Construction, Affinity Maturation, and Biological Characterization of an Anti-tumor-associated Glycoprotein-72 Humanized Antibody*

The tumor-associated glycoprotein (TAG)-72 is expressed in the majority of human adenocarcinomas but is rarely expressed in most normal tissues, which makes it a potential target for the diagnosis and therapy of a variety of human cancers. Here we describe the construction, affinity maturation, and biological characterization of an anti-TAG-72 humanized antibody with minimum potential immunogenicity. The humanized antibody was constructed by grafting only the specificity-determining residues (SDRs) within the complementarity-determining regions (CDRs) onto homologous human immunoglobulin germ line segments while retaining two mouse heavy chain framework residues that support the conformation of the CDRs. The resulting humanized antibody (AKA) showed only about 2-fold lower affinity compared with the original murine monoclonal antibody CC49 and 27-fold lower reactivity to patient serum compared with the humanized antibody HuCC49 that was constructed by CDR grafting. The affinity of AKA was improved by random mutagenesis of the heavy chain CDR3 (HCDR3). The highest affinity variant (3E8) showed 22-fold higher affinity compared with AKA and retained the original epitope specificity. Mutational analysis of the HCDR3 residues revealed that the replacement of Asn97 by isoleucine or valine was critical for the affinity maturation. The 3E8 labeled with 125I or 131I showed efficient tumor targeting or therapeutic effects, respectively, in athymic mice with human colon carcinoma xenografts, suggesting that 3E8 may be beneficial for the diagnosis and therapy of tumors expressing TAG-72.

Monoclonal antibodies (mAbs) 4 are increasingly being used as therapeutic agents for cancer and other diseases. Murine mAbs have limited use as therapeutic agents because of a short half-life, an inability to trigger human effector functions, and the induction of a human antimouse antibody response (1,2). To reduce the immunogenicity of murine antibodies in humans, chimeric antibodies with mouse variable regions and human constant regions were initially constructed (3). Although chimeric antibodies proved to be less immunogenic than murine mAbs, human anti-chimeric antibody responses have been observed (4). To further reduce the immunogenicity of the mouse variable regions, a humanized antibody has been constructed by grafting the complementarity-determining regions (CDRs) of a murine mAb onto the human framework regions (FRs) by a procedure commonly referred to as CDR grafting (5). Simple grafting CDRs, however, often decreased the affinity, because some FR residues directly contact the antigen or support the conformation of the CDR loops (6,7). Therefore, humanized antibody is currently constructed primarily by CDR grafting, while retaining those rodent FR residues that influence antigen-binding activity (8). Since the FR residues often differ from one humanized antibody to another, the identification of key rodent FR residues is a crucial part of the humanization (9). Clinical studies have indicated that such humanized antibodies are, in general, less immunogenic than murine or chimeric antibodies and are tolerated by humans (10,11). Unfortunately, the non-human CDRs of humanized antibodies can induce human anti-humanized antibody responses in patients (12).
Comprehensive analyses of the three-dimensional structures of the antibody-combining sites have revealed that only 20 -33% of the CDR residues participate in antigen-binding (13)(14)(15). Most of the variable positions of CDRs are directly involved in the interaction with antigen (i.e. specificity-determining residues (SDRs)), whereas most of the conserved residues mainly serve to stabilize the structure of the combining site (15). The SDRs are observed primarily in the C-terminal part of light chain CDR1 (LCDR1), the first and sometimes also the middle positions in LCDR2, the middle portion of LCDR3, most of the heavy chain CDR1 (HCDR1), the N-terminal and middle parts of HCDR2, and most of HCDR3 except the terminal position. The SDRs form the center of the antibody-combining site. Therefore, one possible way to minimize the human anti-humanized antibody response is to graft only the SDRs of a murine antibody onto human FRs while maintaining the conformations of the CDRs of the murine antibody.
Tumor-associated glycoprotein (TAG)-72 is expressed by the majority of human adenocarcinomas in the colon, ovary, pancreas, breast, prostate, and lung but not in most normal tissues, except the endometrium in the secretory phase (16,17). A murine mAb B72. 3 (16) that specifically binds to TAG-72 is approved for in vivo imaging in patients with ovarian and colorectal cancers. A second generation antibody to B72.3, CC49 (18,19), which has higher affinity for TAG-72 than B72.3 does, has shown efficient targeting to various carcinomas in clinical trials (20 -22). The epitope for B72.3 or CC49 has been identified as sialyl-Tn (i.e. ␣-sialyl 2-6␣GalNAc), O-linked to Ser/Thr on mucintype glycoproteins (23). However, CC49 elicits human anti-mouse antibody responses in patients (24). To overcome this problem, a humanized CC49 antibody (HuCC49) has been constructed by CDR grafting while retaining the buried FR residues that affect the structure of the antibody and those involved in V L -V H interaction or antigen binding (25). Subsequently, to reduce the immunogenicity of HuCC49, the mouse CDR and FR residues have been replaced individually with corresponding human residues, and the antigen-binding activity and potential immunogenicity of each variant have been analyzed (26 -29).
In the present study, humanized antibody was constructed by grafting only the SDRs onto a human Ig germ line segment with CDRs of the same canonical structures as those of the murine antibody while retaining two key murine FR residues. The resulting humanized antibody AKA showed only about 2-fold lower affinity compared with CC49. Subsequently, antibody affinity was increased by random mutagenesis of HCDR3 residues followed by affinity selection. The resulting humanized antibody (3E8) showed 27-fold lower sera reactivity compared with HuCC49 and higher affinity compared with original CC49. The antibody showed specific tumor targeting and anti-tumor therapeutic effects in athymic mice bearing human adenocarcinoma xenografts expressing TAG-72.

EXPERIMENTAL PROCEDURES
Construction of Humanized Antibody-To select human FRs for SDR grafting, the V H and V L sequences of CC49 were compared with those of human Ig variable and joining region germ line segments. It was found that DP25-J H 4 and DPK24-J K 4 were the most homologous to the V H and V L of CC49, respectively (Fig. 1). To construct the V L of the humanized light chain (HzK), only Tyr 94 in the LCDR3 of CC49 was grafted onto human DPK24-J K 4 (Fig. 1A). The numbering follows Kabat et al. (30).
Expression and Purification of Humanized Antibody-The humanized antibody was transiently expressed in COS7 cells using Lipofectamine (Invitrogen), and the culture supernatant was subjected to ELISA to determine its antigen-binding activity and affinity. For the stable expression of humanized antibody, expression plasmid DNA was transfected into a dihydrofolate reductase-deficient Chinese hamster ovary (CHO) cell line, DG44. After selection in minimal essential medium ␣ containing G418 in 96-well plates, the resistant cell clones were screened for the production of assembled antibody by an indirect ELISA and were subjected to stepwise methotrexate adaptation, as described previously (33).
For the production and purification of antibody, the CHO cell line that stably expressed the antibody was grown in serum-free medium (CHO-SFM II, Invitrogen), and the culture supernatant was subjected to affinity chromatography on a Protein G-Sepharose column (Amersham Biosciences), as described previously (32). The integrity and purity of the purified antibody were determined by SDS-PAGE. For quantification of the purified antibody, an optical density of 1.43 at 280 nm was used for a protein concentration of 1 mg/ml (34).
ELISA-For an indirect ELISA, antibody samples were added to each well that had been coated with 1 g of TAG-72-positive bovine submaxillary mucin (Type I-S; Sigma), as described previously (27), and were incubated at 4°C overnight. After washing, 100 l of goat anti-human IgG-horseradish peroxidase conjugate (1:1000 (v/v); Sigma) were added to each well and incubated at 37°C for 1 h. Finally, 100 l of 0.2 M citrate-PO 4 buffer (pH 5.0) containing 0.04% o-phenylenediamine (Invitrogen) and 0.03% H 2 O 2 were added to each well and incubated for 10 min. The reaction was stopped by the addition of 50 l of 2.5 M H 2 SO 4 , and the optical density was measured at 492 nm in an ELISA plate reader (Titertek Multiskan Plus).
To determine antigen-binding affinity, a competition ELISA was carried out as described previously (32). Briefly, solutions containing 3 ng of each antibody and various concentrations (10 Ϫ11 -10 Ϫ5 M) of bovine submaxillary mucin as a competing antigen were incubated at 37°C for 2 h and were then added to each well that had been previously coated with 200 ng of the antigen. The bound antibody was detected by an indirect ELISA. The dissociation constant (K D ) of each antibody was calculated from a Scatchard plot (35).
To analyze the epitope specificity of 3E8, a competition binding assay between CC49 and 3E8 was performed as described previously (33). Briefly, CC49 or biotinylated 3E8 was incubated with bovine submaxillary mucin (250 ng/well; Sigma) in 96-well plates at 37°C for 30 min in the presence of increasing concentrations of a competing antibody, 3E8.
The bound CC49 or biotinylated 3E8 was detected by goat anti-mouse IgG (Fc-specific)-horseradish peroxidase conjugate (Pierce) or streptavidin-horseradish peroxidase (Sigma), respectively. As a negative control of the competing antibody, humanized KR127, which specifically binds to the preS1 antigen of hepatitis B virus (36), was used.
Serum Reactivity Assay-The potential immunogenicity of AKA in humans was assessed by measuring the reactivity of the humanized antibody to the serum from patient EA, who received 177 Lu-mCC49 in a phase I radioimmunotherapy trial (22). Since the serum from patient EA contained anti-murine variable region antibodies, it was used to compare the reactivity of AKA with that of HuCC49. However, the serum also contained circulating TAG-72 antigen and anti-murine Fc antibodies, which may have interfered with the binding of the humanized antibody to the anti-variable region antibodies in the serum. Therefore, prior to the serum reactivity assay, TAG-72 antigen and anti-murine Fc antibodies were removed from the serum by immunoadsorption to another TAG-72-specific murine mAb, CC92, that recognizes an epitope of TAG-72 that is distinct from the one recognized by CC49 (37). Briefly, CC92 was coupled to Reactigel (HW65F; Pierce). The patient serum was added to the CC92 gel and was incubated overnight at 4°C with end-over-end rotation. The samples were centrifuged at 1000 ϫ g for 5 min, and the supernatants were saved and stored at Ϫ20°C for the serum reactivity assay.
To test the serum reactivity of antibodies, a surface plasmon resonance-based competition assay was carried out using a BIAcore X instrument (BIAcore), as described previously (38). Briefly, proteins were immobilized on the carboxymethylated dextran chip (CM5) by amine coupling. The dextran layer of the sensor chip was activated by injecting 35 l of a mixture of N-ethyl-NЈ-(3-dimethylaminopropyl)carbodiimide hydrochloride and N-hydroxysuccinimide at a flow rate of 5 l/min. Proteins diluted in 10 mM sodium acetate buffer (pH 5.0) at a concentration of 100 g/ml were injected until surfaces of 5000 resonance units were obtained. The remaining reactive groups on the surfaces were blocked by injecting 35 l of 1 M ethanolamine (pH 8.5). To determine the serum reactivity of HuCC49 and AKA, AKA or HuCC49 at different concentrations was incubated with the patient serum cleared of TAG-72 and anti-murine Fc antibodies at 25°C, and then the mixture was applied to the HuCC49-immobilized CM5 chip in flow cell 1 and a rabbit ␥-globulin-immobilized chip in flow cell 2 as a reference to compete for binding to serum anti-variable region antibodies. After the binding was measured for 1000 s, the unbound samples were washed from the surface with running buffer using a flow rate of 100 l/min, and the surfaces were regenerated with a 1-min injection of 10 mM glycine (pH 2.0). The percentage of binding at each antibody concentration was calculated as follows. Percentage of binding ϭ (slope of the signal obtained with competitor (serum ϩ antibody)/slope of the signal obtained without competitor (serum only)) ϫ 100. The IC 50 for each antibody was calculated as the concentration required for 50% inhibition of the binding of the serum to immobilized HuCC49.
Affinity Maturation of AKA-To improve the affinity of AKA, the HCDR3 of AKA Fab was randomly mutated, and the resulting Fab library was subjected to a modified colony lift assay (39). To begin, the Fab expression vector, pC3Q-dgIII, was constructed by deletion of the geneIII from pComb3H (40) and by replacement of the first amino acid residue, glutamic acid, by glutamine. Next, the V H of aka and the V L of HzK were subcloned into the XhoI-ApaI and SacI-XbaI sites, respectively, of pC3Q-dgIII to construct pC3Q-dgIII-AKA. This plasmid DNA was used as a template for random mutagenesis of the HCDR3 by recombinant PCR. The first PCR was performed using the 5Ј and 3Ј primers, VH135 and HCDR3 BACK, respectively, or HCDR3 FOR-WARD and LHS11, respectively. The HCDR3 FORWARD primer was mutagenic and introduced random mutations at the first five amino acid residues of the HCDR3 of AKA. The nucleotide sequences of the primers were as follows: VH135, 5Ј-AGGTGCAGCTGCTCGAGTCTGG; HCDR3 BACK, 5Ј-TCTTGCACAGTAATAGACCGCCGTGTC; HCDR3 FORWARD, 5Ј-GTCTATTACTGTGCAAGA(G/A/T/C)N-(G/C)NNSNNSNNSNNSTACTGGGGCCAAGGCACTCTG; LHS11, 5Ј-CACCGGTTCGGGGAAGT. The two first PCR products were then subjected to recombinant PCR. The final PCR product was digested with XhoI and ApaI and was cloned into the XhoI-ApaI sites of pC3Q-dgIII-AKA to create a mutant Fab library.
After the transformation of Escherichia coli TG1 with the library, 10 6 cells were pipetted onto a nitrocellulose membrane placed on a 2ϫ YTA plate and were grown overnight at 37°C (master membrane). A nitrocellulose membrane was coated with antigen by incubation in PBS containing 10 g/ml bovine submaxillary mucin at 37°C for 6 h, rinsed twice with PBS, and blocked with 5% skim milk at 37°C for 2 h. The blocked membrane was rinsed twice with PBS and soaked in 2ϫ YT broth with 1 mM isopropyl-␤-D-thiogalactopyranoside and 100 g/ml ampicillin (capture membrane). This capture membrane was placed on a 2ϫ YTA plate containing 1 mM isopropyl-␤-D-thiogalactopyranoside, and the master membrane was placed, cell-coated side up, on top of the capture membrane; the membranes were then incubated at room temperature overnight. The capture membrane was rinsed five times with PBS containing 0.05% Tween 20 (PBST), blocked with 5% skim milk at 37°C for 6 h, and incubated with goat anti-human IgG F(abЈ) 2 -horseradish peroxidase conjugate (1:1000 (v/v); Sigma) at 37°C for 1 h. After washing, the membrane was developed by enhanced chemiluminescence, and the spots generated on the film were used to identify the corresponding colonies on the original bacterial plate. Each positive colony was grown in 2ϫ YTA and subjected to another round of selection as described above. Finally, 180 positive colonies were isolated, and the antigen-binding activity of soluble Fab was measured by an indirect ELISA. The Fabs with high affinity were selected, and the nucleotide sequences of the HCDR3 were determined.
Conversion of Fab to Whole IgG-The V H of selected Fab and Ig heavy chain leader sequences (32) were fused by recombinant PCR. The PCR products were digested with EcoRI and ApaI and were subcloned into the EcoRI-ApaI site of pdCMV-dhfr-AKA. The resulting expression plasmid was transfected into COS7 cells, and the antibody secreted in the culture supernatant was analyzed by ELISA to determine the antigen-binding affinity.
Radioiodination of Humanized Antibody-The IODO-BEAD method (Pierce) was used for radiolabeling the purified AKA and 3E8 with 125 I (Amersham Biosciences) or 131 I (Korea Atomic Energy Research Institute). The radiolabeled antibody was purified by gel filtration on a PD-10 column (Amersham Biosciences) and was sterilized by filtration (0.22 m; Millipore Corp.). The radiolabeling yield and radiochemical purity were determined with instant thin layer chromatography-silica gel (Gelman Scientific) in the stationary phase and 70% methanol in the mobile phase.
Biodistribution Study-Female athymic mice (BALB/c-nu/nu; 5-6 weeks old; 17-23 g) were obtained from Japan SLC, Inc. Tumors were grown after a subcutaneous injection of 5 ϫ 10 6 human colon adenocarcinoma cell line (LS174T) in the left thigh. After 14 days, 125 I-3E8 or 125 I-AKA (20 g/740 KBq) were injected into the tail vein of the athymic mice bearing LS174T tumors. For each time point, a group of three mice was sacrificed to collect and weigh blood, tumors, and organs. Radioactivity was measured in a ␥-scintillation counter. The percentage of the injected dose/g of tissue was calculated. MARCH 17, 2006 • VOLUME 281 • NUMBER 11

JOURNAL OF BIOLOGICAL CHEMISTRY 6987
Radioimmunotherapy Studies-The radioimmunotherapeutic efficacy of 131 I-3E8 was evaluated in the athymic mice bearing LS174T tumors. For the tumor growth inhibition study (n ϭ 8), LS174T cells (5 ϫ 10 6 ) were subcutaneously injected into the back of athymic mice (5-6 weeks old; 17-23 g), and, after 14 days, saline or 7.4 MBq of 131 Ilabeled 3E8 (20 g) was intravenously injected into each mouse every week during the first 6 weeks post-treatment. The body weight and tumor size were measured weekly for 42 days. The tumor volume on a given day was expressed relative to the initial volume on day 14. For individual tumors of the mice treated with saline or 131 I-labeled 3E8, the tumor volume doubling time (Td) was calculated as the time required to reach a relative tumor volume of 2. The specific tumor growth delay was defined as (Td in mice treated with 131 I-3E8 Ϫ Td in mice treated with saline)/Td in mice treated with saline.
For the survival study (n ϭ 7), LS174T cells (5 ϫ 10 6 ) were injected intraperitoneally into the abdomen of the athymic mice; after 7 days, saline or 7.4 MBq of 131 I-labeled 3E8 (20 g) was injected intraperitoneally into each mouse every week during the first 6 weeks post-treatment. The survival of the mice was checked daily for 95 days, and their body weights were measured weekly for 76 days. Kaplan-Meier plots of percentage survival as a function of time were generated to assess the effect of treatment on the survival of mice in each group.

Design, Construction, and Affinity Determination of Humanized
Antibody AKA-To design a humanized antibody, the amino acid sequences of the CC49 V H and V L were compared with those of human Ig germ line segments. The result showed that DP25-J H 4 and DPK24-J K 4 were most homologous to CC49 V H and V L , respectively (Fig. 1). The sequence comparison between the CC49 V L and DPK24-J K 4 revealed that only three positions (Leu 27b , Gly 27f , and Gln 29 ) in the LCDR1, one position (Ala 53 ) in the LCDR2, and one position (Tyr 94 ) in the LCDR3 were different, whereas 17 positions in the FRs were different. Their CDR sequences have identical canonical structures, L1-3 (canonical structure 3 of LCDR1), L2-1, and L3-1 (7). For SDR grafting, only one (Tyr 94 in the LCDR3) or three (Gly 27f and Gln 29 in the LCDR1 and Tyr 94 in the LCDR3) residues of CC49 V L were grafted onto human DPK24-J K 4, and the resulting humanized V L sequences were fused to human C to construct humanized light chains, HzK or HzK-GQ, respectively. The FR residues of CC49 V L were not grafted. The V L sequences of the mouse, human, and humanized light chain (HzK) are shown in Fig. 1A.
To design a humanized V H , the amino acid sequence of CC49 V H was compared with those of human DP25 and J H 4. The sequence comparison between CC49 V H and DP25-J H 4 revealed that three positions in the HCDR1 and 11 positions in the HCDR2 were different, whereas 24 positions in the FRs were different. Their HCDR1 and HCDR2 domains have identical canonical structures, H1-1 and H2-3, respectively (7). The HCDR3 is not encoded by human germ line V H genes and has no canonical structure (7). To construct a humanized V H , the SDRs (three residues in the HCDR1, seven in the HCDR2, and five in the HCDR3) were grafted onto DP25-J H 4 while retaining two FR3 residues (Ala 71 and Lys 73 ) that were thought to influence the conformations of HCDR2 and HCDR3. The resulting humanized V H was fused to the human C␥1 (hC␥1) to construct the humanized heavy chain (aka). The amino acid sequences of mouse, human, and humanized V H are shown in Fig. 1B.
The humanized heavy (aka) and light (HzK) chains were transiently expressed in COS cells, and the humanized antibody (AKA) secreted in the culture supernatant was assayed for antigen-binding affinity by a competition ELISA. The dissociation constant (K D ) of AKA was 1.05 ϫ 10 Ϫ8 M (Fig. 2). To evaluate the importance of two mouse FR3 residues (Ala 71 and Lys 73 ) retained in AKA, either Ala 71 or Lys 73 was replaced with Arg 71 (RKA) or Thr 73 (ATA), respectively, of human DP25, or both residues were replaced with the human residues (RTA). The affinities of RKA, ATA, and RTA were 2.13 ϫ 10 Ϫ8 M, 2.95 ϫ 10 Ϫ8 M, and 5.83 ϫ 10 Ϫ8 M, respectively (Fig. 2), indicating that both Ala 71 and Lys 73 are equally important in antigen-binding activity and thus need to be retained in this humanized antibody. The affinity of the humanized antibody (AKA-GQ) that consisted of the humanized heavy chain aka and the humanized light chain HzK-GQ was 0.75 ϫ 10 Ϫ8 M (Fig. 2), which was slightly higher than that of AKA. This suggests that the two LCDR1 residues (Gly 27f and Gln 29 ) may not be SDRs. The affinity of the composite antibody (cH/HzK) that consisted of chimeric CC49 heavy chain and humanized light chain HzK was 0.97 ϫ 10 Ϫ8 M, which is similar to that (1.05 ϫ 10 Ϫ8 M) of AKA (Fig. 2). Therefore, it is likely that the humanized heavy chain aka is almost optimized for affinity. The affinity of AKA was only 2.3-fold lower than that (0.44 ϫ 10 Ϫ8 M) of a chimeric antibody (cH/cK) that consisted of chimeric CC49 heavy and light chains (Fig. 2). The affinity constants of the constructs are summarized in Table 1. The humanized antibody AKA containing the least number of mouse residues was chosen for further study.
Potential Immunogenicity of AKA-To assess its potential immunogenicity, AKA was produced from a stable CHO cell line and purified, as described under "Experimental Procedures." The potential immunogenicity of AKA relative to HuCC49 was analyzed by its reactivity to the serum from the patient who received 177 Lu-CC49 in a phase I radioimmunotherapy trial (22). The serum contained anti-variable region antibodies to CC49 and also circulating TAG-72 antigen plus anti-murine Fc antibodies, which could interfere with the binding of HuCC49 or AKA to the serum anti-variable region antibodies. Therefore, prior to the serum reactivity assay, TAG-72 antigen and anti-murine Fc antibodies were removed from the serum by immunoadsorption to another TAG-72-specific murine mAb, CC92, which recognizes an epitope of TAG-72 that is distinct from the one recognized by CC49 (37), as described under "Experimental Procedures." The serum reactivity of AKA and HuCC49 to anti-variable region antibodies was determined using a surface plasmon resonance-based competition assay. In the assay, AKA or HuCC49 at different concentrations was incubated with the patient serum that had been cleared of TAG-72 and anti-murine Fc antibodies; the mixture was then applied to the HuCC49-immobilized sensor chip, and the binding was measured. By plotting the percentage of binding as a function of competitor concentration (Fig. 3), the IC 50 (i.e. the concentration of the competitor antibody required for 50% inhibition of the binding of the serum to HuCC49) was calculated. A higher IC 50 value indicates a decreased reactivity to the anti-variable region antibodies to CC49. The IC 50 value calculated for AKA was 60 nM, whereas that for HuCC49 was 2.2 nM. This corresponds to a 27-fold decrease in the reactivity of AKA, compared with HuCC49, to the patient serum, suggesting that AKA may be less immunogenic than HuCC49 in patients.
Affinity Maturation of AKA and Generation of the High Affinity Variant 3E8-To improve the affinity of AKA, AKA Fab was subcloned in the soluble Fab expression vector. The SDRs (amino acids 95-99) of the HCDR3 were then randomly mutated and subjected to panning, as described under "Experimental Procedures." Approximately 180 positive Fab clones were selected and analyzed by ELISA for their antigen-binding activities. We were able to identify five different Fab clones (3E8, 3C4, NV, 3D5, and NI) with affinities higher than that of AKA Fab. For the accurate determination of affinity, the Fab format was converted into whole IgG and subjected to a competition ELISA. As shown in Fig. 4A, the affinities of the five variants (3E8, 3C4, NV, 3D5, and NI) were 0.65 ϫ 10 Ϫ9 , 1.33 ϫ 10 Ϫ9 , FIGURE 2. Affinity determination of humanized antibody AKA, AKA variants (RKA, ATA, RTA, and AKA/GQ), and chimeric constructs (cH/hK and cH/cK). cH/hK is a composite antibody that consists of chimeric CC49 heavy chain and humanized light chain HzK. cH/cK is a chimeric antibody that consists of chimeric CC49 heavy and light chains. Each of the constructs was transiently expressed in COS7 cells, and the culture supernatant was subjected to an indirect ELISA followed by a competition ELISA to determine the antigenbinding activity and affinity. The data are presented as a Scatchard plot. v, the fraction of bound antibody; i is the concentration of free TAG-72 at equilibrium. The slope of the straight line represents K A (equal to 1/K D ) of each antibody.  1.66 ϫ 10 Ϫ9 , 2.27 ϫ 10 Ϫ9 , and 3.38 ϫ 10 Ϫ9 M, respectively, which corresponded to 22-, 11-, 8-, 6-, and 4-fold increases in the affinity, respectively, compared with the AKA affinity. The HCDR3 sequences of the affinitymatured variants are shown in Table 2.
To examine whether 3E8 and CC49 bind to the same epitope, 3E8 was produced and purified from a stable CHO cell line, and a competition binding assay with CC49 was performed. As shown in Fig. 4B, the binding of CC49 to an immobilized TAG-72 was inhibited in a dose-dependent manner by increasing concentrations of 3E8 but not by an irrelevant antibody, anti-HBV humanized KR127 (36). This result indicates that 3E8 and CC49 may bind to the same epitope. The IC 50 (i.e. the concentration of the competitor antibody (3E8) required for 50% inhibition of the binding of CC49 to TAG-72) was ϳ5-fold less than that of biotinylated 3E8 to TAG-72, demonstrating that the affinity of 3E8 is higher than that of CC49.
A comparison of the HSDR3 sequences between the AKA and affinity-matured versions revealed that, despite large increases in their affinities, the sequence variations were limited ( Table 2). In more detail, Ser 95 was maintained, Leu 96 was maintained or replaced by Trp 96 , Asn 97 was replaced by Val 97 or Ile 97 , Met 98 was maintained or replaced by Gln 98 , and Ala 99 was replaced by Gly 99 or Gln 99 . Notably, replacement of Asn 97 by Val 97 (NV) or Ile 97 (NI) resulted in an increase in the affinity by 8-or 4-fold, respectively, suggesting that Asn 97 in CC49 is suboptimal for TAG-72 binding. To verify this, Asn 97 of AKA was replaced by Ala 97 (N97A), and its antigen-binding activity was analyzed. Indeed, the activity of the N97A mutant was higher than that of AKA (Fig. 4C). To analyze the effect of the other amino acid changes on the affinity maturation to 3E8, Leu 96 or Ala 99 of AKA was replaced by Trp 96 (L96W) or Gln 99 (A99Q), respectively, and each mutant was analyzed for antigenbinding activity. As shown in Fig. 4C, the single mutation abrogated the activity of AKA, suggesting that the improvement of affinity from AKA to 3E8 may have resulted from the combined effect of the L96W, N97I, and A99Q mutations.
Biodistribution and Tumor Targeting Study-The tumor targeting abilities of 3E8 and AKA were studied in athymic mice bearing human colon adenocarcinoma xenografts. The 125 I-AKA or 125 I-3E8 was injected into the mouse model, and their biodistribution was examined at 4, 24, 48, and 72 h postinjection. The percentage of the injected dose/g of tissue of tumor and normal tissues are shown in Fig. 5. The AKA and 3E8 showed tumor localization, peaking at 24 h, with tumor uptake higher with 3E8 than with AKA at all time points examined. However, the 125 I-AKA accumulated in the tumor was decreased in a time-dependent manner, whereas the 125 I-3E8 in the tumor was almost maintained during the test period. Thus, tumor uptake of 125 I-3E8 at 4, 24, 48, and 72 h postinjection became ϳ197, 167, 224, and 236%, respectively, of that of 125 I-AKA. This is probably due to the increased affinity of 3E8 compared with AKA. The biodistribution study indicated that the 125 I-3E8 showed ϳ2-fold higher tumor uptake than 125 I-AKA.
Radioimmunotherapy with 131 I-3E8-The radioimmunotherapeutic efficacy of 131 I-3E8 was evaluated by the fractionated radioimmunotherapy protocol, because the use of dose-fractionation involving multiple injections of radiolabeled antibody is a strategy for producing more prolonged tumor growth inhibition, reducing hematological toxicity, and allowing higher total doses of radionuclide to be administered than with single-dose therapy (41,42). To determine the therapeutic efficacy of 3E8, the 131 I-labeled 3E8 (7.4 MBq) was injected once weekly for 6 weeks, and its anti-tumor effect in the athymic mice bearing human colon adenocarcinoma xenografts was evaluated by inhibition of tumor growth (Fig. 6A) or extension of the 50% survival rate (Fig. 6B). The  result from the tumor growth inhibition study showed that the Td in the 131 I-3E8-treated mice (13.2 days) was significantly longer than that in the control mice injected with saline (4.9 days; Fig. 6A). The tumorspecific growth delay was calculated to be 1.7. The body weights of control mice increased to 119% of the initial body weight over time, saline-treated mice maintained their initial body weight, and the 131 I-3E8-treated mice showed a reduction of up to 13% of the initial body weight (data not shown).
In the survival study, the 50% survival rate increased from 52.5 days in the control mice to 90.5 days in the 131 I-3E8-treated mice (Fig. 6B), indicating that the 50% survival rate of the 131 I-3E8-treated mice was 1.7 times that of the control mice. The body weight of the 131 I-3E8-treated mice showed a reduction of up to 10% of the initial body weight (data not shown).

DISCUSSION
Murine mAbs induce human anti-mouse antibody responses in humans. To circumvent these problems, murine mAbs have been humanized mostly by CDR grafting. However, the non-human CDRs of humanized antibody can induce human anti-humanized antibody responses in patients (12). Alternative humanization strategies, such as veneering (43) and deimmunization (44), have been developed, but the veneered or deimmunized antibodies contain a higher number of mouse residues than do the CDR-grafted antibodies, and their effectiveness remains unproven in clinical studies. In this study, we constructed a humanized CC49 antibody by grafting only the SDRs within the CDRs of CC49 onto homologous human Ig germ line segments whose CDRs have the same canonical structures as those of CC49 while retaining only two mouse FR3 residues in the heavy chain that support the conformation of CDRs (Fig. 1). The resulting humanized antibody AKA showed only about 2-fold lower affinity compared with CC49 (Fig. 2) and 27-fold lower serum reactivity compared with the humanized anti-body HuCC49 that was constructed by CDR grafting (Fig. 3). The results indicate that SDR grafting is an advanced strategy of antibody humanization.
The affinity of AKA was increased by random mutagenesis of the HCDR3 domain followed by affinity selection. The highest affinity variant, 3E8, showed ϳ22-fold higher affinity (Fig. 4A) and 2-fold better tumor targeting ability (Fig. 5) compared with AKA. In addition, it showed higher affinity compared with parental CC49 while maintaining the epitope specificity of CC49 (Fig. 4B). The sequence comparison and mutational analysis of the HCDR3 between AKA and affinity-matured antibodies revealed that Asn 97 of CC49 is suboptimal for TAG-72 binding and that N97V or N97I is critical for affinity maturation (Fig. 4C, Table 2). The variant 3E8 has the least number of mouse residues and the highest affinity among the humanized versions of CC49 that have been constructed to date (26 -29).
The 3E8 labeled with 131 I showed an anti-tumor therapeutic effect in athymic mice bearing human colon adenocarcinoma xenografts (Fig. 6). The 131 I-3E8-treated mice showed 1.7-fold increases of tumor-specific growth delay and 50% survival rate compared with the control mice treated with saline, and the 131 I-3E8-treated mice showed a 10 -13% reduction of body weight. Since bone marrow toxicity is evaluated by loss of body weight (45), the results suggest that the fractionated radioimmunotherapy protocol resulted in prolonged tumor growth inhibition and low hematological toxicity in the 131 I-3E8-treated mice. To our knowledge, this is the first study to demonstrate the therapeutic efficacy of an anti-TAG-72 humanized antibody conjugated with radioisotope against human colon adenocarcinoma in an in vivo tumor model. The advantages of 3E8, such as high affinity, efficient tumor targeting, and therapeutic efficacy, will make it a useful reagent in the diagnosis and therapy of colon adenocarcinoma expressing TAG-72.