Grafting of “Abbreviated” Complementarity-Determining Regions Containing Specificity-Determining Residues Essential for Ligand Contact to Engineer a Less Immunogenic Humanized Monoclonal Antibody

The mAbs against tumor Ag hold promise for diagnosis and therapy of human cancers (reviewed in Refs. 1, 2, 3). A major impediment to the clinical use of murine mAbs is the human anti-murine Ab (HAMA)⁷ response these mAbs elicit in patients (4, 5, 6, 7). To obviate the potential HAMA response, humanized Abs have been developed by grafting the complementarity-determining regions (CDRs) of the murine Ab onto the frameworks of the variable light (V_L) and variable heavy (V_H) regions of human mAbs (reviewed in Ref. 8). Humanization of a xenogeneic Ab, however, does not necessarily eliminate the immunogenicity of the molecule, because the humanized molecule can evoke anti-V region response (7, 9, 10, 11, 12, 13). It has been proposed that binding of an Ab to an Ag involves only a small number of residues within the CDRs. The latter residues have been designated as specificity-determining residues (SDRs) (14). In an attempt to circumvent anti-V region response in patients, a murine Ab has been humanized by grafting only its SDRs onto human frameworks (13).

The carcinoembryonic Ag (CEA), a cell surface glycoprotein of 180-kDa molecular mass, is one of the most widely used human tumor markers. It is expressed at high levels in the embryonic and fetal digestive epithelial tissue and, to a lesser extent, in normal adult colon and stomach epithelium (15). CEA is overexpressed in ∼95% of gastrointestinal and pancreatic cancers, as well as in most non-small-cell lung carcinomas. Also, it is expressed in breast carcinoma and squamous cell carcinoma of the head and neck (reviewed in Refs. 11 and 16). A number of murine anti-CEA mAbs have been generated (17) and used for measuring CEA levels in blood (18) and for immunohistopathology of tissues from cancer patients (11, 19). In addition, Abs directed to CEA were among the first to be used in clinical trials to successfully localize tumors (20, 21, 22, 23). Also, several clinical trials have been conducted to evaluate the anti-tumor activity of anti-CEA mAbs (24, 25, 26, 27). CEA has some degree of cross-reactivity with several proteins, including nonspecific cross-reacting Ag-1 (NCA-1), normal fecal Ag-1, and the NCA-related proteins present in human granulocytes (17).

The mAb COL-1, an IgG2a, has a high affinity for CEA and has no detectable reactivity to granulocytes or to the CEA-related Ags NCA-1 and normal fecal Ag-1 (28). Murine COL-1 (mCOL-1) reacts with biopsies of colon and gastric carcinomas (80 and 88%, respectively) and with biopsies of human mammary and non-small-cell lung carcinomas (17, 29). Except for some reactivity to skin and gastric and colon mucosa, mCOL-1 does not react with normal tissues (17). In preclinical studies, ¹²⁵I-labeled mCOL-1 achieved good tumor radiolocalization using LS-174T colon carcinoma xenograft in athymic mice (30). In patients, ¹³¹I-labeled mCOL-1 showed good tumor targeting of human gastrointestinal carcinomas in a phase I clinical trial (24). Not unexpectedly, mCOL-1 was found to induce HAMA response in patients (24, 31). Yu et al. (24) reported that 61 and 83% of patients treated with the Ab developed elevated HAMA levels by days 20 and 40, respectively. It was observed by Meredith et al. (31) that 93% of the patients developed Ig response, with most patients developing HAMA by day 14 and some as early as days 7–11 after mCOL-1 administration. The peak of Ab response generally occurred at 4–6 wk after exposure. An attempt has now been undertaken to reduce the immunogenicity of the mCOL-1 mAb by progressively reducing its murine content by genetic manipulation. This report describes the development and characterization of a mouse-human chimeric COL-1 (cCOL-1) mAb, a humanized COL-1 (HuCOL-1) mAb, and variants of HuCOL-1. A final variant of HuCOL-1 was developed by a previously suggested humanization protocol (14) based on transplantation of “abbreviated” xenogeneic CDRs onto the human Ab frameworks. This variant, which retains significant Ag-binding reactivity, is minimally reactive to sera from patients who were earlier administered mCOL-1 during clinical trials.

The long overlapping oligomers and oligonucleotide primers used for DNA amplifications were supplied by Lofstrand Labs (Gaithersburg, MD) and Midland Certified Reagent (Midland, TX). The sequences of the four primers that were used to generate DNA fragments encoding the V_H and V_L domains of the mCOL-1 mAb were as follows: 1) 5′ V_H: 5′-AGTAAGCTTCCACCATGGAGTGGTCCTGGGTCTTCCTCTTCTTCCTGTCCGTGACTACTGGAGTGCACTCCGAGGTTCAGCTGCAGCA-3′; 2) 3′ V_H: 5′-CGATGGGCCCGTAGTTTTGGCAGAGGAGACGGCGACCG-3′; 3) 5′ V_L: 5′-TAGCAAGCTTCCACCATGGATAGCCAGGCCCAGGTGCTCATGCTCCTGCTGCTGTGGGTGAGCGGCACATGCGGCGACATTGTGCTGACACA-3′; 4) and 3′ V_L: 5′-TGCAGCCGCGGCCCGTTTGATTTCCAGCTTGG-3′.

Each of the 5′ primers carries a HindIII site followed by a sequence encoding a signal peptide. The 3′ V_H primer carries an ApaI, whereas the 3′ V_L primer has a SacII site. The four 119- to 133-bp-long oligonucleotides that were used to generate each of the V_H and V_L genes of HuCOL-1 are shown by long arrows in Fig. 1. The sequence of the 20- to 21-bp-long end primers used for DNA amplification were as follows: 5) 5′ V_H: 5′-CGTAAGCTTCCACCATGGAG-3′; 6) 3′ V_H: 5′-TGGGCCCTTGGTGGAGGCTGA-3′; 7) 5′ V_L: 5′-GCAAGCTTCCACCATGGATA-3′; and 8) 3′ V_L: 5′-TGCAGCCGCGGTACGTTTGAT-3′.

The 5′ primers (nos. 5 and 7) carry a HindIII site. Whereas a site for ApaI has been incorporated in the 3′ V_H primer (no. 6), the sequence for the SacII site has been included in the 3′ V_L primer (no. 8). The sequences of two additional mutagenic primers (supplied by Milligen/Bioresearch, Burlington, VT) that were used for the generation and amplification of the genes encoding the V domains of HuCOL-1 variants were as follows: 9) 3′ V_H: 5′-TGCCCTGGAACTTCTGGGCATATTCAGTA-3′; and 10) 3′ V_L: 5′-GCACTGA CACTTTGGCTGGACTTGCAGTTGATGGTGGCCCCTC-3′.

The sequences recognized by the restriction endonucleases are in italics, and the mutagenic changes are underlined.

All PCRs were conducted in a final volume of 100 μl of PCR buffer containing 200 μM dNTPs, 3 U of Taq polymerase (Boehringer Mannheim, Indianapolis, IN), 0.2 μM each of the end primers, and 100 ng of DNA template. Initial denaturation at 94°C for 2 min was followed by 30 cycles of denaturation at 94°C for 30 s, annealing at 55°C for 30 s, and extension at 72°C for 1 min. It was followed by a final primer extension step at 72°C for 10 min. The V_L and V_H genes of the HuCOL-1 were synthesized by the overlap extension PCR that has previously been described (32). Primer-induced mutagenesis was conducted by a dual-step PCR as described by Landt et al. (33).

Two different baculovirus transfer vectors, pAcUW51 (BD PharMingen, San Diego, CA) and pBAC4x-1 (Novagen, Madison, WI), have been used for generating the recombinant viruses and the subsequent coexpression of the Ig H and L chains. In pAcUW51 vector, one of the target genes can be cloned at the BamHI site located downstream of the polh promoter, whereas the other gene can be driven by the p10 promoter by inserting the gene at either the BglII or the EcoRI site located 3′ to the promoter. The pBAC4x-1 baculovirus transfer plasmid contains two of each of the polh and p10 promoters, with a unique cloning site placed downstream of each promoter.

To generate constructs encoding the chimeric H and L chains of mAb COL-1, the V region sequences of the H and L chain genes were PCR amplified using the phagemid constructs of the cDNAs encoding the Fd and the L chain of mCOL-1 as templates. Primers no. 3 and no. 4 were used as forward and reverse primers, respectively, to amplify a 420-bp sequence encoding the V_L domain along with the signal peptide located upstream. The 3′ primer was designed to extend the 3′ end of the amplified sequence to a unique SacII site located 10 bp downstream from the start of the human κ C region. A DNA fragment encoding the human κ C region was excised from a pre-existing construct pLNCXHuCC49HuK (32) by SacII/ClaI treatment. The construct carried an EcoRI site immediately upstream of ClaI site. The V and the C regions of the L chain were joined to the HindIII/ClaI linearized pBluescript II S/K⁺ (pBSc) plasmid (Stratagene, La Jolla, CA) by three-way ligation. Taking advantage of an EcoRI site upstream of the HindIII site in pBSc, the entire L chain sequence was then released from the construct by EcoRI digestion and inserted into the baculovirus expression vector pAcUW51 at the EcoRI site located downstream from the p10 promoter.

For the assembly of the chimeric H chain, a 460-bp sequence encoding the V_H domain and its leader peptide was PCR amplified using primers nos. 1 and 2 as 5′ and 3′ primers, respectively. The design of the 3′ primer facilitated amplification of the V_H sequence to extend to the ApaI site located 17 bp downstream from the start of the C_H1 domain. To assemble the V and C regions, an ApaI/ClaI DNA fragment carrying the human γ1 C region was excised from a pre-existing construct pLgpCXHuCC49HuG1 (32). The ApaI/ClaI fragment along with the 460-bp PCR product was inserted into the HindIII/ClaI linearized pBSc. The DNA encoding the entire H chain was released by HindIII/ClaI treatment of the pBSc construct. The termini of the target DNA were filled in using Klenow fragments of the DNA polymerase, and the DNA fragment was subcloned in the L chain construct of pAcUW51 at the blunt-ended BamHI site located downstream of the polh promoter.

DNA manipulations similar to those described for cCOL-1 were conducted to assemble the humanized and the variant V regions and their respective C regions into pBSc for the subsequent subcloning in baculovirus expression vector pBAC4x-1. After joining the V regions of the H and L chains to their respective C regions in pBSc, the assembled L chain of the HuCOL-1 or its variant was released from the pBSc construct and cloned at the EcoRI site, downstream from the p10 promoter. The entire H chain of the HuCOL-1 or its variant was excised from its pBSc construct by HindIII/XhoI treatment, and it was cloned unidirectionally in the L chain construct of pBAC4x-1 at the HindIII/XhoI site, downstream of polh promoter. Three expression constructs were generated: one containing the variant L chain and the parental humanized H chain, the second containing the variant H chain and the parental humanized L chain, and the third carrying variants of both the L and H chains (Table I).

Serum-free-adapted Sf9 insect cells (Life Technologies, Rockville, MD) were cultured at 27°C in Sf900-II medium (Life Technologies) with 50 μg/ml gentamicin. To develop transfectomas secreting cCOL-1, insect cells were cotransfected with the pAcUW51-derived expression construct and the linearized BaculoGold Baculovirus DNA (BD PharMingen). Transfectomas producing HuCOL-1 and its variants were generated by transfecting insect cells with one of the pBAC4x-1-derived expression constructs and the linearized BacVector2000 Baculovirus DNA (Novagen). A cationic liposome-mediated system (DOTAP; Boehringer Mannheim) was used for all transfections. Harvesting of the recombinant virus, screening for Ig expression, and Ag binding by ELISA have previously been described (13).

ELISAs were conducted by coating 96-well polyvinyl microtiter plates with CEA (100 ng/well; Research Diagnostic, Flanders, NJ) or with Fcγ-fragment-specific goat anti-human IgG (100 ng/well) (Jackson ImmunoResearch Laboratories, West Grove, PA). Anti-human IgG or the CEA-coated plates were used to test for the production of Ig by the insect cells or to assess its Ag reactivity, respectively. The details of the assay procedure have been reported earlier (34).

Three days after infection, the supernatants were collected and made free of cellular debris and any contaminating proteins, before using protein G agarose column (Life Technologies) to purify the desired protein as described earlier (13). The protein was concentrated using Centricon 30 (Amicon, Beverly, MA) and dialyzed in PBS buffer using a Slide-A-Lyzer cassette (Pierce, Rockford, IL). The protein concentration was determined by the method of Lowry et al. (35), and the purity of the eluted proteins was evaluated by SDS-PAGE under reducing and nonreducing conditions, using precast 4–20% Tris-glycine gel (Novex, San Diego, CA). The protein bands were visualized by Coomassie blue staining (Novex).

The relative Ag binding of the mCOL-1 and the recombinant Abs derived from it were determined using competition RIA. Twenty-five microliters of serial dilutions of the Abs to be tested as well as the mCOL-1, prepared in 1% BSA in PBS, were added to microtiter plates containing 200 ng of CEA saturated with 5% BSA in PBS. ¹²⁵I-labeled mCOL-1 or ¹²⁵I-labeled HuCOL-1 (100,000 cpm in 25 μl) was then added to each well. After an overnight incubation at 4°C, the plates were washed and counted in a gamma-scintillation counter. The relative affinity constants were calculated by a modification of the Scatchard method (36).

A previously described method (37) has been used for FACS analysis. To evaluate the ability of HuCOL-1 and its variants to bind to cell-surface CEA, 1 × 10⁶ retrovirally transduced MC38 cells expressing CEA (38) were resuspended in cold Ca²⁺- and Mg²⁺-free Dulbecco’s PBS and incubated with the mCOL-1-derived Abs for 30 min on ice. A human IgG was used as an isotype control. After one washing cycle, the cell suspension was stained with FITC-conjugated mouse anti-human Ab (BD PharMingen) for 30 min on ice. A second washing cycle was performed, and then the samples were analyzed with a FACScan (BD Biosciences, Mountain View, CA) using CellQuest for Macintosh. Data from analysis of 10,000 cells were obtained.

Stored patients’ sera, from a phase I clinical trial (24) that involved the administration of ¹³¹I-labeled mCOL-1 to gastrointestinal carcinoma patients, were used to assess serum reactivity of the mCOL-1-derived Abs. Several sera were tested for the presence of anti-V region Abs to mAb COL-1. The sera, however, contain circulating Ag and anti-murine Fc Abs, which could interfere with the binding of mAb COL-1 and its derivative Abs to the anti-V region Abs. To circumvent this problem, the circulating CEA and anti-murine Fc Abs were removed by sequential preadsorption of the sera with purified mCOL-6 and mCOL-4 mAbs, two Abs that react with epitopes of CEA different from the one recognized by mCOL-1 (28). The mAb mCOL-4 has the same isotype as that of mCOL-1. For preadsorption, serum samples were added to mCOL-6 coupled to Reacti-gel according to the method of Hearn et al. (39). The mixtures were incubated overnight at 4°C with end-to-end rotation and were centrifuged at 1000 × g for 5 min. Preadsorption was repeated until the supernatants displayed no detectable anti-murine Fc activity. The procedure was then repeated using mCOL-4 coupled to Reacti-gel. To detect anti-V region Abs by surface plasmon resonance (SPR), the preadsorbed serum was used as a mobile reactant. Proteins were immobilized on carboxymethylated dextran CM5 chips (BIAcore, Piscataway, NJ) by amine coupling using standard procedure (40, 41). HuCOL-1 was immobilized on the surface of flow cell 1, whereas the surface of flow cell 2 was coated with an unrelated protein, rabbit γ globulin (Bio-Rad, Hercules, CA).

The reactivity of COL-1 variants to anti-V region Abs was determined using a recently developed SPR-based competition assay (58). Competition experiments were performed at 25°C using a CM5 sensor chip containing either mCOL-1 or HuCOL-1 in flow cell 1 and rabbit γ globulin (Bio-Rad), as a reference, in flow cell 2. Typically, mCOL-1, HuCOL-1, or its variants were used at different concentrations, to compete with the Ab immobilized on the sensor chip for binding to serum anti-V region Abs. Patient’s serum with or without the competitor (mCOL-1, HuCOL-1, or its variants) was applied across the sensor surface using a recently developed sample application technique (42) at the unidirectional flow of 1 μl/min. After the binding was measured for 1000 s, the samples were washed from the surfaces with running buffer using a flow rate of 100 μl/min, and the surfaces were regenerated with 10 mM glycine (pH 2.0) for the HuCOL-1 sensor chip or with HCl (pH 2.3) for the mCOL-1 sensor chip. The percent binding at each Ab concentration was calculated as follows: % binding = [slope of the signal obtained with competitor (serum + mCOL-1, HuCOL-1, or HuCOL-1 variants)/slope of the signal obtained without competitor (serum only)] × 100. IC₅₀ for each Ab, the concentration required for 50% inhibition of the binding of the serum to either mCOL-1 or HuCOL-1, was calculated.

The genes encoding the L chain and the Fd region of the H chain of mAb COL-1 were generated by repertoire cloning methodology (43), using synthetic oligonucleotides described in Materials and Methods. The PCR products of the appropriate size were cloned in λ phage vector, and phagemids carrying the target genes were subsequently excised. The cloned genes were sequenced (data not shown) before the phagemids were used as templates for the subsequent PCR amplification.

The mCOL-1 was humanized by grafting the CDRs of the L and H chains onto the V_L and V_H frameworks of the appropriate human Abs, but retaining those framework residues that were deemed essential for preserving the structural integrity of the combining site (44, 45, 46). The Ig CDRs have been defined as comprising residues 31–35b, 50–65, and 95–102 in the H chain and residues 24–34, 50–56, and 89–97 in the L chain (47). The framework residues that were deemed critical were identified on the basis of the atomic coordinates of the Abs of known structures available in the database (for example, see Ref. 48). The human Ab sequences that are most similar to mCOL-1 are VJI′CL (49) (GenBank accession number Z00022) for V_L and MO30 (50) (GenBank accession number A32483) for V_H. The alignment of the V_L sequences of mCOL-1 and VJI’CL, and the V_H sequences of mCOL-1 and MO30 are shown in Fig. 2. Also indicated in Fig. 2 are the locations of the framework residues that are critical for Ag binding. The humanization protocols for the V_L and V_H genes, shown in Fig. 2, are based on putting the CDR sequences of mAb COL-1 together with the frameworks of the human V_L and V_H templates, while replacing some of the human framework residues with those murine framework residues that may be critical for Ag binding.

A nucleotide sequence was then deduced from the amino acid sequence of each of the designed humanized V_L and V_H domains. The nucleotide sequences were refined to provide high frequency usage of codons and by eliminating, with the help of programs FOLD and MAPSORT (51), any self-annealing regions and any sites for restriction endonucleases that might complicate cloning of the designed genes in the desired vectors. Using the four overlapping oligonucleotides (shown by long arrows in Fig. 1) that encompassed, on alternating strands, the entire sequence of either the V_L or V_H region and its leader and the respective end primers described in Materials and Methods, DNA fragments encoding the humanized V_L or V_H regions were generated and amplified by overlap extension PCR technique (32). The humanized V_L or V_H regions thus generated were extended to the ApaI and SacII sites located 10 and 17 bp downstream from the 5′ end of the C regions of the H and L chains, respectively. The PCR products were cloned in pBSc to generate pBScHuCOL-1V_L and pBScHuCOL-1V_H constructs. The inserts were sequenced to check the fidelity of the PCR products.

Examination of the known structures of Ab-Ag complexes reveals that only one-third of the CDR residues are involved in the interaction with the Ag (48). This led to the proposal to redefine the boundaries of the CDRs to 31–35b, 50–58, and 95–101 in the H chain and 27d–34, 50–55, and 89–96 in the L chain (14). Accordingly, we have developed variants of HuCOL-1 H and L chains in which these abbreviated CDRs have been grafted onto the VJI’CL and MO30 frameworks (Table I).

Genes encoding the humanized V_L and V_H domains of the variants ^24,25,27L and ⁶¹H were generated by primer-induced mutagenesis, using pBScHuCOL-1V_L and pBScHuCOL-1V_H constructs, respectively, as templates. Variant ⁶¹H was generated by replacing residue 61 of mCOL-1 H chain CDR2 (numbering convention of Kabat et al. (47)) with the corresponding residues of mAb MO30 H chain CDR2. For the generation of ^24,25,27L variant, residues 24, 25, and 27 of mCOL-1 L chain CDR1 were replaced with the corresponding residues in L chain CDR1 of the human mAb VJI’CL (Table I). The V region sequences were synthesized by a dual-step PCR procedure according to Landt et al. (33). For each L and H chain V region, the mutagenic primer, containing the desired nucleotide changes in the targeted CDR, was used as a 3′ primer, whereas a 20-mer end primer served as a 5′ primer. The resulting PCR product was gel purified and used as a 5′ primer for the subsequent step of the PCR in which a 21-mer oligonucleotide was used as a 3′ primer. The PCR products were cloned in pBSC vector and sequenced. The amino acid sequences of the V_L of the variant ^24,25,27L and V_H of the variant ⁶¹H are shown in Fig. 2, A and B, respectively.

The genes encoding the V_L and V_H domains of mCOL-1, HuCOL-1, and the variants ⁶¹H and ^24,25,27L were assembled with the respective human C region genes (γ1 for the H and κ for the L chain). The expression constructs were introduced into Sf9 insect cells, and the supernatants harvested from the transfectants were assayed for Ig production and Ag-binding reactivity by ELISA, as described in Materials and Methods. All the transfectants and the viral plaques, generated by infecting Sf9 cells with the infectious supernatants, were found to be positive for Ig production as assayed by ELISA. Results of an ELISA for Ag binding also showed that all culture supernatants and the viral plaques were reactive with CEA, albeit with varying degrees. To examine whether the different constructs were expressing comparable levels of Ig molecules, viral plaques were expanded and a large batch of Sf9 cells was freshly infected, at a multiplicity of infection of 5, with infectious supernatant derived from the highest producing clone of each construct, and the infected cells were cultured under identical conditions. The secreted Abs were purified from equal volumes of the culture supernatants. The concentration of the secreted Abs was comparable (2–3 μg/ml) in culture supernatants of all five transfectants.

The apparently lower Ag-binding reactivities of the variant mAbs ^24,25,27L and ^24,25,27L/⁶¹H than those of cCOL-1, HuCOL-1, and the variant ⁶¹H could be attributed either to any possible detrimental effect of genetic manipulations of the combining site of the secreted Abs or to some structural abnormality of the expressed Ig molecules. The latter may be detected on SDS-PAGE by a change in size or mobility of the molecules. To this end, the purified Abs from the culture supernatants and the murine mAb COL-1 were analyzed by SDS-PAGE. The gel profile under nonreducing conditions (data not shown) showed that the mobility of all five recombinant Abs was identical with that of mCOL-1 mAb, which has a molecular mass of ∼160 kDa. Under reducing conditions, all the recombinant COL-1 Abs, like that of mCOL-1, yielded two protein bands of ∼25–28 and 50–55 kDa (data not shown). These molecular masses are in conformity with those of the Ig L and H chains. The results of the SDS-PAGE analysis, together with the ELISA for CEA reactivity of the serially diluted Abs, suggest that the reduced Ag reactivity of the variants ^24,25,27L and ^24,25,27L/⁶¹H may be due to some detrimental effect of the amino acid substitutions in the combining site of HuCOL-1.

A competition RIA was performed to determine the relative CEA-binding affinities of the COL-1-derived mAbs and the parental mCOL-1 mAb. Serial dilutions of unlabeled Abs (murine, chimeric, humanized COL-1 and its variants) were used to compete with the binding of ¹²⁵I-labeled HuCOL-1 (Fig. 3) or ¹²⁵I-labeled mCOL-1 (data not shown) to CEA. All of the COL-1-derived recombinant Abs, like the parental mCOL-1, were able to completely inhibit the binding of ¹²⁵I-labeled mCOL-1 and ¹²⁵I-labeled HuCOL-1 to CEA. The competition profiles of cCOL-1, HuCOL-1, and the variant ⁶¹H were comparable to that of the mCOL-1. In contrast, the competition profiles of mAbs ^24,25,27L and ^24,25,27L/⁶¹H, although of slopes similar to that of the parental mCOL-1 mAb, were shifted to the right. The values of the relative K_a of cCOL-1, HuCOL-1, and ⁶¹H mAbs, calculated from the linear parts of the competition curves in Fig. 3, were 3.45 × 10⁸ M⁻¹, 2.82 × 10⁸ M⁻¹, and 2.64 × 10⁸ M⁻¹, respectively. These relative affinities are ∼1.5- to 2-fold less than that of mCOL-1 (Table I). The relative K_a values of ^24,25,27L and ^24,25,27L/⁶¹H were 1.2 × 10⁸ M⁻¹ and 1.03 × 10⁸ M⁻¹, respectively, ∼4.3- to 5-fold lower than that of parental mCOL-1 (Table I).

Flow cytometric analysis was used to measure the binding of HuCOL-1 and its variants (⁶¹H, ^{24, 25, 27}L, and ^{24, 25, 27}L/⁶¹H) to the CEA expressed on the cell surface of a tumor cell line, MC38, that was retrovirally transduced with CEA (38). No significant differences were found in the mean fluorescence intensity or in the percentage of cells that was reactive with HuCOL-1 and its variants (Fig. 4). The percentages of gated cells, calculated after exclusion of irrelevant binding, were indeed between 54 and 56, whereas the mean fluorescence intensities were between 15 and 16 when 1 μg of each Ab was used.

The immunogenicity of an Ab variant can be assessed only through its clinical trial. A reasonable measure of the potential immunogenicity of the variant Ab, however, is its in vitro reactivity to sera from patients who were administered the parental Ab in a clinical trial. To assess the potential immunogenicity of the mCOL-1, HuCOL-1, and its variants in patients, the Abs were characterized for their reactivity to sera from gastrointestinal carcinoma patients who were administered ¹³¹I-labeled mCOL-1 in a phase I clinical trial (24). As described in Materials and Methods, any circulating CEA and anti-murine Fc Abs were removed from the sera by immunoadsorption with mCOL-6 and mCOL-4, two murine anti-CEA Abs of IgG1 and IgG2a isotypes, respectively. It has been suggested on the basis of epitope mapping of CEA (28) that mCOL-4 and mCOL-6 may react with CEA epitopes that are different from each other and different from the one recognized by mAb COL-1. Preadsorbed sera were tested for the presence of anti-V region Abs to mAb COL-1. Specific binding profiles of HuCOL-1 to the sera from patients EM, JS, and MB (data not shown) show that all three sera have Abs against the variable regions of mCOL-1. Serum reactivity of HuCOL-1 and HuCOL-1 variants was determined by their ability to compete with mCOL-1 or HuCOL-1 immobilized on a sensor chip for binding to the anti-variable region Abs to mCOL-1 present in the serum. IC₅₀, the concentration of the competitor Ab required for 50% inhibition of the binding of mCOL-1 or HuCOL-1 to the patient’s serum, was calculated by plotting the percent inhibition as a function of competitor concentration. A higher IC₅₀ value indicates a decreased reactivity to the serum, suggesting potentially reduced immunogenicity of the Abs in patients. Fig. 5 shows the competition profiles generated by HuCOL-1 and its variants when they were used to compete with the HuCOL-1 immobilized on the sensor chip for binding to the anti-V region Abs to COL-1 present in the sera of patients EM (A), JS (B), and MB (C). The competition profiles were used to calculate IC₅₀ values that are presented in Table II. For serum MB, the IC₅₀ values of all three variants are 2- to 3-fold higher than that of HuCOL-1. Studies with the serum from patient EM show that the variants ^24,25,27L and ^24,25,27L/⁶¹H have 50% higher IC₅₀ values, whereas the variant ⁶¹H has a significantly lower IC₅₀ value than that of HuCOL-1. For the serum from patient JS, the IC₅₀ of the variant ^24,25,27L/⁶¹H is twice as much as that of HuCOL-1, whereas the IC₅₀ values of the variant ⁶¹H and ^24,25,27L are comparable to that of parental HuCOL-1. When mCOL-1 was immobilized on the sensor chip and mCOL-1, HuCOL-1, and the variant ^24,25,27L/⁶¹H were used to compete with it for binding to the anti-V region Abs in the serum of patient MB, the competition profiles shown in Fig. 6 were generated. The data show that the concentration of HuCOL-1 required for 50% inhibition of the binding of the patient’s serum to mCOL-1 is ∼3-fold higher than that of mCOL-1, whereas the concentration of the variant ^24,25,27L/⁶¹H required to attain 50% inhibition of the binding of mCOL-1 to the patient’s serum is ∼5.5- and 17-fold higher than those of HuCOL-1 and mCOL-1, respectively. Moreover, it should be pointed out that the slope of the competition profile of the variant ^24,25,27L/⁶¹H is quite different from that of mCOL-1. Indeed, there was >2 log differential in the concentrations of the ^24,25,27L/⁶¹H and mCOL-1 mAbs required for the 60% inhibition of the binding of the sera anti-V region Abs to mCOL-1 immobilized on the sensor chip (Fig. 6). Sera from two other patients, JS and EM, were used to compare serum reactivity of mCOL-1 and HuCOL-1. Although the IC₅₀ value of HuCOL-1 was ∼6-fold higher than that of mCOL-1 for JS serum, it was not possible to evaluate the difference in the reactivity of the two Abs to EM serum; even 1000 nM HuCOL-1 was unable to attain 50% inhibition of the binding of the serum to mCOL-1 (data not shown).

The HAMA response is a major impediment to the clinical use of the murine mAbs, particularly for the dose fractionation clinical protocols that are likely to be used for greater therapeutic efficacy and reduced toxicity. The advantage of dose fractionation protocols for therapeutic efficiency of the radiolabeled mCOL-1 has been demonstrated in an animal model (30). In addition to causing a possible allergic reaction, the HAMA response results in rapid clearance of the Ab from the circulation, preventing the Ab from reaching the targeted tumor sites. In a phase I clinical trial of ¹³¹I-labeled mCOL-1, the onset of HAMA response prevented administration of a second dose of the radiolabeled Ab in two patients who were otherwise eligible for further treatment. HAMA response was also reported in a phase II clinical trial, when IFN was combined with ¹³¹I-labeled COL-1 and ¹³¹I-labeled CC49, a murine Ab against another pancarcinoma Ag, tumor-associated glycoprotein-72, to achieve more efficient targeting of human colorectal cancer (31).

Development of a mouse-human chimeric mAb is an approach that has been widely used to minimize HAMA response (52). Accordingly, a cCOL-1 mAb was developed by replacing the C regions of the L and H chains of the mCOL-1 with the human κ and γ1 constant regions. The Ag-binding affinity of the cCOL-1 (3.45 × 10⁸ M⁻¹) was found to be comparable to that of murine Ab (5.17 × 10⁸ M⁻¹). A chimeric Ab, though likely to be less immunogenic than the murine Ab, may evoke anti-V region response in patients, because the V region of a murine Ab is potentially immunogenic. Anti-V region responses have been reported after the administration of chimeric Abs in patients (53, 54 ; reviewed in Ref. 55). To attempt to reduce this problem, mAb COL-1 has been humanized following a procedure that involves grafting of the CDRs of a xenogeneic Ab onto the human Ig frameworks (reviewed in Ref. 8).

The most important consideration in humanizing an Ab is the preservation of its Ag-binding property, which depends on the structural integrity of the combining site. This requires selecting the most appropriate human templates for humanization and grafting the CDRs of the target Ab onto the human scaffold, while retaining those murine framework residues that may be involved in ligand contact directly or through their interaction with the CDRs. In the absence of a three-dimensional structure of mCOL-1, the identification of the crucial frameworks residue was facilitated by a search of the database that yielded the V regions of two human Abs, VJI’CL and MO30, respectively, whose sequences were most similar to the mCOL-1 V_L and V_H sequences. There are 76 identities among the 112 overall residues of the V_L of the VJI’CL and mCOL-1, and when comparing the 80 framework residues only, there are 65 identities. Among the 127 overall V_H residues of the MO30 and COL-1 mAbs, there are 75 identities, while there are 57 identities in their 87 frameworks residues. Because the V_L frameworks of VJI’CL and mCOL-1 are so extensively homologous, they differ only by five residues among those deemed crucial for Ag binding. The V_H frameworks of COL-1, however, differ from MO30 at 17 positions that are most probably essential for the ligand binding. This approach of selecting the human frameworks to be used as templates had been successful for the humanization of several Abs, including mAb AUK12–2 (56), mAb 1B4 (57), and mAb CC49 (32). Based on these considerations, humanization of mCOL-1 was conducted using VJI’CL and MO30 as human templates, and the resulting HuCOL-1 showed only 2.5-fold lower affinity than that of the murine COL-1.

Because the murine CDRs of a humanized Ab could still evoke an anti-V region response in patients, CDR grafting is not an ultimate solution of the potential immunogenicity of a xenogeneic Ab. Padlan et al. (14) suggested that not all of the CDR residues are involved in the ligand contact. Tamura et al. (13) humanized the anti-tumor-associated glycoprotein-72 mAb CC49 by grafting only its SDRs onto the human Ab frameworks. The HuCC49 variant, designated HuCC49V10, that was developed by this approach retained the Ag-binding properties of the parental mAb, while reacting with the patients’ sera only minimally. The immunogenicity of an Ab, therefore, could be reduced by transplanting only those parts of the CDRs that contain the SDRs. The “abbreviated” CDRs have been defined (14). Based on this rationale, a variant of each of the L and H chains of the HuCOL-1 was developed. In L chain variant, residues 24, 25, and 27 of the L chain CDR1 were replaced with the corresponding residues of the human Ab VJI’CL, whereas in H chain variant, residue 61 of the H-chain CDR2 was replaced with the residue of the human Ab MO30, located at the same position. The Ag-binding affinity of the H chain variant was comparable to that of the parental HuCOL-1, whereas that of the L chain variant was ∼2.4-fold lower. The slight loss in the affinity of the L chain variant was reflected in the affinity of the HuCOL-1 variant (2.7-fold lower than that of the parental HuCOL-1) that was generated by combining the H and the L chain variants. Flow cytometric analysis of the CEA-transduced MC38 cells that were treated with the variant Abs, however, shows that the slight losses in the relative affinities of the variants had no effect on their binding ability to the CEA expressed on the cell surface.

Humanization by transplanting the “abbreviated” CDRs was undertaken to eliminate any possible idiotopes, present in the CDR-grafted HuCOL-1, that could be the potential targets of patients’ immune response. However, it should be pointed out that engineering of this variant may have generated new idiotopes on the combining site of HuCOL-1 that may be targeted by the patient’s immune response. The immunogenicity of the COL-1-derived Abs can be evaluated only through their clinical trials in patients. Whether the HuCOL-1 variants, generated by grafting of the “abbreviated” CDRs, have reduced the potential to evoke anti-V region response in patients compared with the CDR-grafted HuCOL-1 was assessed by comparing their reactivity to sera from gastrointestinal cancer patients who were administered ¹³¹I-labeled mCOL-1 in a phase I clinical trial (24). The sera were shown to carry anti-V region Abs to COL-1. In lieu of clinical trials, serum reactivity of the variant Ab is as good a measure of its potential immunogenicity as one can get, without administering the variant in a patient. The results confirm an earlier observation using another Ab (13), that the pattern of the anti-V region responses differs from patient to patient. Some patients may elicit more vigorous response to certain idiotopes than other patients; nevertheless, the variant ^24,25,27L/⁶¹H, compared with HuCOL-1, shows 1.5-, 2-, and 3-fold lower reactivity to EM, JS, and MB sera, respectively. The reduction in serum reactivity of the variant is much more significant than these numbers suggest, because the reactivity of HuCOL-1 to serum MB is 3-fold lower than that of mCOL-1. The lower serum reactivity of the variant ^24,25,27L/⁶¹H may be due to the elimination of the immunogenic idiotopes from the L chain variant ^24,25,27L, which shows 2- to 3-fold lower reactivity to EM and MB sera than does HuCOL-1. These results suggest that the variant ^24,25,27L/⁶¹H, compared with HuCOL-1, is significantly less reactive to sera from patients who were administered ¹³¹I-labeled mCOL-1 in a clinical trial and, hopefully, substantially less immunogenic in patients.

We thank Debra Weingarten for her editorial assistance in the preparation of this manuscript.