Izumi Kumagai, Department of Biomolecular Engineering, Graduate School of Engineering, Tohoku University, Aoba 6-6-11-606, Aramaki, Aoba-ku, Sendai 980-8579, Japan Fax: +81 22 795 6164 Tel: +81 22 795 7274 E-mail: firstname.lastname@example.org
We previously reported the construction and activity of a humanized, bispecific diabody (hEx3) that recruited T cells towards an epidermal growth factor receptor (EGFR) positive tumor. Herein, we describe the construction of a second functional, fully humanized, anti-EGFR bispecific diabody that recruits another subset of lymphocyte effectors, the natural killer cells, to EGFR-expressing tumor cells. After we confirmed that an anti-EGFR × anti-CD16 bispecific diabody (Ex16) consisting of a previously humanized anti-EGFR variable fragment (Fv) and a mouse anti-CD16 Fv had growth inhibitory activity, we designed a humanized anti-CD16 Fv to construct the fully humanized Ex16 (hEx16). However, the humanized form had lower activity for inhibition of cancer growth. To restore its growth inhibitory activity, we introduced mutations into the Vernier zone, which is located near the complementarity-determining regions and is involved in their binding activity. We efficiently prepared 15 different hEx16 mutants by expressing each chimeric single-chain component for hEx16 separately. We then used our in vitro refolding system to select the most functional mutant, which had a growth inhibitory effect comparable with that of the commercially available chimeric anti-EGFR antibody, cetuximab. Our refolding system could aid in the efficient optimization of other proteins with heterodimeric structure.
3-(4,5-dimethylthiazole-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium, inner salt
lymphokine-activated killer cells with the NK cell phenotype
peripheral blood mononuclear cells
surface plasmon resonance
T cell receptor
lymphokine-activated killer cells with the T-cell phenotype
Bispecific antibodies (BsAbs) are recombinant antibodies that can bind to two different antigenic epitopes. Bispecificity can be used in cancer immunotherapy to crosslink tumor cells to immune cells such as cytotoxic T cells, natural killer (NK) cells and macrophages. This crosslinking accelerates the destruction of tumor cells by immune cells, which may translate into improved antitumor therapy and lower costs by decreasing the doses needed for therapy [1,2]. However, the use of BsAbs in clinical studies has been hampered by difficulties in producing them on a large scale. Conventional chemical conjugation results in inconsistent quality of the antibodies produced . The production of BsAbs by somatic fusion of two hybridomas to form a quadroma yields BsAbs of more consistent quality, but this results in the formation of various chain-shuffled antibodies. For instance, 10 different antibodies can be generated after random association of two heavy and two light chains [4,5].
Advances in recombinant DNA technology have made it feasible to generate small recombinant BsAb fragments constructed from two different variable antibody fragments. Bispecific diabodies are the smallest available BsAb fragments and the distance between the two antigen binding sites is sufficient to link two cells [6,7]. The effectiveness of bispecific diabodies in cancer therapy has been extensively shown in both in vitro and in vivo models [8–10]. We have constructed functional bispecific diabodies [11,12] and have reported that a humanized bispecific diabody, hEx3, has marked antitumor activity and can retarget lymphokine-activated killer cells with the T-cell phenotype (T-LAK cells) against epidermal growth factor receptor (EGFR) positive cell lines [13,14].
Among immune cells, cytotoxic T lymphocytes (CTLs) are one of the most suitable candidates for targeted immunotherapy, as they participate in the recognition and subsequent killing of tumor cells, virus-infected cells and allogeneic targets . The primary cytotoxic trigger on CTLs is the cluster of differentiation 3/T-cell receptor (CD3–TCR) complex, which is antigen-specific but is restricted by the major histocompatibility complex (MHC). However, BsAbs can react with the CD3–TCR complex to initiate retargeted cytotoxicity without any MHC restriction [1,2]. In contrast, NK cells are a component of the body’s innate immunity and can both lyse target cells and provide an early source of immunoregulatory cytokines . Most human NK cells express high levels of FcγRIII (CD16), which plays a critical role in the induction of antibody-dependent cellular cytotoxicity (ADCC), which is one of the major modes of action of most therapeutic antibodies. Thus, CD16 is an attractive candidate for targeted immunotherapy, and effective CD16-mediated cytotoxicity induced by bispecific diabodies has been documented for malignant tumors [7,9].
Herein, we describe our construction of a second functional bispecific diabody against EGFR that uses an anti-CD16 variable fragment (Fv) to recruit a different subset of lymphocyte effectors, the NK cells. The resultant anti-EGFR × anti-CD16 bispecific diabody (Ex16) showed marked growth inhibitory activity that was clearly dependent on the effector cells. However, humanization of the diabody resulted in a considerable decrease in its growth inhibitory activity due to a reduction in its binding affinity. The Vernier zone , which is located near the complementarity-determining regions (CDRs), plays a role in the binding activity of antibodies. To recover the function of the diabody that had been lost as a result of its humanization, we introduced mutations in the Vernier zone residues by preparing 15 different humanized Ex16 (hEx16) mutants and selected the most functional one by using the in vitro refolding system described in an earlier report . To our knowledge, this is the first report of a functional, fully humanized, bispecific diabody that can retarget EGFR and CD16 on tumor cells and was prepared with the use of our refolding system. Thus, our refolding system has again been demonstrated to be suitable for the efficient optimization of heterodimeric proteins.
Preparation of refolded Ex16
To prepare Ex16 with specificity for EGFR and CD16, we applied the in vitro refolding system used for refolding of hEx3, as described previously [13,14]. The two chimeric single-chain components of Ex16, 3GHh5L and h5H3GL, were separately expressed in Escherichia coli. Results of SDS/PAGE and western blotting showed that each gene product was primarily and substantially present in the intracellular insoluble fractions (Fig. 1). Furthermore, high purity and stoichiometric association of chimeric single-chain components were observed in refolded Ex16 (Fig. 1), demonstrating the successful preparation of Ex16 using our in vitro refolding system.
Growth inhibitory effect of Ex16
To evaluate the inhibition of cancer growth by Ex16, we performed an MTS (3-(4,5-dimethylthiazole-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium, inner salt) assay with TFK-1 human bile duct carcinoma cells by using T-LAK cells or lymphokine-activated killer cells with the NK cell phenotype (NK-LAK) as effector cells. Ex16 inhibited the growth of cancer cells only in the presence of NK-LAK cells (Fig. 2A). Flow cytometric analyses showed that induced NK-LAK cells consisted largely (about 70%) of CD16-positive lymphocytes, whereas T-LAK cells consisted largely of CD3-positive lymphocytes (Fig. 2B). The specific binding of Ex16 to NK-LAK (Ex16) cells was also confirmed by use of flow cytometry (data not shown). These results indicate that the anti-EGFR × anti-CD16 bispecific diabody can inhibit EGFR-positive cancer growth in an effector cell dependent manner.
Humanization of Ex16 to reduce immunogenicity
Because Ex16 demonstrated anti-proliferative activity for cancer cells, we designed and prepared a fully humanized Ex16 (hEx16) to reduce the immunogenicity of Ex16 and make it more suitable for clinical use. The anti-EGFR Fvs used for the construction of Ex16 were already humanized , and we humanized the anti-CD16 Fvs from 3G8 by using the CDR-grafting method, as described in Materials and methods. We selected human sequences as templates for humanized 3G8 Fvs by doing homology searches, taking into account the lengths of the CDRs as well as the residues located at the VH–VL interface  and in the Vernier zone . The sequences of the humanized 3G8 Fvs, shown in Fig. 3, were used to construct the hEx16 expression vectors pRA-h3GHh5L and pRA-h5Hh3GL. hEx16 was successfully prepared by using these vectors, and was subjected to in vitro refolding. However, a substantial decrease was seen in the growth inhibition by hEx16 in the MTS assay with the TFK-1 (Fig. 4A) and A431 (Fig. 4B) cell lines, both of which express high levels of EGFR. Decreased binding affinity due to the humanization of Ex16 may have resulted in the substantially lower growth inhibitory effects of hEx16.
Introduction of mutations into the Vernier zone of humanized 3G8 Fv in hEx16
All residues located at the VH–VL interface whose mutation would affect their interactions with the ligand were conserved in humanized 3G8. However, two residues in each of the Vernier zones of VL and VH were mutated: M4L and G64A in VL, and S29R and R94Q in VH (Fig. 3). When affinity for the ligand is lost after humanization, as seen in the present case, the changed residues often need to be changed back into the corresponding residues in the original murine sequences. However, it is difficult to identify which residues or combinations of residues will critically contribute to the restoration of binding affinity. Therefore, we constructed resubstitution mutants in every possible combination, including single mutations, yielding 15 different hEx16 mutants (Table 1). All mutants were prepared similarly, were of high purity and underwent stoichiometric association of their chimeric single-chain components by using the in vitro refolding procedure described in Materials and methods (data not shown).
Table 1. Mutants constructed in this study and mean fluorescence intensity in flow cytometry. N.C., cells were incubated with NaCl/Pi.
Mean fluorescence intensity
Binding and cancer growth inhibition by hEx16 mutants
Flow cytometric analysis and the MTS assay were performed to evaluate the binding and cancer growth inhibition by hEx16 mutants. Because induced NK-LAK cells contain some CD16-negative cells (Fig. 2B), we established a Chinese hamster ovary (CHO) cell line that stably expressed high levels of CD16 (CD16/CHO) to evaluate more clearly the differences in CD16-dependent binding of the bispecific diabodies. Comparable binding to EGFR-positive TFK-1 cells was observed for Ex16 and hEx16, and each mutant showed a slight increase in binding strength over hEx16 (Table 1). In contrast, the binding strength of hEx16 to CD16/CHO cells was much lower than that of Ex16, and some mutants showed a recovery of binding strength to over 10 times the original strength of hEx16 (Table 1). The cancer cell growth inhibition was also restored for several mutants (Fig. 5). Thus, the function of hEx16 could be restored by changing the residues in the Vernier zones back into the ones present in the original murine sequences of Ex16.
Surface plasmon resonance analysis of fractionated dimers of an hEx16 mutant selected on the basis of its growth inhibitory effect
For further analyses, we limited our studies to certain mutants based on the results of flow cytometry and the MTS assay. Comparable growth inhibitory effects were observed between Mu 1 and 2, Mu 3 and 4, and Mu 11 and 12, respectively, in the MTS assay with TFK-1 cells (Fig. 5). We selected for further analyses those mutants that had fewer mouse sequences, i.e. Mu 1, Mu 3 and Mu 11, as some restoration in their binding strength for CD16 was also seen in flow cytometric analyses (Table 1). Gel filtration of each diabody showed the formation of 58-kDa dimers, but peaks corresponding to monomers and multimers were also observed (data for Ex16, hEx16 and Mu 3 are shown in Fig. 6A as representative examples). Fractionated Mu 3 diabodies (dimer fractions of ∼ 58 kDa) showed the highest growth inhibition among all the selected mutants at any concentration in the MTS assay (Fig. 6B). Therefore, we compared the binding kinetics of Ex16, hEx16 and Mu 3 for CD16 by surface plasmon resonance (SPR) using fractionated diabodies. Marked decreases in both the association and dissociation rates were observed for hEx16, resulting in an affinity constant that was only 4% of the affinity constant for Ex16 (Table 2). In contrast, the binding kinetics of Mu 3 were comparable with those of Ex16 (Table 2), which probably contributed to the restoration of its cancer growth inhibition to a level comparable with that of Ex16.
Table 2. KD values for CD16 evaluated by using surface plasmon resonance.
Comparison of the growth inhibitory effect of Ex16, hEx16 and cetuximab
To compare the growth inhibitory effect of fractionated dimers of Ex16 and the most functional hEx16 mutant, Mu 3, with the US Food and Drug Administration approved therapeutic anti-EGFR antibody cetuximab, we performed the MTS assay with TFK-1 and A431 cells and peripheral blood mononuclear cells (PBMCs) as effector cells. Comparable growth inhibitory effects were observed for Ex16 and cetuximab in both cancer cell lines (Fig. 7A,B). Mu 3 also showed a growth inhibitory effect comparable with that of cetuximab, especially at high doses, thus demonstrating the successful construction of a fully humanized, CD16-targeted bispecific diabody with the potential to be a novel therapeutic antibody fragment.
EGFR is overexpressed in a wide range of human malignancies, and its expression level is correlated with poor clinical outcome in patients with any of several cancers . Therapeutically potent BsAbs targeting EGFR have been prepared by fusion of two hybridomas and by chemical conjugation; however, such classical methods have led to the formation of several nonfunctional analogs [18,19]. We previously reported significant in vitro and in vivo antitumor activity of a humanized bispecific diabody that recruited T cells towards an EGFR-positive tumor (hEx3) . Although studies to examine the usefulness of hEx3 for clinical therapy are currently under way, the most advanced BsAb fragment recruiting T cells, Micromet’s blinatumomab, has shown promising results in a clinical study. However, it has also shown central nervous system symptoms, which have led to permanent discontinuation of the study drug in some cases . Thus, in the present study, we tried to construct a second, functional, fully humanized, anti-EGFR diabody that would recruit another subset of lymphocyte effectors, the NK cells, to EGFR-expressing tumors.
To confirm the usefulness of an anti-EGFR × anti-CD16 diabody, we first prepared Ex16, which consists of a previously humanized, anti-EGFR 528 Fv and a mouse anti-CD16 3G8 Fv. Although downsizing antibodies enables their large-scale preparation by using bacterial expression systems, efficient preparation of functional small antibody fragments has often been hampered by the formation of insoluble aggregates in the cytoplasmic or periplasmic space . We had previously developed an in vitro refolding system to prepare functional, bispecific diabodies from insoluble intracellular aggregates in E. coli [12,14], and we applied it in this study to the preparation of Ex16. We observed both a stoichiometric association of chimeric single-chain components and effector cell dependent growth inhibition with refolded Ex16 (Figs 1, 2), demonstrating the successful preparation of Ex16 with our in vitro refolding system and the usefulness of Ex16 for inhibiting the growth of cancer cells.
We then designed a humanized version of the anti-CD16 Fv (Fig. 3) to construct a fully humanized Ex16 (hEx16) in order to reduce the immune response against murine antibodies in human hosts. However, the resultant hEx16 showed a substantial decrease in the growth inhibition of EGFR-expressing cancer cells (Fig. 4). Grafting of the CDRs of murine antibodies onto appropriate human frameworks has often resulted in reduced affinity or specificity for the target antigen [22,23]. Because the CDR-grafting method is widely used for humanizing murine antibodies, there are a few general strategies for the recovery of the binding affinity of humanized antibodies, although these strategies often require several trial-and-error approaches [24–26]. The humanized Ex16 had reduced binding strength (Table 2); hence, we constructed hEx16 mutants to restore the ability of the diabody to inhibit the growth of EGFR-expressing cancer cells.
Previous work has suggested that residues in the β-sheet framework underlying the CDRs play a critical role in the adjustment of the loop structures of the CDRs; these residues are referred to as the Vernier zone . Although residues located at the VH–VL interface have also been reported to be important , we focused only on the Vernier zone because the residues at the VH–VL interface were all conserved in humanized 3G8. Within the Vernier zone, two residues in VL and two in VH were mutated as a result of humanization. Because it is difficult to identify or predict which residue or combinations of residues can critically contribute to the recovery of binding affinity, we constructed mutants in every combination (i.e. 15 different hEx16 mutants).
Bispecific diabodies are usually prepared using coexpression vectors containing two chimeric single-chain components from the culture supernatant or the periplasmic fraction [8,28]. In the present study, we would have needed to construct 15 different expression vectors if we had used a coexpression vector system. Instead, we were able to do this efficiently by combining four kinds of h5Hh3GL and h3GHh5H (Table 1). The most functional hEx16 mutant, Mu 3, demonstrated binding affinity comparable with Ex16 in SPR studies (Table 2) and a growth inhibitory effect comparable with that of cetuximab (Fig. 7). Thus, this method of separately expressing each chimeric single-chain component and refolding the two chimeric single-chain components after stoichiometric mixing could be used to efficiently prepare diabody mutants, especially when many mutants need to be constructed. All BsAb fragments currently in clinical trials, such as blinatumomab  and Affimed’s AFM13 (http://www.affimed.com/afm13), are produced in expensive mammalian expression systems even though they are small-format antibody derivatives. Our in vitro refolding system could provide a low-cost alternative production method based on a bacterial expression system for similar BsAb products.
The values of affinity (148 nm) and EC50 (1.2 nm calculated for A431) for hEx16 are lower than those previously reported for BsAb fragments based on anti-CD16 Fv [30–32], and a more homogeneous preparation is desired. To date, several different small BsAb fragments have been proposed to increase efficacy; in the cases of BsAb fragments targeting CD16, the formats designed with two binding sites for tumor cells especially have shown remarkably low EC values [31,32]. However, they are also produced in mammalian expression systems, like the advanced BsAb fragments described above. A previous report  and our unpublished data showed that the orientation of the variable domains of the diabody can influence the expression and formation of active binding sites. Thus, we are working to further improve hEx16 by converting the orientation of the variable domains and to optimize its homogeneous preparation by creating additional mutations.
In conclusion, we efficiently constructed a humanized anti-EGFR × anti-CD16 diabody with a growth inhibitory effect comparable with that of cetuximab by using a proven in vitro refolding method. Our refolding system may allow industrial-scale production of bispecific diabodies and also aid in the efficient optimization of additional proteins with heterodimeric structures.
Materials and methods
Preparation of Ex16 diabodies
The mouse hybridoma cell line 3G8 was used as the source of the anti-human CD16 variable region genes. The VH and VL genes of the anti-human CD16 Fv were cloned with primers synthesized based on the sequences of genes described in an earlier report  and designated 3GH and 3GL, respectively. We previously reported the construction of the bacterial expression vectors pRA-hOHh5L and pRA-h5HhOL, which were designed to express the two chimeric single-chain components h5HhOL and hOHh5L, respectively, of a humanized anti-EGFR × anti-CD3 bispecific diabody (hEx3) . The genes coding for the VH and VL regions of the humanized anti-CD3 antibody OKT3 Fv, hOH and hOL respectively, were replaced with 3GH and 3GL, which code for the VH and VL regions of the anti-human CD16 Fv. The resulting vectors, pRA-3GHh5L and pRA-h5H3GL, were used to prepare Ex16 using the in vitro refolding system described in our earlier report . Briefly, 3GHh5L and h5H3GL chimeric single-chain components were individually purified through a TALON Metal Affinity Resin column (Clontech, Palo Alto, CA, USA) from the intracellular insoluble fraction after solubilization with 6 m guanidinium hydrochloride/phosphate-buffered saline (Gu-HCl/NaCl/Pi). Each of the chimeric single-chain component solutions was diluted to 15 μm with 6 m Gu-HCl/NaCl/Pi, and the resulting solutions were then mixed in a 1 : 1 ratio. The denatured chimeric single-chain component mixture (5 mL) underwent stepwise dialysis into NaCl/Pi through solutions of 3, 2, 1 and 0.5 m Gu-HCl/NaCl/Pi.
Preparation of effector cells
T-LAK cells were induced as previously reported . Briefly, PBMCs were cultured for 48 h at a density of 1 × 106 cells·mL−1 in a growth medium supplemented with 100 IU·mL−1 of recombinant human interleukin-2 (kindly supplied by Shionogi Pharmaceutical Co., Osaka, Japan). The culture flask (A/S Nunc, Roskilde, Denmark) containing the T-LAK cells had been precoated with an anti-CD3 mAb (10 μg·mL−1). For the induction of NK-LAK cells, PBMCs were co-cultured for 48 h at a density of 1 × 106 cells·mL−1 with MUC1/B7-cotransfected K562 cells (an irradiated cancer vaccine cell line) at a density of 2 × 105 cells·mL−1 in a growth medium supplemented with 200 IU·mL−1 of recombinant human interleukin-2 and 1% human plasma.
Flow cytometric analyses
CHO cells were transfected with the CD16 expression vector pcDNA-CD16 by using Lipofectamine 2000 (Invitrogen, Groningen, The Netherlands) for the establishment of a CHO clone that stably expressed CD16 (CD16/CHO). Test CD16/CHO cells (1 × 106) were incubated on ice with 200 pmol of bispecific diabodies for 30 min. The incubated cells were washed with NaCl/Pi containing 0.1% NaN3 and were then incubated for 30 min on ice with a fluorescein isothiocyanate (FITC) conjugated secondary antibody with an affinity for the c-myc tag (9E10; Santa Cruz Biotechnology, Santa Cruz, CA, USA). The FITC-labeled cells were subsequently analyzed by use of flow cytometry (FACSCalibur; Becton Dickinson, San Jose, CA, USA) .
In vitro growth inhibition assay
In vitro growth inhibition of TFK-1 (human bile duct carcinoma) cells and A431 (human epidermoid cancer) cells was assayed with an MTS assay kit (CellTiter 96 AQueous Non-Radioactive Cell Proliferation Assay; Promega, Madison, WI, USA), as reported previously . Half-maximal effective concentration (EC50) values were calculated by using a sigmoidal dose–response curve fit.
Construction and preparation of hEx16 diabodies
An anti-CD16 Fv was humanized by using CDR-grafting methods, as described previously [37–39]. Based on homology searches with human antibodies using the blast sequence program, the sequence of sc77u-41 VH [NCBI protein database (http://www.ncbi.nlm.nih.gov/protein); accession no. AAD53863]  was chosen as the template for 3G8 VH, and the sequence of Hod VL (accession no. AAD03723)  was chosen as the template for 3G8 VL. VH and VL sequences containing the CDRs were designed by substituting the 3G8 CDRs with the chosen sequences and were then constructed by using PCR overlap methods and synthesized primers that had been optimized for E. coli. The resulting genes were designated h3GH and h3GL and were used to replace 3GH and 3GL in the pRA-3GHh5L and pRA-h5H3GL to create the constructs pRA-h3GHh5L and pRA-h5Hh3GL, respectively, as described above. The hEx16 diabodies were prepared using the aforementioned bacterial expression and in vitro refolding systems.
Construction and preparation of hEx16 mutants
Site-directed mutagenesis was performed as described previously . Three kinds of mutants were constructed for each of the h3GHh5L and h5Hh3GL chimeric single-chain components. For the preparation of hEx16 mutants, eight chimeric single-chain components, including two wild-type chimeric single-chain components, were prepared individually. The denatured h3GHh5Ls and h5Hh3GLs were then mixed stoichiometrically in every possible combination and subsequently refolded. The hEx16 mutants constructed in this study are summarized in Table 1.
Gel filtration chromatography
Gel filtration analysis was done with a Hiload Superdex 200-pg column (10/300; GE Healthcare Bio-Science Corp., Piscataway, NJ, USA) to fractionate the dimers of each bispecific diabody. The column was equilibrated with NaCl/Pi and then 0.25 mL of purified bispecific diabodies were loaded onto the column at a flow rate of 0.5 mL·min−1.
Surface plasmon resonance
The interactions between CD16 and IgG or antibody fragments were analyzed by using SPR spectroscopy with Biacore 2000 (GE Healthcare). CD16 was prepared in accordance with the method in a previous report  and immobilized onto the cells in a CM5 sensor chip to achieve levels of up to 3183 resonance units. Solutions of various concentrations of IgG or antibody fragments in 0.005% NaCl/Pi–Tween 20 were passed over the immobilized CD16. The data were normalized by subtracting the response of a blank cell after blocking. biaevaluation software (GE Healthcare) was used to analyze the data. Kinetics parameters were calculated by using a global fitting analysis with the assumptions of the 1 : 1 Langmuir binding model .
This work was supported by Grants-in-Aid for Scientific Research from the Ministry of Education, Science, Sports, and Culture of Japan (R.A. and I.K.) and by grants from the New Energy and Industrial Technology Development Organization of Japan. Additional support was provided through the Program for Promotion of Fundamental Studies in Health Sciences of the National Institute of Biomedical Innovation.