Antitumor Efficacy of Anti-GD2 IgG1 Is Enhanced by Fc Glyco-Engineering

Antibody-based immunotherapy is an accepted modality in cancer treatment. In the presence of IgG antibodies, many tumor cells are susceptible to Fc-dependent killing, which includes natural killer (NK) cell–mediated antibody-dependent cytotoxicity (ADCC), granulocyte-mediated ADCC, monocyte-macrophage mediated cytotoxicity, and complement-mediated cytotoxicity (CMC; reviewed in ref 1). All the IgG1 antibodies approved by the FDA to date for cancer immunotherapy rely partly or solely on their Fc effector functions. The affinity of their Fc for the FcγRIIIA (CD16A) receptor on NK cells or macrophages and the affinity for FcγRIIA (CD32A) receptor on myeloid cells (2) are critical determinants of the cytotoxic potential. The affinity for the neonatal Fc receptor, FcR(n), can influence the serum half-life, thereby changing the cumulative drug exposure as measured by the area under the concentration curve (AUC) over time. Because of the importance of Fc affinity to therapeutic effects, we sought to improve the affinity of an antibody to GD2, a ganglioside prevalent on neuroendocrine tumors, by modifying either their amino acid sequence or their carbohydrate moieties.

In humans, FcγR polymorphism stratifies patients into those with higher affinity for human IgG1, FCGR3A(158V) or FCGR2A(131H), and those with lower affinity FCGR3A(158F) or FCGR2A(131R). Because of this affinity difference, patients benefit differentially from human IgG1 antibody treatment. Among lymphoma patients treated with rituximab, both higher-affinity FCGR2A and FCGR3A alleles were shown to have better response with survival advantage in most but not all studies (3–6). When the receptor CD16A was the higher-affinity allele, in vitro cytotoxicity could increase several fold (7), which translated into a 2-fold advantage in clinical response and in time to progression (3). When metastatic breast cancer was treated with trastuzumab, patients with the higher-affinity FCGR3A allele had better overall response (83% vs. 35%, P = 0.03), and longer progression-free survival (P = 0.005; ref. 8), although in other studies no significant advantage was found (6, 9, 10). For metastatic colorectal cancer treated with cetuximab, patients with lower-affinity FCGR2A and FCGR3A alleles had worse hazard ratios comparable with those with mutated KRAS (11).

Site-directed mutagenesis has been successfully applied to the Fc region to enhance affinity for FcγR, creating antibodies with enhanced clinical potential (12). The triple mutation (S239D/I332E/A330L, DEL) is one example which increases FcγR affinity by 10- to 100-fold (13). However, mutations can increase the immunogenicity of IgGs. An alternative approach to enhancing FcγR affinity is by modifying Fc glycosylation (14). A key enzyme for the N-linked glycans is UDP-N-acetylglucosamine: α-3-d-mannoside-β-1,2-N-acetylglucosaminyltransferase I (GnT1) encoded by the MGAT1 gene. GnT1^−/− CHO cells synthesize antibodies carrying Man₅-GlcNAc₂ without α1,6-fucose, exhibiting increased ADCC (15). Although antibodies carrying Man₅ are susceptible to clearance through the mannose receptors on scavenger cells, serum half-lives of these IgGs have not been substantially affected (16). Recently, a WT1-TCR mimic antibody produced in GnT1^−/− CHO cells, specific for the WT1 RMF peptide/HLA-A*02:01 complex, was shown to have enhanced ADCC both in vitro and in vivo (17).

Anti-GD2 antibody immunotherapy has proven efficacy (18, 19). Besides NK-ADCC, granulocyte-ADCC appears to be important for clinical benefit (20, 21). Using the humanized version of 3F8 (hu3F8; ref. 22) currently in clinical trials, we tested the effect of GnT1^−/− and triple DEL mutations, alone and together, on FcR affinity, in vitro tumor cytotoxicity, and in vivo antitumor effects.

The neuroblastoma cell line LAN-1 and the melanoma line M14 were obtained from the University of California, Los Angeles in 1983 and 1984, respectively. The following cell lines were established at the Memorial Sloan Kettering Cancer Center (MSK): neuroblastoma line SKNCM in 2007, and Ewing sarcoma line SKELP in 2004. The following cell lines were purchased from ATCC: neuroblastoma IMR-32 in 1983; melanoma SKMEL1 in 1984; small cell lung cancer (SCLC) NCI-H524, NCI-H69, and NCI-H82 all in 2005. Ewing's sarcoma line TC71 was from DSMZ in 2005. All cell lines have been tested and authenticated within 6 months of experiments by short tandem repeat profiling using PowerPlex 1.2 System (Promega), by comparison with either a public database, our original tumor samples, or the original batch of cells from outside sources. All cell lines were periodically tested for Mycoplasma using a commercial kit (Lonza). The luciferase-labeled tumor cell line IMR-32-Luc was generated by stably expressing an SFG-GFLuc vector into the parental cells. All cell lines were grown in F10 medium (RPMI 1640 medium supplemented with 10% fetal bovine serum, 2 mmol/L glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin) at 37°C in a 5% CO₂ incubator.

The antibody hu3F8-IgG1 was produced as previously described (22). Nucleotide sequence encoding the parental hu3F8 (22) and the mutant form hu3F8-IgG1-DEL (S239D/I332E/A330L in the human IgG1-Fc region) was synthesized by GenScript, and was subcloned into a mammalian expression vector. The antibodies hu3F8-IgG1 and hu3F8-IgG1-DEL were produced in wild-type CHO-S cells, whereas hu3F8-IgG1n and hu3F8-IgG1n-DEL were produced in GnT1-deficient CHO cells (Eureka Therapeutics, Inc.). Antibodies were purified by protein A affinity column chromatography. The purity of these antibodies was evaluated by both SDS-PAGE and size-exclusion high-performance liquid chromatography (SE-HPLC) as previously described (22, 23).

Lectin PHA binding to CHO cells was performed as previously described (24), using FITC-conjugated L-PHA (EY Laboratories, Inc.) at 2 μg per 10⁶ cells.

Biacore T-100 Biosensor, CM5 sensor chip, and related reagents were purchased from GE Healthcare. Human recombinant CD16A-158V, CD16B, and CD32B were purchased from R&D Systems; CD16A-158F was from Sino Biological. CD32A-131R and CD32A-131H were produced in Escherichia coli as previously described (25). Human FcR(n) was purchased from R&D Systems, while mouse FcR(n) was purchased from ACRO Biosystems. All recombinant proteins as the active surface and blank as the reference were immobilized using the Amino Coupling kit (GE Healthcare).

The antibodies were diluted in standard HBS-EP buffer (0.01 mol/L HEPES pH 7.4, 0.15 mol/L NaCl, 3 mmol/L EDTA, 0.05% v/v Surfactant P20) for CD16/CD32 study, and in FcR(n) running buffer (50 mmol/L sodium phosphate pH 6.0, 0.15 mol/L NaCl) for FcR(n) study, at varying concentrations (42–667 nmol/L) prior to analysis. Samples were injected over the sensor surface at a flow rate of 30 μL/minute over 1 to 2 minutes. Following completion of the association phase, dissociation was monitored for 3 minutes at the same flow rate. At the end of each cycle, the surface was regenerated using 10 mmol/L NaOH for CD16/CD32 study, and FcR(n) regeneration buffer (0.1 mol/L Tris pH 8.0, 0.2 mol/L NaCl) for FcR(n) study, at a flow rate of 50 μL/minute over 1 to 2 minutes. Prior to kinetics analysis, the control curves obtained with samples injected over reference surfaces were subtracted from the biosensor curves obtained after injection of the samples over active surfaces. The results were analyzed using Biacore T-100 evaluation software. The data for CD16 and CD32 were fitted either with two-state reaction model: A + B = AB = AB*, K_D = (k_d1/k_a1)/(1 + k_a2/k_d2), or with steady-state (SS) equilibrium analysis as noted in the table footnotes; FcR(n) data were fitted with SS equilibrium analysis.

Cell cytotoxicity was assayed by ⁵¹Cr release as previously described (22), and EC₅₀ was calculated using GraphPad Prism software. ADCC assays were performed using NK92MI cells stably transfected with the human CD16A-158V or human CD32A-131H Fc receptors, and effector-to-target ratio was kept at 20:1. CMC assays were performed using human serum from healthy donors.

Blood was collected from the tail vein of nude mice at 0.5, 2, 4, 8, 24, 48, 72, 96, and 168 hours after a bolus injection of 100 μg of antibodies. Quantitation of serum Hu3F8 was carried out by ELISA. In brief, antibody A1G4, specific for the 3F8 idiotype, was used to capture serum hu3F8 (human IgG1 antibody) on 96-well microtiter plates, and it was quantified using a hu3F8-IgG1 standard curve ranging from 1.25 to 40 ng/mL. Diluted serum samples were allowed to bind to the ELISA plates, followed by reaction with peroxidase-conjugated goat antibody to human IgG Fc. Color reactions were developed using hydrogen peroxide and chromogen o-phenylenediamine. Optical density was read with ELISA plate reader at 490 nm. Based on the standard curve, serum hu3F8 for each sample was calculated in ng/mL. Pharmacokinetics analysis was carried out by noncompartmental analysis of the serum concentration–time data using the WinNonlin software program (Pharsight Corp.).

All animal procedures were performed in compliance with the Institutional Animal Care and Use Committee (IACUC) guidelines. The immunodeficient mice colony BALB-Rag2^−/−IL2R-γc-KO (DKO) was maintained at MSK under sterile conditions and provided with Sulfatrim food. In vivo experiments were performed with 6- to 10-week-old mice, and each treatment group had 5 mice. Human peripheral blood mononuclear cells (PBMC) were prepared as previously described (23), and they had ∼15% CD56⁺ NK-cell subpopulations.

Fresh PBMCs (4 × 10⁶ cells) were mixed with IMR-32-Luc cells (6 × 10⁶ cells) in Matrigel (BD Biosciences) and implanted subcutaneously (s.c.) into the right flank of DKO mice. Treatment schedules are detailed in the Results section. Tumor size was measured weekly by calipers, and tumor volumes were calculated using the formula V = 0.5 (length × width × width). The percentage of tumor growth was the change in growth from pretreatment size.

Differences between samples indicated in the figures and tables were tested for statistical significance (P < 0.05) by the Student t test.

The antibodies hu3F8-IgG1 and hu3F8-IgG1-DEL were produced in CHO-S cells, whereas hu3F8-IgG1n and hu3F8-IgG1n-DEL were produced in GnT1-deficient CHO cells. Antibodies were purified by protein A affinity chromatography, and tested by SEC-HPLC with 99%, 90%, 92%, and 98% purity, respectively. On reducing SDS-PAGE, heavy chain and light chain of all four antibodies migrated to the same expected molecular size (data not shown). All four antibodies remained stable after multiple freeze and thaw cycles when assayed by SDS-PAGE and SEC-HPLC (data not shown).

Loss of PHA binding could demonstrate that these GnT1-deficient CHO cells had the GnT1-deficient glycosylation phenotype (24). Thus, wild-type CHO-S cells or GnT1-deficient CHO cells were analyzed by flow cytometry with FITC-conjugated L-PHA. As shown in Fig. 1, wild-type parental and hu3F8-IgG1 stably transfected CHO-S cells had PHA binding (MFI at 50–60), whereas hu3F8-IgG1n− and hu3F8-IgG1n-DEL–producing GnT1-deficient CHO cells lost most of PHA binding (MFI at 8–10, control baseline MFI set at 5). Polymerase chain reaction (PCR) of genomic DNA from these GnT1-deficient CHO cells with primers flanking the single exon encoding MAGT1 gene produced no detectable PCR band, whereas a single band at the right size (∼1.4 kb) was detected from wild-type CHO-S cells (ref. 24; data not shown), suggesting GnT1-deficient CHO cells either totally or partially lost MAGT1 gene segment.

Binding kinetics of the four antibodies to human recombinant FcRs were compared by SPR using Biacore T-100, with different FcRs coated onto CM5 chips. One primary goal of our study was to compare kinetics that seemed critical for clinical efficacy. The method that gave the best fit was chosen. The two-state reaction model fit medium-ranged affinities best (as shown in Supplementary Fig. S1). In these studies, after the initial fast association phase, a slow increase in binding continued, suggestive of a two-state mechanism, instead of a flat plateau (typical for 1:1 model), presumably due to the potential multivalent binding and/or conformational changes on the chip surface. However, when the overall affinity was too low (K_D ≥ 1 μmol/L, see footnote of Table 1), the two-state kinetics analysis started losing accuracy. In these cases, the data were also fitted using the SS equilibrium method. As expected, hu3F8-IgG1 had higher overall affinity K_D for CD16A(158V) than CD16A(158F), and higher K_D for CD32A(131H) than CD32A(131R) (Table 1). The inhibitory CD32B had lower affinity than their activating counterparts, translating into a more favorable activating:inhibitory (A:I) ratio.

The comparison between the four antibodies is shown in Table 2 (the data were fitted with either two-state reaction or steady state as noted), in which we used low-affinity polymorphic alleles (CD16A-158F and CD32A-131R, respectively) as reference. GnT1^−/− CHO cell–produced antibodies (IgG1n and IgG1n-DEL) contained only mannose and acetylglucosamine, gaining affinity preferentially for CD16, but not for CD32B, resulting in higher A:I FcγR affinity ratios than parental IgG1. In contrast, IgG1-DEL mutant gained affinity for both CD16 and CD32B, resulting in A:I ratio comparable with parental IgG1. IgG1n-DEL mutant gained more affinity for CD16 but not for CD32B, resulting in the highest A:I ratio. As shown in Table 2, only the A:I ratio of CD16A-158V/CD32B was calculated, although the trends were similar irrespective of any of the three activating CD16 receptors (CD16A-158V, CD16A-158F, or CD16B). The data were also fitted with the two-state reaction only, as shown in Supplementary Table S1. Although the numbers were different, the overall trends were still the same, especially for A:I ratio.

To examine the cytotoxic potency of these four antibodies for individual activating FcγR (CD16A-158V or CD32A-131H) in the absence of inhibitory FcγR (CD32B), NK92MI cells that did not carry human FcRs were stably transfected with either of these human higher-affinity FcγRs, and ADCC was measured using a standard 4-hour ⁵¹Cr release assay. A representative panel of human tumor cell lines, including neuroblastoma (LAN1, IMR-32, and SKNCM), Ewing sarcoma (TC71 and SKELP), SCLC (NCI-H524, NCI-H69, and NCI-H82), and melanoma (M14 and SKMEL1), was tested for cytotoxicity, and the potencies relative to hu3F8-IgG1 (EC₅₀ for IgG1/EC₅₀ for IgG1 variants) were summarized in Fig. 2. Both IgG1n and DEL mutants had significantly enhanced killing potency when engaging CD16A-158V compared with parental IgG1 (Fig. 2A). In contrast, DEL mutant maintained potency, while IgG1n lost potency, when engaging CD32A-131H (Fig. 2B). All the ADCC results were consistent with the affinity data summarized in Table 2. Interestingly, both IgG1n and DEL mutants had decreased CMC activity, whereas IgG1n-DEL lost almost all of its CMC activity (Fig. 2C).

Serum pharmacokinetics analysis was performed in nude mice after intravenous (i.v.) injections of these antibodies, where serum antibody levels were quantified by the ELISA assay. As shown in Table 3 and Supplementary Fig. S2, while IgG1, IgG1n, and IgG1-DEL had comparable serum half-life (t_1/2), C_max, AUC, and clearance (Cl), IgG1n-DEL had markedly shorter serum half-life, lower AUC, and faster clearance. In contrast, all four antibodies had similar binding affinity to both human FcR(n) and mouse FcR(n) (Supplementary Table S2), suggesting other mechanisms might be involved in the inferior pharmacokinetics of IgG1n-DEL in mice.

For in vivo therapy studies, BALB-Rag2^−/−IL2R-γc-KO (DKO) mice were used for xenografting human tumors (26). Using a humanized mouse xenograft model (ref. 23; s.c. effector cells plus s.c. tumors to simulate effector cells residing within tumor), i.v. antibodies showed high activity against established tumors. No clinical toxicities were observed. IMR-32-Luc cells were mixed with fresh PBMCs (unactivated, from buffy coat) and implanted s.c. On day 7, treatment with antibodies (50 μg i.v., twice per week for 4 weeks) was started and tumor size was measured. As shown in Fig. 3, human IgG control antibody palivizumab failed to shrink tumors. Although the antitumor effects of the IgG1, IgG1-DEL, and IgG1n-DEL forms of hu3F8 were similar, but modest, hu3F8-IgG1n was significantly more effective (P < 0.05). The experiment was repeated once with similar settings, and similar results were observed.

Cancer immunotherapy using human IgG relies on its Fc to mediate ADCC. Affinity of Fc for the CD16A (FcγRIIIA) receptor on white cells can be substantially enhanced through either site-directed mutagenesis of the Fc or expression of IgG in GnT1^−/− CHO cell lines. However, they have different effects on the affinity for CD32, especially on the inhibitory FcγR allele CD32B, translating into a superior A:I ratio when using antibodies generated with the GnT1^−/− approach, which gives rise to a loss of fucose and a gain of Man₅-GlcNAc₂ (27). Yet, in vitro FcR affinity and in vitro cytotoxicity assays cannot accurately predict the in vivo antitumor effect in mouse models, nor the varying influence of different Fc modifications on their in vivo pharmacokinetics. For hu3F8-IgG1n-DEL, despite its highest A:I FcγR affinity ratio, its fast clearance, and low AUC (Table 3), presumably because of lower stability of the antibody in vivo, and higher affinity for FcR-bearing auxiliary cells circulating in the blood, appeared to negate all the advantages of enhanced FcγR affinities. The fact that hu3F8-IgG1n-DEL lost nearly all of its CMC capability (Fig. 2C) might also partially contribute to its inferior in vivo antitumor effect.

Besides the GnT1^−/− phenotype, defucosylation alone (FUT8^−/−; ref. 28) and bisected oligosaccharides (29) are other examples in which changes in the Fc oligosaccharides have led to substantial changes in the affinity of IgG1 for the FcγR. Removing fucose from asparagine = 297 of the heavy chain results in a 5-fold increase in the affinity of Fc for CD16A, and a 2- to 11-fold improvement in the EC50 in ADCC assays (30). Affinities for CD32B and for the 131R allele of the CD32A also increase slightly, but not for the 131H allele (31). Binding to CD64(FcγRI), C1q, and FcR(n) is unaffected (31). Human NK cells bearing CD16A stimulated with a defucosylated antibody to CD20 (obinutuzumab) exhibit downstream signaling through the Vav-1, MAPK, and PI3K pathways, resulting in an increase in actin rearrangement and degranulation (32). The cytotoxic potential of each individual NK cell is improved, resulting in better antitumor effects in mouse models (30, 33). Macrophages also express CD16A, and their ADCC may also benefit from defucosylated antibodies (2, 34). Because serum antibodies compete with fucosylated therapeutic antibodies for FcγRs (35), antibodies with enhanced affinity for CD16A have a definite affinity advantage in dwell time and productive signaling (36), especially when antigen density is low (37, 38). Based on superior clinical responses, obinutuzumab was approved by the FDA for treatment of chronic lymphocytic leukemia in 2013 (39, 40). Bisected carbohydrates can also be introduced into IgGs using the enzyme β1,4-N-acetylglucosaminyltransferase III to enhance affinity and ADCC activity (29). Bisected, afucosylated carbohydrates have been made with the additional Golgi α-mannosidase II enzyme in CHO cells (41), thereby enhancing their affinity for CD16A, as well as their in vitro and in vivo antitumor effects (42). Because the IgG produced in GnT1^−/− CHO cells also lacks the core fucose residues, one might question the additional value of Man₅-GlcNAc₂ over defucosylation alone. Here, we show that antibodies from GnT1^−/− CHO cells gained affinity for activating CD16, but not for inhibitory CD32B, thereby enhancing the A:I ratio, which translated into superior antitumor effects in vivo. In contrast, IgG-DEL affinities for both CD16 and CD32B were increased, resulting in an unfavorable A:I ratio. Glycoproteins containing high mannose structures bind the mannose receptor during clearance, resulting in a shorter serum half-life, and an adverse effect on antibody pharmacokinetics (43, 44). Unexpectedly, in mouse studies, these high mannose IgG preparations had half-lives similar to those in wild-type IgGs (45), a finding that was confirmed in our current study. High mannose containing IgGs bind to CD16A with increased affinity, enhancing ADCC functions while losing CMC capacity (46).

Besides CD16A and CD16B, CD32A is another key receptor on granulocytes and myeloid effectors in performing their ADCC functions (47), and its higher-affinity allele FCGR2A(131R) for murine mAb 3F8 was associated with a more favorable patient outcome (48–50). In our studies of mouse models humanized with human PBMCs, human granulocytes were removed during Ficoll gradient separation, and only monocyte-macrophages were recovered; hence, the antitumor effect of granulocyte-ADCC could not be ascertained. Although the IgG1n antibodies showed decreased affinity for CD32A, the enhanced affinity by 2-fold for CD16B on granulocytes should compensate for the net ADCC ability of these IgG1n antibodies.

In summary, this study systematically compared two common ways to enhance affinity of Fc for CD16A (FcγRIIIA), through either GnT1-deficient glyco-engineering or site-directed mutagenesis of the Fc. Hu3F8-IgG1n produced in GnT1-deficient CHO cells turned out to have more promising potential as a therapeutic antibody than the parental hu3F8-IgG1 or its other Fc enhanced variants (summarized in Table 4), in terms of both in vitro and in vivo antitumor effects. This platform may be useful for enhancing the antitumor potential of other IgG antibodies, and may have broad implications for therapeutic antibodies in cancer therapy.

H. Xu was named a co-inventor on a patent application for humanized anti-GD2 antibodies hu3F8 filed by the Memorial Sloan Kettering Cancer Center. N.K. Cheung has ownership interest (including patents) in scfv constructs of anti-GD2 antibodies, use of mAb 8H9, bispecific HER2 and CD3 binding molecules, anti-chondroitin sulfate proteoglycan 4 antibodies and uses thereof, ROR2 antibodies, T-cell receptor–like antibody agents specific for EBV latent membrane protein 2A peptide presented by human HLA, therapy-enhancing glucan, methods for preparing and using scfv, GD2 peptide mimics, methods for detecting MRD, anti-GD2 antibodies, generation and use of HLA-A2–restricted peptide-specific mAbs and CARs, high-affinity anti-GD2 antibodies, and multimerization technologies. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Xu, N.K. Cheung

Development of methodology: H. Xu, I Y. Cheung, N.K. Cheung

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

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Xu, I.Y. Cheung, N.K. Cheung

Writing, review, and/or revision of the manuscript: H. Xu, I.Y. Cheung, N.K. Cheung

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Xu, H. Guo, N.K. Cheung

Study supervision: H. Xu, I.Y. Cheung, N.K. Cheung

The authors thank Dr. Mamoru Ito of Central Institute for Experimental Animals (Kawasaki, Japan) for providing the DKO mice, Dr. Gloria Koo, Hospital for Special Surgery (New York, NY), for her advice in handling DKO mice, and Dr. Cheng Liu of Eureka Therapeutics Inc. for his expert advice and service in providing the GnT1-deficient CHO cell line. We also thank Dr. Mahiuddin Ahmed for his expertise in antibody structural design.

This study was supported, in part, by grants from the Band of Parents, Arms Wide Open Cancer Foundation, Kids Walk for Kids with Cancer NYC, Robert Steel Foundation, and Enid A. Haupt Chair Endowment Fund. Technical service provided by the MSK Small-Animal Imaging Core Facility was supported in part by the NIH 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.