tissues. Despite some success with existing agents and combination regimens, including those targeting angiogenesis and EGFR, the prognosis of patients with metastatic colorectal cancer (mCRC) remains extremely poor overall (1).
Recent reports however have suggested that newly diagnosed colorectal cancer patients with evidence of immune infiltration in their tumors may have a more favorable clinical prognosis (2–4). Collectively, these observations have led to the generation of a set of criteria, known as the Immunoscore (5), that is now being evaluated as both a prognostic and predictive biomarker in patients with colorectal carcinoma. Furthermore, programmed death-1 (PD-1) blockade with pembrolizumab has demonstrated clinical efficacy in the subset of patients with colorectal cancer that are mismatch repair deficient (6). However, the majority of patients with colorectal cancer (>85%) are mismatch repair proficient and do not respond to immune checkpoint blockade, indicating the need for alternate strategies to leverage T-cell–mediated responses for antitumor activity. One such approach being explored across multiple cancer types are bispecific antibodies designed to recruit host T cells and harness their cytolytic activity to selectively eradicate the target cancer cell population (7). Indeed, Blinatumomab, an antibody-based bispecific T-cell engager (BiTE)–directed against CD3 and CD19, is an approved immunotherapeutic strategy capable of enabling the patient's own T lymphocytes to eliminate leukemia cells (8). However, the need for continuous intravenous infusion due to its short half-life represents significant limitations of blinatumomab clinical use.
To address the functional and structural limitations of existing bispecific molecules, we have developed the DART platform that displays favorable structural and biological properties, including stability and optimal redirect T-cell killing of malignant tumor cells (9–11). Previous studies have also demonstrated successful application of DART molecules for the targeting of B cells through either recruitment of natural killer cells (12) or by co-ligation of inhibitory and activating receptor pathways (13). Importantly, DART molecules can also be tailored to incorporate an Fc domain to enhance half-life and support convenient intermittent dosing rather than continuous infusion (14, 15). In the present study, we have explored the potential for DART molecules to recruit the cytolytic activity of human T cells against colorectal cancer cells.
Dissection of colorectal cancer progression has revealed compelling evidence for the critical role played by the cancer stem cell (CSC), a self-renewing, immortal cell that maintains cancer by supporting tumor growth and differentiation (16, 17). Although these observations have fueled interest in targeting eradication of the colorectal CSC (18–20), this is countered by the apparent plasticity of human cancer in the face of changing tumor niche to revert from non-CSC to CSC (21, 22) calling into question strategies that solely target the putative CSC population. With this in mind we set out to identify a colorectal cancer cell surface target expressed on both the CSC and differentiated tumor cell population. Characterization of mAbs generated from murine immunization of a colorectal cancer–derived cancer stem-like cell (CSLC) line identified glycoprotein A33 (gpA33), a cell surface antigen previously subjected to therapeutic mAb targeting (23, 24) as being expressed across a panel of CSLCs derived from a various stages of colorectal cancer, in addition to exhibiting 100% penetrance across >50 primary and metastatic colorectal cancer specimens. To enable therapeutic targeting of gpA33, we generated MGD007, gpA33 x CD3 DART protein designed to co-engage gpA33-expressing colorectal cancer cells with CD3-expressing T cells and herein characterized its biological activity in vitro and in vivo.
Materials and Methods
Cancer stem-like cell line generation and characterization
All human specimens were obtained from Kaiser Permanente (KSPC #4548), the National Disease Research Interchange (NDRI) or the Cooperative Human Tissue Network (CHTN) with institutional review board approval and appropriate written informed consent for tissue acquisition and use. Colorectal cancer tissue was shipped on ice and received within 24 hours of excision. Tissue with visible microbial contamination on receipt was discarded. Tumors were minced into small (<1mm) fragments and treated with collagenase/dispase (Roche Applied Science) until small aggregates of dissociated cells were observed; dispersing all tissue to single cells severely decreased viability. The conditions appropriate for tissue stem/progenitor cells were previously optimized using a stepwise approach (25, 26); these conditions were further modified empirically for the optimal isolation and growth of colon cancer cells. Viable dissociated cells were plated on tissue culture plates pre-coated with fibronectin and laminin in F12/DME (50:50 v/v) medium supplemented with: recombinant human insulin (10 μg/mL), transferrin (10 μg/mL), EGF (5 ng/mL), ethanolamine (10−9 mol/L), phosphoethanolamine (10−9 mol/L), tri-iodothyronine (T3, 10−12 mol/L), selenium (2.5 × 10−8 mol/L), hydrocortisone (2.5 × 10−8 mol/L), vitamin E (5 μg/mL), glucagon (50 ng/mL), and gastrin-1 (100 ng/mL). Cell line authentication was determined by sixteen-locus short tandem-repeat (STR) analysis performed on the original tumor specimen and longitudinally during cell line development using the AmpFℓSTR Identifiler polymerase chain reaction (PCR) Amplification Kit (Applied Biosystems; Supplementary Table S1). Mutation analysis of KRAS exon 2, BRAF exon 15, and PIK3CA exons 9 and 23, and the Mutation Cluster Region (MCR) of APC exon 15 was performed by sequencing amplified genomic DNA (27, 28). For flow cytometry analyses, CSLC were harvested with collagenase/dispase or trypsin/EDTA, washed, resuspended in HBSS (Invitrogen) with 1% BSA (Rockland Immunochemicals) and stained with indicated antibodies. To enable in vitro cell differentiation, CSLCs were plated on the basis of conditions previously reported for the differentiation of LGR5+ (29) or EphB2-sorted (30) murine crypt cells. Briefly, small aggregates or sorted single cells were mixed with Matrigel overlaid with F12/DME supplemented with B27 and N2 media supplements (Invitrogen). Organoids were prepared for frozen tissue immunohistochemistry and stained for carbonic anhydrase II (CAII, enterocytes); chromogranin A (enteroendocrine cells); and MUC2 (goblet cells). To evaluate tumorigenicity, CSLC lines were implanted under sub-renal capsule of immune deficient mice. Initial patient sample from which CSLC was derived and excised tumor xenograft tissue were fixed in 10% neutral formalin, embedded in paraffin blocks and sectioned for H&E staining.
Identification of anti-gpA33 mAb RECA47
The mouse RECA47mAb was derived from an intact whole-cell immunization of the RECA020108 CSLC performed as previously described (31). Briefly, Balb/c mice received a primary footpad immunization with 50 μL of RECA020108 (2 x 106 cell/mL) in RIBI adjuvant followed by a booster dose one week before lymph node collection and fusion with LB653 myeloma cells. Hybridoma supernatants demonstrating reproducible binding to RECA020108 by flow cytometry were selected for purification and additional flow cytometry and IHC analyses, leading to identification of RECA47mAb. To identify the binding antigen for RECA47mAb, a mammalian cDNA expression library of RECA020108 CSLC was constructed, a battery of DNA pools prepared and expressed in CHO cells, and screened by immunohistochemistry with RECA47mAb by limited dilution cloning. A positive clone was identified and DNA sequencing revealed identity to gpA33. Humanized RECA47 VH and VL sequences are based on the CDRs from mouse RECA47mAb and human frameworks (FR) derived from human germline.
Characterization of MGD007 binding properties
Binding of MGD007 to soluble human and cynomolgus monkey CD3 and gpA33 receptors was evaluated by surface plasmon resonance (SPR): MGD007 binding to immobilized CD3 protein was analyzed (in duplicates) at concentrations of 0, 6.25, 12.5, 25, 50, and 100 nmol/L. Binding to gpA33 was evaluated using human (12.5 nmol/L) gpA33 His-tagged proteins expressed and purified from Chinese hamster ovary (CHO) cells. For flow cytometry analyses, MGD007 or control DART with 0.5 to 1 x 106 cells/mL (100 μL) of cancer cells, engineered CHO cells, or human or cynomolgus T cells were analyzed by flow cytometry with biotin-conjugated anti-EK antibody mixed with APC-streptavidin.
Redirected killing of gpA33-expressing cell lines using human PBMC (E:T = 30:1) or T-cell subsets (E:T ≤ 10:1) was evaluated as previously described using either LDH readout or cell lines engineered to constitutively express luciferase (32, 33). Colo205, SW948, and LS174T cancer cell lines were obtained from the ATCC; JIMT-1 was obtained from DSMZ (Braunsschweig, Germany); RECA0201-GF CSLC line, stably transduced to constitutively express luciferase and GFP was generated at MacroGenics. Cell lines were authenticated by STR analyses, determined to be mycoplasma free, and passaged for less than 3 months. T-cell activation, intracellular granzyme B and perforin levels and T-cell proliferation were determined as described previously (32). Human T cells, CD4+ T cells and CD8+ T cells were isolated via negative selection from PBMCs using Untouched human T-cell isolation, CD4+ T-cell isolation, or CD8+ T-cell isolation kits (Life Technologies). Suppressive T cells were isolated using human regulatory T-cell enrichment kit (Stem cell technologies) and expanded ex vivo as previously described (34). For indicated experiments a previously described (32) control DART protein (single arm CD3) comprising an anti-fluorescein mAb 4-4-20 specificity rather than anti-gpA33 was employed. Non-linear regression analyses were used to fit curves in GraphPad Prism6. Flow cytometry data were performed using Flowjo v9.3.3 software (Treestar, Inc.).
Co-mix in vivo models
All studies were reviewed and approved by MacroGenics' Institutional Animal Care and Use Committee (IACUC). Female NOD/SCID mice were used for this study (n = 8/group). Human T cells were isolated from heparinized whole blood using RosetteSep T-cell isolation kit (Stemcell technologies). The purified T cells were subsequently activated by exposing the cells to anti-CD3 (OKT-3; 1 μg/mL) and anti-CD28 (66 μg/mL) antibodies for a period of 48 hours and maintained in presence of IL-2 (7.6 ng/mL) for up to 3 weeks. The human T cells and tumor cells were combined at a ratio of 1:1 (5 x 106 cells each) and suspended in 200 μL of sterile saline and injected subcutaneously on day 0. The vehicle control (0.5% sterile saline containing 0.5% BSA) or MGD007 was administered intravenously via tail vein injections (100 μL) on Days 0, 1, 2, and 3. Individual tumor dimensions (length × width) were measured using calipers daily from Day 7 to 24 and calculated as described previously (10, 32).
Monkey study/PK assay
Study was performed at Charles River Laboratories (Reno, NV) under IACUC guidelines. Pharmacokinetics was evaluated in cynomolgus monkeys (n = 4/sex/group) after intravenous administration of 30 or 100 μg/kg MGD007 over 2 hours once weekly for 4 weeks. All animals also received vehicle control for the 1st infusion on Day 1. A two-compartment model with constant intravenous infusion was used for PK analysis.
Identification of gpA33 as a cell surface antigen universally expressed on putative colorectal cancer stem cell populations
To enable identification of cell surface antigens expressed on the putative colorectal cancer stem cell population, novel cell lines were developed from freshly resected human colorectal cancer samples using conditions designed to enrich and maintain putative stem cell populations (detailed in Materials and Methods). A panel of 9 CSLC lines were derived, representative of various stages of colorectal cancer differentiation, including both KRAS WT and mutated cancer, two lines derived from metastatic lesions and one from a high microsatellite instable (MSI-H) specimen (Supplementary Table S2). Under serum-free, growth factor-supplemented defined conditions, the CSLC can be maintained indefinitely as semi-adherent cultures on a fibronectin and laminin matrix, while under conditions that promote colon crypt differentiation, the CSLC have capacity to differentiate into organoids comprising the principal cell types observed in the colon and differentiated colorectal tumors (Fig. 1A; Supplementary Fig. S1). When implanted in immune deficient mice under the sub renal capsule, the RECA CSLC form tumors that fully recapitulate the morphologic and phenotypic characteristics of the patients' original tumors (Fig. 1A). Taken together, the CSLC lines therefore exhibit the key properties associated with the cancer stem cell: self-renewal, differentiation and tumor initiation. To identify antibodies that recognize cell surface antigens expressed on the CSLC, a whole-cell murine immunization was performed with the RECA020108 CSLC. Hybridoma supernatants demonstrating cell surface binding to the immunizing CSLC were purified and analyzed by flow cytometry across the broader panel of CSLC lines and by IHC to determine their reactivity toward CSLC xenograft, primary and metastatic colorectal cancer tissues. One said mAb, RECA47mAb, displayed such a profile indicative of reactivity to a cell surface antigen universally expressed in colorectal cancer (Fig. 1B). Expression cloning identified the antigen for this antibody as glycoprotein A33 (gpA33), which was confirmed by immunoprecipitation and western blotting, with SPR analyses revealing high-affinity binding to recombinant gpA33 (Fig. 1C). The expression of gpA33 across the CSLC panel, regardless of their mutational status or histological derivation is distinguished from CD133 and CD44, generally regarded as “canonical” CSC markers, which are not expressed on a subset of the CSLC (Fig. 1D). Furthermore, IHC analyses of a panel of approximately 50 individual colorectal cancer tumor specimens (both primary and metastatic, Fig. 1E; Supplementary Table S3) revealed all were reactive at a 2–3+ level to RECA47mAb whereas IHC across a panel of normal adult human tissues confirmed reactivity restricted to intestinal epithelium (Supplementary Fig. S2), a collective binding profile consistent with that of a gpA33 reactive antibody (35, 36). On the basis of its universal expression profile across models of colorectal cancer stem cells together with homogenous expression within tumor specimens, gpA33 was selected for a DART-based T-cell immunotherapy for the treatment of metastatic colorectal cancer.
gpA33 x CD3 DART (MGD007) design and biophysical properties
MGD007, a gpA33 x CD3-bispecific antibody–based molecule, was constructed using the DART platform, incorporating monovalent binding for both gpA33 and CD3 in addition to an IgG1 Fc domain to provide antibody-like pharmacokinetics. It is heterotrimeric—composed of three chains co-expressed and assembled through disulfide bonds and non-covalent forces (Fig. 2A). Chains 1 and 2 provide the dual-antigen-binding portion of MGD007 with the anti-CD3 Fv region derived from humanized XR32 (10) and the anti-gpA33 specificity based on humanized RECA47mAb. Chains 1 and 3 provide the Fc chains. Chains 1 and 2 form a heterodimer through a disulfide bond at the C-termini of the chains and by virtue of oppositely charged coiled–coil sequences (E-coil and K-coil). Chains 1 and 3 are covalently linked by two disulfide bonds in the Fc hinge region. To prevent homodimerization of Chains 1 and 3, the knob (T366W) and hole (T366S/L368A/Y407V) mutations have been incorporated in the respective CH3 region of each Fc. To facilitate the removal of any remaining homodimers of Chain 3 during purification, the H435R mutation has been incorporated in the Fc CH3 region. The Fc CH2 regions in Chains 2 and 3 contain Ala, Ala substitutions (L234A/L235A) to markedly reduce or eliminate FcγR and complement binding, while maintaining neonatal FcR (FcRn) binding to take advantage of the IgG salvage pathway mediated by this receptor. MGD007 was expressed in CHO cells and purified using Protein A–based chromatography with purity and structural integrity demonstrated by reduced capillary electrophoresis and SE-HPLC (Fig. 2B). The reduced capillary electrophoresis analysis confirmed that MGD007 is primarily composed (99%) of three protein chains and the SE-HPLC analysis demonstrated the correct apparent molecular weight (approximately 110 kDa). Concentration-dependent binding of MGD007 to CHO cells transfected with either human or cynomolgus monkey gpA33 was confirmed, while no MGD007 binding was observed on parental CHO cells (Fig. 2C). Likewise, MGD007 showed similar binding to both human and cynomolgus monkey T cells (Fig. 2D). SPR analysis using recombinant soluble human antigens revealed MGD007 binds with KD values of 2.2 nmol/L to gpA33 and 23 nmol/L to CD3 (Supplementary Table S4).
MGD007-mediated T-cell lysis of gpA33-expressing colon cancer cells and inhibition of murine xenografts
MGD007-mediated in vitro lysis of gpA33-expressing colorectal cancer cells in the presence of either human PBMC donor or freshly isolated T cells is shown in Fig. 3A–D. Although gpA33+ cell lines (LS174T, Colo205) and a representative CSLC (RECA020108-GF) line were all efficiently lysed by MGD007, no cytotoxicity was observed against a human breast cancer cell line (JIMT-1) lacking gpA33 expression (Supplementary Fig. S3) or with the single-arm, CD3-engaging control DART molecule lacking gpA33 targeting specificity. Notably, both Colo205 or the RECA020108-GF CSLC line were effectively completed lysis with MGD007 consistent with uniform expression of gpA33 (Fig. 3B-C). In addition, evaluation of multiple independent donor T cells demonstrated all were supportive of MGD007 activity against Colo205 cells with EC50 values spanning single to double digit ng/mL (Fig. 3D). Inhibition of tumor growth by MGD007 was evaluated in vivo in NOD/SCID mice implanted SC with Colo205 or LS174T human colorectal tumor cells in the presence of activated human T cells. Although the lowest MGD007 dose evaluated (0.8 μg/kg) resulted in no tumor growth inhibition in either model, treatment with MGD007 at doses ≥4 μg/kg resulted in significant inhibition of both Colo205 and LS174T tumor growth (Fig. 3E), with mice treated at ≥ 20 μg/kg showing no apparent tumor growth in either model. LS174T or Colo205 tumors in the vehicle or control DART molecule–treated groups demonstrated a monophasic growth profile for the duration of the study.
MGD007-mediated cytotoxicity is associated with T-cell activation and can be supported by various T-cell subsets
To determine the relationship of MGD007-mediated cytotoxicity with T-cell activation and cytokine release, a parallel analysis of each functional activity using two gpA33-expressing model colorectal cancer cell lines (LS174T and SW948) was performed (Fig. 4A). Flow cytometry analyses confirmed MGD007 cell surface binding to both cells with slightly higher-level binding observed on SW948 (Supplementary Fig. S3). Concomitant with MGD007 mediated cytotoxicity (left hand), the CD69 T-cell activation marker was upregulated on CD4 and CD8 T cells in a dose-dependent manner upon exposure to gpA33-expressing cells (Fig. 4A, middle). In addition, the level of TNF-α, and IFNγ in supernatants of PMBC co-cultured with colorectal cancer cell lines also increased in presence of MGD007 (Fig. 4A, right) but not with PBMC alone, consistent with the restricted expression of gpA33 to intestinal cells (36).
To evaluate the contribution of T-cell subsets to MGD007-mediated target cell killing, T cells from the same donor were purified and CTL assays were performed in parallel with total T cells (CD3+) as well as CD4+ or CD8+ subpopulations as effector cells and Colo205 as target cells. As shown in Fig. 4B, all three T-cell populations supported lysis of luciferase-expressing Colo205 target cells with potency in the relative order of CD8 > CD3 > CD4. The enhanced potency with CD8 T cells was mirrored with increased dose-dependent upregulation of granzyme B and perforin levels in human CD8+ T cells compared with CD4+ T cells (Fig. 4C). Evaluation of the lytic activity of an in vitro expanded population of CD4+Foxp3+ Tregs, with confirmed suppressive activity against conventional T cells (Supplementary Fig. S4), revealed they can also be leveraged by MGD007 to lyse gpA33-expressing target cells albeit with an approximate 10-fold potency reduction (Fig. 4D) compared with unfractionated T cells. In contrast, control DART (single arm-CD3) mediated no cytotoxicity with either conventional CD3 or suppressive T cells.
Prolonged exposure to MGD007 drives expansion of T cells with memory phenotype that retain cytolytic activity but reduced cytokine response
T-cell expansion is a hallmark response to T-cell activation supported by CD3 bispecifics following co-engagement of CD3 with tumor-specific antigen; consistent with this anticipated response T cells underwent replication in the presence of MGD007 and gpA33 target cells as determined following 72- and 96-hour incubation (Fig. 5A). In contrast, no proliferation of T cells was observed in the presence of the control DART molecule. During this timeframe, expression of cell surface markers associated with both T-cell activation and checkpoint inhibition (CD25, PD-1, LAG-3) were also upregulated (Fig. 5B), whereas evaluation of the T-cell phenotype following prolonged incubation over 7 days demonstrated a drive away from naïve T-cell phenotype to central and effector memory (TCM, TEM) phenotypes (Fig. 5C). To determine whether the prolonged 7-day incubation of T cells with MGD007 diminished their ability to support MGD007-mediated activity, T-cell–mediated responses were evaluated following re-exposure to MGD007 and freshly presented gpA33 target cells. As shown in Fig. 5D, T cells expanded following 7-day incubation with gpA33 target cells and MGD007, retained effective cytolytic activity when re-exposed to MGD007 and Colo205 (gpA33+) target cells even at a relatively low E:T ratio (1:1), with efficacy comparable with that supported by T cells incubated for 7 days with control DART (single-arm CD3) and gpA33 target cells. Interestingly however, evaluation of cytokine levels indicated that T cells pre-exposed for 7 days to MGD007 and gpA33 target cells were now inert in supporting IL-2, IFN-γ or TNF-α release. In contrast, the control-exposed T cells maintained cytokine response.
Prolonged circulating half-life following repeat doses in cynomolgus monkeys
MGD007 pharmacokinetics were evaluated in cynomolgus monkeys administered 4 weekly doses of 30 or 100 μg/kg MGD007 by short intravenous infusion. Mean serum concentration–time profiles and PK parameters for 30 and 100 μg/kg MGD007 are shown in Fig. 6 and Supplementary Table S5. Cmax and AUC increased in proportion to dose, indicating linear PK. The mean clearance of MGD007 ranged from 0.7 to 0.8 mL/h/kg, lower than the glomerular filtration rate (GFR) in cynomolgus monkeys (∼125 mL/h/kg), indicating that MGD007 is not cleared by the kidney, as expected for a large molecular weight protein. Importantly, MGD007 demonstrated a prolonged mean beta half-life (t1/2,β) of 146 to 162 hours (6.1–6.8 days) and mean residence time (MRT) of 181–186 hours (7.5–7.8 days), consistent with an IgG Fc-bearing molecule. MGD007 was considered well tolerated in all animals treated at these dose levels, with no treatment-related increases in serum cytokine levels observed (IFN-γ, IL-2, IL-4, IL-5, IL-6, or TNF-α; Supplementary Fig. S5).
Remarkable clinical responses have been observed in subsets of patients with cancer through therapeutic strategies leveraging the antitumor properties of T lymphocytes, including immune checkpoint inhibition (37), chimeric antigen receptor–expressing T cells (38) or bispecific molecules designed to co-engage T cells with cancer cells (8). A member of the latter category, the DART protein MGD007 was designed to co-engage T lymphocytes with colorectal cancer cells through the cell surface antigens, CD3 and gpA33, respectively, promote T-cell recruitment and antitumor activity.
The selection of the colorectal cancer targeting arm of MGD007 was based on a desire to target an antigen expressed on both the putative CSC and the differentiated malignant tumor cell populations. Recent studies have demonstrated the critical importance of the CSC population to tumor initiation, metastases and resistance to conventional therapies of colorectal cancer (39). Given their plasticity, CSC are not only able to differentiate into more mature “daughter” tumor cells, but may also switch back from daughter cells to CSC on cues from the tumor microenvironment niche (21, 22); therefore, eradication of both the CSC and the non-CSC populations appears necessary to ensure optimal therapeutic response and effectively curtail tumor progression.
To identify cell surface antigens expressed on the putative CSC, a panel of colorectal CSLC lines was developed. Various approaches have been previously used to isolate and characterize the CSC population from colorectal cancer, including direct cell sorting, tumor spheroid isolation and expansion, as well as organoid formation employing stem cell culturing media (17, 18, 40). Here, we have used a direct culturing method in defined media together with enrichment on a matrix system and without sorting. We previously employed a similar culturing strategy evolved from conditions developed for isolation and culturing of tissue stem cells, to develop clonal lung cancer-derived CSLC lines that exhibit properties of self-renewal, tumor initiation and differentiation (41). The panel of colorectal cancer CSLC lines we have currently developed covers a range of cancer stages and includes both KRAS WT and mutated specimens (Supplementary Table S2). Consistent with the characteristics ascribed to CSC, these lines can be maintained indefinitely, undergo a differentiation program under conditions that support tissue stem cell differentiation and, upon implantation in NOD-SCID mice, form tumors that morphologically match the histological features of the original cancer (Fig. 1A).
Whole-cell immunization with one of the CSLC lines yielded a panel of antibodies among which one mAb, RECA47, displayed the desired feature of homogeneous reactivity with the entire panel of CSLC lines (interestingly, the putative CSC markers, CD133 and CD44, displayed variable expression). The antigen, subsequently identified as gpA33, was found to be universally expressed across over 50 primary and metastatic colorectal cancer patient samples (Supplementary Table S3). The high penetrance and homogenous expression profile of gpA33 across patients with colorectal cancer observed with the RECA47 mAb is consistent with prior published studies (35). Indeed, gpA33 was originally defined as the target antigen for the A33 mAb, which was identified via a serological screen for antibodies that recognize colorectal cancer antigens (23). gpA33 is a 43 kDa membrane-bound glycoprotein comprising two immunoglobulin-like domains with homology to tight junction proteins CAR and JAM (36). Although the functional significance of gpA33 is unclear, studies in knockout mice indicate a role in colonic mucosal repair, consistent with its exclusive expression on normal human colon and small bowel epithelial cells (42).
Considering its attractiveness as a colon cancer antigen, several clinical studies evaluating antibody-based modalities targeting gpA33 have been attempted (23, 24, 43). Although initial studies with radiolabeled A33mAb were thwarted by immunogenicity and bone marrow toxicity associated with the radio-isotopes used, imaging analyses provided clear evidence of differential tumor retention and penetration compared with normal tissue (44). Several explanations for this selective targeting to tumors compared with normal tissue have been provided, but appeared on the basis of the more rapid turnover of gpA33 expressed by the colon epithelial cells compared with cancer cells (45). More recently, clinical evaluation of KRN330, a fully-human anti-gpA33 mAb, revealed initial evidence of clinical activity among patients with advanced and metastatic colorectal cancer, with effective tumor targeting at doses that were well tolerated (24). Unfortunately, KRN330 in combination with irinotecan failed to achieve significant clinical outcome in a subsequent study (46), suggesting the need for alternative gpA33 targeting modalities to yield benefit.
The appeal of T-cell–based therapies for cancer treatment, the high penetrance of gpA33 in colorectal cancer together with its expression by the putative CSC population and the prior clinical experience demonstrating selective tumor targeting by gpA33-antibody–based modalities all provided a strong rationale for the development of MGD007, a gpA33 x CD3-bispecific DART molecule that leverages T cells for colorectal cancer cell cytolysis. The DART platform has been previously demonstrated to support efficient redirected T-cell killing to target both hematological malignancies (9, 10, 32) and solid tumors (15). The compact nature of the DART molecule, as confirmed by its crystal structure (15), appears ideal for supporting optimal cell-cell association relative to other bispecific platforms (9). MGD007 is comprised of three polypeptide chains that self-assemble to provide monovalent binding to both gpA33 and CD3; it includes a modified IgG1 Fc domain to enhance its circulating half-life that is devoid of Fcγ receptor and complement binding. Flow-cytometry analyses confirmed the ability of MGD007 to bind natively expressed cell surface CD3 on human and cynomolgus monkey T cells and gpA33 expressed across a panel of human colorectal cancer cell lines or CHO cells engineered to express either the human or cynomolgus monkey gpA33.
Upon co-engagement with gpA33 on target cells, including model colorectal cancer CSLC lines, MGD007 mediates potent in vitro killing in the presence of T cells (Fig. 3); furthermore, MGD007 demonstrated antitumor activity in murine tumor xenograft models with gpA33-expressing human colorectal cancer cells co-implanted subcutaneously with human T cells, with doses ≥4 μg/kg resulting in significant tumor growth inhibition. Both CD8 and CD4 T cells can contribute to the killing, with enhanced activity observed with CD8 cells, consistent with prior reports and with the higher level of granzyme B/perforin of this subpopulation (Fig. 4). Considering the preponderance of Tregs in the tumor microenvironment (47), it is notable that MGD007 can also leverage a suppressive T-cell population for gpA33 target cell killing, an observation consistent with prior reports that CD3 directed bispecifics can leverage Tregs for cell-targeted lysis (48). As previously reported and consistent with the mechanism of action of CD3-based bispecific molecules, redirected T-cell killing of gpA33-positive tumor cells was accompanied by T-cell activation and concomitant cytokine production (Fig. 4A). Importantly, no cytokine release was observed with PBMC alone, attesting to both the absence of gpA33 expression on circulating immune cells and the lack of FcγR interaction by the crippled Fc domain. Considering that MGD007 was designed to maintain prolonged exposure through elevated pharmacokinetics, the effects of prolonged exposure to MGD007 on T-cell biology was also modelled in vitro. As anticipated, this led to T-cell expansion, with increases in the level of checkpoint inhibitor molecules, including PD-1 and LAG-3, and an effector-memory phenotype. Importantly, despite chronic exposure to MGD007, T cells still maintained significant killing but were drastically impaired in their ability to support a cytokine response, suggesting a potential for enhancing the cytolysis–v-cytokine release treatment window during the course of MGD007 exposure. These observations are indeed consistent with previous studies demonstrating that seven daily doses of a short serum half-life murine reactive EpCAM/CD3 bispecific BiTE (muS110) led to increased tolerability coinciding with reduced cytokines without compromise of effector function or antitumor activity (49).
Extended MGD007 serum half-life (∼1 week) was confirmed in cynomolgus monkeys, where both the CD3 (Fig. 2D) and gpA33 (Supplementary Fig. S2)–binding specificities are cross reactive. Upon administration of 4 weekly doses of either 30 or 100 μg/kg MGD007, serum concentrations were linear, with confirmed exposure throughout the 4-week dosing period. At these dose levels, no clinical symptoms, toxicity or increased circulating cytokine levels were observed. These dose levels far exceed the dose required to inhibit xenograft formation in mice (wherein toxicity analyses were not feasible due to lack of MGD007 reactivity to mouse gpA33 and CD3) but more importantly, Cmax levels observed at both dose levels were well above the concentration required to achieve maximal redirected T-cell killing of gpA33 expressing tumor cells in vitro. In subsequent studies, performed at exaggerated pharmacological doses of MGD007 (≥300 μg/kg) evidence of GI toxicity has been observed (J.G. Brown; unpublished observations) consistent with on-target engagement with gpA33 on intestinal cells.
Taken together, the high penetrance and uniformed expression of gpA33 in colorectal cancer, the potent lytic activity of MGD007 against gpA33+ colorectal cancer through leveraging of human T cells, including suppressive T-cell populations, and its favorable PK profile support the clinical evaluation of MGD007 in patients with colorectal cancer. A phase I dose-escalation and optimization study in advanced metastatic colorectal cancer is currently enrolling patients (NCT02248805).
Disclosure of Potential Conflicts of Interest
R. Alderson is a director at In Vivo Modeling. S. Gorlatov is a scientist at Macrogenics. L. Liu has ownership interest (including stock, patents, etc.) in Stock option. All authors were employees of MacroGenics during course of study and as condition of employment receive stock options. No additional conflicts of interest were disclosed by authors.
Conception and design: P.A. Moore, K. Shah, R. Alderson, K.L. King, G.R. Chichili, L. Liu, S. Koenig, J. Mather, E. Bonvini, S. Johnson
Development of methodology: Y. Yang, R. Alderson, V. Long, D. Liu, J.C. Li, S. Burke, V. Ciccarone, K.L. King, P. Young, D. Loo, G.R. Chichili, L. Liu, F.Z. Chen, J. Mather, S. Johnson
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Yang, R. Alderson, J.C. Li, H. Li, J. Hooley, A. Easton, M. Licea, D. Loo, L. Liu, D.H. Smith, J.G. Brown
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.A. Moore, Y. Yang, R. Alderson, V. Long, D. Liu, J.C. Li, M. Licea, K.L. King, P. Young, D. Loo, G.R. Chichili, L. Liu, D.H. Smith, J.G. Brown, F.Z. Chen, E. Bonvini
Writing, review, and/or revision of the manuscript: P.A. Moore, R. Alderson, J.C. Li, A. Adami, G.R. Chichili, L. Liu, D.H. Smith, J.G. Brown, F.Z. Chen, S. Koenig, J. Mather, E. Bonvini, S. Johnson
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Roberts, D. Liu, S. Gorlatov, A. Adami, D.H. Smith
Study supervision: P.A. Moore, F.Z. Chen, S. Koenig, E. Bonvini
Other (acquisition of data—antibody discovery and characterization): C.B. Fieger
Other (tumor study in animals and histology of tumor development): P. Young
We thank M. Lewis, C. Sung, and S. Sharma for help in article assembly and data analysis.
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.