A Novel Antibody Engineering Strategy for Making Monovalent Bispecific Heterodimeric IgG Antibodies by Electrostatic Steering Mechanism*

Background: Bispecific heterodimeric antibody consisting of two different heavy chains and two different light chains requires heterodimerization of heavy chains and cognate light-heavy chain pairings. Results: Cognate light-heavy chain pairing can be achieved by an antibody engineering approach. Conclusion: Bispecific hetero-IgG antibodies can be made in mammalian cells. Significance: The technology could be used in the production of bispecific antibodies for many biotechnological applications. Producing pure and well behaved bispecific antibodies (bsAbs) on a large scale for preclinical and clinical testing is a challenging task. Here, we describe a new strategy for making monovalent bispecific heterodimeric IgG antibodies in mammalian cells. We applied an electrostatic steering mechanism to engineer antibody light chain-heavy chain (LC-HC) interface residues in such a way that each LC strongly favors its cognate HC when two different HCs and two different LCs are co-expressed in the same cell to assemble a functional bispecific antibody. We produced heterodimeric IgGs from transiently and stably transfected mammalian cells. The engineered heterodimeric IgG molecules maintain the overall IgG structure with correct LC-HC pairings, bind to two different antigens with comparable affinity when compared with their parental antibodies, and retain the functionality of parental antibodies in biological assays. In addition, the bispecific heterodimeric IgG derived from anti-HER2 and anti-EGF receptor (EGFR) antibody was shown to induce a higher level of receptor internalization than the combination of two parental antibodies. Mouse xenograft BxPC-3, Panc-1, and Calu-3 human tumor models showed that the heterodimeric IgGs strongly inhibited tumor growth. The described approach can be used to generate tools from two pre-existent antibodies and explore the potential of bispecific antibodies. The asymmetrically engineered Fc variants for antibody-dependent cellular cytotoxicity enhancement could be embedded in monovalent bispecific heterodimeric IgG to make best-in-class therapeutic antibodies.

Producing pure and well behaved bispecific antibodies (bsAbs) on a large scale for preclinical and clinical testing is a challenging task. Here, we describe a new strategy for making monovalent bispecific heterodimeric IgG antibodies in mammalian cells. We applied an electrostatic steering mechanism to engineer antibody light chain-heavy chain (LC-HC) interface residues in such a way that each LC strongly favors its cognate HC when two different HCs and two different LCs are co-expressed in the same cell to assemble a functional bispecific antibody. We produced heterodimeric IgGs from transiently and stably transfected mammalian cells. The engineered heterodimeric IgG molecules maintain the overall IgG structure with correct LC-HC pairings, bind to two different antigens with comparable affinity when compared with their parental antibodies, and retain the functionality of parental antibodies in biological assays. In addition, the bispecific heterodimeric IgG derived from anti-HER2 and anti-EGF receptor (EGFR) antibody was shown to induce a higher level of receptor internalization than the combination of two parental antibodies. Mouse xenograft BxPC-3, Panc-1, and Calu-3 human tumor models showed that the heterodimeric IgGs strongly inhibited tumor growth. The described approach can be used to generate tools from two pre-existent antibodies and explore the potential of bispecific antibodies. The asymmetrically engineered Fc variants for antibody-dependent cellular cytotoxicity enhancement could be embedded in monovalent bispecific heterodimeric IgG to make best-in-class therapeutic antibodies.
Pancreatic cancer is the fourth leading cause of cancer death in western countries with a 5-year survival of less than 10% (1); there is a pressing need for developing therapeutic agents to improve the survival rate. Overexpression of EGFR 3 in 40 -60% of cases and overexpression of HER2 in some subsets were observed in pancreatic cancer patients (2). Heterodimeric HER2-EGFR is more active than either the HER2 or EGFR homodimer in transducing proliferative signals (3,4). Blocking the EGFR alone by cetuximab increased HER2 signaling via amplification of HER2 or increased levels of the HER3/HER4 ligand heregulin (5,6), leading to resistance to the treatment. Dual inhibition of HER2 and EGFR was proposed as a plausible therapeutic strategy to improve treatment outcome (7,8), because this strategy could take advantage of blocking both HER2-and EGFR-mediated signaling pathways and reduce the chance for tumor cells to develop drug resistance. A promising approach is to make bsAbs against both HER2 and EGFR.
bsAbs that can target two antigens or two epitopes on the same antigen have long been considered as an attractive approach to combine the additive or potentially synergistic effects exhibited by the combination of two monoclonal antibodies (9,10). Over the past 2 decades, more than 45 different formats of bsAb, including DVD-Ig (11), crossover Ig (12), dualacting Fab (13), and BiTE (14), have been developed for different biological applications (15). Nevertheless, many of these formats have been limited by some of their liabilities, such as instability, short half-life, poor manufacturability, and immunogenicity.
Heterodimeric IgG, which is based on the heterodimerization of two different IgG molecules in the Fc region, is a promising bsAb format because it maintains the overall size and natural structure of the regular IgG with good bioavailability and pharmacokinetics profile (16). When co-expressing two different HCs and two different LCs in the same cell to generate a functional IgG bsAb, only 1 in 10 combinations has the correct configuration (17). Engineering the C H 3 domain of antibody by knob-into-hole (18), or SEED (19), or charge pair residues (20) can promote HC heterodimerization and reduce the combinations to four, but the LC-HC mispairing issue still exists. One solution is to use a common LC (21), but it is time consuming to identify and engineer a promiscuous LC that can accommodate two different HCs while maintaining the desirable functional properties. Catumaxomab, a mouse IgG2a and rat IgG2b hybrid mAb, was produced by quadroma technology by fusing two hybridoma cell lines (22,23). However, bsAbs derived from quadromas need extensive purification steps and are typically of rodent origin, which limit their applications.
Other approaches have been explored to make monovalent bispecific antibodies. Strop et al. (24) and Labrijn et al. (25) have engineered the hinge region and C H 3 domain of Fc, separately expressed the parental antibody, and then assembled to full size bsAb by partial reduction and oxidization. This approach requires generation of two master cell lines and extra postproduction purification steps to clean up the final products. Ideally, one would like to produce the bsAb using a single cell line. Spiess et al. (26) have made IgG bsAbs by co-culturing two transformed Escherichia coli cell lines, but the antibodies produced by this approach lack the carbohydrate in the Fc region, which is important for effector functions.
Here, we describe a new method of generating bsAbs from two different pre-existent antibodies by electrostatic steering mechanism. The format of bsAb, which we refer as hetero-IgG, was produced by engineering the HC and LC of the two different antibodies in such a way that they can assemble exclusively into a bsAb without other contaminating species. This was based on the charge pair strategy for C H 3 engineering (20) so that two different HCs form a heterodimer exclusively. Similarly by applying an electrostatic steering effect to engineer interface residues between LC and HC, the potential mispairing of LCs to noncognate HCs can be prevented.
The strategy described herein can be used to efficiently produce a full-length bsAb from two pre-existent antibodies in mammalian cells without using any artificial linkers. The resulting bsAbs are stable and amenable to commercial manufacturing without excessive aggregation or loss of yield. As this new version of bsAb can target two different antigens or two different epitopes on the same antigen, it may have significant potential for treating serious diseases with unmet needs such as pancreatic cancer.
Computational Analysis-Antibody crystal structures were identified from the Protein Data Bank. Two methods were used to identify the residues involved in the light-heavy chain interaction: 1) contact as determined by distance limit criterion and 2) solvent-accessible surface area analysis. According to the contact based method, interface residues are defined as residues whose side chain heavy atoms are positioned closer than a specified limit (5 Å) from the heavy atoms of any residues in the second chain. The second method involves calculating solventaccessible surface area of the residues in the presence and absence of the second chain. The residues that show a difference of Ͼ1 Å 2 in accessible surface area between the two calculations are identified as interface residues. Both methods identified a similar set of interface residues. The following criteria were further applied to select VH-VL and C H 1-CL interface residue pairs for mutagenesis: 1) they should not be in CDRs and not make contact with the CDR residues; 2) they are highly conserved among IgG antibody subtypes; 3) they are mostly solvent-inaccessible (i.e. buried or partially buried); and 4) they have minimum interference for BiP-C H 1 binding (27)(28)(29)(30)(31). For immunogenicity predictions, the TEPITOPE algorithm was utilized to identify potential nontolerant agretopes as described by Sturniolo et al. (32). The nontolerant agretopes could bind to HLA class II molecules to elicit immune responses. Tertiary structural information of the antibody was not considered in this analysis. Rather, an exhaustive search of all linear 9-residue peptides that could bind to HLA class II molecules was performed. Such peptide-HLA complexes could drive T lymphocyte-dependent immune responses. The analysis focused on the eight most common HLA-DRB1 alleles (0101, 0301, 0401, 0701, 0801, 1101, 1301, and 1501), which cover Ͼ95% of human populations.
Gene Synthesis and Mutagenesis to Make Variants-The amino acid sequences of anti-HER2 (trastuzumab, clone humAb4D5-8) (33), anti-HER2 (pertuzumab, clone humAb2C4) (34), and anti-EGFR (panitumumab, clone E7.6.3) (35) were used to design the DNA sequences after codon optimization for mammalian expression using GeneArt program (Invitrogen). The DNAs encoding V1 O2/O12 signal peptide and variable regions with flanking sequences for restriction enzyme digestion were synthesized by Invitrogen. PCRs using PfuTurbo Hot Start were carried out to amplify the inserts, which were then digested by SalI ϩ NheI and SalI ϩ BsiWI for VH and VL, respectively. The double-digested VH fragments were ligated with SalI ϩ NheItreated pTT5 expression vector (36) in which the human IgG1 C H 1 ϩ hinge ϩ C H 2 ϩ C H 3 domains were already inserted. The double-digested VL fragments were ligated with SalI ϩ BsiWItreated pTT5 vector in which the human constant domain was already inserted. Plasmid DNAs were verified by double strand DNA sequencing. For proof-of-concept studies, an Fn3 tag was inserted in-frame at the N terminus of anti-EGFR HC, and an Fn3-FLAG-His 6 tag was fused in-frame to the C terminus of anti-EGFR LC. Mutagenesis reactions were carried out to introduce the pairs of charged residues by using QuikChange Lightning multisite-directed mutagenesis kit according to the manufacturer's recommendations. Double strand DNA sequencing reactions were conducted to confirm the mutant sequences.
Expression and Purification of Variants-For 20 ml of medium scale expression testing, a total of 10 g of plasmid DNAs in pTT5 (1.5 g of HC1, 3.5 g of LC1, 1.5 g of HC2, and 3.5 g of LC2) were mixed in a 1.5-ml Eppendorf tube; 1 ml of 293 SFM medium containing 10 l of 3 mg/ml polyethyleneimine, pH 7.0, was added and incubated at room temperature for 20 min. The mixture of DNA/polyethyleneimine was loaded into 19 ml of 2936E cells at 1-2 ϫ 10 6 /ml in a 125-ml shaking flask. 0.5 ml of 20% Yeastolate was added to each flask to a final 0.5% the next day. Cells were shaken for 6 more days. The supernatant was harvested by centrifuging cells at 3000 rpm for 15 min. For 5-ml small scale chain drop-out transfections in 24-well plates or for 1-liter large scale production in shaking flasks, the above conditions were scaled down or up proportionally. The harvested supernatant at large scale was purified by protein A column followed by a polishing with Superdex 200 size exclusion column. For chain drop-out experiments, only two plasmid DNAs (either matched or mismatched) were cotransfected in 2936E cells using the same condition as above, 6 days post-transfection, and 10 l/lane supernatant was loaded in 8 -16% Tris-glycine SDS-PAGE and subjected to Western blotting.
Dual Antigen-binding Plate ELISA-huHER2-Fc fusion protein was coated with 100 l/well at 2 g/ml in 1ϫ PBS, pH 7.4, in Maxisorp plates at 4°C overnight. The plates were washed five times with 1ϫ PBS containing 0.05% Tween 20 (1ϫ PBST) and then blocked with 3% nonfat milk/1ϫ PBST at 200 l/well with shaking at room temperature for 1 h. 100 l/well of normalized crude supernatants in 1ϫ PBS at 1:3 series dilutions were added. The plates were incubated at room temperature for 1 h with shaking, followed by five washes with 1ϫ PBST. 100 l/well of 1 g/ml biotin-huEGFR protein in 1ϫ PBS was added, and plates were shaken at room temperature for 1 h. After five washes with 1ϫ PBST, 100 l/well of 0.1 g/ml streptavidin-conjugated HRP in 1ϫ PBS was added. The plates were shaken at room temperature for 1 h followed by five washes with 1ϫ PBST. 100 l/well 2,2Ј-azino-bis(3-ethylbenzthiazoline-6-sulfonic acid) substrate in 1ϫ PBS was added for color development. The data were collected in Victor II machine by reading at 405 nm for 0.1 s per well.
Stable CHO-K1 Cell Line Development-A DNA fragment encoding a furin recognition site (RRRRRR) and a spacer and a self-cleaving peptide (scp) (37,38) were linked between the C termini of HC2 and N termini of HC1 by overlapping PCR reactions; the full-length DNA encoding HC2-R6-spacer-scp-HC1 was purified from 1.5% agarose gel and subcloned in pSLX240_puromycin vector treated with SalI ϩ NotI restriction enzymes. Similarly, the same R6-spacer-scp DNA fragment was inserted between the C termini of LC2 and N termini of LC1, and the full-length DNA encoding of LC2-R6-spacerscp-LC1 was subcloned in pSLX240_hygromycin vector treated by SalI ϩ NotI restriction enzymes. Suspensionadapted CHO-K1 cells were transfected with Lipofectamine using a 1:1 ratio (by mass) of expression vectors. Transfected cells were cultured in proprietary media and selected at 10 g/ml puromycin and 600 g/ml hygromycin B. Cells were cultivated in suspension format using disposable shake flasks and rotated in a humidified incubator (37°C 5% CO 2 ) at 150 rpm. The selection media were replaced every 3-4 days for a duration of ϳ3 weeks until the pools were fully recovered with Ͼ90% viability. Protein production was carried out by inoculating a nutrient-rich proprietary media with 2 ϫ 10 6 cells/ml and incubating at 37°C 5% CO 2 for 7 days. Culture media were harvested from the production by centrifugation (3,000 ϫ g, 5 min) followed by 0.22-m sterile filter. Concentration of the protein expression in the harvested media was determined by an Octet RED96 (ForteBio) using protein A biosensors.
SDS-PAGE and Western Blotting-SDS-PAGE was carried out using NuPAGE 8 -16% Tris-glycine gels and corresponding running buffer. Samples were prepared by combining the harvested supernatant with 2ϫ SDS sample buffer and heating for 5 min at 95°C. Preparation of reduced samples included the addition of NuPAGE reducing agent prior to heating. After electrophoresis, proteins in the gel were stained with either Coomassie Blue or transferred to nitrocellulose membrane using an iBlot (Invitrogen). For supernatant from stable CHO-K1 cells, the nitrocellulose membrane was blocked with fluorescent Western blocking buffer and probed with IR Dye 700 DX-conjugated donkey anti-human IgG antibody. The nitrocellulose membrane was then thoroughly washed in 1ϫ PBST, and images were acquired by using an Odyssey infrared imaging system from LI-COR Biosciences. For chain drop-out experiments, the nitrocellulose membrane was blocked with 3% milk/1ϫ PBST and probed with HRP-conjugated goat anti-human IgG (Fc-specific) and developed with SuperSignal West Pico chemiluminescent substrate.
Mass Spectrometry Analysis-Treatment for deglycosylation and complete reduction and/or partial reduction of bsAb were carried out as follows. 20 g of bsAb was deglycosylated by incubation with 1 l of PNGase F in 20 l of 50 mM Tris buffer, pH 7.2, at 37°C for 18 h. 5 g of deglycosylated or nondeglycosylated bsAb was completely reduced by 9 mM DTT in 20 l of 4 M guanidine HCl, 50 mM Tris buffer, pH 8.0, at 55°C for 15 min. 10 g of deglycosylated bsAb was partially reduced by incubation with 2-fold (at molar ratio) TCEP in 20 l of 50 mM Tris buffer, pH 7.2, at 37°C for 50 -120 min. Intact mass analysis of deglycosylated and nondeglycosylated whole bsAb and completely reduced or partially reduced bsAb was done in an HPLC-ESI-TOF system (Agilent 6210 TOF mass spectrometer in combination with an Agilent 1200 liquid chromatography system, Santa Clara, CA). A 2.1 ϫ 150-mm Pursuit diphenyl column with 5 m particle size (Agilent-Varian Inc, Santa Clara, CA) was connected to the liquid chromatography system and operated at 400 ml/min. The column temperature was 75°C; solvent A was 0.1% TFA in water, and solvent B was 0.1% TFA in acetonitrile. The gradient started at 25% B and increased linearly to 80% B over 30 min. The TOF mass spectrometer was tuned and calibrated in the range of 100 -4500 m/z. The capillary voltage was set at 4500 V, drying gas at 12 liters/min, drying gas temperature at 300°C, nebulizer gas flow at 40 liters/min, and fragmentor voltage at 375 V for intact antibodies and 300 V for reduced antibodies.
Thermal Stability Analysis by Differential Scanning Calorimetry (DSC)-The DSC measurements were obtained using a VP-Capillary DSC system (Microcal Inc., Northampton, MA) equipped with tantalum 61 cells, each with an active volume of 125 l. Protein samples were diluted to 0.5 mg/ml, and the corresponding buffer was used as a reference. The samples were scanned from 20 to 110°C at a rate of 20°C/h with an initial 15 min of equilibration at 20°C. A filtering period of 16 s was used, and data were analyzed using Origin 7.0 software (OriginLab Corp., Northampton, MA). Thermograms were corrected by subtraction of buffer-only blank scans. The corrected thermograms were normalized for protein concentration. The melting temperatures represent peaks in the experimental thermograms, and the enthalpy of unfolding was obtained using the Origin 7.0 software by integration of the area under the melting curves.
Surface Plasmon Binding Analysis to Measure the Affinity of Hetero-IgG1 Variants to Antigens-Biosensor analysis was conducted at 25°C in an HBS-EP buffer system (10 mM HEPES, pH 7.4, 150 mM NaCl, 3 mM EDTA, and 0.05% surfactant P20) using a ProteOn XPR36 optical biosensor equipped with a GLC sensor chip (Bio-Rad). Goat anti-human IgG capture antibody was immobilized to all channels in the horizontal direction of the sensor chip using standard amine coupling chemistry to a level of 5600 -6000 response units. Channels 1-6 in the vertical direction were used to capture antibodies (ϳ100 response units). Five different rhuHER2 or rhuEGFR concentrations ranging from 25.0 to 0.309 nM (3-fold series dilutions) were prepared in running buffer. Each of the five analyte sample concentrations was injected simultaneously over the chip surface in triplicate in the horizontal direction, as a means of assessing the reproducibility of binding and managing potential systematic bias. Blank buffer injections were run simultaneously with the five analyte concentrations and were used to assess and subtract system artifacts. The association phases were monitored for 420 s each, at a flow rate of 50 l/min, and the dissociation phases were monitored for 3600 s, at a flow rate of 50 l/min. The surface was regenerated with 10 mM glycine, pH 1.5, for 30 s, at a flow rate of 50 l/min. The data were aligned and double-referenced using the ProteOn Manager 3.1.0 version 3.1.06© software (Bio-Rad). The data were then fit using Scrubber version 2.0© software (BioLogic Software Pty. Ltd., Campbell, Australia), which is an SPR nonlinear least squares regression fitting program. First, a dissociation rate coefficient (k d ) was determined from the 25 nM rhuHER2 or rhuEGFR 3600-s dissociation phase data. Second, this value was applied as a fixed parameter in the global fit of the 420-s association phase data to a 1:1 binding model, to determine the association rate coefficient (k a ) and the R max value.
ADCC Assay-Cultured human tumor cells were harvested using pre-warmed trypsin, washed two times with sterile PBS, and seeded at 1 ϫ 10 5 cells per well in 96-well black/clear bottom plates in 100 l of their respective growth media. Plates were incubated at 37°C/5% CO 2 overnight. Hetero-IgG1 antibodies, parental antibodies, and irrelevant antibody were titrated from 6.7 nM to 0.67 fM in Immune Cell Media (ICM) (RPMI 1640, 10% heat-inactivated FBS, 5 mM L-glutamine, 1ϫ sodium pyruvate, 20 mM HEPES, 55 M 2-mercaptoethanol). After a wash with PBS, 25 l of each antibody titration was added to the well containing target cells. Primary human NK cells (Fc␥RIIIA 158F/F genotype) were washed two times in pre-warmed ICM. 50 l of NK cells were added to each well at an effector/target ratio of 10:1. Target cells alone and effector cells alone (as controls) were included in each plate. ADCC activity was determined after an overnight incubation at 37°C/5% CO 2 . After washing out the NK cells with 1ϫ PBS, viable target cells were detected using CellTiter-Glo as per the manufacturer's instructions and using a Wallac VICTOR microplate reader. Percent specific lysis was calculated as (RLU values of treated samples Ϫ average RLU value of effector alone)/((the average RLU of untreated cells (effector ϩ target) Ϫ average RLU of effector alone)⅐100. Percent specific lysis values were transferred to a graphic program (GraphPad Prism) where the data were transformed in a sigmoidal curve fit graph.
EGF-induced EGFR Phosphorylation Assay and Basal Level HER3 Phosphorylation Assay-MSD assay plates were coated with 30 l/well of EGFR capture antibody or HER3 capture antibody Mab3481 at 1 g/ml in 1ϫ TBS at 4°C overnight. Plates were washed three times with 200 l/well 1ϫ TBST and then blocked with 200 l/well of 3% blocker A in 1ϫ TBST. Plates were incubated at room temperature with shaking for 1 h followed by three washes with 200 l/well 1ϫ TBST. Plates were tapped to dry waiting for the addition of cell lysate. CHO-huEGFR stable cells or BT-474 cells both at 50,000 cells/well were seeded in 160 l of complete growth medium in 96-well plates and incubated at 37°C 5% CO 2 overnight. The next morning, a 1:3 series of diluted control human IgG1, or anti-HER2 humAb4D5-8 IgG1 alone, or anti-EGFR E7.6.3 IgG1 alone, or the combination of two parental Abs or anti-HER2 ϫ EGFR bsAb variants in serum-free DMEM plus 0.5% BSA were added in triplicate and incubated at 37°C with 5% CO 2 for 30 min. The CHO-huEGFR stable cells were stimulated with 5 nM EGF for 15 min at 37°C. The liquid was tossed, and 50 l/well complete cell lysis buffer (10 mM Tris, 150 mM NaCl, 5 mM EDTA, 1% Triton X-100, 0.1 mg/ml PMSF, 1 mg/ml aprotinin, 1 mg/ml leupeptin, and 1 mM sodium vanadate) was added to lyse the cells. Plates were incubated on ice for 30 min. 30 l/well cleared cell lysate was added to the above treated MSD assay plates that were then shaken for 1 h at room temperature. Plates were washed three times with 200 l/well 1ϫ TBST, 50 l/well MSD SULFO-TAG-labeled pan-tyrosine (Tyr(P)-20) detection antibody diluted to 1.5 g/ml in 1% Blocker A/1ϫ TBST was added to wells containing the lysate from CHO-huEGFR stable cells. Similarly, rabbit anti-human HER3-pY1289 antibody followed by MSD SULFO-TAG-labeled goat anti-rabbit IgG detection antibody were added to wells containing the lysate from BT-474 cells. Plates were shaken at room temperature for 1 h followed by three washes with 200 l/well 1ϫ TBST. 150 l/well of 1ϫ Read Buffer was added, and the data were collected by reading the plates in MSD Sector Imager 6000.
Inhibition of EGFR Phosphorylation on BxPC-3 Cells and of HER2, HER3, and AKT Phosphorylation on MCF-7 Cells-96-Well ELISA plates were coated with respective anti-EGFR or HER2 or HER3 capture antibody at 2 g/ml for EGFR and HER2, or 4 g/ml for HER3 in 1ϫ PBS at room temperature overnight. The plates were washed three times with 1ϫ PBST and then blocked with 300 l/well 3% BSA in 1ϫ PBS at room temperature for 1 h followed by three washes with 300 l/well 1ϫ PBST. BxPC-3 cells at 20,000 cells/well and MCF-7 at 50,000 cells/well were seeded in 96-well tissue culture plates in their completed growth medium and were incubated at 37°C, 5% CO 2 overnight. The next day, BxPC-3 cells were treated with either 15 g/ml anti-HER2 humAb4D5 IgG1 plus 15 g/ml huIgG1 isotype, 15 g/ml anti-EGFR E7.6.3 IgG1 plus 15 g/ml huIgG1 isotype, 15 g/ml anti-HER2 humAb4D5-8 IgG1 plus 15 g/ml anti-EGFR E7.6.3 IgG1, 30 g/ml anti-HER2 ϫ EGFR hetero-IgG1 V23, or 30 g/ml anti-HER2 ϫ EGFR hetero-IgG1 V23_W165 at 1:3 serious titration at room temperature for 1 h. MCF-7 cells were treated with either 15 g/ml anti-HER2 humAb4D5-8 IgG1 plus 15 g/ml huIgG1 isotype, 15 g/ml anti-HER2 humAb2C4 IgG1 plus 15 g/ml huIgG1 isotype, 15 g/ml anti-HER2 humAb4D5-8 IgG1 plus 15 g/ml anti-HER2 humAb2C4 IgG1, 30 g/ml anti-HER2 (humAb4D5-8) ϫ HER2 (humAb2C4) hetero-IgG1 V23, or 30 g/ml anti-HER2 (humAb4D5-8) ϫ HER2 (humAb2C4) hetero-IgG1 V23_W165 at 1:3 serious titration at room temperature for 1 h. At the end of the antibody treatment, BxPC-3 and MCF-7 cells were stimulated with 16.67 nM EGF and 12.5 nM neuregulin 1-␤1 at 37°C for 7 min, respectively, followed by one wash with ice-cold 1ϫ PBS. The 1ϫ PBS was tossed, and 60 l/well of ice-cold lysis buffer containing phosphatase and protease inhibitors was added to lyse the cells on ice for 30 min. Aliquot of 50 l/well cell lysate in triplicate was transferred into ELISA plates that were coated with the capture antibodies. The plates were incubated at 4°C overnight followed by three washes with 300 l/well of 1ϫ PBST. Anti-Tyr(P) detection antibody (4G10) at 0.5 g/ml in 1ϫ PBST at 50 l/well was added and incubated at room temperature for 2 h. After three washes with PBST, 1 g/ml HRP-conjugated goat anti-mouse IgG Fc at 50 l/well was added. Plates were incubated at room temperature for 1 h. After three washes, 50 l/well of 3,3Ј,5,5Јtetramethylbenzidine substrate was added for color development. 25 l/well of 1 N HCl was added to stop the reactions. Absorbance at 450 nm was recorded in Thermo Multiskan Ascent reader. For the detection of pAKT in MCF-7 cells, MSD plates that have been pre-coated with anti-AKT capture antibody were blocked with 3% BSA in Tris wash buffer at room temperature for 1 h, followed by three washes with Tris wash buffer. MCF-7 cells were treated the same manner as above except MSD SULFO-TAG-labeled pAKT (Ser-473) detection antibody was used. The percent specific inhibition was calculated as (1 Ϫ RLU value of treated sample/average RLU of untreated cells)⅐100.
Internalization Assay-Tumor cell monolayers in a flat bottom 96-well plate were exposed to 100 l/well of either control human IgG1, anti-HER2 humAb4D5-8 IgG1 alone, anti-EGFR E7.6.3 IgG1 alone, anti-HER2 ϫ EGFR hetero-IgG V23 at final concentration of 5 g/ml (34 nM), or the combination of anti-HER2 humAb4D5-8 IgG1 and anti-EGFR E7.6.3 IgG1 each at final concentration of 2.5 g/ml (17 nM). Cells were incubated at 4°C for 30 min. After washing once with assay buffer (1ϫ PBS containing 5% FBS), a mixture of Alexa 488-conjugated anti-human IgG (HϩL) at a dilution of 1:1000 and Hoechst 33342 at a dilution of 1:2000 were added to cells that were then incubated at 4°C for 30 min. For the cells designated as time point 0 h, they were washed twice with assay buffer, then fixed, and permeabilized with BD cytofix/cytoperm buffer (BD Biosciences). For the cells designated as time point 1, 2, or 4 h, the cells were washed twice with assay buffer and incubated at 37°C 5% CO 2 for 1, 2, or 4 h, respectively, then fixed, and permeabilized with BD cytofix/cytoperm buffer (BD Biosciences). Cells were analyzed by an ArrayScan VTI HCS reader (Cellomics of Thermo Fisher Scientific) with BioApplication "Spot Detector" set at ϫ40 objective. Images were captured with a Leica fluorescent microscope (DMI6000B) connected to Leica digital camera (CTR6500).
BxPC-3, Panc-1, and Calu-3 Xenograft Murine Models-Female CB-17 SCID mice (7-8 weeks old) were implanted subcutaneously with 5 ϫ 10 6 BxPC-3 cells mixed 2:1 with Matrigel (BD Biosciences) in a total volume of 100 l. Ten days posttumor implantation, the tumor volume was measured, and mice were randomly distributed to control and treatment groups with 10 mice in each group, so that the mean tumor size was similar across groups at the beginning of the treatment. Starting on day 10, the mice were treated intraperitoneally in a volume of 200 l once a week for 5 weeks with either anti-HER2 humAb4D5-8 IgG1 (250 g), anti-EGFR E7.6.3 IgG1 (250 g), the combination of these two parental antibodies (250 g each), anti-HER2 ϫ EGFR hetero-IgG1 (V23, 500 g), or ADCC-enhanced anti-HER2 ϫ EGFR hetero-IgG1 (V23_W165, 500 g). Animals receiving saline served as the vehicle control. Rag2 Ϫ/Ϫ / mFc␥R4 Ϫ/Ϫ /huCD16a ϩ C57 BL/6 mice (8 -9 weeks old) were implanted subcutaneously with 5 ϫ 10 6 Panc-1 cells mixed 2:1 with Matrigel in a total volume of 100 l. Six days post-tumor implantation, the tumor volume was measured, and the mice were randomly distributed to six groups with eight mice in each group. One hour prior to the antibody treatments, mice were injected intraperitoneally with 10 mg of huIVIG plus 0.2 mg of mouse Fc␥R II/III blocker 2.4G2, followed by intraperitoneal treatment with either 250 g of anti-HER2 humAb4D5-8 IgG1 plus 250 g of huIgG1 isotype, 250 g of anti-EGFR E7.6.3 IgG1 plus 250 g of huIgG1 isotype, 250 g of anti-HER2 humAb4D5-8 IgG1 plus 250 g of anti-EGFR E7.6.3 IgG1, 500 g of anti-HER2 ϫ EGFR hetero-IgG1 V23, 500 g of ADCC enhanced anti-HER2 ϫ EGFR hetero-IgG1 V23_W165, or 500 g of huIgG1 isotype. Treatments were administered once a week for 3 more weeks. Female NSG mice (7-8 weeks old) were implanted subcutaneously with 5 ϫ 10 6 Calu-3 cells mixed 1:1 with Matrigel in a total volume of 100 l. Fourteen days posttumor implantation, the tumor volume was measured, and the mice were randomly distributed to either control or treatment groups with eight mice in each group. The mice were treated similarly as the above Panc-1 xenograft model except that anti-HER2 humAb4D5-8 and anti-HER2 humAb2C4 derived IgG1 antibodies were used.
For all studies, tumor-bearing mice were monitored for weight and for tumor volume twice a week. Tumor volume was calculated using Equation 1.

Percent tumor growth inhibition (% TGI) was determined based on Equation 2
.
where ⌬C or ⌬T indicates the difference between the average tumor volumes on the last day and the day of initial measurement for the control (⌬C) or treatment groups (⌬T). Animals receiving saline or isotype huIgG1 served as the control group for these calculations. Antitumor activity is defined as percent TGI Ն50%.
In Vitro Stability of Hetero-IgG1 Antibodies in Human Serum-Hetero-IgG1 bsAbs and their parent antibodies at 150 g/ml in 90% pooled normal human serum were incubated at 37°C. An equal portion of incubated samples was taken out at different time points (24,48,72,96, and 168 h) from the same tube. At each time point, the samples were briefly spun and stored at Ϫ20°C until the test. 50 l/well of biotinylated huHER2 (ECD) or huEGFR (ECD)-Fc (rabbit) at 125 ng/ml was added in the NeutrAvidin-immobilized 96-well assay plates. After three washes with 1ϫ PBST, a 1:2 series of diluted hetero-IgG1 or parental IgG1 at 50 l/well was transferred to the plates, which were then shaken at room temperature for 2 h. After three washes, the HRP-conjugated goat anti-human (Fcspecific) detection antibody was added. The plates were washed again. 3,3Ј,5,5Ј-Tetramethylbenzidine substrates were added for color development. The concentration of the antibodies was deduced from the standard curve of each antibody collected and frozen at t ϭ 0. The stability of each antibody was determined by analyzing the percentage retention concentrations (Ab concentration at test time point/Ab initiation concentra-tion⅐100) over the incubation. The paired Student's t test was used to compare retention concentration at the end of incubation (t ϭ 168 h) to the initiation concentration (t ϭ 0).
Statistical Analysis-Tumor growth was expressed as the means Ϯ S.E. and plotted as a function of time. Statistical comparison of groups was performed at both overall level and at last measurement time point using the analysis of variance test followed by Dunnett's or multivariate t test adjusted for multiple comparisons. Statistical calculations were made through the use of JMP software version 7.0 interfaced with SAS version 9.2 (SAS Institute, Inc., Cary, NC).

Selection of Residue Positions for Introducing Charged Amino
Acid Pairs-We aimed to make bsAbs in hetero-IgG format from mammalian cells by introducing two different HCs and two different LCs in the same cells. The two different HCs were engineered to strongly favor the formation of heterodimers by using the charge pair mutations in the C H 3 regions (20). Because of the presence of the C H 1 domain, the HCs are retained in the endoplasmic reticulum by BiP proteins before they are engaged by LCs for the assembly of full IgG. To ensure the correct pairing of LC with its cognate HC in the hetero-IgG molecule, it was critical to find an effective way to control the kinetics of LC-HC assembly process so that the LC strongly favors its cognate HC and disfavors the noncognate HC. We attempted to achieve this by engineering the VH-VL and C H 1-CL interfaces as they are both involved in the HC-LC recognition and engagement process (27)(28)(29)(30)(31).
Examination of the VH-VL and C H 1-CL interface structures revealed that hydrogen bonds and van der Waals interactions are dominant. Unlike the C H 3-C H 3 interface (20), electrostatic charge-charge residue interaction is rare between the LC and HC. For example, C H 1-CL interface has one and C H 1-CL interface has two positive-negative charge interactions involving Lys and Glu/Asp residues. All three charged residue pair interactions involve partially or fully solvent-exposed positions. Because of the solvent molecules interacting with the charged moieties, the electrostatic interaction would be considerably weakened. Hence, to utilize the electrostatic steering effect to drive specific pairing of LC and HC, it is essential to switch the polar or hydrophobic residues with the charged residues at the VH-VL and/or C H 1-CL interfaces.
To select appropriate positions for maximal electrostatic steering effect, the following criteria were applied: 1) they should not be in CDRs and not make contact with the CDR residues; 2) they are highly conserved among IgG antibody subtypes; 3) they are mostly solvent-inaccessible (i.e. buried or partially buried); and 4) they have minimum interference with BiP-C H 1 binding (27)(28). The interface residues that meet the criteria for engineering are listed in Table 1 and were explored to make hetero-IgG antibodies.
As shown in Fig. 1, A and B, Gly-44 and Gln-105 in VH are spatially close to Gln-100 and Ala-43 in VL, respectively, regardless of antibody germ lines. Gly-44 (VH) to Gln-100 (VL) and Gln-105 (VH) to Ala-43 (VL) have been widely mutated to Cys to make a disulfide-stabilized fragment of variable regions (29). The Gln-39 (VH) to Gln-38 (VL) pair, which is located near the center of hydrophobic core of the VH-VL interface, has been mutated to the charged residue pairs to stabilize the diabody (39). In the constant regions shown in Fig. 1, C and D, Ala-141, Pro-171, and Ser-183 in C H 1 region are close to residues Phe-116, Ser-162, and Ser-176 in C, respectively.
Proof-of-Concept Studies to Validate the Feasibility of Hetero-IgG Approach-The VH and VL regions of the anti-HER2 trastuzumab and anti-EGFR panitumumab were selected for proofof-concept studies because trastuzumab and panitumumab are well characterized and are approved drugs, and dual inhibition of HER2 and EGFR may have therapeutic potential for treating pancreatic cancers. To facilitate validation, an Fn3 tag (12 kDa) was inserted at the N termini of anti-EGFR HC2 and an Fn3-FLAG-His 6 tag (14 kDa) was fused in-frame to the C termini of anti-EGFR LC2, whereas anti-HER2 HC1 and LC1 were kept as wild type. Four different LC-HC combinations will yield products at three different sizes in the SDS-polyacrylamide gel as follows: 162 kDa for LC1ϩHC1::LC1ϩFn3_HC2; 176 kDa for the wanted LC1ϩHC1:: LC2_Fn3-FLAG-His 6 ϩFn3_HC2 or the unwanted LC2_Fn3-FLAG-His 6 ϩHC1::LC1ϩFn3_HC2; 190 kDa for LC2_Fn3-FLAG-His 6 ϩHC1::LC2_Fn3-FLAG-His 6 ϩFn3_HC2. A product with a size of 176 kDa implies the possibility of correct LC-HC pairings. A dual antigen-binding plate ELISA was utilized to quickly screen the favorable variants that may have the wanted LC-HC pairings.
A total of 80 variants in which two pairs of charged residues in VH-VL only; two pairs of charged residues in C H 1-CL only; one pair of charged residues in C H 1-CL only; two pairs of charged residues in VH-VL and 1 pair of charged residues in C H 1-CL (Fig. 1E); two pairs of charged residues in VH-VL; and two pairs of charged residues in C H 1-CL (Fig. 1F) were investigated to find variants with high dual antigen binding after nor- The amino acid residues closely located at the VH-VL and CH1-C interfaces were selected for substitution with charged residues Germline residues in VH and VL are numbered by different numbering systems. The bolded residues are dominant ones. The spatially close residues at VH-VL interface of most antibody germlines are arrayed in the same row as indicated by a double-headed arrow. The spatially close residues in human IgG1 CH1 domain and C region are also bolded and arrayed in the same row. FW: Framework. Ref# is the extended AHo # into constant region. malization based on Fc titers. Their sequence variations are listed in patent application WO2014081955. As shown in Fig.  2A, neither the parental anti-HER2 humAb4D5-8 IgG1 alone nor anti-EGFR E7.6.3 IgG1 alone generated a binding signal. The supernatant from cells that were transfected with wild type HC1 and LC1 from anti-HER2 and wild type HC2 and LC2 from anti-EGFR showed a dose-dependent binding. Anti-HER2 ϫ EGFR hetero-IgG variants 1C02, 1C04, 2A05, 2B05, and 5D03 (Table 2 and Fig. 2A) showed significantly improved dual antigen binding as their curves shifted to the left. In the nonreducing SDS-polyacrylamide gel, all five variants appeared as a single band with the size matched to the middle band of antibody mixture that was made from random LC-HC pairings of anti-HER2 and anti-EGFR (Fig. 2B). Both results indicated that these five variants may have correct LC-HC pairings.
The above five variants were scaled up by transiently transfecting 2936E cells, purified with a protein A column, and then polished with Superdex 200 size exclusion column. From 900 ml of conditioned medium, 5-10 mg of final products with ϳ100% purity by analytical size exclusion chromatography were obtained. In the nonreducing SDS-polyacrylamide gel (Fig. 2C), all variants had a dominant band of full-length IgG1. Variants 2B05 and 5D03 were the purest with a residual level of contaminated bands. Under reducing conditions (Fig. 2D), four different chains (Fn3_HC2 at 61 kDa; HC1 at 50 kDa; LC2_Fn3-FLAG-His 6 at 36 kDa; LC1 at 23 kDa) were separated due to their different sizes. The four different chains appeared to be at 1:1:1:1 ratio in the assembled full-length IgG1 antibody.
The five purified hetero-IgG variants were analyzed by mass spectrometry to verify they have the predicted components and correct LC-HC pairings. Variant 2B05 is given as an example and is shown in The intact hetero-IgG variant 2B05 was partially reduced by heating at 37°C for 80 min in the presence of 2-fold molar excess of TCEP. TCEP preferably breaks up the inter-chain disulfide bonds, yielding five different products (Fig. 3, F-I) consisting of HC1 ϩ LC1 (1/2 Ab1); HC1 ϩ Fn3_HC2; HC1 ϩ Fn3_HC2 ϩ LC1 (3/4 Ab1); HC1 ϩ Fn3_HC2 ϩ LC2_Fn3-FLAG-His 6 (3/4 Ab2); Fn3_HC2 ϩ LC2_Fn3-FLAG-His 6 (1/2 Ab2). The presence of residual full-length antibody indicates that partial reduction happened in the reaction. All components had their theoretical mass with an error of Ͻ100 ppm. The carbohydrate attached at Asn-297 was normal as usual (Fig. 3, G and H). Most importantly, no LC1-HC2 or LC2-HC1 product was observed. The 1:1:1:1 stoichiometry in variant 2B05 showed by mass spectrometry also matched the same band intensity in SDS-polyacrylamide gel under reducing conditions (Fig. 2D). Similar results were obtained for four other purified hetero-IgG variants.
Because the anti-HER2 ϫ EGFR hetero-IgG variants 2B05 and 5D03 (in the presence of Fn3 and Fn3-FLAG-His 6 tags) have embedded with ADCC-enhancement mutations W165 by asymmetrical Fc engineering (40), ADCC killing assay was carried out by using human NK cells (Fc␥RIIIA 158F/F genotype) as effector cells and NCI-N87 cells, a human gastric tumor cell line expressing high levels of HER2 and moderate levels of EGFR, as target cells. As shown in Fig. 2E, at 1 g/ml the irrelevant human IgG1 control antibody had a background lysis of 30% and did not show a dose-dependent response when it was titrated down, but both 2B05 and 5D03 had much higher specific lysis and showed a dose-dependent manner of response with EC 50 at 0.10 and 0.19 pM, respectively. The data suggested that the hetero-IgG variants 2B05 and 5D03 can bind to targets HER2 and/or EGFR with their Fab arms and induce strong killing to NCI-N87 cells by engaging the NK cells.
CHO cells stably expressing human EGFR induce the phosphorylation of EGFR upon EGF stimulation. Although irrelevant human IgG1 did not inhibit the phosphorylation of EGFR (data not shown), the parental anti-EGFR E7.6.3 IgG1 inhibited the phosphorylation of receptor EGFR at IC 50 ϭ 2.7 nM (Fig.  2F). The combination of anti-EGFR E7.6.3 IgG1 and anti-HER2 FIGURE 1. Residues at VH-VL and C H 1-CL interfaces for substitutions are buried, conserved, and spatially close. Anti-HER2 trastuzumab crystal structure 1N8Z from Protein Data Base is illustrated as an example. Structure 1N8Z pdb was loaded into Molecular Operating Environment (Chemical Computing Group, Montreal, Canada). Amber 10:EHT force field was set. Missing atoms were repaired, and charges were applied to the termini as appropriate at pH 7.0 automatically by the structure preparation function. Ribbon rendering was selected with a different color for each of the four domains in the following manner: VH, magenta; VL, orange; C H 1, blue; C1, green. Selected residues were rendered as ball-and-stick, and the Site View Function isolated a region within 5 Å of all selected residues. Kabat numbering was used for the variable region residues, whereas Eu numbering was used for the constant region residues. Distances were measured between each ␣C of the indicated (circled) residue pair by the distance tool. Circled pairs are substituted with charged residues to drive the electrostatic steering mechanism. E, configuration of monovalent bispecific hetero-IgG antibody variants using the electrostatic steering approach. Two pairs of charged residues, KK-DD, in the variable regions binding to antigen A combined with one pair of charged residues, D-K, in C H 1-CL drives the LC1 to pair with its cognate HC1.Similarly, two pairs of charged residues, DD-KK, in the variable regions binding to antigen B combined with one pair of charged residues, K-D, in C H 1-CL drives the LC2 to pair with its cognate HC2. The charged residues for heterodimerization in the C H 3 domains are also indicated. F, configuration of monovalent bispecific hetero-IgG antibody variants using the electrostatic steering approach. Two pairs of charged residues, KK-DD, in the variable regions binding to antigen A combined with two pairs of charged residues, DD-KK, in C H 1-CL drives the LC1 to pair with its cognate HC1.Similarly, two pairs of charged residues, DD-KK, in the variable regions binding to antigen B combined with two pairs of charged residues, KK-DD, in C H 1-CL drives the LC2 to pair with its cognate HC2. The charged residues for heterodimerization in the C H 3 domains are also indicated. More configurations can be found in patent application WO2014081955. The symbol "ᮍ" represents positively charged residues and "ᮎ" represents negatively charged residues.
Taken together, the above results suggested that electrostatic steering mechanism allows us to generate monovalent bispecific hetero-IgGs with cognate LC-HC pairings, and both Fab arms in the hetero-IgGs are functioning properly.
Optimization of Hetero-IgG Format in the Absence of Any Tags-The tags of anti-HER2 ϫ EGFR hetero-IgG1 variants 2B05 and 5D03 (Table 2) were removed and then re-tested by chain drop-out transient transfection in mammalian 2936E cells (Fig. 4A). When all four engineered chains (LC1 and HC1 from anti-HER2; LC2 and HC2 from anti-EGFR) were co-transfected, the main full-size hetero-IgG antibody appeared in the nonreducing SDS-polyacrylamide gel with a smaller amount of half-size antibody. Transfections with two plasmid DNAs encoding the matched LC1 ϩ HC1 or LC2 ϩ HC2 produced the full-size homodimer antibody with a significant amount of halfsize antibody. No product was observed when LC1 was co-transfected with the noncognate HC2 for both variants 2B05 and 5D03. However, when the LC2 was co-transfected with the noncognate HC1, there was a faint band at full-size Ab for variant 2B05 but two obvious bands (at full-size Ab and half-size Ab) for variant 5D03, suggesting LC2-HC1 mispairing would occur if two different HCs and two different LCs were present during the production of hetero-IgG. To further improve the design, we explored a series of new variants ( Table 3) in most of which symmetrical oppositely charged residue pairs were introduced. Chain drop-out transient transfections and Western blotting were carried out to assess the tolerance of LC-HC mispairings. The mutually reciprocal polarities of charged residues at the same positions of LC-HC interfaces could lead to more stringent LC-HC pairings.
For variants V15 and V20 (Fig. 4B), co-transfection with all four plasmid DNAs or matched two plasmid DNAs produced full-size and half-size antibodies. No product was seen in nonreducing SDS-polyacrylamide gel when LC2 was co-transfected with noncognate HC1, indicating the anti-EGFR LC2 was not tolerated by anti-HER2 HC1. However, high level expression of full-size and half-size antibodies was observed when anti-HER2 LC1 was co-transfected with anti-EGFR HC2. Hetero-IgG variants V21 and V22 had more stringent LC-HC pairings (Fig. 4C), whereas variants V23 and V25 did not tolerate any mis-matched LC-HC pairings (Fig. 4D). Variants (V12, V23, V24, and V25) with strict LC-HC pairings were scaled up by transient transfections; mass spectrometry analysis demonstrated that four different chains were correctly assembled in these hetero-IgG variants (data not shown).
Hetero-IgG Antibody Targeting Two Different Epitopes on the Same Antigen-It was reported that in xenograft HER2-positive human tumor models the combination of anti-HER2 trastu-

1C02
Q39K ϩ Q105K K147D ϩ S183D Q38D ϩ A43D S131K ϩ S176K G44D ϩ Q105D S183K G100K ϩ A43K S131D ϩ S176D 1C04 Q39K ϩ Q105K K147D ϩ S183D Q38D ϩ A43D S176K ϩ T180K G44D ϩ Q105D S183K zumab and anti-HER2 pertuzumab showed strongly enhanced antitumor activity than trastuzumab alone or pertuzumab alone (42,43). Anti-HER2 trastuzumab binds to the domain IV of HER2, whereas anti-HER2 pertuzumab binds to the domain II of HER2. We questioned whether anti-HER2 ϫ HER2 hetero-IgG consisting of trastuzumab and pertuzumab could block the signaling pathways synergistically by binding to two different epitopes simultaneously, leading to higher efficacy than the combination of two parental antibodies. The same variant V23 (Table 3) in which two pairs of charged residues in VH-VL and one pair of charged residues in C H 1-CL were reciprocally introduced was tested by either transfecting with four DNAs to make full-length antibody or with only two DNAs to assess the tolerance of mismatched LC-HC pairings. Similar to anti-HER2 ϫ EGFR hetero-IgG variant V23, the anti-HER2 ϫ HER2 hetero-IgG antibody V23 (Fig. 5, V23A) was mainly expressed as the intact antibody after four different chains were translated and assembled. In the presence of two matched chains (anti-HER2 humAb4D5-8 LC1 ϩ HC1 or anti-HER2 humAb2C4 LC2 ϩ HC2), half-antibody and homodimer antibody were produced, indicating that the LCs were compatible with their cognate HCs. In the presence of mismatched chains LC2 ϩ HC1 or LC1 ϩ HC2, no product was formed, indicating that the engineered LCs were not tolerated by their noncognate HCs.

Different Combinations of Charged Residues Affect the Hetero-IgG Expression and LC-HC Pairings-To investigate
whether different combinations of charged residues could result in different expression and/or affect the LC-HC pairings, we made and expressed four anti-HER2 (humAb4D5-8) ϫ HER2 (humAb2C4) hetero-IgG1 variants by introducing charged residue pairs with different combinations at the same positions (Table 4 and Fig. 5). V23B also had strict LC-HC pairings as V23A, but the expression level decreased. However, V23C did not produce any antibody when the matched LC2-HC2 was co-transfected, and it produced a low level of antibody when mis-matched LC1-HC2 was co-transfected. V23D produced a low level of antibody when the matched LC2-HC2 was co-transfected, but LC1 was well tolerated by noncognate HC2 as both full-size and half-size antibodies were observed in nonreducing SDS-polyacrylamide gel. This set of data suggested that the electrostatic steering is not the only mechanism that controls the wanted LC-HC pairings, other mechanisms such as shape complementarity may play a role in this process.
Anti-HER2 ϫ EGFR Hetero-IgG1 Variants Showed Good Thermal Stability-The temperature-induced unfolding of anti-HER2 trastuzumab IgG1, afucosylated anti-HER2 humAb4D5-8 IgG1, anti-EGFR E7.6.3 IgG1, and four anti-HER2 ϫ EGFR hetero-IgG1 variants V12, V23, V24, and V25 having ADCC enhancement Fc (W165) were assessed under the same solvent condi-tions by differential scanning calorimetry (Fig. 6). The thermogram of each protein consisted of two or three transitions. Anti-HER2 trastuzumab showed a T m of Fab/C H 3 at 83°C and a T m of C H 2 at 71°C; the afucosylated anti-HER2 humAb4D5-8 IgG1 did not change the T m of separate domains but decreased the enthalpy slightly. The anti-EGFR E7.6.3 IgG1 antibody had a similar profile of temperature-induced unfolding. All four anti-HER2 ϫ EGFR hetero-IgG1 variants had a slightly decreased T m of merged C H 2/C H 3 at ϳ69°C as they all have the ADCC enhancement substitutions in C H 2 domains and heterodimerization substitutions in the C H 3 domain. In terms of T m of Fab domains, variants V12 and V24 had the most significant decrease from 83 to 75°C; variant V25 had two separate peaks at 74°C and 79°C, and variant V23 had a single peak at 79°C. Overall, the four hetero-IgG variants showed good thermal stability. The data suggested the selected positions for substitutions with charged residues in the Fab regions do impact the stability of intact hetero-IgG1 antibodies to some extent, with T m of separate domains above 68°C.
Stable Expression of Hetero-IgG Antibodies in Mammalian CHO-K1 Cells-As the anti-HER2 ϫ EGFR hetero-IgG variant V23 showed the balanced expression for each half-Ab and strict LC-HC pairings by transient transfection (Fig. 4D) and good DSC profile (Fig. 6), it was chosen to explore the strategy on how to stably express hetero-IgGs in mammalian cells and obtain a large amount of material for further characterizations and animal studies. We linked the open reading frame of anti-EGFR E7.6.3 HC2 and anti-HER2 humAb4D5-8 HC1 with a DNA sequence encoding furin cleavage site (R6), a spacer, and scp (37,38). Similarly, we inserted the same R6-spacer-scp between anti-EGFR E7.6.3 LC2 and anti-HER2 humAb4D5-8 LC1. The two different HCs are designed to integrate into the same chromosome loci to balance the expression of two different HCs as HC heterodimerization is required to form hetero-IgGs. The constructs with opposite orientation (anti-HER2 humAb4D5-8 in front of anti-EGFR E7.6.3) were also made. However, transient transfection revealed that, for some unknown reasons, the Fc titer was significant lower (data not shown). Similar constructs were made for anti-HER2 ϫ HER2 hetero-IgG1 with anti-HER2 humAb2C4 in front of anti-HER2 humAb4D5-8. Constructs for hetero-IgGs, either having regular Fc variant V23 or ADCC enhancement Fc variant V23_W165, were transfected in CHO-K1 cells in duplicate and selected under the pressure of puromycin and hygromycin for ϳ3 weeks. Cell viability decreased to ϳ10% at day 7 and quickly recovered to Ͼ90% at day 22 (data not shown). Compared with transient transfections, CHO-K1 stable pools boosted the Fc titers from 20 to 70 mg/liter to 200 -320 mg/liter. The crude supernatant from two separate stable pools was examined together with the purified hetero-IgG1 from tran- sient transfection in SDS-polyacrylamide gel and Western blotting. Under nonreducing condition (Fig. 7A), the dominant band of full-size IgG and bands for half-size IgG, LC dimer, and LC monomer were detected. Under reducing conditions (Fig.  7B), two HCs were separated due to their different size and LCs migrated concurrently due to their identical size. The results indi-cated that hetero-IgG1 can be stably produced from CHO-K1 cells, although minor half-size IgGs and LCs are present.
The conditioned medium was purified by a standard protein A column followed by Superdex 200 size exclusion chromatography. The final products showed ϳ100% purity by analytical size exclusion chromatography. Mass spectrometry analysis  Table 2. Variants V15, V20, V21, V22, V23, and V25 are indicated in Table 3. LC1 and HC1 are derived from anti-HER2 trastuzumab, and LC2 and HC2 are derived from ant-EGFR panitumumab. The ؉ symbol indicates the presence of the particular plasmid DNA for transfection, and the Ϫ symbol indicates its absence.

TABLE 3 New variants to improve the cognate LC-HC pairings by negative design principle
Hetero-IgG1 variants in the absence of Fn3 and Fn3-FLAG-His 6 tags were produced by transient transfection of 2936E cells. The tolerance of LC-HC pairings were assessed by Western blotting. Most variants take the negative design principle by introducing opposite charged residues at the same positions of opposing chain. Asterisk on V02 indicates the disulfide bond instead of charge pairs was introduced in the respective antibody.

Anti-HER2 humAb4D5-8 HC1 (K392D ؉ K409D) Anti-HER2 humAb4D5-8 LC1
Anti-EGFR E7.6.3 HC2 (E356K ؉ D399K) Anti-EGFR 7.6.3 LC2 Q39K S183D Q38D S176K Q39D S183K Q38K S176D V16 Q105K S183D A43D S176K Q105D S183K A43K S176D V17 Q39K S183D Q38D S176K Q105D S183K A43K S176D V18 Q105K S183D A43D S176K Q39D S183K Q38K S176D V19 Q39K A141D ϩ S183K Q38D F116K ϩ S176D Q39D A141K ϩ S183D Q38K F116D ϩ S176K V20 G44K ϩ Q105K S183D Q100D ϩ A43D S176K G44D ϩ Q105D S183K G100K ϩ A43K S176D  Table 4. LC1 and HC1 are derived from anti-HER2 trastuzumab, LC2 and HC2 are derived from ant-HER2 pertuzumab. The ؉ symbol indicates the presence of the particular plasmid DNA for transfection, and the Ϫ symbol indicates its absence. and complete reduction by DTT, the intact hetero-IgG1 was shown to contain four different chains. The two different HCs were correctly and efficiently processed (Fig. 7, I and J). A ladder of 1-3 extra Arg were found to be retained at the C termini of anti-EGF LC2 (Fig. 7L), whereas anti-HER2 LC1 had been correctly processed (Fig. 7K). Most importantly, no mis-matched HC-LC pairing (LC1 ϩ HC2 or LC2 ϩ HC1) was identified (Fig.  7H). More work is required to improve the homogeneity of hetero-IgG antibodies expressed from stably transfected mammalian cells. rhuHER2 and rhuEGFR Bind Simultaneously to Anti-HER2 ϫ EGFR Hetero-IgG1 Antibodies with Comparable Affinity as the Parental Antibodies-SPR binding analysis was conducted using a ProteOn XPR36 optical biosensor equipped with a GLC sensor chip. Channels in the vertical direction were used to stably capture hetero-IgGs and parental antibodies using a goat anti-human IgG capture surface. 75 nM rhu-EGFR was injected over the captured antibody surfaces in the horizontal direction at time 0 -420 s followed by a second injection of 75 nM rhuHER2 at time 800 -1220 s. Both the anti-HER2 ϫ EGFR hetero-IgGs V23 (Fig. 8A) and V23_W165 (Fig. 8B) demonstrated additive and simultaneous binding as the binding signal increased after the rhuEGFR injection and further increased after the rhuHER2 injection, whereas the parental anti-EGFR and anti-HER2 IgG1 antibodies demonstrated an increase in binding signal only for their respective antigen injection (Fig. 8,  C and D). The reciprocal experiments by injecting rhuHER2 first followed by rhuEGFR produced similar results (data not shown). Collectively, these results demonstrated that both Fab arms in the hetero-IgG1 can bind to their specific antigens simultaneously and irrespective of the order of addition in this protein based assay.
SPR binding analysis was used to measure the binding affinity of the hetero-IgGs and parental antibodies to rhuHER2 and rhuEGFR. A 3-fold rhuHER2 and rhuEGFR dilution series ranging from 25.0 to 0.309 nM was injected over the captured hetero-IgGs and parental antibody surfaces. The association phase was monitored for 420 s, and the dissociation phase was monitored for 3600 s. Both the anti-HER2 ϫ EGFR hetero-IgG1 V23 (Fig. 8E) and V23_W165 (Fig. 8F) showed similar association and dissociation rates as the parental anti-HER2 humAb4D5-8 IgG1 antibody (Fig. 8G) when binding to rhuHER2 with an affinity of 61.89, 65.95, and 60.08 pM, respectively (Table 5). Both the anti-HER2 ϫ EGFR hetero-IgG1 V23 (Fig. 8H) and V23_W165 (Fig. 8I) showed similar association and dissociation rates compared with the parental anti-EGFR E7.6.3 IgG1 antibody (Fig. 8J) when binding to rhuEGFR, with an affinity of 115.06, 118.28, and 116.50 pM, respectively ( Table  5). The comparable binding kinetics of the hetero-IgGs to that of parental antibodies suggested that the introduced charge pairs and the extra 1-3 Arg ladder did not impact antigen binding. Likewise, the ADCC enhancement mutations in the Fc variant W165 (40) had no impact on antigen binding. rhuHER2 binds to Anti-HER2 ϫ HER2 Hetero-IgG Antibodies with an Intermediate Affinity Compared with the Parental Antibodies-To investigate the binding kinetics of the anti-HER2 ϫ HER2 hetero-IgGs that are derived from anti-HER2 humAb4D5-8 and anti-HER2 humAb2C4 parental antibodies, SPR analysis was similarly conducted using the ProteOn XPR36 optical biosensor with an antibody capture format. Interestingly, both anti-HER2 ϫ HER2 hetero-IgGs V23 (Fig. 9A) and V23_W165 (Fig. 9B) (Table 6), respectively, when compared with the parental antibody kinetics. The parental anti-HER2 humAb4D5-8 IgG1 (Fig. 9C) has a relatively faster association rate with a higher affinity of 60.08 pM (Table 6) compared with the hetero-IgGs, whereas the parental anti-HER2 humAb2C4 IgG1 has a slower association rate (Fig. 9D) with a lower affinity of 285.9 pM ( Table 6). The results confirmed that both the Fab arms from anti-HER2 humAb4D5-8 and humAb2C4 within the hetero-IgG1 context can bind rhuHER2.

TABLE 4 Charged residue pairs at the VH-VL and C H 1-CL interfaces were swapped to different combinations to explore the electrostatic steering effect
The amino acid residues in variable regions (VH or VL) are numbered by Kabat numbering system and amino acid residues in constant regions (C H 1 or C ) are numbered by Eu numbering system.
Asymmetrically Engineered Hetero-IgG Antibodies Elicit Potent ADCC Killing to Tumor Cells-One objective of hetero-IgG antibodies is to kill tumor cells with high potency by the enhanced ADCC effector function, which was achieved by asymmetrical Fc engineering (40). Asymmetrical Fc variant W165 was integrated in anti-HER2 ϫ EGFR and anti-HER2 ϫ HER2 hetero-IgG V23 individually. NCI-N87 cells expressing a high level of HER2 and a moderate level of EGFR were used as target cells for ADCC killing assay (Fig. 10A). The human IgG1 isotype control did not show meaningful killing. The anti-EGFR E7.6.3 IgG1 had 20% specific lysis at the highest concentration of 10 nM, with lower activity when it was titrated down. The anti-HER2 humAb4D5-8 IgG1 showed 80% specific lysis at 10 nM and titrated down in a dose-dependent manner with EC 50 ϭ 17.75 pM. The combination of anti-EGFR E7.6.3 IgG1 and anti-HER2 humAb4D5-8 IgG1 showed slightly lower killing (EC 50 ϭ 33.64 pM) compared with anti-HER2 humAb4D5-8 IgG1 alone, indicating the ADCC killing was mainly driven by anti-HER2 humAb4D5-8 and is in line with higher HER2 expression on NCI-N87 cells. The anti-HER2 ϫ EGFR hetero-IgG1 V23 having regular Fc showed the killing between the two parental antibodies, whereas a potent killing was observed for the ADCC-enhanced anti-HER2 ϫ EGFR hetero-IgG1 V23_W165 (EC 50 ϭ 2.55 pM). JIMT-1 cells, which express high levels of HER2 but low levels of EGFR and are resistant to anti-HER2 trastuzumab treatment (44), showed low killing by all Ab treatments except that the ADCC-enhanced anti-HER2 ϫ EGFR hetero-IgG1 V23_W165 had strong killing at EC 50 ϭ 4.173 pM (Fig. 10B).
SK-BR-3 cells expressing high levels of HER2 were used to assess the activity of different anti-HER2 antibodies (Fig. 10C). Although the human IgG1 isotype control did not show meaningful killing, anti-HER2 humAb2C4 IgG1 alone had 70% specific lysis at the highest dose with EC 50 ϭ 220.4 pM. The anti-HER2 humAb4D5-8 IgG1 alone had stronger killing at EC 50 ϭ 19.71 pM. The combination of anti-HER2 humAb2C4 and humAb4D5-8 IgG1s killed SK-BR-3 cells intermediately with EC 50 ϭ 46.16 pM. The anti-HER2 ϫ HER2 hetero-IgG1 V23 also killed SK-BR-3 cells at intermediate EC 50 ϭ 49.85 pM. However, the anti-HER2 ϫ HER2 hetero-IgG1 V23_W165, which had incorporated ADCC enhancement of Fc variant W165, strongly killed the SK-BR-3 cells with EC 50 ϭ 2.272 pM. Similar results were found with BT-474 cells (Fig. 10D). These results confirmed that the asymmetrically engineered Fc variant W165 can enhance the ADCC effector function.

Hetero-IgG Antibodies Inhibit the Phosphorylation of Receptors and AKT in Downstream
Signaling Pathway-To test how well anti-HER2 ϫ EGFR hetero-IgG1 antibodies inhibit the signaling pathways, BxPC-3 cells were treated with either the titrated parental anti-HER2 humAb4D5-8 IgG1, anti-EGFR E7.6.3 IgG1, the combination of two parental antibodies, anti-HER2 ϫ EGFR hetero-IgG1 V23, ADCC-enhanced anti-HER2 ϫ EGFR hetero-IgG1 V23_W165, or the isotype human IgG1 control. For a proper comparison, the total dose of single parental antibody, the combination of the two parental antibodies, and hetero-IgG1s were normalized by their binding valences to each target receptor; hence they have equal binding capacity to the receptors. As shown in Fig. 10E, the phosphorylation of EGFR was neither inhibited by isotype human IgG1 control nor by anti-HER2 humAb4D5-8 IgG1, but it was strongly inhibited by the parental anti-EGFR E7.6.3 IgG1, with 82.11% of inhibition at the plateau dose (IC 50 ϭ 1.49 nM). The inhibition from the combination of two parental Abs was slightly weaker, with 77.60% at the plateau dose (IC 50 ϭ 5.48 nM). A comparable inhibition of pEGFR was observed for either anti-HER2 ϫ EGFR hetero-IgG1 V23 (IC 50 ϭ 4.85 nM) or ADCC-enhanced anti-HER2 ϫ EGFR hetero-IgG1 V23_W165 (IC 50 ϭ 4.45 nM).
The basal expression level of HER2 on BxPC-3 cells was too low to have enough window for the detection of pHER2, regardless of the cells being stimulated with or without 100 ng/ml neuregulin 1, and even the treatment by the combined parental anti-HER2 humAb4D5-8 and anti-EGFR E7.6.3 IgG1s (10 g/ml each) revealed no effect on pHER3 and pAKT in BxPC-3 cells (data not shown).
when compared with the parental anti-HER2 humAb2C4 IgG1. Only about a 40% inhibition on pHER2 (Fig. 10F), although a strong inhibition on pHER3 (Fig. 10G, IC 50 ϭ 5.7 nM) and pAKT (Fig. 10H, IC 50 ϭ 6.78 nM), was observed for the combination treatment of the two parental Abs at the plateau dose. Single agent treatment by anti-HER2 humAb2C4 IgG1 had IC 50 ϭ 5.44 nM for pHER3 and IC 50 ϭ 6.08 nM for pAKT, but single agent treatment by the anti-HER2 humAb4D5-8 IgG1 had a low impact on pHER3 and pAKT (Fig. 10, G and H). IgG1, or anti-HER2 ϫ EGFR hetero-IgG1 V23 bound strongly (Fig. 11A) and comparably (Fig. 11F) to BxPC-3 cells. Anti-HER2 humAb4D5-8 IgG1 alone bound with low intensity to BxPC-3 cells (image not shown), correlating with the fact that moderate level of EGFR (about 300,000 sites per cell) and a low level of HER2 (about 20,000 sites per cell) are expressed on BxPC-3 cells (45). At the 4-h time point, no formation of punctate spots was detected for anti-HER2 humAb4D5-8 IgG1 treatment (Fig. 11B), although many punctate spots were observed inside cells for the treatment with anti-EGFR E7.6.3

TABLE 5 Kinetic rate binding constants for rhuHER2 and rhuEGFR binding to anti-HER2 ؋ EGFR hetero-IgG1 and parental antibodies
Kinetic rate coefficients were determined from binding analysis experiments performed with a ProteOn XPR36 biosensor. Five concentrations of monomeric rhuHER2 or rhuEGFR ranging between 25.0 and 0.309 nM were run in triplicate against the captured antibodies on a GLC surface. The k d values were determined by fitting the respective 3600-s dissociation phase data and then using this value as a fixed parameter in the global fits of the association phase data to a 1:1 binding model to calculate the respective k a and R max values. The error is the standard error of the global fit for triplicate surfaces. rhuHER2, monomeric recombinant human HER2 extracellular domain; rhuEGFR, monomeric recombinant human EGFR extracellular domain.

Antibody
Analyte k a k d K d   IgG1 alone (Fig. 11C) or the combination of anti-EGFR E7.6.3 IgG1 and anti-HER2 humAb4D5-8 IgG1 (Fig. 11D). An increased amount of punctate spots was observed for the treatment with the bispecific anti-HER2 ϫ EGFR hetero-IgG1 V23 (Fig. 11E). Semi-quantitation by the ArrayScan VTI reader showed that the anti-HER2 ϫ EGFR hetero-IgG1 induced ϳ4-fold higher levels of target internalization than anti-EGFR E7.6.3 IgG1 alone and ϳ2-fold higher internalization than the combination of two parental antibodies when the incubation time reached 1 h. At the 4-h time point the anti-HER2 ϫ EGFR hetero-IgG induced ϳ5-fold more target internalization than either anti-EGFR E7.6.3 IgG1 alone or the combination of two parental antibodies (Fig. 11F). Similar results were observed for Panc-1 cells and Colo699 cells (data not shown). The data suggested that the bispecific anti-HER2 ϫ EGFR hetero-IgG1 may have the unique capability of down-regulating HER2 and EGFR receptors from the cell surface by internalization. A431 cells, which express high levels of EGFR but almost undetectable levels of HER2, did not show accelerated internalization rates compared with either parental antibody alone or a combination of parental antibodies (data not shown), suggesting that the engagement of both EGFR and HER2 and heightened clustering of receptors are required for a higher internalization rate.
hanced anti-HER2 ϫ EGFR hetero-IgG1 V23_W165 significantly reduced the tumor size at the end of study when compared with the combination of two parent IgG1 antibodies (p ϭ 0.0473), whereas the ADCC-norm anti-HER2 ϫ EGFR hetero-IgG1 V23 did not show any significant difference (p ϭ 0.2701). These results showed that anti-HER2 ϫ EGFR hetero-IgG1s strongly inhibited the tumor growth, and ADCC enhancement helps antibody to overcome K-RAS mutation, which is in line with the report by Schlaeth et al. (46). The results also agreed with the finding that anti-HER2 ϫ EGFR hetero-IgG1 induces higher internalization of receptors than single parental Ab or the combination of the two parental Abs in this report.
The effect of anti-HER2 ϫ HER2 hetero-IgG1 on tumor growth was tested in a xenograft tumor model of Calu-3 cells which highly express HER2 at about 300,000 sites per cell. Tumor-bearing NSG mice were treated similarly as those in Panc-1 xenograft tumor model. As shown in Fig. 12C, significant tumor growth inhibition was observed for all antibody treatment groups when compared with the huIgG1 isotype control (p Ͻ 0.0001). However, the differences among the treatment by single parental antibody, the combination of two parental antibodies, or anti-HER2 ϫ HER2 hetero-IgG1 could not be demonstrated at the concentrations tested in this study.
Stability of Hetero-IgG1 bsAbs in Human Serum Is Comparable with That of Their Parent Antibodies-To assess the in vitro stability of hetero-IgG1 bsAbs, the bsAbs and their parental antibodies were incubated in 90% human serum at 37°C up to 168 h. The stability of anti-HER2 ϫ EGFR and anti-HER2 ϫ HER2 hetero-IgG1s was determined by their binding capacity to the captured EGFR and HER2 antigens using ELISA as readout. As shown in Fig. 13A, anti-HER2 ϫ EGFR hetero-IgG1 V23 retained similar binding capacity to EGFR as the parental anti-EGFR E7.6.3 IgG1 over the incubation in human serum. However, in the HER2 binding assay anti-HER2 ϫ EGFR hetero-IgG1 V23 lost ϳ30% of HER2 binding after 168 h of incubation in human serum, and a similar trend was observed for the parental anti-HER2 humAb4D5-8 IgG1 (p ϭ 0.7208) (Fig.  13B). Similarly the anti-HER2 ϫ HER2 hetero-IgG1 V23 lost ϳ30% of HER2 binding at the end of incubation, which tracked the parental anti-HER2 humAb4D5-8 IgG1 because the HER2 binding of a second parental anti-HER2 humAb2C4 IgG1 stayed the same over the incubation (Fig. 13C). There was no statistical difference when comparing anti-HER2 ϫ HER2 hetero-IgG1 V23 with either parental anti-HER2 humAb4D5-8 IgG1 (p ϭ 0.7027) or humAb2C4 IgG1 (p ϭ 0.0534) at the 168-h time point. The data suggested that the hetero-IgG1s had maintained comparable stability in human serum as the parental antibodies.
Prediction of Immunogenicity for Hetero-IgGs-To explore the possibility that the bispecific hetero-IgGs could elicit immunogenicity after the charged residue pairs are introduced in different domains of antibody, we utilized an in silico TEPITOPE algorithm to predict the potential immunogenicity for the parent IgG1 and hetero-IgG1 antibodies (Table 7). Although the parent anti-EGFR E7.6.3 IgG1 did not have any predicted 9-mer peptides binding to DRB1 alleles, one peptide (173-181) in C and one peptide (36 -44) in the VH of anti-EGFR E7.6.3 in the context of hetero-IgG1 were predicted to FIGURE 13. In vitro stability of anti-HER2 ؋ EGFR and anti-HER2 ؋ HER2 hetero-IgG1 bsAbs in human serum. A, retention of the parent anti-EGFR E7.6.3 IgG1 and anti-HER2 ϫ EGFR hetero-IgG1 V23 when binding to the captured biotin-EGFR. B, retention of the parent anti-HER2 humAb4D5-8 IgG1 and anti-HER2 ϫ EGFR hetero-IgG1 V23 when binding to the captured biotin-HER2. C, retention of the parent anti-HER2 humAb4D5-8 IgG1, parent anti-HER2 humAb2C4 IgG1, and anti-HER2 ϫ HER2 hetero-IgG1 V23 when binding to the captured biotin-HER2. The data were derived from two separate ELISA tests.
bind to DRB1_0401 and DRB1_0301, respectively. The parent anti-HER2 humAb4D5-8 IgG1 had three potential DRB1binding peptides. There were two additional linear peptides (173-181 and 175-183) in C, one additional peptide (36 -44) in VH, and two additional linear peptides (404 -412 and 406 -414) in C H 3 of the hetero-IgG1 that could bind to DRB1 alleles. Similarly, the parent anti-HER2 humAb2C4 IgG1 had five potential DRB1-binding peptides. There was one additional linear peptide (173-181) in C and one additional linear peptide (36 -44) in VH of the hetero-IgG1 that were predicted to bind to DRB1_0401 and DRB1_0301, respectively. It should be noted that the introduced charge residues in all additional DRB1binding peptides are buried at the VH-VL and C H 1-CL interfaces (Fig. 1); therefore, the engineered residues are not likely to present B cell epitopes. It remains to be tested whether the predicted additional linear 9-mer peptides in hetero-IgG1 have high enough binding affinity to HLA molecules and induce any nascent immunogenicity.

DISCUSSION
In this report we described an antibody engineering strategy to produce monovalent bispecific heterodimeric IgG from two pre-existent antibodies in mammalian cells. The hetero-IgG bsAbs produced in this format have the same overall size and natural structure as the regular IgG antibody. In particular, we tested multiple designs to engineer LC-HC interface residues so that LC strongly favors its cognate HC during the assembly of bispecific heterodimeric IgGs. This was achieved by introducing charged residues with opposite polarity at the selected positions between VH-VL and C H 1-C interfaces.
During the assembly process of bsAb inside cells, each HC is retained in the endoplasmic reticulum by the BiP protein through the interaction in the C H 1 region. The BiP protein can be exchanged by either of the two LCs to form the IgG molecules that are subsequently secreted from cells. It is reasonable to assume that the relative assembly kinetics between HC and the two LCs will determine the final pairing of the HC with each LC. The interactions of both VH-VL and C H 1-CL domains contribute to final recognition and assembly of HC and LC (27)(28)(29)(30)(31).
We selectively placed charged residues with opposite polarity at VH-VL and C H 1-CL interfaces of the cognate HC-LC so that LC is attractive to its cognate HC due to the presence of contacting residues with opposite charge polarity. This electrostatic steering strategy is likely to impact the kinetic of LC-HC assembly process by affecting the "on-rate" of LC-HC recognition, which is very sensitive to the charge-based interactions. To completely avoid the LC-HC mispairing issues, it is very important to apply a negative design principle in the engineering process so that neither of the two LCs can efficiently assemble with its noncognate HC. For any given pair of antibodies, this can be achieved empirically by using different designs as exemplified in this report. We also developed a very stringent LC-HC mispairing assay in which each design of different LCs is introduced with its noncognate HC by co-transfection of HEK293 cells and tested for its ability to support the secretion of IgG.
The residues, which we selected to apply in the charge pairbased engineering strategy, are well conserved among most germlines of VH and VL families of all IgG isotypes. In principle, this engineering strategy could be applied in many, if not all, of the pre-existent human antibodies. The fewer numbers of charged residue pairs that are introduced in hetero-IgG, the less impact on antibody expression and stability. Among the many bsAbs produced using this strategy, we found that engineering both VH-VL and C H 1-CL interfaces is generally required to completely avoid mispaired LC-HC species. A similar observation was reported recently by Lewis et al. (47) who described different antibody engineering solutions to address LC-HC mispairing issues for the production of heterodimeric IgG. To minimize the impact on antigen binding, it is preferable to keep the variable regions (VH-VL) untouched. We found that in some particular cases by introducing charged residue pairs only in constant regions (C H 1-CL) the cognate LC-HC pairings could be achieved (data not shown).
Immunogenicity is always a concern when engineering any proteins, including antibodies. However, this question cannot be completely answered until the engineered protein is tested in

Prediction of potential immunogenicity by in silico analysis for parent and hetero-IgG1 antibodies
The TEPITOPE algorithm was utilized to identify potential non-tolerant agretopes which could bind to HLA class II molecules and elicit immune responses. Tertiary structural information of the antibodies was not considered in this analysis. The linear 9 residue peptides that could bind to HLA class II molecules are listed. The 8 DRB1 alleles cover 95% of human populations. The red colored "D" and "K" indicate the engineered residues. human subjects. A fully human anti-TNF antibody adalimumab can elicit immune responses in patients (48). Because the positions for charged pair residue substitutions in this report are buried inside the hydrophobic core of the VH-VL and C H 1-CL interfaces, it is reasonable to speculate that minimal humoral immune response could be elicited due to the introduction of charged residue pairs. However, more work is needed to determine whether the hetero-IgG1s induce T celldependent immune responses in humans.
Most of the LC variants described in this report were based on C chain in the constant region. However, because the C and C constant domains are structurally conserved, particularly at the LC-HC interfaces, the negative design principle could also be applied in addressing the LC-HC mispairing issues when the LC(s) is engineered. Alternatively, the constant region can be switched from C to C to make a V-C chimeric LC. We did observe a slight decrease of expression titer for some antibodies when V-C chimeric LC was used. The main advantage of using the same LC constant regions is that this strategy could achieve a maximum electrostatic steering effect by the negative design principle. When a residue in LC is changed to a positively charged residue to accommodate a negatively charged residue in its cognate HC, it is also designed to repel from the positively charged residue in its noncognate HC. Similarly, a negatively charged residue is introduced at the same position of the opposing LC to accommodate a positively charged residue in its cognate HC; it is designed to repel from the negatively charged residue in its noncognate HC as well. Introducing a bulky and charged residue (e.g. Arg) to exert both steric clash and electrostatic steering effect could disfavor the unwanted LC-HC pairings (data not shown).
The bispecific anti-HER2 ϫ EGFR hetero-IgG1 induces a higher level of receptor internalization than the combination of two parental antibodies. This is likely due to the unique feature of two different Fab arms that could simultaneously bind to two different receptors on the cell surface. Down-regulation of receptors EGFR and HER2 has been proposed as one of the antitumor mechanisms for the combined anti-EGFR and anti-HER2 antibody treatment in mouse xenograft models (7). The bispecific anti-HER2 ϫ EGFR hetero-IgG1 may have the same effect. Each Fab arm in anti-HER2 ϫ HER2 hetero-IgG1 can individually bind antigen HER2. Anti-HER2 ϫ HER2 hetero-IgG1 V23 functions comparably as the combination of two parental antibodies in the ADCC killing assay but less potently than the combination of two parental antibodies in pHER2/ pHER3/pAKT assays. It still remains to be tested whether the two different Fab arms in anti-HER2 ϫ HER2 hetero-IgG1 can simultaneously bind to two different epitopes on HER2.
The bispecific anti-HER2 ϫ EGFR hetero-IgG1s strongly inhibited tumor growth in the mouse xenograft BxPC-3 human tumor model. Higher antitumor activity of anti-HER2 ϫ EGFR hetero-IgG1s was observed than that of the combination of two parental antibodies in the Panc-1 xenograft bearing Rag2 Ϫ/Ϫ / mFc␥R4 Ϫ/Ϫ /hCD16a ϩ transgenic mice, suggesting that the bispecific anti-HER2 ϫ EGFR hetero-IgG1s may have advantages over the combination of two parental Abs. Furthermore, the ADCC-enhanced anti-HER2 ϫ EGFR hetero-IgG1 V23_ W165 had higher antitumor activity than anti-HER2 ϫ EGFR hetero-IgG1 V23, indicating that ADCC enhancement can elicit potent killing to tumor cells to overcome K-RAS mutation in Panc-1 cells (46). In another aspect, all anti-HER2 antibodies worked well in the Calu-3 xenograft tumor model in female NSG mice, different from the studies by Scheuer et al. (42) in which BALB/c nu/nu female mice were tested. One possible explanation for the increased activity observed in our study is that, unlike BALB/c nu/nu mice, the NSG mice lack serum Ig to block Fc␥Rs on effector cells such as neutrophil, monocytes, and macrophages. Therefore, the effector cells in NSG mice could more effectively bind to Fc fragments of human IgG1, leading to a strong tumor growth inhibition to Calu-3 cells by engaging murine Fc␥Rs.
In summary, we developed an antibody engineering approach to produce bispecific hetero-IgG1 antibodies in mammalian cells. The hetero-IgG1 we generated contains the predicted four different chains, binds to two different antigens with comparable affinity when compared with their parental antibodies, retains their functionality of parental antibodies, and induces higher receptor internalization than the combination of two parental anti-HER2 and anti-EGFR antibodies. Although BxPC-3, Panc-1, and Calu-3 human tumor xenograft models showed that the hetero-IgG1s strongly inhibited the tumor growth, more studies are warranted with different antibodies to demonstrate that our observations are expandable in this hetero-IgG format. The approach described in this report could be applied to the evaluation of bsAb using two preexistent antibodies as well as for the development of novel therapeutic molecules for treatment of many diseases such as cancers and infectious diseases. The asymmetrically engineered Fc variants for ADCC enhancement could be embedded in monovalent bispecific hetero-IgG1 to make best-inclass therapeutic antibodies.